The Turkish Economy
Including contributions from noted international scholars, this collection of papers provides a strong theoretical and empirical underpinning for the discussion of major public policy issues facing Turkey today. Matters addressed include: ● ● ● ● ● ●
determinants of growth and productivity education and human capital accumulation income inequality corporate control and government performance of the government sector impact of major public policy issues on the future growth prospects of the Turkish economy.
This volume relates the impact of major public policy issues on the future growth prospects of the Turkish economy. At a time when Turkey is currently attempting to gain membership to the European Union, this pertinent reference questions whether the country’s economy is in fact ready for EU accession and membership. Sumru Altux is professor at the College of Administrative Sciences and Economics at Koç university in Istanbul. Her fields of interest include business cycles, investment, productivity and growth, intertemporal models of consumption and labour supply, and financial markets. Her previous publications include Dynamic Choice and Asset Markets (Academic Press, 1994) with Pamela Labadie, and Dynamic Macroeconomic Analysis: Theory and Policy in General Equilibrium (Cambridge University Press, 2003) with Jagjit Chadha and Charles Nolan. Alpay Filiztekin is an Associate professor in the Faculty of Arts and Social Sciences at Sabanci University in Istanbul. His fields of research include growth theory, regional economics, productivity dynamics and real costs of inflation.
Routledge Studies in Middle Eastern Economies
The Egyptian Economy Performance policies and issues Khalid Ikram The Turkish Economy The real economy, corporate governance and reform Edited by Sumru Altux and Alpay Filiztekin
The Turkish Economy The real economy, corporate governance and reform
Edited by Sumru Altux and Alpay Filiztekin
First published 2006 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN Simultaneously published in the USA and Canada by Routledge 270 Madison Ave, New York, NY 10016 This edition published in the Taylor & Francis e-Library, 2006.
“To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” Routledge is an imprint of the Taylor & Francis Group © 2006 Sumru Altux and Alpay Filiztekin, selection and editorial matter; the contributors, their own chapters All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data A catalog record for this book has been requested ISBN 0–415–36592–9 (Print Edition)
To my grandfathers Sumru Altux To my father Alpay Filiztekin
Contents
List of figures List of tables List of contributors Preface 1 Introduction
ix xi xiv xvi 1
SU MRU ALT U X
PART 1
The real economy 2 Productivity and growth, 1923–2003
13 15
SU MRU ALT U X A ND A L PAY F I L I Z TE K I N
3 Income and consumption inequality in Turkey: what role does education play?
63
BURC U DU YGAN A ND NE Z I H G Ü NE R
4 Female labor supply in Turkey
92
WN SAN T UNAL I A ND CE M BA V L E VE NT
5 A disaggregated analysis of price dynamics
128
MUSTAFA Ç AX L AYA N A ND A L PAY F I L I Z T E K I N
PART 2
The corporate sector 6 Capital structure decisions of Turkish firms YIL MAZ G ÜN E Y, AY D I N Ö Z K A N, A ND K Ü R VAT YA LÇIN E R
147 149
viii
Contents
7 Firm-level profitability, liquidity, and investment
172
B. BU R Ç IN YU RTOX LU
8 Ownership concentration and corporate performance of Turkish companies
199
HAL IT GÖN E NÇ
PART 3
The government sector and reform 9 Privatization in Turkey: what has been achieved?
225 227
Ç AXL A Ö KT E N
10 Political economy of pension reform in Turkey
252
SE R DAR SAYA N
11 The Turkish banking sector: a rough ride from crisis to maturation
276
AL I T Ü KEL , M U RAT Ü ÇE R, A ND CA RO L I NE VAN R IJCKE G HE M
Index
305
Figures
2.1 2.2 2.3 2.4 3.1 3.2 3.3 4.1 4.2 4.3 4.4 4.5 5.1 5.2 5.3 5.4 5.5 7.1 7.2 8.1 8.2 8.3 9.1 9.2 9.3 10.1 10.2 11.1
Aggregate labor productivity Relative sectoral labor productivity Capital deepening Labor productivity and capital productivity Marital sorting and skill premium Male–female literacy rate difference, ages 15 Public expenditure on education as a percentage of GDP Labor force participation rates by location and sex Hours of work per week by employment status, females residing in urban areas Distribution of wife’s annual hours of work Distribution of wife’s wages Distribution of husband’s wages Cost of living index inflation The distribution of price durations Monthly percentage of stores with no price changes The distribution of real prices The distribution of heterogeneity controlled real prices An example of a pyramidal ownership structure An example of dual class shares Market capitalization to GDP ratios for several selected countries Association between both earnings and dividends and concentrated ownership Association between market-to-book ratio and concentrated ownership Revenues from privatization Share of public and private sector in total manufacturing value-added Market concentration measured by Herfindahl Index SSK’s pension balances under different policy scenarios SSK’s pension balances under reform scenario 2 Banking sector total assets and credit
17 18 39 39 80 82 84 99 102 109 109 110 132 138 142 143 144 182 182 200 210 211 233 242 246 267 272 278
x
Figures
11.2 11.3 11.4 11.5 11.6
Saving-investment (non-financial) balances Real interest rate Net trading gains Inflation: the long view Consumer credit growth
278 286 286 288 292
Tables
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 2.12 2.13 2.14 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11
Accounting for growth, private manufacturing industries, 1970–2000 Accounting for growth, private manufacturing industries, 1981–2000 Sectoral allocation effects, private manufacturing industries, 1970–2000 Public manufacturing industries Small- and medium-sized firms Value added and employment by region Sectoral shares by region Regional convergence Regional convergence by sector Regional cycles and specialization Productivity and employment growth Decomposition of labor productivity Technical efficiency in selected years Decomposition: average annual changes Distribution of annual total household income: income shares of quintiles and Gini coefficients Distribution of annual total household income: average income by quintiles Distribution of annual total household income per capita, 2002 Distribution of household per capita income, 1990s International comparisons Distribution of consumption expenditures per person Distribution of consumption under alternative equivalence scales Household characteristics of rich and poor (income) Household characteristics of rich and poor (consumption) Relationship between education and earnings Sources of annual household income by per capita income quintiles and geographic area (percentage shares)
29 30 32 35 42 45 45 47 47 48 49 50 52 53 65 65 66 67 68 69 71 72 73 75 76
xii
Tables 3.12 Education expenditure per child by per capita income quintile, 2002, households with children 3.13 Share of education in total government expenditure 3.14 Private education expenditure as a percentage of total education expenditure 4.1 Total employment (age 15 and over) by sector and sex 4.2 Participation by marital status, females aged 15 and over residing in urban Turkey 4.3 LFPR for persons aged 15–64 by sex and age group, Turkey and other OECD countries 4.4 Sample means (std. devs.) of variables by labor force participation status 4.5 Maximum likelihood estimates of reduced form participation equations 4.6 Least squares estimates of the wage equation 4.7 Least squares estimates of the annual hours equation 4.8 Maximum likelihood estimates of structural participation equations 5.1 Distribution of products across vendor types and categories 5.2 The duration of price quotations and the size of adjustments 5.3 Within group duration of prices 5.4 The duration of prices across vendors 5.5 Heterogeneity controlled real price distribution: food items 5.6 Heterogeneity controlled real price distribution: all goods 6.1 Selected indicators 6.2 Ownership structure of firms in Turkey 6.3 Descriptive statistics 6.4 Alternative estimates of dynamic capital structure model 6.5 Alternative GMM estimates of dynamic capital structure model 6.6 Long-run multipliers 6.7 Static capital structure 7.1 Share of gross investment in GDP and the growth rate of total gross fixed capital formation 7.2 Evidence from World Business Environment Survey: “Measuring Conditions for Business Operation and Growth” private enterprise questionnaire 7.3 Ownership and control structure of Turkish listed companies 7.4 Summary statistics 7.5 OLS estimates of the impact of group membership on the investment-cash flow sensitivity of Turkish firms 7.6 GMM estimates of the impact of group membership on the investment-cash flow sensitivity of Turkish firms 7.7 Investment of group firms
81 83 85 95 100 101 111 113 116 118 120 132 134 139 140 143 144 152 153 157 160 162 166 167 174
176 181 186 187 188 190
Tables 7.8 8.1 8.2 8.3 9.1 9.2 9.3 9.4 10.1 10.2 10.3 10.4 10.5 10.6 10.7 10.8 11.1 11.2 11.3 11.4 11A.1
Further tests of the impact of ownership structure as a determinant of investment Sample descriptive statistics on ISE listed firms Changes in concentrated ownership and its categories Empirical results on the relationship between ownership and firm performance Privatization through block sales Public share offering privatizations International public offering privatizations Privatizations through block sales and public offerings Statutory entitlement ages in Turkey before 1999 Pension coverage by status and agency Pre-1999 growth of social security deficit in Turkey Minimum entitlement ages for female workers Transfers to publicly managed social security agencies funded by the treasury Pension deficits of SSK under different policy scenarios Results under scenario 1 with the Genetic Algorithm Results under scenario 2 with the Genetic Algorithm The cost of the banking sector crisis NPL ratio and provisions in the Turkish banking sector Selected items from the balance sheet of the banking sector Performance of ISE-listed banks Efficiency indicators
xiii 192 205 214 218 234 236 237 238 254 256 257 261 262 268 270 271 281 283 284 285 296
Contributors
Sumru Altux, Professor of Economics, College of Administrative Sciences and Economics, Koç University, Istanbul, Turkey and Research Fellow, Centre for Economic Policy Research, London, UK Cem Bavlevent, Assistant Professor, Department Economics, Bilgi University, Istanbul, Turkey Mustafa Çaxlayan, Reader, Department of Economics, University of Leicester, Leicester, UK Burcu Duygan, Research Fellow, Department of Economics, Finance and Consumption Programme, European University Institute, Florence, Italy Alpay Filiztekin, Associate Professor, Faculty of Arts and Sciences, Sabanci University, Istanbul, Turkey Halit Gönenç, Associate Professor, Department of Business Administration, Hacettepe University, Ankara, Turkey Nezih Güner, Assistant Professor, Department of Economics, Pennsylvania State University, University Park, Pennsylvania, USA Yılmaz Güney, Senior Lecturer, School of Management, University of Surrey, Surrey, UK Çaxla Ökten, Assistant Professor, Department of Economics, Bilkent University, Ankara, Turkey Aydın Özkan, Senior Lecturer, Department of Economics and Related Studies, University of York, York, UK Serdar Sayan, Associate Professor, Department of Economics, Bilkent University, Ankara, Turkey and the Department of Agricultural, Environmental and Development Economics, Ohio State University, Columbus, Ohio, USA Ali Tükel, Boxaziçi University, Istanbul, Turkey Wnsan Tunalı, Associate Professor of Economics, College of Administrative Sciences and Economics, Koç University, Istanbul, Turkey
Contributors xv Murat Üçer, EuroSource Economist and Adjunct Faculty, Koç University, Istanbul, Turkey Caroline Van Rijckeghem, Faculty of Arts and Sciences, Sabanci University, Istanbul, Turkey Kürvat Yalçıner, Assistant Professor, Faculty of Economics and Administrative Sciences, Gazi University, Ankara, Turkey B. Burçin Yurtoxlu, Professor, Department of Economics, University of Vienna, Vienna, Austria
Preface
This volume arose from the idea of creating a fresh approach to policy discussions in Turkey. In contrast to much other work on Turkey, we wished to take the “real” economy as the basis for our discussions. Second, we wanted to present a detailed examination of the behavior of two of its main actors, namely, the corporate sector and the government sector. Third, we wished to phrase the discussion in the language of modern economic analysis. As a reading of the various chapters will show, there exists a significant body of existing scholarship regarding the behavior of the Turkish economy, be it from the pen of economic historians or through the efforts of statisticians, national planners, and economists active in policy circles. Yet the analysis of the Turkish economy from a comparative context and employing the techniques of general equilibrium or dynamic economic analysis is still relatively rare. In this volume, we have sought to collect the work of individuals who are creating precisely this form of pioneering scholarship. Another distinguishing aspect of our volume is that it contains the viewpoint of corporate governance and corporate finance. The future is always created on the legacies of the past, and it is our pleasure to acknowledge the inspiration of Professor Tuncer Bulutay of the Faculty of Political Science at Ankara University. Professor Bulutay and his students have laid the foundations of the modern analysis of the Turkish economy. This tradition also informs the work of Dr Wnsan Tunalı, a leading expert in Turkish labor economics. We would also like to acknowledge the influence of Dr Hasan Ersel, formerly of Boxaziçi University and currently at Sabanci University, especially for his role as founder and co-editor of the Yapi Kredi Economic Review, where many of the recent studies on the Turkish economy have appeared. Another influence on some of the analysis in this volume stems from the outstanding scholarship of Professor Halil Inalcık, Bilkent University, and Professor Vevket Pamuk, Boxaziçi University, who have written extensively on the economic history of Turkey and the Ottoman Empire. We would like to end this brief preface by acknowledging the helpful comments and refereeing advice from various colleagues, especially Dr Hasan Ersel, Kemal Saatçioxlu, Aysit Tansel, Ercan Uygur, Murat Usman, and Atakan Yalçin. The editors would like to extend their special thanks to Nezih Güner and Wnsan Tunalı for reading and reviewing some of the chapters. Finally, we are indebted to Ms Nadia Seemungal for her editorial assistance.
1
Introduction Sumru Altux
The Turkish economy is known for its inflation in the 1990s and, more recently, its successful stabilization. However, relatively few studies have attempted to understand its “real” economy – the factors that kept the Turkish economy functioning in the face of high and chronic inflation and that have underwritten its dramatic recovery of recent months. The present volume seeks to remedy this oversight through a series of in-depth studies on the “real” side of the economy, along with some on the organization and structure of the corporate sector. This volume will also consider some government sector reform issues. Further distinguishing the volume is its use of state-of-the-art analytic techniques. In sum, the present work seeks to understand the state of the Turkish economy by examining the behavior of its main actors, the “real” or productive sector, the corporate sector, and the government. In Turkey and elsewhere, considerable popular debate revolves around the link between policy choices and economic outcomes. Several prominent US economists have written extensively on the Turkish economic experience from this perspective. The early study by Krueger (1974) on trade regimes and economic performance is among the most widely known. Turkey’s trade liberalization in the 1980s also attracted several studies (e.g. Aricanli and Rodrik, 1990a,b or Krueger and Aktan, 1992). Numerous other studies examine Turkey’s economic experience in the context of such issues as debt, exchange rate regimes, and stabilization policies (e.g. Rodrik, 1990, or the studies in Sachs and Collins, 1990).1 Although these studies have provided much insight into Turkey’s growth prospects, one cannot help thinking that some important outstanding issues lie in the “real” economy. A brief enumeration of the salient facts indicates that major challenges are present. The OECD (2004) reports that Turkey’s labor productivity is just 30 percent of the level of United States. Further, Turkey’s recent growth has lagged behind peer countries. The 2000–2001 crisis brought a GDP decline of 7.5 percent – one of the largest in Turkish Republican history – even in the midst of an IMF-sponsored stabilization plan. Whereas many OECD countries witnessed robust sectoral productivity growth during the 1990s, the productivity growth for Turkey in the same period came primarily from sectoral reallocation, with labor moving out of relatively less productive sectors such as agriculture (see table 2.12, Altux and Filiztekin, 2005). The share of the
2 Sumru Altux labor force working in agriculture is 34 percent, but these workers contribute just 12 percent of GDP (OECD, 2004, p. 154). Turkey also suffers from high levels of income inequality: the Gini index for total income was about 0.42 in 1994 (World Bank, 2002). Although extreme examples are higher – Brazil or South Africa – Turkey is nevertheless close to some very unequal countries, such as the Russian Federation (table 3.5, Duygan and Güner, 2005). Education statistics for Turkey are also alarming. In 2001, education expenditures were only 2.2 percent of gross national income, less than half that of many developing countries with similar per capita income (World Bank, 2003). Illiteracy remains shockingly high, especially among women. In the 2001 data, illiteracy was 15 percent of the general population aged 15 years or more and 25 percent among women aged 15 years or more. By comparison, illiteracy in Brazil was 12 percent, despite the fact that this country has a much higher percentage of its population below the one-dollar-a-day poverty line (World Bank, 2003). Even more troubling is the trend. Between 1990 and 2000 the share of education expenditures in total public expenditure declined by half, plummeting from about 18 to 9 percent (SIS, 2002). Recent Turkish labor market trends are also disconcerting. The analysis in Tunali and Bavlevent (2005) reveals that employment growth has been weak. This is true even before the crisis of 2000–2001 hit the economy. Employment growth from 1988 to 1998 was 1.4 percent, despite 3 percent growth in working-age population (20–54 years). When the period 1998–2003 was brought under scrutiny, average aggregate employment growth is negative at 0.6 percent per annum. The Labor Force Participation Rate (LFPR) is also very low in Turkey: whereas the LFPR is near 70 percent for OECD countries, this figure is 51 percent for Turkey. The corresponding figure for women in Turkey is even worse, only 28 percent compared to nearly 60 percent as the OECD average. (See table 3.3, Tunali and Bavlevent, 2005.) Perhaps the most surprising aspect of these findings is that they occurred at a time when policy changes were meant to unleash the growth potential of the economy. The purpose of this volume is to account for these observations and to examine their implications for the future. We examine the determinants of productivity and growth, the role of educational attainment in income and consumption inequality, female labor supply and participation decisions, and relative price variability. The second topic considered in this volume pertains to the impact of corporate governance and corporate structure on firms’ investment decisions and performance. As the chapters on this topic make abundantly clear, Turkish listed firms often have very weak corporate governance, exhibiting problems such as highly concentrated family ownership, divergence of cash-flow and voting rights, pyramidal structures, and business groups. Linking these factors to investment behavior can help answer questions regarding the determinants of Turkey’s overall economic performance. This research can also have potential implications for the impact of corporate governance on real outcomes more generally. Our third topic is the government sector, considering three main areas that usually inform structural adjustment programs: privatization, social security
Introduction
3
reform, and banking sector reform. The purpose of these studies is to describe the nature of decision making at the governmental level and to understand the political economy considerations that have helped to shape real outcomes in Turkey. Different facets of the recent policy experience in Turkey have been studied already, for example, Kibritçioxlu, A. et al., 2002, discusses inflation and alternative disinflation programs. Whereas such studies consider various endogenous outcomes as a source of macroeconomic risk, the approach in this volume is to identify the underlying factors lead to macroeconomic instability and risk. In our view, the importance of the present volume stems from its potential to illuminate upcoming policy decisions regarding a key set of issues now facing the Turkish economy. The December 17, 2004 summit in Brussels where Turkey obtained a date to begin membership negotiations with the European Union marks a turning point. The question of European Union (EU) accession and European Union membership is becoming the focal point of policy discussions in Turkey.2 It is also beginning to receive attention in the mainstream Economics literature. (See, for example, Flam, 2003.) The question of whether the Turkish economy can converge to the levels of per capita income observed in European Union countries is a key part of the recent discussion (see Derviv et al., 2004; see also Altux and Filiztekin, 2005). The entry of the Republic of Cyprus into the European Union on May 1, 2004 is leading to new debate regarding the reunification of Cyprus. The issue of economic convergence between the formally separated parts of Cyprus figures in this debate. (See the report by Eichengreen et al., 2004 on the economic aspect of the Annan Plan.) Our analysis contributes to this discussion by examining broad characteristics of the Turkish economy. Turning to the individual chapters in more detail, the topic of the first chapter is a new study on productivity and growth for the Turkish economy since the inception of the Republic in 1923. The chapter describes the impact of policies such as import-substituting industrialization practiced throughout the 1970s versus the program of trade and financial liberalization instituted in the 1980s on productivity growth. It includes an analysis of regional convergence and growth as well as an international productivity comparison. Altux and Filiztekin, Chapter 2, use both long time series and sectoral data to conduct a growth accounting exercise, and analyze the behavior of the broad sectors of the economy using production frontier methods. They argue that productivity growth is a key indicator of an economy’s long-run growth prospects, and it also has ramifications for cyclical fluctuations. In the Turkish context, they note that productivity growth is crucial for the sustainability of economic reform programs and that the rate of productivity growth will also be an important determinant of the ability of the Turkish economy to converge to per capital income levels in developed countries. The chapter utilizes the modern approach to productivity measurement to analyze the determinants of productivity growth for the Turkish economy. The study begins by providing existing theoretical definitions of productivity. It also seeks to incorporate new decompositions and approaches. Total factor productivity (TFP) is measured as the difference between the growth rate of output and the share-weighted growth rates of inputs. An alternative approach to the
4
Sumru Altux
growth-accounting framework for productivity measurement is based on production frontier methods, which allow for differences between technological catch-up (efficiency improvements) and technological change (shifts in the production frontier). Various authors have also suggested that productivity movements may be accounted for by the sectoral re-allocation of inputs to more productive uses rather than an increase in productivity in individual sectors. The chapter discusses the implications for exogenous versus endogenous models of growth for the purpose of explaining the Turkish experience. It also discusses the recent literature that seeks to incorporate the effects of greater openness and trade on productivity growth and convergence. The chapter presents an analysis of other determinants of productivity such as firm size, capital accumulation, and the role of macroeconomic/political instability observed in the 1990s. One of the main findings of the chapter has to do with the relatively weak record of capital accumulation in the Turkish experience. The chapter also highlights the role of macroeconomic instability in Turkey’s recent productivity performance and argues that these factors have been influential in determining the nature of productivity and growth as well as the process of structural transformation for the Turkish economy. Duygan and Güner, Chapter 3, provide evidence about Turkey’s future growth prospects from a different perspective. Specifically, they examine income and consumption inequality in Turkey and the role of education in determining changes in the distribution of income. Using primarily the 1994 and 2002 Household Income and Consumption Expenditures Surveys, they begin by documenting the extent of income and consumption inequality and then present the results within a broader international context. Moving behind these aggregate statistics, they also provide a detailed characterization of the country’s poor and the rich. Education is found to be an important determinant of inequality both because it is distributed very unequally and because it has a large effect on earnings. They argue that unequal distribution of education is often viewed as a major obstacle to the inequalities in labor markets as well as a key factor in the intergenerational transmission of inequality. Duygan and Güner move beyond the current, shortterm discussions on inequality by highlighting some key determinants of its intergenerational transmission – household formation, private and public spending on education, and women’s education. The goal is to understand what today’s distribution of resources and institutions might imply for tomorrow’s inequality. Tunali and Bavlevent, Chapter 4, provide an analysis of labor market trends and female labor supply and participation in Turkey over the period 1988–2003. They relate the trends in employment growth to the structural and demographic changes that have been taking place in the Turkish economy during the same period. Among the structural changes they discuss are the shift to a more open trade regime, the declining share of public enterprises, and the modifications of the social security system. They then provide a micro-econometric analysis of married women’s labor supply and participation choices for a given year. They employ a multi-category model of labor force participation. Their results imply that the distinction between self-employment and labor income is an important
Introduction
5
one. They also find that small changes in real wages elicit a large supply response from women. Their analysis brings to bear different pieces of evidence on a complex issue that is likely to have important effects on Turkey’s productivity and growth experience in the coming decades. Çaglayan and Filiztekin, Chapter 5, investigate the duration and the size of relative price changes in Istanbul over the period 1994–2000. The underlying motivation for their study lies in the recent theoretical literature and the empirical results that inflation induces higher relative price variability. Considering that the average inflation rate was about 60 percent per annum during the sample period in Turkey, their results contribute to an understanding of the effects of inflation on the pricing behavior in Turkey. Their analysis is carried out using a unique set of data collected from 15 neighborhoods in Istanbul. These data allow for a study of the micro-level data price adjustments by providing monthly price observations, which is mainly for foodstuffs, from three distinct store types: bakkals (convenience stores), pazars (open-air markets), and Western-style supermarkets over the period 1994–2000. These store types are expected to differ in obvious ways in terms of market power, transaction costs, and other characteristics such as fixed costs. Thus, their analysis helps to contribute to an understanding of the microfoundations of the transaction costs of inflation, as well as effects of market structure on price dynamics. Their data also provides prices of several other products from all 15 neighborhoods. These data help to provide a broader perspective of the underlying price dynamics. The findings reveal that on average price duration is 3.1 months, leading to an average reduction of 14.5 percent in real prices in between price changes should prices remain fixed for so long. They also find that price changes are not synchronized, which implies that vendors do not act simultaneously in response to changes in the economy. This behavior implies that the price of a good in relative terms could be substantially different across vendors. Indeed, the chapter finds that the average price dispersion across vendors for food items is around 10 percent and that for clothing is around 25 percent. The average price dispersion is 15 percent when all goods are considered. The findings also raise some questions about the effectiveness of monetary policies pursued by the government. Particularly, nominal price stickiness suggests that it will take a longer time for the monetary policies to have the desired effect on the economy. The chapters on corporate governance analyze the interaction of real and financial decisions for the Turkish economy. Güney, Özkan, and Yalçiner, Chapter 6, examine the capital structure of firms in Turkey, which has a significantly different legal and regulatory framework and a relatively young stock market. The approach taken in the chapter attempts to provide insights into the following questions. First, do the factors that influence the financing decisions of firms in industrialized countries exert similar impact on the capital structure decisions of firms operating in a different environment? Second, if similar factors exert a different impact, can we determine why this is so? The analysis in the chapter focuses on the dynamics of the capital structure decision. In so doing, the analysis captures important aspects of borrowing behavior of firms arising from market imperfections such as adjustment costs
6 Sumru Altux and financial constraints, which may prevent firms from adapting to new circumstances and immediately offsetting the effects of random shocks. The chapter provides some interesting findings regarding capital structure. First, in contrast to previous studies, there is strong evidence that size exerts a negative impact on borrowing decisions of firms, whereas there is a positive relation between growth opportunities and leverage. Second, firms do not adjust to their desired leverage ratios immediately, which might suggest that the costs of adjustment are significant. Finally, in line with the prediction of theory, profitability and asset structure of firms appear to influence leverage decision. The authors argue that the distinct features of the Turkish corporate sector, such as a high degree of ownership concentration and the absence of an institutional framework to facilitate sufficient protection for investors, may provide some explanation for these findings. The chapter by Yurtoxlu, Chapter 7, tackles the problem of the allocation of economy-wide investment more directly. Yurtoxlu argues that investment in capital equipment and R&D are important determinants of long-run economic performance. Since investment must be financed at the firm level, the efficiency of financial markets, corporate finance, and corporate governance arise as critical determinants of investment behavior of firms. Despite a strong theoretical case established by Modigliani-Miller for the independence of investment and financing decision, there is a large literature that has found a consistent and often strong relationship between investment spending and liquidity at the firm level. Yurtoxlu argues that two quite different explanations for this relationship have been given: one, the existence of asymmetric information (AI); and two, managerial discretion or empire building (EB) These two explanations have dramatically different predictions concerning the deviation of observed investment from levels that would maximize firm value. While the AI explanation points to financing constraints and the resulting under-investment, EB results in over-investment. A large number of studies have shown the importance of these two factors by exploiting different characteristics of firms, such as the dividend pay-out ratios, size, or ownership structure. Most of these studies, however, analyze firms from Anglo-Saxon countries, with relatively efficient capital markets and corporate governance regimes. Yurtoxlu studies the determinants of capital expenditures and R&D expenditures for a sample of publicly listed Turkish firms over the 1990–2002 period. He focuses on the specificities of a variety of corporate governance structures of firms that are likely to influence the degree of AI and the likelihood of pursuing goals other than shareholder wealth maximization. The chapter starts with a short review of the existing theoretical models of firm-level investment. After giving a detailed account of the corporate governance structures, it formulates and tests several hypotheses concerning the relationship between different corporate governance structures and the sensitivity of investment to the availability of internally generated funds. The final section is devoted to the implications of the findings on the growth process of Turkey’s corporate sector. Among Yurtoxlu’s important findings is the fact that the structure of corporate ownership for Turkey is characterized by the presence of Business Groups (BGs). To control this
Introduction
7
phenomenon, Yurtoxlu examines the response of investment decisions by group versus non-group firms to measures such as Tobin’s q versus cash flow. He finds that the impact of cash flow is weak to insignificant for group firms, suggesting that part of the role of BGs is to function as internal capital markets. However, further tests reject the notion that BGs function efficiently in terms of allocating capital to group companies with better investment opportunities. Gönenç, Chapter 8, also considers the impact of ownership structures on firm-level behavior, but his focus is on the impact of ownership concentration on firm market performance. He argues that one of the key elements for the development of financial markets is better protection of outside investors. Corporate governance is a set of mechanisms that allows outside investors to protect themselves against expropriation by the insiders (both managers and controlling shareholders). Investors would want to make sure that their funds would not be expropriated and would prefer to not only be repaid but also compensated for the time and risk of their investments. One view argues that better legal protection of outside investors leads to the expansion of financial markets with a higher number of listed firms and more valuable stock markets. Since laws protect outside investors, they are willing to invest financial assets such as equity and debt (La Porta et al., 1997). Another view suggests that if legal protection is poor, investors could get more effective control rights by being large (Schleifer and Vishny, 1997.) Thus, ownership concentration is accepted as an alternative of the legal protection for a corporate governance mechanism in the economies that are legally inadequate for the protection of the investors. However, there may be some costs associated with ownership concentration besides benefits of it. Gönenç investigates whether concentrated equity ownership creates an agency problem or better monitoring on expropriation of minority shareholders by using Turkish industrial firms as a laboratory. His analysis depends on the period from 1992 to 1998. Most of Turkish corporations listed in stock exchange have highly concentrated equity ownership and are affiliated with each other within a business group. Thus, pyramidal equity ownership structure and cross-shareholdings are dominant. The effect of the concentrated ownership on the firm performance in this context depends on the divergence between control and cash flow rights. Most previous studies assume that equity ownership of the controlling shareholder is exogenously determined and depends on the history and life cycle of the firm. By contrast, the analysis in the chapter employs a simultaneous equation system in which concentrated ownership measured by the total percentage of shares owned by the three largest shareholders is assumed to be endogenous. To support this assumption, Gönenç provides evidence for the changing percentage of controlling shareholders, showing that the largest shareholders vary their holdings of stock based on future performance expectations. The evidence shows that concentrated ownership in about 75 percent of sample firms changes on average from the lowest 0.6 percent to the highest 13 percent for the entire sample period. His results also show that holding companies, family members, the Turkish State, foreign investors, non-financial firms, and managers are seen as one of the three largest shareholders and their ownership changes during the sample period.
8
Sumru Altux
The main findings of this study are as follows: Ownership structure of Turkish firms is highly concentrated. The average percentage of shares owned by the three largest shareholders is above 50 percent for 162 out of 185 Turkish industrial firms. He also finds that variation in the firm performance has a positive effect on the variation in the percentage of shares owned by the three largest shareholders. Controlling for the endogeneity of ownership structure, he finds some evidence that ownership concentration is affected by the firm accounting performance rather than it affecting the performance. Taken together, the results in Chapters 6, 7, and 8 emphasize the role that corporate governance and corporate ownership can have on the economy-wide allocation of capital. In the last part of the volume, there are three chapters that examine issues that are relevant for reforming the government sector and implementing long-lasting policy measures to ensure macroeconomic stability. The first is a study of the privatization process in Turkey. The second is a study of pension reform in Turkey and the political economy considerations that underlie it. The third is a study of the behavior of the banking sector during the crises of 2001–2002 with an eye towards measures that will help prevent similar future events. Ökten, Chapter 9, provides an evaluation of the privatization process for Turkey. Ökten shows that while the privatization process in Turkey started earlier than other developing countries, its progress – measured by the size of divestiture – has been slower compared with the principal Latin American and Eastern European cases. From its start in 1985 up to 2005, the total proceeds from privatization efforts have amounted to $9.4 billion, with more than half of this sum being realized in the 2000–2005 period after the 1999 IMF Stand By agreement that placed a particular emphasis on privatization. As Ökten argues, the economic theory of privatization is a subset of the vast body of literature on the economics of ownership and the role for government ownership of productive resources. Within this literature, there are two main views: The Social View and the Agency View. According to the Social View, stateowned enterprises (SOEs) are capable of curing market failures by implementing pricing policies that take into account of social marginal costs and benefits of production. By contrast, the Agency View of firm ownership presents a strong critique of this theory. According to one stand of the Agency View, managers of SOEs may lack high-powered incentives or proper monitoring. According to another stand, political interference in the firm may result in excessive employment, poor choices of product and location, lack of investments, and ill-defined incentives for managers. Ökten reviews privatization experiences in light of this theory and discusses the results of empirical studies on privatization. She identifies common findings as well as conflicting results in the literature. She examines the impact of privatization in terms of its impact on productive and allocative efficiency, on investment and employment, market structure, and revenue generation. Her analysis is interesting in its own right as it describes the historical record of privatization in Turkey. However, it also complements the analysis by Altux and Filiztekin, Chapter 2, in examining the impact of privatization on productivity. The conclusions that
Introduction
9
emerge from Ökten’s analysis are that Turkey faces a mixed record in terms of its privatization experience. Sayan, Chapter 9, considers the process of pension reform for Turkey. As Sayan notes, insuring workers against future loss of income due to aging is a primary task for any social security system. For this reason, many countries legally require workers to have some old-age insurance coverage provided through publicly managed pension schemes that are typically run on a Pay-As-You-Go (PAYG) basis. However, the financial health of public PAYG pension schemes is sensitive to reductions in the relative sizes of retirees collecting benefits and active workers continuing to pay contributions. If such reductions are not matched with increases in contributions and/or reductions in old-age benefits over extended periods of time, the resulting deterioration in revenue–expenditure balances of public pension schemes eventually turns into contingent liabilities for the fiscal authority. Sayan notes that different political economy issues shape up policy makers’ attitude towards growing pension deficits and the timing of policy responses. First, incumbent governments tend to delay taking action to curb the growth in pension deficits as long as possible in response to the gradual decline in the number of workers as increasing contributions and/or lowering pension benefits are not likely to prove popular among voters. A second and more extreme form of myopic policy action involves interventions that change the incentive structure driving individuals’ work/retirement decisions or the eligibility criteria for retirement in such a way as to speed up the deterioration in pension balances for political rent-seeking purposes. The chapter by Sayan discusses some of these political economy issues in the context of recent efforts to rehabilitate the Turkish pension system, and investigates alternative parametric reform scenarios that could be adopted for that purpose. Similar to the chapter by Ökten, this chapter provides a closer look at the processes that determine governmental decision-making on some important public policy issues. As Sayan notes, the Turkish experience with pension crisis and reform is interesting to study because it differs from the experience of other countries where similar pension systems have faced financial difficulties largely due to population aging. By contrast, the crisis of the Turkish system prior to 1999 was due primarily to the entitlement ages that were exceptionally low by international standards and to other populist interventions of governments that wanted to improve their chances of re-election. Sayan’s analysis thus makes abundantly clear the short-term political motivations that transformed what was originally a system with ample surpluses into one of the biggest holes in Turkish public finances in recent years. While his proposals for parametric pension reform are also worth reading, perhaps the more interesting part of his analysis derives from obtaining a better understanding of the potential roadblocks to a sounder social security system in Turkey. Tükel, Üçer, and Van Rijckeghem, Chapter 11, discuss the state of the Turkish Banking Sector (TBS) in the period prior to and following the banking and financial crisis of 2000–2001. Their analysis emphasizes the fact that the TBS has been in a state of transition, with consolidation and the prospects of future growth
10
Sumru Altux
remaining as future challenges. The chapter describes in detail the state of affairs that prevailed in the Turkish banking sector – the prevalence of duty losses, open positions, a lax and weak supervisory environment, the existence of outright fraud – that eventually led to a bank structuring plan costing over a third of 2001 Turkish GNP. The chapter also describes the various banking sector operations undertaken to re-capitalize the banking system, as well as regulatory changes introduced since the disinflation program of late 1999. The chapter continues by exploring the growth prospects for the TBS. The discussion is based, in part, on the results of a questionnaire conducted by Steinherr et al. (2004) with a number of Turkish bankers. The chapter argues that long-standing impediments to growth of banking in Turkey, notably under-capitalization of the sector and chronic macroeconomic instability, have lost in importance. Another major impediment, the high taxation of the sector, also appears to have become less acute in recent years due to the decline in interest rates and plans to harmonize taxes across financial instruments beginning from 2006. The absence of a clear strategy on state banks, which still make up about (one-third) of the sector’s assets, appears to be a key problem. However, the chapter argues that the greatest challenge for the sector is achieving further consolidation and growth, in a way that does not undermine financial stability. In this regard, the chapter concludes by noting that sustaining the single digit inflation-high growth environment through structural reform that lies at the root of the current government’s economic policies provides the only way out of the conundrums that face the TBS. Several important topics not covered in this volume deserve serious discussion, especially agriculture. In Turkey’s future deliberations with the European Union, it is doubtless that this issue will acquire major importance. The analysis of productivity and convergence for the agricultural sector would constitute another book, but we would like to offer some comments before concluding. Many recent structural reform programs for Turkey aim to increase efficiency and the primacy of market-based institutions. There is a transformation underway in agriculture, as is already being documented by various academic studies. This transformation involves changes in the patterns of production, ownership, trade, and organization of agricultural activity.3 Nevertheless, the state of agriculture remains critical. As we argued earlier, over one-third of the workforce is still employed in agriculture.4 Productivity growth in agriculture is typically low and relative productivity is among the lowest in Europe (table 4.18, OECD, 2004, p. 154). An important topic for future analysis is whether the current changes and reforms will be sufficient to move the level of agricultural productivity closer to those in Europe, and whether they will effect the reallocation labor out of agriculture. A second issue warranting further study is the recent “jobless growth” phenomenon in Turkey. Whereas there has been a rapid decline in inflation and strong output growth since 2002, the increase in employment has been disappointing. Turkey’s demographic pressures make this an issue important for the fortunes of politicians both at home and abroad.5 One of the salient characteristics of Turkey’s past economic performance is its relatively weak record in factor accumulation and factor use.6 The recent stabilization effort led many to expect
Introduction
11
that this trend would be reversed. Some preliminary evidence shows that this is not the case (see Gürsel and Tunali, 2005). The large informal sector doubtless makes rapid employment expansion in the formal sector problematic for Turkey. As privatization continues and the share of public employment shrinks, a source of jobs is eliminated, at least in the short-run. Changes in firms’ employment policies, whether in response to changes in the international competitive environment or to technological progress, also suggest that growth in jobs will not necessarily follow on the heels of growth in output. Many of these factors are poorly understood. Yet their ramifications for Turkey’s future growth prospects appear crucial. We highlight these issues as important areas for further research. The future of the Turkish economy is also predicated on political developments. Recently, Turkey has witnessed some major political changes. The Justice and Welfare Party came to power on the back of such changes. Öniv (2005) provides a careful look at these new political players and the transformation of Turkish politics. As the different studies in the volume suggest, this is not the first major shift in Turkish politics, nor is it likely to be the last. How the various political players and their agendas evolve is likely to have important effects in determining the path of the Turkish economy. In this volume, we have not provided a detailed examination of the period since the Justice and Welfare Party took power. The implications of this change for Turkey’s democratization, the maturation of its institutions, and for its overall transformation also remain as key issues that will receive much future attention.
Notes 1 More recently, see the NBER Project on Exchange Rate Crises in Emerging Market Countries – Turkey, July 18, 2001, Jeffrey Frankel and Dani Rodrik. 2 See the Conference on Macroeconomic Policies for EU Accession May 6–7, 2005, organized jointly by the Central Bank of the Republic of Turkey, Bonn University, and Bilkent University, Ankara. 3 See, for example, the studies in the Turkish Economic Association publication Trade and Agriculture in the Southeast Anatolian Project Area (2004). Earlier studies on the state of Turkish agriculture include Kazgan (1992). 4 In some areas such as the Black Sea region and the eastern Anatolian region, the share of employment in agriculture is over 60 percent. 5 Turkey’s potential entry into the European Union is already emerging as one of the several critical issues that is dividing electorates in Europe. 6 On this point, see the report by McKinsey Global Institute (2003) as well as the studies in this volume.
References Altux, S. and A. Filiztekin (2005). “Productivity and Growth, 1923–2003,” Chapter 2, this volume. Arıcanlı, T. and D. Rodrik (eds) (1990a). The Political Economy of Turkey: Debt, Adjustment and Sustainability, New York: St. Martin’s Press. Arıcanlı, T. and D. Rodrik (1990b). “An Overview of Turkey’s Experience with Economic Liberalization and Structural Adjustment,” World Development 18: 1343–1350.
12
Sumru Altux
Derviv, K., D. Gros, F. Öztrak, F. Bayar, and Y. Ivik (2004). “Relative Income Growth and Convergence,” CEPS-EDP Final Report. Duygan, B. and N. Güner (2005). “Income and Consumption Inequality in Turkey: What Role Does Education Play?,” Chapter 3, this volume. Eichengreen, B., R. Faini, J. von Hagen, and C. Wyplosz (2004). Economic Aspects of the Annan Plan for the Solution of the Cyprus Problem, Report to the Government of the Republic of Cyprus. Flam, H. (2003). “Turkey and the EU: Politics and Economics of Accession,” Institute for International Economic Studies, Stockholm University. Gürsel, S. and W. Tunali (2005). “Ivlendirme Sorunu, Politikalan, ve Çalivma Ekonomisi Yazısından Çıkarılacak Dersler (Employment Creation Problem, Policies, and Lessons from the Labor Economics Literature),” Report for the Turkish Academy of Sciences. Kazgan, G. (1992). “Currents Trends and Prospects in Turkish Agriculture,” Middle East Technical University Studies in Development 19: 37–60. Kibritçioxlu, A., L. Rittenberg, and F. Selçuk (2002). Inflation and Disinflation in Turkey, Aldershot, UK and Burlington, VT: Ashgate Publishing. Krueger, A. (1974). Foreign Trade Regimes and Economic Development: Turkey. New York: Columbia University Press. Krueger, A. and O. Aktan (1992). Swimming Against the Tide: Turkish Trade Reforms in the 1980s, San Francisco, CA: International Center for Economic Growth. La Porta, R. Lopez-De-Silanes, F., Shleifer, A. and Vishny, R. (1997). “Legal Determinants of External Finance,” Journal of Finance 52: 1131–1150. McKinsey Global Institute (2003). Turkey: Making the Productivity and Growth Breakthrough. OECD (2004). Economic Surveys: Turkey, October. Öniv, Z. (2005). “The Political Economy of Turkey’s Justice and Development Party,” in Hakan Yavuz (ed.) The Transformation of Turkish Politics: The Justice and Development Party of Turkey. Salt Lake City: University of Utah Press (2006, forthcoming). Rodrik, D.(1990). “Premature Liberalization, Incomplete Stabilization: The Özal Decade in Turkey,” in M. Bruno, S. Fischer, E. Helpman, and N. Liviatan (eds) Lessons of Economic Stabilization and Its Aftermath, 323–353, Cambridge: MIT Press. Sachs, J. and S.M. Collins (1990). Developing Country Debt and Economic Performance: Country Studies – Indonesia, Korea, Philippines, Turkey, National Bureau of Economic Research Project. Schleifer, A. and Vishny, R. (1997). “A Survey of Corporate Governance,” Journal of Finance, 52: 737–783. State Institute of Statistics (SIS) (2002). Demography and Development Indicators, available at http://nkg.die.gov.tr (June 1, 2004). Steinherr, A., A. Tukel, and M. Üçer (2004). “The Turkish Banking Sector: Challenges and Outlook in Transition to EU Membership,” CEPS EU-Turkey Working Papers No. 4 (August). Tunalı, W. and C. Bavlevent (2005). “Female Labor Supply in Turkey,” Chapter 4, this volume. Uygur, E. and I. Civcir (2004). GAP Bölgesinde Ticaret ve Tarim (Trade and Agriculture in the Southeast Anatolian Project Region), Turkish Economic Association Publication, Ankara. World Bank (2002). World Development Report. World Bank (2003). World Development Report.
Part 1
The real economy
2
Productivity and growth, 1923–2003 Sumru Altux and Alpay Filiztekin
Introduction Productivity growth is a key determinant of an economy’s long-run growth prospects. In Solow’s neoclassical growth model, the source of long-run growth is given by the rate of exogenous technological progress. Endogenous growth theories attribute long-run growth to increasing returns that arise within the model. These may derive from a broad measure of capital as in Romer (1986) or from human capital as in Lucas (1988). Productivity shocks may also have ramifications for cyclical fluctuations. Following Kydland and Prescott (1982), the Real Business Cycle approach attributes a significant contribution of productivity changes to business cycles. According to this approach, procyclical productivity movements are due to exogenous technological shocks. This approach has generated a large literature on testing for the sources of cyclical fluctuations.1 In the Turkish context, productivity growth is a key indicator for the sustainability of economic reform programs. The rate of productivity growth will also be an important determinant of the ability of the Turkish economy to converge to per capita income levels in the European Union and other developed countries and to eliminate regional income disparities. Labor productivity provides the basis for discussions involving convergence of per capita incomes and the standard of living. Total factor productivity (TFP) growth allows a decomposition of a country’s (or sector’s) growth rate in terms of a component involving the growth rate of the inputs and the growth rate of technology. It is often used in discussions involving the determinants of a country’s growth rate, as in recent discussions involving the so-called Asian miracle.2 There is a wide literature surrounding productivity measurement. The measurement of TFP typically follows Robert Solow’s seminal (1958) contribution. According to this approach, TFP growth is measured as the difference between the growth rate of output and the share-weighted growth rates of inputs. The Solowian model assumes that firms maximize profits subject to a constant returns to scale production function and that product and factor markets are perfectly competitive. The simple growth-accounting framework has acquired widespread use in studies of growth, convergence, and the determinants of productivity change.
16
Sumru Altux and Alpay Filiztekin
Measures of TFP growth provide a comparison with a country’s (or industry’s) past performance. An alternative approach to productivity measurement constructs a best-practice frontier, and measures how far each country (or industry) is away from this frontier. This approach is based on the Malmquist index introduced by Caves et al. (1982). This approach does not require cost or revenue shares to aggregate inputs or outputs, and allows measurement of total factor productivity change in multiple output settings. In this chapter, we examine the sources of productivity and growth in the Turkish economy. We employ the modern theory of productivity measurement to examine the determinants of productivity changes. The Turkish experience is interesting, because it includes an extensive period of import-substitution policies followed by a period of trade and financial liberalization. There is also a period beginning with the 1990s that is characterized by volatility and macroeconomic instability. Another reason for examining the Turkish experience resides in issues of European Union integration and enlargement. In this chapter, we also examine the evidence regarding convergence for the Turkish economy. The predictions from the Solow growth model for long-run convergence have been tested extensively. (See, for example, Barro and Sala-iMartin, 1992.) One outgrowth of this literature is that economies may converge to their own unique steady state equilibrium that is characterized by differences in population growth rates, differences in government policy, societal preferences, and technology. This approach has become known as conditional convergence. Another outgrowth of this approach is that there may exist convergence clubs, that is, countries that are similar in their structural characteristics and initial conditions will converge to each other. However, there is no presumption that there will be convergence among the different clubs. (See Quah, 1996.) We take as our starting point the establishment of the Turkish Republic in 1923 for our productivity analysis. The growth experience of the economies in the Mediterranean and the Middle East has become the subject of increasing study by economic historians.3 At various points, we seek to provide a comparative perspective based on historical experience. The remainder of this chapter is organized as follows. In the section on Historical experience of the Turkish economy, we provide a brief description of historical patterns in labor productivity. In the section on Productivity and growth, we provide some definitions of productivity, and in the section on Growth accounting for Turkey, we discuss the application of growth accounting to Turkish manufacturing industries. The section on Determinants of productivity provides a further analysis of the determinants of productivity while the section on Regional convergence and growth is devoted to a discussion of regional growth and convergence. The section on EU and Turkey: ready or not? contains an international comparison of productivity. The final section concludes.
Historical experience of the Turkish economy One of the most comprehensive sets of data on the Turkish national accounts is due to Bulutay et al. (1974) for the period 1923–1948. These data were
Productivity and growth, 1923–2003
17
7 6 5 4 3 2 1 0 1923 1928 1933 1938 1943 1948 1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003
Figure 2.1 Aggregate labor productivity, 1923–2003.
subsequently updated and linked to form data on sectoral accounts by Temel (1998) for the period 1923–1967. Using these data, Figure 2.1 displays aggregate labor productivity or real GDP per employee for the Turkish economy for the period 1923–2003. This measure of labor productivity is defined as the ratio of aggregate GDP in 1987 prices to aggregate employment.4 The historical data show that aggregate labor productivity increased nearly sixfold between 1923 and 2003. There are various episodes that characterize productivity growth. Figure 2.1 shows that there are significant productivity increases for the early Republican years defined loosely as the period from 1928–1929 to 1938–1939. It is well documented that prior to 1929, the Turkish economy was one that mainly produced and exported agricultural products under a liberal trade regime. However, in 1929, Republican policy-makers began undertaking new policies in the face of the drastic fall in commodity prices. These include more protectionist measures in foreign trade, and the adoption of a new system of state-led import-substituting industrialization known as etatism. In Figure 2.2, we report relative sectoral productivity, or the ratio of labor productivity in each sector to aggregate labor productivity, for the main sectors of the economy. From this figure, we observe relative productivity increases in such key sectors as manufacturing, services, and utilities.5 These relative productivity increases are no doubt due to the new measures to stimulate output in manufacturing and utilities, and the ensuing activity in services. However, the large increases in relative sectoral productivity for manufacturing, utilities, and services are also due to the low levels of employment in these sectors in the early Republican years. Thus, we observe declines in relative productivity as these sectors develop at the expense of agriculture. The early Republican years are characterized by both impressive increases in labor productivity and GDP growth. Using the data underlying our productivity
18 Sumru Altux and Alpay Filiztekin (a) 7 6 5 4 3 2 1 0 1923 1928 1933 1938 1943 1948 1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003 Agriculture
Manufacturing
Services
(b) 30 25 20 15 10 5 0 1923 1928 1933 1938 1943 1948 1953 1958 1963 1968 1973 1978 1983 1988 1993 1998 2003 Mining
Utilities
Transportation
Figure 2.2 Relative sectoral labor productivity, 1923–2003 (Aggregate 1).
calculations, it is possible to show that output growth was 6.5 percent on average for the period from 1928–1929 to 1938–1939. It has generally been assumed that the interventionist policies followed by the new regime were responsible for stimulating output during the period from 1928–1929 to 1938–1939. Using data that is similar to ours in construction, Pamuk (2000) disputes this claim. Pamuk (2000) argues instead that the expansion in agriculture, which accounted for 50 percent of the economy, was at least as important for Turkey’s strong growth performance during this time.6,7 Boratav (1997) also attributes a significant role to agriculture
Productivity and growth, 1923–2003
19
to the development of a national economy. His analysis refers to the pre-Republican period 1908–1922 when the Ottoman Empire was being wracked by uprisings, rebellions, and the First World War itself. The war years 1940–1945 are generally viewed as constituting a break in this set of favorable developments. Despite Turkey’s neutrality during the Second World War this period is characterized by widespread mobilization of the male working age population and high military expenditures. Due to these developments, it is documented that wheat production fell by 50 percent during this period, and national income itself declined on average by 5 percent per year. It is often claimed that the deprivations experienced by various classes during this period laid the foundations for many of the political and economic developments in the years that followed. The postwar era is distinguished by several new phenomena. One of these is the shift to a multiparty system in 1950. Part of the reasons for this shift can be found in a combination of external and internal factors. The Democrat Party (DP), founded by large landowners and merchants, initially followed a program favoring agriculture and private enterprise. Indeed, from Figure 2.2a, we see some evidence of the relative decline in manufacturing productivity and the relative increase in agricultural productivity in the post Second World War era. Aggregate labor productivity also shows a strong increase up until 1953, corresponding to the end of Second World War and the commodity boom associated with the Korean War. However, with the end of favorable world conditions after 1953 and the decline in exports, the DP government soon began implementing more restrictive economic and political measures, including policies to restrict foreign trade. The worsening economic situation culminated in 1958 with the government agreeing to a major devaluation and other measures under the auspices of an IMF-sponsored program.8 Once the negative effects of the military coup staged to oust the Bayar–Menderes Democratic Party regime in 1960 had been overcome, the Turkish economy started growing rapidly around 1963. The post-1963 period also witnessed the pursuit of importing-substituting industrialization (ISI) policies in a more comprehensive manner, which reputedly obtained the backing of even some international agencies such as the OECD. The State Planning Organization (SPO) was established and five-year plans were drawn up to coordinate investment decisions. In contrast to the etatism of the 1930s, the ISI regime of this period featured State Economic Enterprises (SEEs) investing in large-scale intermediate goods production while the large private conglomerates took as their base the more lucrative consumer goods industries. For much of the 1960s, labor productivity shows a steady increase, and Turkey is typically mentioned among the group of more rapidly growing developing economies. By the early 1970s, however, most commentators agree that the continued pursuit of ISI policies was beginning to take its toll of the Turkish economy.9 In 1970, a balance-of-payments crisis erupted and was overcome as in the earlier case with IMF support. This date is also coincident with the second time that the military assumed power since 1960. After 1970, a favorable foreign exchange situation based on large foreign
20
Sumru Altux and Alpay Filiztekin
worker’s remittances allowed the Turkish economy to invest and to grow at a rate of 8–9 percent until 1975 or 1976. Around 1975 or 1976 the effects of the first oil shock began to make themselves felt in the Turkish economy. Even a cursory look at the data shows that things were not going right for the Turkish economy after 1975 or 1976. From Figure 2.1, we observe aggregate labor productivity falling relative to its level in 1974 or 1975. As Krueger (1995) shows, GDP grew by 15 percent between 1975 and 1980. This is nearly the same as the rate of population growth during the same period. Filiztekin (2000) provides a breakdown of real value-added growth in manufacturing for different periods between 1970 and 1996. According to these data, real value-added grew by 7.13 percent between 1970 and 1976 in manufacturing industries. By contrast, manufacturing value-added had a negative growth rate of 7.64 percent between 1976 and 1980. As various authors have recounted, the Turkish authorities had failed to provide an adequate response to the first oil shock in 1973.10 This created various external and internal imbalances in the Turkish economy. Some have argued that beginning with the 1970s Turkish industry had failed more fundamentally to change its orientation towards exports and to more capital-intensive sectors.11 In any case, the combination of events during this period also spelled the end of the era of ISI. The severe economic and political crisis at the end of 1979 and the beginning of 1980 eventually led to a sweeping set of changes to liberalize the Turkish economy. The initial measures were enacted by the government headed by Süleyman Demirel, with Turgut Özal as the deputy prime minister in charge of economic affairs. However, even as initial measures were being put in place to end the balance-of-payments crisis, the military took power for the third time in September 1980, installed a new government, and enacted additional measures to restrict trade union activity and other forms of political organization. Alongside, at times, highly controversial political measures, a major set of economic reforms were passed. From the period beginning with 1980, many of the former quantitative restrictions on imports were almost totally abolished. Deposit interest rates were eliminated and other measures to liberalize financial markets were put in place. The first democratically held elections in 1983 brought to power Turgut Özal and the newly formed Motherland Party. The reforms continued under the new Özal government. Capital account liberalization soon followed, as did other measures to liberalize capital markets such as the reopening of the Istanbul Stock Exchange, the beginning of Treasury auctions for marketing new government debt, the formation of an inter-bank money market, and other changes.12 Sustained increases in productivity are observed during 1983–1989 following the program of trade and financial liberalization instituted in the early 1980s. Filiztekin (2000) notes that real value-added in manufacturing grew by 9.77 percent between 1980 and 1988. The pattern of output is mirrored in the behavior of labor productivity. Whereas labor productivity in manufacturing fell by 10.82 percent during the period 1976–1980, it increased by 4.86 percent during 1980–1988. Nevertheless, there are various studies that have questioned the basis of the growth and productivity performance during the 1980s. Voyvoda and Yeldan
Productivity and growth, 1923–2003
21
(2001) note that the orientation of the economy to export-led growth was achieved at the expense of cost savings on wage labor which were then directed to export markets by means of a generous export subsidy program. They argue that the subsequent shortfall in investment rates circumscribed to a large extent the policy of export-led growth in manufacturing.13 Onaran (2002) examines the question of labor market flexibility as part of the adjustment of the Turkish economy from import-substitution to export-led growth. She shows that whereas productivity increases were typically reflected in real wages during the 1963–1979 period, after 1980 real wages became insensitive to productivity increases. During the latter period, real wages also became more responsive to the unemployment rate. She argues that relying on wage suppression to enhance competitiveness was not only unsustainable but also unsuccessful in overcoming high unemployment rates and generating effective domestic demand. There is a more erratic pattern in aggregate labor productivity after 1989. From Figure 2.2a, we also observe that relative sectoral productivity in manufacturing falls after this date. Commenting on a peculiar feature of the relative productivity displayed in Figure 2.2b, we note the large decline in relative productivity in utilities after 1990. As in Figure 2.2a, this is a relative decline, and it has to do with the large expansion of employment in newly established telecommunications industries. In contrast to the first part of the 1980s, there are some other noteworthy changes in the latter part of the 1980s. One of these has to do with the increased share of government expenditures in current GDP. As Krueger (1995) demonstrates, this share increased from a low of 18 percent in 1982 to 24 percent by 1990. In the early 1980s, the Özal government had embarked on an ambitious program of infrastructure investment, financed by domestic and external borrowing. The increasing deficits were due to the costs of these expenditures together with the cost of their financing. After 1983, as the Turgut Özal and the Motherland Party began losing popularity at the polls, the deficits also arose from transfer payments made prior to elections. As successive governments followed, the government expenditures begun during the Özal years were increasingly financed by monetary expansion. The liberalized capital account regime allowed for unfettered capital movements. Instead of providing funding for long-term investment projects, as Özal had hoped, international capital flows turned out to be short-term, volatile “hot money.” A financial crisis erupted in 1994 with an overvalued exchange rate and rising current account deficits. A speculative attack against the Turkish Lira (TL) was followed by a devaluation of 150 percent, skyrocketing overnight interest rates, and inflation reaching 107 percent. Average inflation in the 1980s had been in the range 50.4 percent whereas it increased to 73.2 percent in the 1990s. Throughout much of this period, real interest rates in Turkey have traversed in the 20–25 percent range. Turkish GDP during the 1990s also showed a highly volatile path, declining by 6 percent in 1994 and by 5 percent again in 1999 as a result of the Marmara earthquake. Turkey’s GDP registered one of its largest declines in Republican history of 7.5 percent during the banking and financial crisis of
22
Sumru Altux and Alpay Filiztekin
2000–2001 that erupted in the midst of an IMF-sponsored stabilization plan. As the OECD (2004a) notes, the 1990s represents the “lost decade” in Turkey. The combination of high real interest rates, high and chronic inflation, and macroeconomic and political instability also make themselves evident in the low growth rate for aggregate labor productivity during the 1990s. The period of the 1990s is witness to another set of developments. These pertain to Turkey’s relation with the European Union. In 1996, Turkey entered into the Customs Union Agreement with the European Union and in 1999, Turkey’s candidate status was confirmed at the European Union summit in Helsinki. Following the election of the Justice and Welfare Party in 2002, Turkey has witnessed a decline in inflation to single-digit levels in 2004, strong output growth in 2003 and 2004, and a dramatic expansion of trade volume. The recent December 17, 2004 Summit in Brussels in which Turkey obtained a date to begin membership negotiations also marks a turning point.
Productivity and growth: some further definitions In this section we discuss issues related to productivity measurement and to convergence and growth of an economy. We briefly outline the implications of exogenous versus endogenous models of growth. We also discuss alternative methods of productivity measurement. Exogenous versus endogenous growth models The neoclassical growth model advocated by Solow (1956) has long constituted one of the primary frameworks for analyzing countries’ growth experience. Recent models of endogenous growth constitute an alternative approach to modeling growth. To briefly describe the implications of the Solow growth model, consider an aggregate production for output that is of the Cobb–Douglas variety: Yt K t AtL1 , t
0 1,
(2.1)
and L, K, and A denote labor, capital, and the level of technology, respectively. Assume that labor and technology grow exogenously at rates n and g as: Lt L0ent,
(2.2)
At A0e gt.
(2.3)
The model assumes that a constant fraction of output, s, is saved. Defining kˆ as the stock of capital per effective labor, kˆ KAL, and ˆy as the level of output per effective labor, yˆ YAL, the evolution for kˆ is given by: dkˆt s yˆ (n g )kˆt, dkˆt skˆt (n g )kˆ t,
(2.4)
Productivity and growth, 1923–2003
23
where is the rate of depreciation. Equation (2.4) implies that kˆ converges to a steady state defined as kˆ* where kˆ* [s(n g )]1(1).
(2.5)
Substituting this relation into the production and taking logs yields the expression for the steady state level of per capita income as ln
YL ln A gt 1 ln(s) 1 ln(n g ). t
0
(2.6)
t
Thus, the Solow growth model predicts that countries will converge to a steadystate level of per capita income that depends on the nature of societal preferences, population growth, government policy, and technology. This expression can be used to derive quantitative predictions about the speed of convergence of per capita income. Let y* be the steady-state level of income per effective worker given by equation (2.6) and let yt be the actual value at time t. Approximating around the steady state, the speed of convergence is d ln( yt) (ln( y* ) ln( yt)), dt
(2.7)
where (ng)(1 ). This equation can be written as ln( yt) (1 et ) ln( y* ) et ln( y0),
(2.8)
where y0 is income per effective worker at some initial date. Subtracting ln(y0) from both sides and substituting for y* yields ln( yt) ln( y0) (1 et )
ln(s) (1 e t ) ln(n g ) 1 1
(1 et )ln( y0)
(2.9)
We can determine the rate of convergence by evaluating the convergence coefficient under some plausible assumptions. Suppose that the share of capital in aggregate income 0.33 and n g 0.06.14 Then the value of 0.04, and the economy moves halfway to the steady state in 17 years. We denote this by the half-life of convergence and refer to it in versions of 2.9 that we estimate for the Turkish economy. The Solow growth model also has predictions for convergence for countries or regions that differ in such characteristics as their saving rates, population growth rates, etc. This is known as conditional convergence and states that countries will converge to their own steady states, irrespective of their initial conditions. Endogenous growth models have widely differing predictions compared to the Solow growth model. One important variant of endogenous growth models
24
Sumru Altux and Alpay Filiztekin
which delivers sustained growth has become known as the AK-model. These models dispense with the assumption of diminishing returns to capital and lead to a representation of output that has the form Y AK, where A denotes the given level of technology and K the aggregate capital stock. We can derive a version of the AK model by assuming that the production function at the level of the firm has the form Yt AtK jt L1 jt ,
(2.10)
where At denotes the common level of technology. If all firms face the same technology and factor prices, then the aggregate production function can be written as Yt AtK t L1 , t
(2.11)
where At A(Kt Lt),
01.
(2.12)
Thus, the aggregate stock of knowledge At is proportional to the capital per person in the economy. Furthermore, the stock of knowledge is endogenous to the economy but it is taken as given by firms.15 If 1, then the aggregate production function has the form Y AK. Furthermore, assuming a constant saving rate, the evolution of the aggregate capital stock can be written as dKt sAKt Kt.
(2.13)
Thus, this model implies that capital grows at the constant rate sA and output also grows at this rate. Unlike the Solow growth model, the simple endogenous growth model presented in this example also shows that the growth rate of output depends on the saving rate. Furthermore, endogenous growth models do not predict conditional convergence: depending on differences in parameters that determine saving rates, the AK model predicts that there will be permanent differences in growth rates of per capita income. Another implication of endogenous growth is that the growth rate of per capita income may depend on the size of the economy. Ignoring the labor input in (2.11) and (2.12) and setting 0, if we take At as equal to the total (rather than average) stock of accumulated capital, then the growth rate of capital and output become equal to g sA(NK ) where N denotes the number of firms in the economy. Thus, the larger the number of firms, the greater the potential for creating potential knowledge externalities and the faster the growth rate of the economy. The relationship between the size of the economy and growth also provides a rationale for trade liberalization, the extent of the market having a positive effect on overall growth.
Productivity and growth, 1923–2003
25
The growth accounting approach The literature of productivity measurement has defined productivity in terms of an index.16 The growth-accounting approach decomposes output growth into components arising from the growth of inputs versus changes in total factor productivity. The (primal) measure of TFP gives output per unit of factor inputs whereas the dual TFP measure gives marginal cost per unit of input prices. The output-based primal productivity measure constitutes the basis for much of the growth-accounting literature. However, the price-based dual measure has also featured in many recent debates.17 To derive the standard output-based primal measure of TFP, consider a production function for gross output Yit in sector i at date t as Yit F i(Ait, Kit, Lit, Mit). In this expression, Lit denotes man-hours, Kit denotes services from capital, Mit denotes a vector of materials inputs, and Ait denotes the possibly sector-specific technology shock. The function F i is assumed to be homogeneous of degree i in L, K, and M. Totally differentiating the production function, assuming that technological progress is Hicks neutral, and making use of the first-order conditions for the firm’s cost-minimization problem yields: d(Yit Kit) d( Jit Kit) d( Jit Kit) sJit (it 1) sitj Yit Kit Jit Kit Jit Kit JL,M JL,M ( 1)
dKit dAit , Kit Ait
(2.14)
where sitj denote the revenue shares of the inputs, it is the markup of price over marginal cost, and is the returns to scale parameter. Notice that if it 1, then we obtain the Solow residual: (dAit Ait) (dYit Yit)
JK,L,M
sJit (dJit Jit),
(2.15)
which is just the residual growth in output after the contribution of the inputs have been removed.18 Comparing (2.14) with (2.15), the Solow residual overstates technical progress when there is imperfect competition or endogenous changes in efficiency due to increasing returns to scale.19 Decomposing productivity changes: changes in efficiency versus technical change An alternative approach to productivity measurement seeks to differentiate the impact of changes in productivity arising from the impact of technological progress versus changes in efficiency. Deterministic production frontier methods have been used to provide international productivity comparisons. See, for example, Kumar and Russell (2002). In the Turkish context, these techniques have been used to analyzing the sources of productivity growth for public and private Turkish firms. In a
26
Sumru Altux and Alpay Filiztekin
later section, we employ the Malmquist productivity index to examine changes in efficiency versus technical change for Turkey relative to a set of OECD countries. The Malmquist productivity index is based on quantity-based distance functions. Distance functions are function representations of multiple-output, multipleinput production technologies that require data only on outputs and inputs. For each time period t 1, . . . ,T, the production technology S t models the transforn into outputs y t Rm as S t {(x t, y t ): x t can produce y t}, mation of inputs x t R where S t satisfies certain axioms that serve to define meaningful distance function. The output distance function at date t is defined as the reciprocal of the maximum proportionate expansion of the output vector yt, given the inputs xt, that is, as Dt0(xt, yt ) inf{ : (xt, yt/ )S t}.20 An output-based Malmquist productivity index is defined as M0(xt1, yt1,xt, yt )
Dt0(xt1, yt1) D0t (xt, yt )
12 21
D0t1(xt1, yt1) D0t1(xt, yt )
.
(2.16)
This index allows a decomposition of productivity change into efficiency change and technical change as M0(xt1, y t1,x t, y t ) efficiency change technical change,
(2.17)
where efficiency change measures the change in how far is observed production from maximum potential production between years t and t 1 and technical change captures the shift in technology between years t and t 1 evaluated at xt and xt1, that is, efficiency change
technical change
DD (x(x , y, y ) )DD (x(x,y,y) )
D t0(xt1, yt1) D t0(xt, yt )
(2.16)
and t t1 t1 0 t1 t1 t1 0
t t t 0 t1 t t 0
12
.
(2.16)
Notice that the first term measures relative technical efficiency at t and t 1, that is, whether production is getting closer or further way from the frontier over time. The second term measures the shifts in technology at the vector of inputs x t, x t1, respectively; thus, technical change is the geometric mean of the two shifts. In terms of the decomposition, changes in relative efficiency are interpreted as evidence of catching up while technical change is interpreted as evidence of innovation.
Growth accounting for Turkey In this section, we provide a growth-accounting exercise for Turkish manufacturing industries. We begin our analysis of productivity growth for Turkey with
Productivity and growth, 1923–2003
27
manufacturing industries because they have been typically identified as an “engine of growth.” We consider the evidence for productivity growth for the import-substituting and export-led growth eras separately. One of the earliest studies to implement a growth accounting exercise for Turkey is the influential study by Krueger and Tuncer (1982). These authors examine the behavior of productivity using TFP measures for Turkish manufacturing for 1963–1976, and consider the impact of trade policy on relative TFP growth for both private and public firms. Despite the continuous implementation of import-substitution policies during this period, the authors note that there are sub-periods when, for example, foreign exchange difficulties forced the authorities to follow trade policies that were more restrictive. For this purpose, they divide the entire period into four sub-periods. The first and third sub-periods, 1963–1967 and 1970–1973, are ones in which there are plentiful foreign exchange reserves and relatively less restrictions while the second and fourth sub-periods, 1968–1970 and 1973–1976, are characterized by more restrictions. Krueger and Tuncer estimate TFP growth for the manufacturing sector over the entire period as 2.10 percent. For the four sub-periods in question, they find TFP growth of 3.2, 1.31, 2.51, and 1.18 percent, respectively. Thus, they conclude that TFP growth for Turkish manufacturing industries was negatively affected during periods that correspond to foreign exchange shortages and greater import restrictions. Krueger and Tuncer also compare the performance of public versus private enterprises. They find that the public enterprises are about 5 percent less efficient than the latter. Somewhat to their surprise, however, they find that TFP growth for public enterprises is 2.65 percent whereas it is 1.84 percent for private enterprises. Yildirim (1989) examines TFP growth in two-digit public and private manufacturing industries. He considers the period 1963–1983, which he splits according to the duration of the first four Five-Year Plans. His findings corroborate the decelerating TFP growth over the entire period. He argues that the higher productivity growth observed for public firms during this period can be attributed to differences in capital stocks per worker and due to their size.22 There are relatively few papers that have examined the sources of productivity differentials in public versus private firms. Zaim and Tavkin (1997) use the Malmquist index to decompose productivity growth for public and private Turkish firms between 1974 and 1991 in terms of technical change and changes in efficiency. Using data on 28 sub-sectors to develop a manufacturing production frontier, they find that the total private sector performed better than the public sector. Zaim and Tavkin (1997) also allow for a breakdown of efficiency changes into changes involving scale efficiency versus a pure efficiency change indicating developments in managerial efficiency over time, and find that public enterprises were especially deficient in the latter. Schmitz (2001) uses a simple two-sector growth model in which public and private investment goods are substitutes to examine theoretically the impact of government production of investment goods on overall labor productivity. Schmitz shows that aggregate labor productivity may be lower for two reasons: if government production is less efficient than
28
Sumru Altux and Alpay Filiztekin
private production, then the presence of the government creates both a direct efficiency loss in a sector and also an indirect efficiency loss. Under some plausible parameter values, Schmitz reports that TFP would have been 30 percent higher in countries such as Turkey or Egypt in the absence of any government intervention in the investment sector. An extensive program of trade and financial liberalization characterizes the period after 1980. We consider the experience of the private manufacturing sector over the period 1970–2000. We also discuss some of the trends regarding public sector manufacturing. The data are from the Annual Survey of Manufacturing Industries from the State Institute of Statistics. We employ productivity measures based on real value-added due to the lack of a reliable price index for materials. Sectoral price deflators that are available for 1981–1997 are extrapolated to the earlier period using a regression technique. The labor input is measured as persons engaged to provide a consistent measure for 1970–2000. The capital stock measure is derived by cumulating past real investment flows. Table 2.1 displays the results of the growth-accounting exercise for the sub-periods 1970–2000, 1981–1995, and 1981–2000 for private manufacturing industries. The growth of real value-added for the entire 1970–2000 period is 5.54 percent. This is significantly lower than the growth of value-added in many East Asian countries during their take-off period. For example, value-added growth for Korean manufacturing industries was 11.02 percent during the same period. However, when we take 1981 as the starting point for our analysis, real value-added growth nearly doubles following the trade and financial liberalization program undertaken in 1980. Turning to a decomposition of real value-added growth, the growth in real value-added for the entire sample period 1970–2000 can be attributed solely to factor accumulation, with capital deepening being the more important cause. By contrast, the contribution of TFP growth is measured to be negative.23 However, comparing the pre- and post-1980 periods yields very different conclusions. For the 1981–1995 period, TFP growth accounts for close to 54 percent of real value-added growth. Including the years up to 2000 implies a lower contribution of TFP growth. Nevertheless, TFP growth accounts for close to 39 percent of real value-added growth for the period 1981–2000. Table 2.2 provides a further breakdown of the period between 1981 and 2000. Both labor and capital accumulation are low in the 1981–1988 period. By contrast, TFP growth accounts for close to two-thirds of real value-added growth. During this period, real wages fell relative to their level in 1980, as restrictions on trade union activity and repression of civil servant salaries prevented any substantial increases in nominal wages. By 1988, however, real wages increased back to 1980 levels as various prohibitions on political life were eased and the loss of popularity of the Motherland Party led it to grant higher nominal wage increases to employees at SEEs. Table 2.2 shows that for 1989–1995, capital deepening accounts for nearly twice as much real value-added growth relative to the earlier period, as capital is substituted for labor in response to higher real wages. In the period leading up to 1996, capital deepening also occurs as an anticipatory response to the Customs Union Agreement that was concluded in 1996. Despite
Food Beverage Tobacco Textiles Clothing including footwear Leather Wood Furniture Paper Printing Chemicals Miscellaneous products of petroleum Rubber Pottery, glass and minerals Iron and steel Fabricated metal Machinery Electrical machinery Motor vehicles Plastics not elsewhere classified and instruments Total
3.86 10.52
17.88 12.80 43.65 14.87
9.28 16.86
16.81
6.16 8.22
7.25 6.09 2.57 12.83
15.64 7.75
5.54
112.90
56.79 101.58
104.99 73.84 173.67 51.31
65.55 81.67
135.15 48.06 123.86 233.27 44.63 144.23 98.88
13.73 12.64
9.93
29.71
11.09 9.78 9.34 15.08
22.87 13.36 117.32 33.82 33.92 18.44
8.87 8.85
10.14 9.32 11.92 10.01 4.29 8.75 3.53
46.35 45.42 46.94 166.00 52.70 54.02 15.68
11.20 6.52 23.07 32.73 2.67 9.79 16.80
7.60 10.72 5.94 3.33 12.98 4.25 10.27 30.59 7.81
7.69 9.05 19.48 7.42 19.19
255.74 77.28 107.16 142.94 86.16
198.88 20.09 4.12 65.15 9.24
43.14 2.63 3.03 22.21 23.08
2.23 9.17 9.69 3.82 14.48
7.32
8.50 7.82
3.42 3.97 3.41 3.62
0.33 3.65
8.71 0.77 12.61 10.29 17.63 11.41 4.59
6.88 1.81 1.30 6.83 16.77
Labor
Value added
TFP
Capital
Value added
Labor
1981–1995
1970–2000
Table 2.1 Accounting for growth, private manufacturing industries, 1970–2000
38.90
48.67 53.63
10.15 59.08 13.73 30.27
87.45 52.32
0.70 36.32 42.68 39.16 148.26 28.35 108.94
59.40 56.75 105.72 39.14 59.22
Capital
53.78
42.83 38.55
86.43 36.95 110.33 66.11
12.22 44.03
90.59 62.91 44.71 50.55 65.89 60.24 13.53
33.72 41.44 7.01 54.02 24.01
TFP
8.90
12.24 12.10
9.72 10.09 8.89 13.64
5.52 7.11
6.73 9.27 12.91 9.84 6.31 7.83 1.08
6.09 5.01 10.82 7.52 13.76
Value added
7.61
8.48 8.27
3.88 6.35 4.04 5.76
1.92 5.40
5.86 2.57 13.06 8.86 9.12 3.64 31.17
7.93 2.17 2.84 10.28 16.93
Labor
1981–2000
53.50
52.43 68.67
21.62 43.83 13.86 33.79
122.54 68.01
94.45 57.15 63.13 54.56 100.59 44.50 444.17
92.93 152.35 166.13 61.31 75.08
Capital
38.90
39.09 23.07
74.51 49.82 82.10 60.45
24.46 26.58
0.31 40.28 23.81 36.58 9.71 51.86 375.34
0.86 54.52 68.97 28.41 7.99
TFP
Food Beverage Tobacco Textiles Clothing including footwear Leather Wood Furniture Paper Printing Chemicals Miscellaneous products of petroleum Rubber Pottery, glass and minerals Iron and steel Fabricated metal Machinery Electrical machinery Motor vehicles Plastics not elsewhere classified and instruments Total
13.85 8.01 2.16 12.89 18.16
19.76 10.02 11.46 13.44 14.05 6.67 9.23
9.30 11.96
8.45 10.91 10.12 8.84
10.98 16.64
12.01
7.19 8.47 14.76 6.68 27.17
7.51 3.79 17.29 10.35 11.63 11.47 4.10
10.22 8.28
12.35 8.92 9.17 12.11
13.67 8.06
9.63
24.36
47.97 36.06
0.08 4.20 52.49 41.58
63.63
41.05 47.30
91.47 93.29 142.37 49.58
31.87 79.03
99.66 185.61 73.62 25.53 88.10 59.85 201.75
19.42 95.63 14.92 61.03 2.15 33.48 110.99 58.84 9.01
27.28 36.93 6.17 43.00 56.43
58.87 128.92 104.01 44.11 25.41
TFP
10.28
13.80 17.87
9.66 10.75 9.54 18.48
7.33 9.49
13.14 15.63 5.78 9.62 4.10 5.65 2.88
8.27 9.71 24.86 8.28 10.08
24.86 119.11 28.82 21.80
3.93 2.61 3.96 0.30
2.31
54.45
49.47 62.69
133.04 95.52
13.95 4.65
5.69 3.27
13.85 72.93 137.63 12.27 339.62 16.46 466.02
59.92 15.15 106.87 34.56 163.39
0.04 4.37 0.71 1.25 12.52 1.48 1.80 16.52 6.40 6.01 22.41 2.95
Capital
Labor
Value added
Capital
Value added
Labor
1989–1995
1981–1988
Table 2.2 Accounting for growth, private manufacturing industries, 1981–2000
43.24
44.85 34.04
79.07 16.50 75.14 78.50
19.09 9.13
84.67 28.87 54.15 81.33 433.60 61.13 363.07
40.12 119.52 7.58 64.18 75.90
TFP
5.79
7.78 10.50
5.62 11.04 7.53 9.29
4.52 1.91
3.48 9.13 15.90 9.35 12.36 5.06 6.29
1.26 7.10 15.14 7.80 2.54
Value added
9.06
8.38 9.89
6.61 12.68 6.38 16.22
7.43 29.85
30.72 8.08 14.08 4.26 0.25 36.70 13.62
27.03 0.79 3.11 20.10 13.29
Labor
1996–2000
128.65
72.32 122.95
89.53 3.34 116.59 50.91
84.17 286.19
724.18 120.89 109.13 104.01 50.94 128.30 120.74
706.72 213.13 67.07 124.56 284.85
Capital
37.71
19.29 32.85
3.86 83.99 22.98 32.87
191.60 216.05
793.46 28.97 23.21 8.28 48.81 8.40 234.37
633.75 312.34 170.18 44.66 371.56
TFP
Productivity and growth, 1923–2003
31
the financial crisis in 1994, real value-added continues growing at a rate over 10 percent during the 1989–1995 period, and the fraction that can be attributed to TFP growth is around 43 percent. In the period after 1995, however, real valueadded growth falls to 5.79 percent, and the contribution to TFP growth becomes negative. Thus, the initial positive response of trade liberalization on TFP becomes dissipated in the latter part of 1990s, as the Turkish economy becomes increasingly subject to political and macroeconomic instability and the impact of external shocks such as the 1998 Russian crisis and the 1999 Marmara earthquake. Various authors have noted that productivity increases may be due to sectoral re-allocation of inputs to more productive uses rather than an increase in productivity in individual sectors (see, for example, Bernard and Jones, 1996).24 We now examine the experience of individual industries relative to the aggregate by using this decomposition. The results are summarized in Table 2.3. In this table, PGE refers to the Productivity Growth Effect, which arises from TFP improvements within the industry, and SE refers to the Sectoral (Share) Effect, which arises from the allocation of total factor inputs from low to high productivity industries. This table shows that the sectoral effect contributed negatively to TFP growth in all sub-periods except in the periods 1981–1995 and 1981–1988.25 In particular, we observe that the positive impact of the sectoral re-allocation effect is largely confined to the period immediately following the trade liberalization, namely, 1981–1988. A negative sectoral effect arises when industries that are relatively more productive enjoy less than average factor accumulation. There is a negative sectoral effect for the period 1981–2000 because industries such as chemicals and electrical machinery with the highest TFP growth had negative factor accumulation in the postliberalization period. This suggests that the resource allocation in Turkish manufacturing is inefficient. The motor vehicles industry displays moderate TFP growth for 1981–2000. Nevertheless, its contribution to aggregate TFP is nearly as large as that of electrical machinery because it enjoys a positive shift effect. There is a positive sectoral shift effect on aggregate TFP for the first period after liberalization (1981–1988). In this period, within-sector productivity is augmented with greater factor accumulation for such high productivity industries as chemicals to lead to a positive effect to TFP growth. Voyvoda and Yeldan (2001) present a decomposition of sectoral labor productivity for the periods 1970–1976 and 1981–1996. They identify a “leading” sector as one that has a high value of its reallocation weight due to a low labor share and a high output share. For 1970–1976, they show that leading sectors such as petroleum industries with a high reallocation weight did not achieve sufficient productivity increases to enable the economy to achieve sustained growth. For the outward-oriented 1981–1996 period, they find that 15 out of the 19 subsectors display negative productivity gains from labor reallocation, echoing in part our findings for 1981–2000. In contrast to the earlier periods analyzed by Krueger and Tuncer (1982), the period 1980–2000 is characterized by the diminishing importance of public
Food Beverage Tobacco Textiles Clothing including footwear Leather Wood Furniture Paper Printing Chemicals Miscellaneous products of petroleum Rubber Pottery, glass and minerals Iron and steel Fabricated metal Machinery Electrical machinery Motor vehicles Plastics not elsewhere classified and instruments Total 0.2 1.2 0.6 7.2 0.7 13.4 0.5 1.4 3.9 0.6 3.6 26.7 15.2
0.3 0.7 0.6 2.6 0.1 0.7 0.7 2.9 2.1 4.3 6.1 10.4 3.7 5.7 3.0
13.5
0.5 1.9 1.1 9.8 0.8 12.7 0.2
1.5 1.8
3.7 2.6 16.3 11.5
7.0 2.8
86.5
100.0
12.6 0.2
41.9 0.7 0.0 43.8 3.1
8.7 0.2 0.1 11.2 4.1
33.2 0.5 0.1 32.6 1.0
98.9
6.3 3.1
9.7 3.5 13.1 17.5
0.5 7.7
0.7 1.2 0.4 1.8 0.3 9.8 0.1
5.2 1.3 0.4 15.6 2.3
PGE
Total
PGE
SE
1981–1995
1970–2000
Table 2.3 Sectoral allocation effects, private manufacturing industries, 1970–2000
1.1
100.0
10.9 5.1
6.9 4.6 6.1 18.4
2.8 1.1 7.0 1.0 4.7 2.0
1.6 8.0
0.4 1.3 0.5 1.8 0.2 7.5 0.8
5.7 1.7 2.8 10.7 6.5
Total
1.1 0.3
0.3 0.1 0.1 0.0 0.5 2.3 0.6
0.4 0.4 3.2 4.9 4.2
SE
101.3
9.2 3.1
11.4 8.9 14.4 23.9
0.7 5.4
0.0 1.3 0.5 2.1 0.1 12.1 1.4
0.1 1.1 1.6 13.5 0.7
PGE
1981–2000
16.3 6.9
100.0
1.3
6.6 6.9 6.2 21.8
0.5 4.9
0.2 1.8 1.2 2.1 0.7 7.7 1.9
1.6 0.1 0.3 12.9 4.2
Total
7.0 3.8
4.8 1.9 8.2 2.1
0.3 0.5
0.2 0.5 0.7 0.0 0.8 4.4 0.5
1.7 1.2 1.9 0.6 3.5
SE
Food Beverage Tobacco Textiles Clothing including footwear Leather Wood Furniture Paper Printing Chemicals Miscellaneous products of petroleum Rubber Pottery, glass and minerals Iron and steel Fabricated metal Machinery Electrical machinery Motor vehicles Plastics not elsewhere classified and instruments Total 0.2 0.1 0.9 2.1 1.7 17.8 0.6 2.7 7.9 9.8 2.8 6.7 10.0
0.2 0.9 0.0 1.2 0.4 4.0 3.7 1.2 4.0 0.3 4.3 7.1 3.4 7.1 0.4
2.6
0.4 1.0 1.0 0.9 1.3 13.8 3.1
1.6 11.9
10.1 7.0 13.8 6.6
5.0 1.5
97.4
100.0
12.0 1.1
5.1 2.1 0.8 7.8 9.0
1.4 3.1 0.8 2.3 3.0
3.7 1.0 0.1 10.1 6.0
1981–1988
100.8
10.2 4.1
3.9 6.5 5.4 27.3 9.8 9.3
100.0
0.4 5.2 0.8
0.5 8.1
0.7 2.5 0.1 1.4 1.3 3.3 0.9
6.2 1.3 4.9 13.7 4.0
6.7 8.5 4.5 6.0
1.3 6.2
0.8 1.9 10.7 2.0 9.9 33.3
0.3 1.6 0.3 2.0 1.4 12.0 1.9
0.9 3.9 5.5 7.7 11.8
0.9 0.9 0.3 3.5 2.7 8.7 2.8
7.1 5.1 0.6 21.4 7.8
1989–1995
91.5
5.9 6.7
0.5 23.7 4.6 18.2
6.6 16.6
4.6 1.5 1.0 0.7 1.3 1.7 4.3
29.5 13.9 10.5 28.2 14.0
1996–2000
8.5
5.5 9.1
8.9 18.6 0.8 22.5
3.8 4.2
3.0 2.6 3.5 0.2 0.3 8.4 0.5
6.9 5.6 2.7 26.6 2.4
100.0
11.3 2.3
8.4 5.1 5.4 4.3
10.3 20.8
1.6 1.1 2.6 0.5 1.6 6.6 3.9
22.7 8.4 13.1 1.7 16.4
34
Sumru Altux and Alpay Filiztekin
manufacturing enterprises. Table 2.4 shows that the share of employment in public sector manufacturing industries also follows a declining trend. Whereas 35–45 percent of the total employment in manufacturing industries is in public sector firms up until 1980, after this date there is a rapid decline in this proportion, which falls to only 11 percent by year 2000. Likewise, the share of total valueadded in manufacturing industries never falls below 50 percent until 1980. After this date, however there is a rapid decline in value-added, with the fraction of value-added being attributed to public sector firms falling below 20 by year 2000. In terms of value-added growth, public sector firms have shown relatively strong value-added growth and that value-added growth has actually increased across the three sub-periods, no doubt as a result of privatization and attrition of the least productive firms. Whereas TFP growth is negative for the period that encompasses the ISI era, public sector firms have shown positive TFP growth in the post-1980s period. However, in contrast to private sector firms, factor growth in public sector firms is negative in the post 1980s period, with large discernible declines in the rate of capital accumulation from 1.81 percent during 1970–1999 to 1.22 percent during 1981–2000. Finally, considering the impact of sectoral re-allocation, with the exception of one sub-period, productivity growth in public sector firms is mainly attributable to within-sector productivity growth. Several other studies decompose output growth into changes in factor utilization, markups, economies of scale, and productivity or technology. Bayar (2002) considers a version of equation (2.14) and uses 28 cross sections of the ISIC threedigit industrial data on private firms employing more than ten people over the period 1974–1994 in her estimation. To allow for the effects of trade liberalization on productivity growth, she splits the sample into two, the first period corresponding to 1974–1983 and the second period to 1984–1994. She finds a positive markup of 29 percent and some evidence for scale economies (with the estimated value of 1.198) before trade liberalization. After liberalization, there is evidence that price-cost margins declined by around 12 percent and decreasing returns to scale become more prevalent. Finally, whereas TFP growth is negative in the period before 1984, it becomes positive, with a net gain equal to 2.1 percent after the trade liberalization. Akdede (2002) also reports similar findings based on the behavior of private and public firms in three-digit manufacturing industries. He considers the sample period 1980–1991 and splits the sample as 1980–1984 and 1985–1991. Akdede estimates numerous specifications allowing for departures from constant returns to scale and imperfect competition, where markups and scale effects may depend on the trade regime. He finds that markups are smaller in public firms and that most industries display constant or decreasing returns to scale. One of his main findings is that TFP growth increases between 2.5 and 4 percentage points for private and combined industries as a consequence of liberalization. By contrast, there appears to be evidence that the scale of operations declined after liberalization, as in Bayar (2002).26 In summary, a detailed examination of Turkey’s productivity performance in manufacturing industries across the pre- and post-1980 periods suggests that
(c) Sectoral reallocation effects
(b) Growth accounting
(a) Employment and value-added Share in employment (%) Share in value-added (%)
100.0
49.3
149.3
112.1
PGE
Total
SE
PGE 12.1
SE
1989–1995
9.93
1981–1988
3.14
25.89
1.77
6.97
100.0
53.78
TFP
36.09 40.54
1980
91.6
PGE
8.4
SE
1981–2000
7.09
Capital
34.95 50.55
1975
Total
8.18
Labor
Value added
TFP
Capital
36.33 54.27
1970
Labor
— 57.13
1965
Value added
42.50 59.11
1960
1981–1995
39.72 50.41
1955
1970–2000
45.95 58.33
1950
Table 2.4 Public manufacturing industries, 1950–2000
24.29 31.28
1990
100.0
Total
11.04
Value added
11.01
Capital
16.59 23.11
1995
112.0
PGE
12.0
SE
1981–2000
7.99
Labor
1981–2000
29.47 38.03
1985
100.0
Total
50.76
TFP
11.02 18.95
2000
36
Sumru Altux and Alpay Filiztekin
trade and financial liberalization led to increased TFP growth in the latter period. While there is a significant contribution of TFP growth to value-added growth in the post-1980 period, factor accumulation is the more important source of valueadded growth in the pre-1980 period. Trade and financial liberalization also appear to have had the effect of reducing markups and also economies of scale in public and private manufacturing. In a related point, while public firms were typically less efficient than private firms, any advantages that they might have enjoyed during the import-substituting era such as size or capital accumulation are reversed in the period after 1980.
Determinants of productivity Our earlier discussion indicates that there are important productivity differences in the pre- and post-liberalization periods. In this section, we examine in more detail the type of factors that are likely to affect productivity and growth. Productivity and trade Much popular debate often associates a more open trade regime with sustained increases in growth rates across countries. Yet, as Lucas (1988) notes in his elegant treatise, the standard neoclassical growth model has no such prediction. There may exist one-time effects of the removal of trade barriers but one must look elsewhere for the sources of sustained growth. The literature in openness and trade has postulated various channels but not all of them will lead to sustained growth. If certain sectors are needed to provide the “big-push” for industrialization, exports in such sectors could lead to growth or alternatively, exports may break the vicious cycle of low demand leading to low income, and low income leading to lack of industrialization. This is the coordination failure channel (see Rosenztein-Rodan, 1943), and it finds resonance in some endogenous growth models that attribute a role for the size of the market. Other explanations stress such factors as the role of exports and foreign aid in generating the foreign exchange to allow the import of capital goods and intermediate goods (MacKinnon, 1964) or increases in efficiency through increased competition (Balassa, 1978). More recent work argues that there may exist diffusion of technology through foreign buyers’ suggestions and learning-by-doing (Grossman and Helpman, 1991). Edwards (1998) conducts a cross-sectional analysis examining the role of greater openness on TFP growth using data on 90-odd countries (including Turkey) for 1960–1990. He controls for the effects of other factors such as human capital, which signifies a country’s ability to absorb new technology, and the country’s initial GDP per capita level. Holding other factors, he finds that countries that are more open tend to experience faster TFP growth. Nishimizu and Robinson (1984) provide an influential early study on the determinants of productivity and the impact of trade regimes. These authors consider the experience of Korea, Turkey, Yugoslavia, and Japan for the 1960s and 1970s using sectoral data. Mirroring some of our earlier findings, they find
Productivity and growth, 1923–2003
37
that TFP growth is important for Japan and Korea whereas it is the growth of inputs that primarily drives output growth for Turkey and Yugoslavia. The chapter finds that substantial proportions of TFP growth are explained by output growth allocated to export expansion and import substitution for Turkey, Yugoslavia, and Korea. In many analyses of trade and openness on growth, Granger causality tests between exports and growth have been used. Hatemi-J and Irandoust (2000) examine Balassa’s export-led growth hypothesis using Granger causality and cointegration analysis. They consider the experience of Greece, Ireland, Mexico, Portugal, and Turkey for the period 1960–1997. They find that exports and growth are causally linked for Ireland, Portugal, and Mexico. However, they cannot find a causal link for Turkey or Greece. The causal link for Mexico and Ireland runs from exports to growth, but the opposite is true for Portugal. One interpretation of the results for Turkey is that contrary to expectations, export-led growth has not materialized. Bahmani-Oskooee and Domac (1995) employ co-integration and error correction for exports and output for the long sample 1923–1990, and find evidence for bi-directional causality between exports and growth. Doxanlar and Fisunoxlu (1999) also conduct Granger causality tests between exports and growth for seven Asian countries. The countries are India, Pakistan, Philippines, Singapore, Thailand, South Korea, and Turkey using annual observations between 1960 and 1995. They find evidence for bi-directional causality between export and output growth for Turkey, South Korea, Singapore, and the Philippines, and that causality runs from exports to growth for Thailand, and from growth to exports for Pakistan.27 The impact of openness and the extent of the market are also associated with recent endogenous growth theories that suggest a role for initial wealth on the subsequent growth rates of an economy.28 In a closed economy, larger initial wealth can increase subsequent growth by making possible investment in fixed cost (or other increasing returns) projects. In models with learning-by-doing, initial wealth may be correlated with subsequent growth if such learning-by-doing is greater for producing more advanced products. On the other hand, if trade causes less advanced countries to specialize in products where there is little learning-by-doing, then it may lead divergence in growth rates between rich and poor countries. Ades and Glaeser (1999) examine these hypotheses using cross-country regressions for a set of poorer countries since 1960 and for the US states in the nineteenth century. They find that openness and initial development are substitutes in generating growth, suggesting that growth may be a function of the size of the market. However, they do find that specialization for the poorest countries tends to decline with openness, and that too much specialization is bad for growth. A number of papers have examined convergence using dynamic panel data analysis (see, for example, Islam, 1995). Miller and Upadhyay (2000) consider panel data regressions for 83 countries (including Turkey) for the period 1960–1989. Following the analysis in Mankiw et al. (1992), they enter human capital as an input into the production function. Miller and Upadhyay find that openness (as measured by the ratio of exports to GDP) has a positive effect on
38
Sumru Altux and Alpay Filiztekin
TFP. In contrast to Islam, they also find human capital has a significant positive effect on output when it is included as a factor of production. However, their results on the interaction of openness and human capital on TFP suggests that too little openness does not allow a country to efficiently utilize its human capital stock. It is only when openness (as measured by some threshold in the exports–GDP ratio) is sufficiently high that a country’s stock of human capital can have a positive effect on TFP. In an interesting historical aside, Lains (2002) examines the reasons behind the growth experience of a set of southeastern European countries such as Portugal, Greece, and the Balkan countries for the period 1870–1913. First, he shows that the initial income levels of these countries by 1870 were lower than the income levels of the Scandinavian countries that eventually converged to the levels of the high-income countries. Second, he shows that a variety of experiences regarding openness to trade or capital imports characterized these countries. Yet there were no important differences in their economic performance. He concludes that the reasons for economic backwardness in southeastern Europe, despite much accepted wisdom to the contrary, were outside the realm of options that could be overcome with simple policy devices such as low tariffs or financial discipline. Capital productivity and capital deepening The impact of factor growth and, in particular, capital accumulation has played a significant role in the controversies regarding productivity growth for different countries. The OECD (2004a) cites the limited extent of capital deepening as one potential cause of low labor productivity in Turkey. Whereas the labor productivity gap between Turkey and the United States is 70 percent as of 2004,29 the study by Farrell et al. (2003) estimates non-residential capital inputs per worker and per capita as 25 and 17 percent of US levels. In Figures 2.3 and 2.4, we present the behavior of the capital–labor ratio and aggregate labor and capital productivity for the period 1972–1999.30 We find that the average growth rate of the capital–labor ratio was 4.7 percent for 1972–1999. As a consequence, the capital–labor ratio increased by over three times between 1972 and 2000. Comparing the pre- and post-1980 periods, we find that the growth rate of the capital–labor ratio was over 7.7 percent during the former and fell to 3.8 percent in the latter. Aggregate capital productivity typically shows a declining trend. The exception is the period following the liberalization of the Turkish economy beginning in 1980 and lasting until 1987. As we discussed earlier, the period between 1980–1989 is characterized by wage suppression with severe restrictions against collective bargaining and unionization. By contrast, real wages started increasing after 1989, thus inducing firms to substitute capital for labor in production. From Figures 2.3 and 2.4, we also observe increased variability in capital productivity and capital deepening during the 1990s. For the East Asian countries, Krugman (1994) and Young (1995) instigated a wide debate by arguing that East Asian growth has been due to capital
Productivity and growth, 1923–2003
39
140 120 100 80 60 40 20 0 1972
1975
1978
1981
1984
1987
1990
1993
1996
1999
Figure 2.3 Capital deepening.
0.08
6
0.07
5
0.05 0.04
3
0.03
2
Capital productivity
Labor productivity
0.06 4
0.02 1
0 1972
0.01 0 1975
1978
1981
1984
Labor productivity
1987
1990
1993
1996
1999
Capital–labor ratio
Figure 2.4 Labor productivity and capital productivity, 1972–2000.
accumulation and not TFP growth. Indeed, the calculations in Young (1995) showed remarkable increases in the capital–output ratios for these countries, with growth rates of 3.4 percent for Korea between 1966 and 1990, 2.8 percent for Singapore, 0.7 percent for Hong Kong, and 2.9 percent for Taiwan. By contrast, TFP growth is typically found to be low, and even negative for Singapore. If these findings are held to be correct, then the East Asian growth “miracle” would simply be the observed transition of a neoclassical economy to a given steady state. Hsieh (2002) has used the dual measure of TFP growth to challenge these findings.
40
Sumru Altux and Alpay Filiztekin
In particular, he argues that if the capital–output ratio had increased at such a high rate, then we should have observed comparable declines in the rental rate of capital. Based on new calculations using the dual measure of TFP growth, he finds that TFP growth in Singapore is positive and of a comparable rate as the other East Asian countries, the discrepancies arising from the over-statement of investment in the Singaporean national accounts.31 By contrast, the dual TFP estimates for the other East Asian countries and, in particular, for Korea are found not to differ significantly from the earlier primal estimates. Thus, the finding remains that growth in the East Asian countries has not been achieved through significant increases in TFP. It is of interest to examine further the evidence for Turkey in light of the findings for East Asia. Considering the behavior of the capital–output ratio for Turkey, our results indicate that this grew at a substantially lower rate than the East Asian countries. For the pre-1980 period, the capital–output ratio grew at a rate of 4 percent, which is comparable to some of the East Asian countries but this growth rate fell to only 1.1 percent for the 1980–2000 period. Hence, the growth rate was only 1.8 percent over the period 1972–2000. In a study of aggregate TFP and labor productivity growth, Saygili et al. (2001) report estimates of TFP growth for Turkey, which are equal to 0.29 percent for 1972–1979 and 0.44 percent for 1980–2000. By contrast, the average TFP growth during the period 1966–1990 is around 2 percent for the East Asian countries. Saygili et al. (2001) also examine Turkish productivity performance relative to a selected group of OECD countries for the period 1970–2000. They find that the contribution of capital to GDP growth is substantial across countries, and especially important for Turkey. They find no evidence of catching up for Turkey based on the performance of TFP. They do find some evidence of catching up based on the performance of labor productivity, indicating the importance of capital accumulation or capital deepening. However, even this factor ceases to contribute after the financial crisis of 1994.32 These findings suggest that capital accumulation has been relatively low for Turkey, compared to high-performers such as the East Asian countries, for example. Moreover, relative to OECD countries, Turkey also has a weak record of TFP growth. We now discuss some additional factors that may have affected this performance. In the context of developed economies, various authors have turned to the increasingly important role of investment in ICT technologies for TFP growth and capital formation.33 Saygili et al. (2001) focus on factors related to the Knowledge Economy and intangible investments to explain low TFP growth in Turkey. The OECD (2004b, p. 30) also cites Turkey’s poor record in ICT investment and R&D activities as leading to a low level of technical know-how. In a related analysis, Dönek (1998) examines the competitiveness of the Turkish economy in the new technology goods. These are typically based on electronic and semi-conductor technologies, and have constituted the mainstay of the Japanese and East Asian growth and export performance. From the viewpoint of the Turkish experience between 1970–1990, he defines these goods to be chemicals
Productivity and growth, 1923–2003
41
and related products, manufactured goods, and machinery and transport equipment. Dönek tests a set of hypotheses that determine an economy is competitive in these goods. He finds that the hypotheses are not, on the whole, supported for the Turkish economy. Fiscal policy and infrastructure investment has been considered by the new growth literature as potentially important determinants of growth.34 Ismihan, Metin-Özcan, and Tansel (2002) study Turkey’s investment performance and its relation to growth and macroeconomic instability over the period 1963–1999. They distinguish between the behavior of private and public investment and demonstrate that there are significant differences in investment rates during the pre- and post-1980s.35 They argue that one of the important effects of macroeconomic instability in the Turkish economy has been to reduce public investment, especially its infrastructure component, and to reverse the complementarity between private and public investment in the long run. Examining the dynamic response of the variables, they find that both private investment and output respond positively to increases in public investment but that the response of private investment is significant and large. Aside from public investment, one component of fiscal policy that has received some attention in the Turkish case is defense and military expenditures. Turkey’s military expenditures have typically been high, reaching 4.42 percent for the decade of the 1990s. Sezgin (2001) estimates a demand–supply model for Turkey’s defense expenditures for the period 1954–1994. He finds that military expenditures have a positive effect on growth but the impact of defense spending is not significant for saving or the trade balance.36 By contrast, Dunne et al. (2001) find no relationship between the growth rates of Greek military expenditures and output for the period 1963–1996 when the cointegration between these series is taken into account, and that military expenditures in Turkey have a negative effect on GDP growth. Sezgin (2000) also provides a disaggregated analysis of defense spending and economic growth for Turkey for the period 1985–1999. He finds that defense equipment spending affects output growth positively in the short- and long-run but the effect of personnel expenditures on output growth is negative. One aspect of capital investment decisions that tends to get overlooked in many discussions has to do with their political context. The dynamic fiscal policy framework suggests that one objective of constructing long-lived capital projects today may be to control policy-makers’ options in the future.37 In our earlier analysis, we noted that the Özal years were characterized by an ambitious program of public investments. We conjecture that some of the new investment projects were seen as a commitment device by the Ozal administrations. Whether they succeeded in this dimension is a topic that we leave for future study. The impact of firm size on productivity In this section, we consider the impact of firm size on productivity. Table 2.5 shows the share of private firms by size in total manufacturing value-added and
(c) Productivity Small (10–24) Small (10–49) Medium (25–99) Medium (50–99) Large (100) Large (200) Manufacturing
8.01 19.82 23.50 11.69 68.49 54.04
6.51 18.32 24.41 12.60 69.08 53.50
0.41 0.48 0.60 0.77 1.22 1.31 1.00
0.27 0.51 0.66 0.83 1.22 1.29 1.00
0.45 0.44 0.57 0.74 1.23 1.34 1.00
0.65 0.51 0.60 0.78 1.19 1.29 1.00
1996–1999
10.20
3.45
0.80 1.62 3.63 3.75 4.01 3.88
No. of firms
6.15 6.52 5.60 6.75 6.49 6.62 6.58
1981–1988
45.25 71.19 39.37 13.43 15.38 7.55
0.61 1.30 1.43 6.77 4.84 5.31 4.78
1981–1988
4.91
2.13 0.78 3.99 4.46 6.39 6.74
1981–1999
1980–1988
1989–1995
10.33
9.94 15.50 14.50 9.44 9.44 8.53
1980–1999
9.69
10.59 10.86 10.56 10.28 10.28 9.87
Productivity growth
10.03
Manufacturing
1.52 2.08 5.42 11.23 11.23 12.05
1996–1999
11.80 23.55 23.69 11.94 64.51 51.94
Relative productivity
5.35 8.14 9.23 10.51 10.51 10.50
1989–1995
9.42 21.20 23.76 11.98 66.82 52.99
1981–1999
1981–1988
2.79 9.30 14.73 8.22 82.48 68.90
1981–1999
3.00 8.72 13.34 7.62 83.66 72.05
Employment
5.28 12.24 15.69 8.73 79.02 67.20
Value added
3.99 10.42 14.68 8.24 81.34 69.24
(b) Growth rates Small (10–24) Small (10–49) Medium (25–99) Medium (50–99) Large (100) Large (200)
(a) Share according to size Small (10–24) Small (10–49) Medium (25–99) Medium (50–99) Large (100) Large (200)
Employment
42.18 69.51 41.13 13.80 16.69 8.11
10.30 8.98 8.27 8.65 8.48 8.47 8.53
1989–1995
1.80
0.30 1.89 2.29 1.64 1.80 1.40
1989–1995
51.81 75.20 35.40 12.02 12.79 6.48
9.99 12.67 9.25 3.40 6.31 6.00 6.77
1996–1999
3.43
0.05 2.83 5.25 6.04 3.13 2.53
1996–1999
35.88 65.13 45.22 15.97 18.90 8.99
1980–1999 1980–1988 1989–1995 1996–1999 1980–1999 1980–1988 1989–1995 1996–1999 1980–1999 1980–1988 1989–1995 1996–1999
Value added
Table 2.5 Small- and medium-sized firms, 1980–1999
Productivity and growth, 1923–2003
43
employment for the period 1980–1999. According to these results, the share of small firms, those employing 10–29 or 10–49 workers, has been shrinking in the share of both manufacturing value-added and employment and the share of large firms has been increasing over this period. By contrast, the share of medium-sized firms appears to have fallen in total value-added but increased slightly in employment over the same period. Interestingly, relatively smaller firms, namely, those with workers 10–29, 10–99, and 25–99, have shown more rapid value-added growth in the 1989–1999 period whereas the relatively larger firms have witnessed a decline in their growth rate during this period. Reflecting the more capital-intensive production in larger firms, employment growth at large firms has also been relatively weak. The second part of Table 2.5 shows that productivity growth has increased for firms in all firm size categories for the 1989–1995 period. Surprisingly, productivity growth has increased or remained high during the period 1996–1999 for relatively smaller firms. As we show later, this may be related to survival by small firms in the face of increased macroeconomic instability during this period. Erzan and Filiztekin (1997, 2003) examine productivity and employment growth for private small- and medium-sized establishments (SMEs). Both papers examine the impact of the Customs Union Agreement on the competitiveness of SMEs. Whereas the first paper refers to the period 1980–1992, the second refers to 1980–1999. This paper also estimates an error components model for valueadded, productivity, employment, and real wage growth controlling for sectorspecific and size-specific effects. It finds that size matters and the smaller firms have poorer performance. However, industry specific factors turn out to be more important than size, explaining up to 39.3 percent of the variation in the different variables. The paper also finds that small firms are widely affected from changes in the economic environment. By contrast, there is no impact of macroeconomic variables on value-added growth for large firms. For productivity growth, declines in interest rates and increases in unit labor costs have positive effects for large firms, suggesting that large firms substitute capital for labor when unit labor costs increase and lower interest rates make this substitution more desirable. However, changes in the economic environment do not induce productivity gains for small firms. The paper notes that the greater sensitivity of small firms to macroeconomic conditions might be interpreted as greater flexibility by small firms. However, in the absence of productivity gains under more adverse conditions, such flexibility is more likely for “survival,” not growth. In the earlier analysis, no account was made of the process of entry, exit, and the dynamics of firm survival and growth. Taymaz (2005) examines productivity growth for SMEs for the period 1987–1997 by taking this process into account. The paper utilizes a stochastic production frontier model to estimate the impact of technical efficiency, scale economies, and the rate of technical change for 4-digit manufacturing industries. It models the entry and exit decisions of firms as a function of expected productivity, expected prices, and the opportunity cost of operation. The paper finds that entrants are significantly smaller than incumbents, and the exit rate is quite high for small entrants. Entrants are less efficient
44
Sumru Altux and Alpay Filiztekin
than incumbents, and the efficiency differential is wider for smaller entrants. However, the mean efficiency of entrants increases over time. Likewise, those entrants that increase their efficiency and/or scale are more likely to survive, but surviving establishments find it difficult to overcome efficiency differentials. These results suggest that the dynamics of firm growth and survival depend on size. There are various policy measures that could be put in place to imply a greater survival probability for small firms. These include greater access to information and communications technology (ICT) and financial credit.38 It is well known that macroeconomic instability in Turkey during the 1990s has led to reduced business lending to SMEs and also curtailed investment in productive public infra-structure (including ICT). Our analysis suggests that smaller firms have shown significant growth potential despite such instability. Thus, our results indicate that a more stable macroeconomic and political environment would lead to conditions for productivity improvements through the enhanced performance of SMEs.
Regional convergence and growth Regional patterns of convergence and growth are the result of a complex set of historical, cultural, institutional, and policy factors. No doubt geographical and political factors also play a role. Throughout its history, there were many centers of trade and commerce in Turkey, which lay on the paths of interregional and international trade.39 In this section, due to data limitations, we focus on the period 1975–2000, and examine regional patterns of growth and convergence for 65 Turkish cities or provinces. Regional convergence is typically analyzed from the viewpoint of the neoclassical growth model. Regions that have lower per capita income ought to grow faster, yielding the standard growth regressions in the literature. In a variety of studies, Barro and Sala-i-Martin have examined regional convergence empirically for the US states and regions of Europe, Japan, and the Canadian provinces.40 One of the well-known results from these studies is that different regions all tend to converge at a rate of about 2 percent per year, a result known as the 2 percent rule.41 Various authors have also argued that there may be the effects of location on regional convergence.42 Table 2.6 shows the regional growth rates in real value-added and the regional shares of real value-added and employment.43 We note Turkey has experienced 3.8 percent growth in value-added over the sample period. However, value-added per person has grown by 1.7 percent, and value-added per labor force has grown by even less. Thus, we find that Turkey has achieved a relative improvement in productivity during the 1975–2000 period. However, it has not been as successful in creating new employment for its working age population. Table 2.6 also shows that there are significant differences among the regions based on all three indicators. Regional patterns of growth and convergence may be due to region-specific factors such as the distribution of factors of production, the impact of interregional income transfers, or other social or cultural factors. Another important factor may have to do with the sectoral allocation of production across regions.
Productivity and growth, 1923–2003
45
Table 2.6 Value added and employment by region
Turkey Marmara Aegean Medittarenean Black Sea Central Anatolia Eastern Anatolia
Regional value added growth, 1975–2000 (annual %)
Share of value added and and employment, 2000 (%)
Value added
Value added per capita
Value added per labor force
Labor productivity
Value added
Employment
3.77 4.29 3.92 4.26 2.33 3.77 2.72
1.70 1.22 1.90 1.67 1.61 2.09 0.68
1.04 0.77 1.34 0.89 0.80 1.35 0.11
2.16 1.67 2.21 1.99 1.84 2.46 1.63
38.9 15.1 11.8 9.3 16.5 8.4
25.6 15.1 12.8 14.0 16.5 16.0
Table 2.7 Sectoral shares by region, 2000 (%)
Share of sectoral production by region Agriculture Mining Manufacturing Utilities Construction Trade Transportation Finance Services Share of sectoral employment by region Agriculture Mining Manufacturing Utilities Construction Trade Transportation Finance Services
Turkey
Marmara
Aegean
Medittarenean
Black Sea
Central Anatolia
Eastern Anatolia
14.0 1.4 24.8 3.3 5.2 23.3 13.7 7.5 6.7
5.7 0.3 34.0 3.3 4.0 26.2 12.2 8.7 5.6
16.6 3.0 24.7 4.6 4.6 22.1 12.2 6.5 5.8
19.6 0.9 19.3 2.7 5.1 25.1 15.2 5.7 6.5
24.0 1.7 21.2 2.7 4.8 18.1 16.3 4.3 6.9
14.4 1.7 15.7 2.4 9.0 22.3 16.4 10.0 8.0
27.9 3.6 11.8 4.4 5.7 17.0 13.4 5.1 11.1
48.4 0.4 12.6 0.4 4.6 9.7 3.3 3.1 17.6
25.3 0.2 25.1 0.4 5.8 14.3 4.8 5.4 18.8
50.5 0.6 12.8 0.4 4.2 10.1 3.0 2.6 15.7
55.0 0.1 8.3 0.4 4.5 10.5 3.0 2.3 16.0
66.1 0.8 6.2 0.3 4.1 6.0 2.2 1.4 12.9
46.8 0.4 9.6 0.5 4.8 8.9 3.4 4.0 21.5
64.0 0.1 4.6 0.3 3.4 5.3 2.2 1.1 18.9
Table 2.7 shows that there are indeed significant differences in this dimension. First, while agriculture generates only 14 percent of total real value-added, its share in total employment is 48.4 percent. By contrast, the Marmara region stands out in terms of its low share of agricultural real value-added and employment. These facts imply that sectoral considerations are likely to have an important impact on regional convergence and growth.
46
Sumru Altux and Alpay Filiztekin
Long-run regional convergence The issues of long-run convergence and growth across regions have been studied by a number of authors. Filiztekin (1998) finds only conditional convergence for the period 1975–1995. Temel et al. (1999) find evidence for a bi-polar distribution characterizing the long-run behavior of labor productivity across regions. Their results indicate that a majority of the provinces move to a low productivity level while a few move towards a high productivity level. Temel et al. (1999) also characterize the spatial pattern of productivity using nonparametric methods, and show that there is a pattern of high productivity around three industrialized provinces – Istanbul, Izmir, and Adana – that has lasted throughout the period 1975–1990. Table 2.8 presents the results on long-run regional convergence for completeness sake. We consider regressions of value-added growth on the initial level of per capita value-added. The results can be summarized as follows. First, value-added per capita shows a tendency for absolute divergence, the convergence coefficient being positive and significant. By contrast, value-added per labor force shows a tendency for absolute convergence. However, the estimated convergence rate across different provinces is very low. For the value-added per employee regression, the convergence coefficient is positive but it is not statistically estimated. Thus, we can infer neither convergence nor divergence. To allow for differences in the steady state values of the different variables, we also consider regressions that include province dummies. In this case, each province is found to converge to its own steady state but the convergence rate for labor productivity is nearly half the rate for the other measures. Thus, even when we consider conditional convergence, the results in Table 2.8 indicate that the tendency for labor productivity to show convergence across different provinces is quite low. In Table 2.9, we examine the regional convergence of labor productivity by sector. Aside from agriculture, all sectors indicate that there is statistically significant conditional convergence of labor productivity by province. However, the convergence rate for manufacturing is found to be quite low. Likewise, trade and transportation industries also have a large half-life. Taken together, the results in Tables 2.8 and 2.9 indicate that convergence of various measures of per capita value-added by province appears weak. When sectoral considerations are taken into account, we find that convergence across regions has not occurred because of the absence of convergence of agricultural productivity and differences in the sectoral allocation of real value-added by regions. Regional business cycles No doubt business-cycle type phenomena also influence regional growth and convergence. Filiztekin (2004) provides the only study that examines the relationship between sectoral growth and business cycles. Table 2.10 presents the weightedaverages of variances over time of regional value-added growth and their covariances with aggregate value-added growth. The variances of value-added growth of all the regions outside of the Marmara and Aegean regions appear to have declined in the 1990–2000 period compared to 1980–1990. Given two major financial crises
Productivity and growth, 1923–2003
47
Table 2.8 Regional convergence, 1975–2000
OLSa Convergence coefficient
Value added per capita
Value added per labor force
Labor productivity
0.0064 (0.0028) 0.0761 108
0.0056 (0.0005) 0.6734 125
0.0013 (0.0028) 0.0036 517
0.0571 (0.0065) 0.3969 12 0.429
0.0260 (0.0041) 0.2997 27 0.483
R-squared Half-life Least squares with dummy variablesb Convergence coefficient 0.0643 (0.0100) R-squared 0.3027 Half-life 11 Coefficient of variation, 0.466 fixed effects
Notes Let xit denote the growth rate of the relevant variable at date t in province i and xi,1975 its value in 1975. Estimated regressions: a xit a b xi,1975 b xit a b xi,1975 Di where Di is a province dummy. Standard errors in parentheses.
Table 2.9 Regional convergence by sector, 1975–2000
Aggregate productivity Agriculture Construction Utilities Finance Manufacturing Mining Services Transportation Trade
Convergence coefficient
Standard error
R-squared
Half-life
0.0013 0.0075 0.0407 0.0445 0.0446 0.0089 0.0220 0.0282 0.0144 0.0105
(0.0028) (0.0051) (0.0034) (0.0055) (0.0046) (0.0047) (0.0207) (0.0039) (0.0056) (0.0049)
0.0036 0.0327 0.6975 0.5132 0.5974 0.0531 0.0176 0.4494 0.0954 0.0672
517 93 17 16 16 78 32 25 48 66
Notes Let xit denote the growth rate of the relevant variable at date t in province i and xi,1975 its value in 1975. Estimated regressions: xit a b xi,1975 Di; where Di is a province dummy.
and other external factors, this suggests that the recent crises are a ‘western’ phenomenon. By contrast, the remaining regions appear to have been affected more by the policies during the transition from import-substituting industrialization to export-led growth. Another surprising result is that the co-variances of regional value-added growth with aggregate value-added growth have increased for all regions except the eastern Anatolian region during the 1990s.
48
Sumru Altux and Alpay Filiztekin
Table 2.10 Regional cycles and specialization
Marmara Aegean Medittarenean Black Sea Central Anatolia Eastern Anatolia
Variance of regional value-added growth
Covariance with aggregate value-added growth
1980–1990
1990–2000
1980–1990
1990–2000
0.2992 0.2975 0.9821 0.6416 0.5446 1.5543
0.4713 0.3317 0.3228 0.5956 0.2984 0.5537
0.2978 0.3624 0.3933 0.1362 0.6424 0.3395
0.8055 0.6941 0.6962 0.5598 0.6973 0.3152
Dependent variable: correlation of regional value-added with total value-added growth ( i) 1980–2001 1980–1990 1990–2000 Specialization Share of agriculture R-squared
1.7082 (0.5620)*** 1.3356 (0.2507)*** 0.2282
1.4374 (0.5580)*** 1.0716 (0.3403)*** 0.0991
1.3020 (0.9554) 1.3281 (0.3121)*** 0.2225
Notes *** denotes statistical significance at the 1% level. The regression is i a bSi cAgrshi, where Si Ui Ss1(I˙KD si/I˙ KDi 1/(J 1) ji(I˙KD sj/I˙KDj)) and Agrshi is share of agricultural value added in provincial total. Standard errors in parentheses.
We also examine the impact of the sectoral distribution of value-added on the correlation between regional growths and aggregate value-added growth. For this purpose, we consider a regression that includes a measure of specialization and the share of agriculture.44 The results in Table 2.10 show that as regional specialization increases, the correlation between regional and aggregate value-added growth falls. However, this relationship falls in magnitude over time, and the significance of the relationship disappears. By contrast, as the share of agriculture increases, the correlation between regional and aggregate value-added falls, and this relation tends to become more important over time. These results indicate that cyclical phenomena at the regional level have witnessed some changes in the 1990s relative to the 1980s. Whereas in the earlier period, the correlation between output growth for different cities or provinces was quite weak, this correlation has grown in recent years. Likewise, the more specialized a region is, the less correlated will be its business cycle with the aggregate. However, this result tends to weaken in recent years. How these changes will affect regional convergence and growth in Turkey is a topic that we leave for future study.45
European Union and Turkey: ready or not? The growth prospects for Turkey depend, in many crucial ways, on the course and outcome of future membership deliberations with the European Union. Whether these deliberations will be concluded successfully, in turn, depend
Productivity and growth, 1923–2003
49
crucially on Turkey’s productivity performance. In this section, we provide an international comparison of productivity and convergence. International productivity comparisons Table 2.11 shows the growth rates of labor productivity and employment for a selected group of OECD countries for the period 1980–2002.46 It is of interest to note that Turkey has the highest rate of productivity growth of 3.41 percent among the entire sample of countries for the period 1980–1991. The only country that comes close is Portugal, a relative latecomer to the European Union. Both productivity growth and employment growth decline during the 1991–2002 period for Turkey. Productivity growth for Turkey in the 1991–2002 is still comparable to that in some high-performing OECD countries such as Finland and Sweden as well as Greece. However, employment growth declines from nearly 2 percent to less than 1 percent during the 1991–2002 period for Turkey.47 Table 2.12 further decomposes labor productivity growth into productivity growth in a given sector versus re-allocation of labor across sectors. One of the salient findings from Table 2.12 is that the contribution of intra-sectoral productivity growth to productivity growth falls dramatically during the 1991–2002 period for Turkey. Table 2.11 Productivity and employment growth Productivity
Turkey Austria Belgium Czech Republic Denmark Finland France Germany Greece Hungary Italy Netherlands Norway Poland Portugal Spain Sweden UK USA Average
Employment
1980–1991
1991–2002
1980–1991
1991–2002
3.41 2.66 2.08 — 1.43 2.46 2.48 — — — 1.63 0.72 2.08 — 3.13 2.31 — 1.92 1.21 2.12
2.46 1.92 1.06 3.84 1.76 2.93 1.12 1.62 2.53 3.5 1.24 0.69 2.09 3.61 2.12 0.82 2.23 2.11 1.78 2.06
1.99 0.29 0.11 — 0.2 0.05 0.28 — — — 0.71 3.34 0.43 — 0.41 0.87 — 0.33 0.74 0.74
0.97 0.24 0.56 2.07 0.41 0.02 0.8 0.05 0.78 0.34 0.31 1.93 1.17 0.07 0.33 1.37 0.67 0.51 0.48 0.48
Available data: Czech Republic 1995–2000; Greece 1995–2000; Hungary 1992–2000; Netherlands 1986–1991 and 1991–2001; Poland 1992–2002; Portugal 1980–1991 and 1991–1999; Spain 1980–1991 and 1991–2001; Sweden 1993–2002; UK 1980–1991 and 1991–2001; US 1980–1991 and 1991–2001.
Turkey Austria Belgium Czech Republic Denmark Finland France Germany Greece Hungary Italy Netherlands Norway Poland Portugal Spain Sweden United Kingdom United States Average
2.75 1.97 1.81 — 1.06 1.83 1.93 — — — 0.37 0.61 1.54 — 1.99 1.71 — 2.05 0.93 1.58
0.61 1.35 0.96 3.84 1.61 2.78 0.9 1.03 2.53 3.04 0.88 0.73 1.86 3.68 1.79 0.65 1.8 2.38 1.71 1.76
0.67 0.69 0.27 — 0.37 0.64 0.56 — — — 1.26 0.12 0.54 — 1.13 0.6 — 0.12 0.28 0.54
1.85 0.57 0.1 0 0.14 0.15 0.23 0.59 0.78 0.46 0.36 0.04 0.23 0.07 0.32 0.17 0.42 0.27 0.07 0.3
1991–2002
1980–1991
1980–1991
1991–2002
Inter-sectoral employment shift
Intra-sectoral productivity growth
Table 2.12 Decomposition of labor productivity
80.45 74.08 87 — 73.98 74.06 77.54 — — — 22.59 83.97 74.06 — 63.76 73.9 — 106.34 76.97 74.52
1980–1991 24.81 70.18 90.68 99.88 91.8 94.88 79.87 63.43 76.53 86.85 71.06 105.16 88.93 101.9 84.75 78.89 81.1 112.95 96.06 82.91
1991–2002
Contribution of intra-sectoral productivity growth
19.55 25.92 13 — 26.02 25.94 22.46 — — — 77.41 16.03 25.94 — 36.24 26.1 — 6.34 23.03 25.48
1980–1991
75.19 29.82 0.32 0.12 8.2 5.12 20.13 36.57 23.47 13.15 28.94 5.16 11.07 1.9 15.25 21.11 18.9 12.95 3.94 17.09
1991–2002
Contribution of inter-sectoral employment shift
Productivity and growth, 1923–2003
51
This is in contrast to the remaining countries in our sample, all of which show high and even increasing intra-sectoral productivity growth rates over both the 1980–1991 and 1991–2002 periods. By contrast, intra-sectoral productivity growth for Turkey during 1991–2002 is nearly one-quarter of its value for the earlier period 1980–1991, and the major contribution to productivity growth during 1991–2002 comes from shifts in employment to faster-growing sectors. No doubt part of the reason for this finding is that Turkey is still experiencing the movement of employment out of agriculture and into more productive sectors of the economy. However, another factor is the weak overall productivity performance for Turkey during 1991–2002. Decomposing international productivity movements: changes in efficiency versus technical change In this section, we construct an international production frontier, and examine Turkey’s performance relative to this frontier. This approach also allows us to observe movements of different countries towards this frontier (changes in efficiency) versus movements in the frontier itself (technical change). We construct the international production possibilities frontier for labor productivity using data on 18 OECD countries and outputs in four broad sectors in each country for the period 1981–1999. Due to data availability problems, the four broad sectors are defined as agriculture, manufacturing, mining and utilities, and a fourth sector encompassing construction, trade, transportation, finances, and the services sector. The novel feature of our approach compared to some of the other approaches that we discussed earlier refers to the fact that we use multiple outputs. Hence, our efficiency calculations reflect the efficiency with which the labor input is used in producing outputs in different sectors. The frontier is constructed using deterministic production frontier methods that we described in the section on the Growth accounting approach. As in Färe et al. (1994) and other papers, we calculate the technical efficiency index under both constant returns to scale and also variable returns to scale.48 The scale effect is defined as the ratio of the index or efficiency scores under constant returns versus variables returns to scale as SEt tCRS/ tVRS. Table 2.13 reports the technical efficiency indices for selected years. Turkey’s efficiency score is the worst among the selected group of OECD countries, suggesting that Turkey is farthest from the frontier for all the years in the sample.49 This is not surprising as the selected countries are those which have completed the transformation from agriculture to a much larger extent than Turkey. Comparing Turkey’s performance with Korea and Portugal, both of these countries have achieved impressive increases in efficiency between 1980 and 1999. However, for the latter, the increase in efficiency is mostly due to improvements in scale. Correcting for changes in scale efficiency, Turkey also displays catching-up between 1980 and 1990, as discussed earlier. However, its distance from the constructed production frontier appears to have increased during 1990–1999, indicating the role of macroeconomic instability and various external shocks during this period. Table 2.14 also reports the average annual change in the Malmquist productivity index and its components over the entire sample period 1981–1999. In
Turkey Austria Belgium Denmark Finland France Italy Netherlands Portugal Spain Sweden United Kingdom Norway Australia Korea Japan Canada USA
0.3985 0.6826 0.5077 0.9398 0.5749 0.9659 0.6516 0.8809 0.5229 1.0000 0.5333 0.4465
1.0000 0.5087 0.6576 1.0000 0.6395 0.7332
0.3598 0.5147 0.4162 0.6713 0.6233 0.6659 0.5844 0.7019 0.3630 0.8836 0.3567 0.4120
1.0000 0.3932 0.5323 1.0000 0.6544 0.6282
1.0000 0.5190 0.7146 1.0000 0.6186 0.7397
0.3559 0.8243 0.5124 1.0000 0.6407 1.0000 0.7699 0.8402 0.4753 0.8032 0.5788 0.4945 1.0000 0.4485 0.5867 1.0000 0.7129 1.0000
0.4090 0.7382 0.6381 1.0000 0.9933 0.6921 0.6139 0.7428 0.5555 0.9342 0.5613 0.4170
1980
1999
1980
1990
VRS level ( tVRS)
CRS level ( tCRS)
Table 2.13 Technical efficiency in selected years
1.0000 0.5101 0.6614 1.0000 0.6519 1.0000
0.4103 0.8574 0.7189 1.0000 0.8356 0.9683 0.6558 0.8976 0.5997 1.0000 0.5862 0.4523
1990
1.0000 0.5325 0.7166 1.0000 0.6875 1.0000
0.3633 0.9547 0.7364 1.0000 1.0000 1.0000 0.7725 0.8557 0.5233 0.8232 0.7321 0.5003
1999
1.0000 0.8767 0.9072 1.0000 0.9180 0.6282
0.8796 0.6972 0.6523 0.6713 0.6276 0.9621 0.9519 0.9449 0.6534 0.9459 0.6355 0.9878
1980
Scale effect (SEt)
1.0000 0.9974 0.9942 1.0000 0.9811 0.7332
0.9713 0.7961 0.7063 0.9398 0.6880 0.9975 0.9937 0.9814 0.8720 1.0000 0.9097 0.9871
1990
1.0000 0.9746 0.9973 1.0000 0.8998 0.7397
0.9797 0.8634 0.6958 1.0000 0.6407 1.0000 0.9967 0.9818 0.9084 0.9758 0.7907 0.9884
1999
Productivity and growth, 1923–2003
53
Table 2.14 Decomposition: average annual changes
Turkey Austria Belgium Denmark Finland France Italy Netherlands Portugal Spain Sweden United Kingdom Norway Australia Korea Japan Canada USA
Malmquist index
Technical change
Efficiency change
Pure efficiency change
Scale change
1981–1999
1981–1999
1981–1999
1981–1999
1981–1999
0.9692 1.0164 1.0077 1.0185 1.0083 1.0203 1.0165 1.0054 1.0142 1.0015 1.0203 1.0059 1.0000 1.0147 1.0127 1.0000 0.9994 1.0123
0.9699 0.9883 0.9934 0.9943 1.0056 0.9959 0.9994 0.9953 0.9985 1.0061 0.9923 0.9937 1.0000 0.9983 0.9960 1.0000 1.0024 1.0036
1.0019 1.0286 1.0144 1.0243 1.0044 1.0248 1.0174 1.0106 1.0173 0.9968 1.0291 1.0127 1.0000 1.0211 1.0210 1.0000 0.9980 1.0119
0.9952 1.0141 1.0077 1.0000 1.0011 1.0222 1.0144 1.0087 0.9986 0.9946 1.0147 1.0167 1.0000 1.0137 1.0140 1.0000 1.0009 1.0000
1.0062 1.0130 1.0064 1.0243 1.0044 1.0021 1.0025 1.0022 1.0205 1.0020 1.0150 1.0014 1.0000 1.0061 1.0055 1.0000 1.0014 1.0119
contrast to our analysis in the section on the Growth accounting approach, this decomposition also includes the breakdown of efficiency changes into a pure efficiency change and changes in scale. Turkey has the lowest rate of growth in technical change in the sample, which is comparable to the change in productivity measured by the Malmquist index. The productivity increases in Turkey reflect efficiency changes, with the scale efficiency changes being the most important. This is also true for a high-performing country such as Korea. By contrast, for productivity leaders such as Finland and the United States, there are positive rates of growth for technical change. In conclusion, taking sectoral considerations into account, there is evidence of catch-up and efficiency increases for Turkey. However, macroeconomic instability and external shocks have a negative effect on this performance. Our results indicate that technical change appears to have occurred for rich countries whereas productivity increases for middle-income countries are of the efficiency variety.50 These results indicate that Turkey has witnessed improvements in its relative productivity performance, whether this is judged on the basis of the aggregate economy or its broadly defined sectors. However, its performance has lagged behind other countries at comparable stages of development. Some recent papers have raised the role of structural transformation in discussions surrounding overall convergence.51 More recently, Ventura (1997) has argued that the process of structural transformation may be also linked to capital deepening or capital accumulation when the interdependence between different economies is taken into account. In the context of Turkey, this model suggests that macroeconomic
54
Sumru Altux and Alpay Filiztekin
instability and low rates of savings and capital accumulation may be reasons behind the relatively slow process of structural transformation and the inability of the Turkish economy to achieve sustained increases in its growth rate.
Conclusion The conclusions from our study are varied and interesting. We find a number of factors at work that have led to increases in productivity and per capita income growth in the period since 1923. In the very early years of the Republic, the initial gains came from liberalizing and invigorating the agricultural and manufacturing sectors through a sustained policy effort. In later years, Turkey practiced import-substituting industrialization. Import-substitution policies were followed throughout the 1960s and 1970s. Probably their most important outcome was to lead to growth through increases in capital accumulation. However, the historical record suggests that unlike other rapidly industrializing countries of this period, Turkey did not experience significant increases in TFP growth. The import-substituting period came to an end with the onset of the 1970s oil shocks, and the political and social turmoil that ensued. The 1980s witnessed a change in the orientation of the Turkish economy to export-led growth. Trade and financial liberalization measures were enacted, and the initial response of the economy was an increase in exports, output, and productivity growth. The 1990s constitute a disappointing episode; macroeconomic and political instability and two major financial crises mark this period. TFP growth became negative, and labor productivity growth declined to a fraction of that observed in high-performing countries. In this chapter, we have examined a variety of determinants of productivity and growth. We have also presented an international productivity comparison and discussed issues related to regional convergence and growth. Our analysis has not provided direct evidence on the importance of exogenous versus endogenous theories of growth. Nevertheless, we have examined how openness affects productivity and growth, and discussed the role of capital accumulation in an open economy. We have also pinpointed deficiencies in Turkey’s performance in such areas as development of the Knowledge Economy and investment in infrastructure. Policies that lead to a stable political and macroeconomic environment – by creating the conditions for capital accumulation through foreign direct investment, for the better allocation of domestic credit, and for renewed investment in such areas as ICT and infrastructure improvements – also appear key. There are other determinants of productivity and growth that we have not examined in this chapter. One of these is corruption.52 Another factor is the existence of the informal economy.53 We leave these issues for future analysis. It appears that Turkey is at a crossroad in terms of its future prospects. In this chapter, we have provided a wide-ranging overview of the events and issues that have shaped the Turkish experience in the 80 years since the inception of the Republic. In the chapters that follow, our remaining contributors will provide their own assessments of factors that are likely to determine Turkey’s economic, social, and political agenda for the years ahead.
Productivity and growth, 1923–2003
55
Notes 1 Altux (1989) provides an early example. 2 See, for example, Young (1995). 3 See, for example, the volumes by Owen and Pamuk (1998) or Pamuk and Williamson (2000). 4 The derivation of aggregate GDP for the period 1968–1986 is described in the State Institute of Statistics publication (1994). The remaining data on GDP for the period 1987–2003 are from the State Institute of Statistics. The employment data for 1988–2003 are from the Treasury publication “Treasury Statistics” (2003). Employment data for the period 1923–1987 are obtained by applying the growth rates in Bulutay (1995) in a backward interpolation scheme to the Treasury employment data. 5 We note that relative sectoral productivity may be decreasing even when aggregate labor productivity is increasing. 6 Pamuk (2000) cites the expansion of arable lands, greater labor input by peasant households to sustain a given level of income, and the stimulating effects of urban demand as some reasons behind the performance of agriculture during this period. 7 For example, whereas agriculture and manufacturing employ 5.183 million and 268,000 workers in 1929, real output in these sectors is 2.996 billion TL and 440.7 million TL in 1987 prices, respectively. By 1939, both agricultural and manufacturing output register large increases, the former increasing to 4.485 billion TL and the latter to 1.180 billion TL. However, employment in agriculture also increases 6.389 million workers whereas employment in manufacturing remains at 427,000, explaining the large differences in relative productivity. 8 These measures included import liberalization, the lifting of price controls and curbs on fiscal deficits and monetary expansion, and were to feature in similar programs implemented in the 1970s and 1980s. 9 As some examples, GDP growth between 1963 and 1966 is 6.8 percent. However, it falls to 5.7 percent for 1963–1976. 10 See Derviv et al. (1982) for an analysis of the various factors that contributed to the buildup of the crisis. 11 See Owen and Pamuk (1999, Chapter 5). 12 The Turkish financial system was essentially opened to the rest of the world with the liberalization of foreign exchange operations and international capital movements in 1989 (as well as the establishment of convertibility of the lira). For a further discussion of the financial liberalization in this period, see Atiyas and Ersel (1992). 13 In contrast to investment rates for the private sector which reached 14.1 percent during the 1983–1997 period, they show that investment growth in manufacturing industries remained at 7.7 percent, while the major gains in investment (reaching 24.5 percent growth) were obtained by the private housing sector. 14 See Mankiw et al. (1992). 15 This is similar to the notion of externalities to the aggregate stock of knowledge capital that Romer (1986) uses. 16 For an early discussion of the theory of index numbers, see Diewert (1976). 17 Hsieh (2002) uses the dual measure in the debate surrounding the sources of growth for the East Asian countries. We describe his results in more detail in a later section. 18 Using the terminology of index theory, the Solow residual represents a Tornquist index when it is defined using the approximation: ln(Yi,t 1) JK,L,M(1/2)(sJi,t1 sJi,t)ln ( Ji,t1). 19 For a further analysis of the impact of imperfect competition and scale economies on productivity measurement, see, for example, Altux and Filiztekin (2002). 20 We note that D0t (x t,y t ) 1 if and only if (x t, y t )S t. Furthermore, D t0(xt, yt ) 1 if and only if (x t, y t ) is on the frontier or boundary of technology. If observed production at t is interior to technology at t, we say that x t, y t is inefficient.
56
Sumru Altux and Alpay Filiztekin
21 In this expression, D t0(xt1, yt1) measures the maximum proportionate change in outputs required to make (x t1, yt1) feasible with respect to the technology at t t date t. Likewise, Dt1 0 (x , y ) measures the maximum proportionate change in outputs that t t will make (x , y ) feasible with respect to the technology at date t 1. 22 Using data from both the Census of Manufacturing Industries and the Annual Survey of Manufacturing Industries, Yıldırım argues that government enterprises employed 567.7 workers per firm in 1967 and 612.1 worker per firm in 1975. By contrast, the private sector employed 75.7 and 79.1 workers per firm, respectively. 23 From equation (2.14) (assuming that the materials share is equal to zero), we note that measured TFP growth might be negative if there is capital deepening with significant market power in product markets, that is, dKit/Kit relatively large with it 1, or if there are decreasing returns with 1 or equivalently, the existence of production inefficiencies. 24 To decompose aggregate TFP growth into a sectoral TFP growth component and into a sectoral shift component, let denote the share of labor in output, and define aggregate TFP as the weighted sum of 1 sectoral TFPs as follows: 1 Lt ) i zitTFPit , where zit (K it Lit )/(K t Lt ). Taking first TFPt Yt/(K 1 t differences yields TFPt izi,t1 TFPit iTFPi,t1 zit, where the first term denotes the within-in sectoral productivity effect and the second term denotes the sectoral shift effect. 25 Since TFP growth is negative during 1970–2000 or 1996–2000, a positive sectoral effect implies a decline in aggregate TFP. 26 The results of Kim (2000) on Korea for the period 1966–1988 can be compared with this experience. Kim examines the impact of trade liberalization on TFP growth for 36 Korean manufacturing industries by allowing for imperfect competition and scale effects. He finds that markups were quite substantial for almost all of Korean manufacturing industries but they tended to decline over the sample period, and that trade liberalization has the effect of reducing both markups and the returns to scale. He argues that the process of trade liberalization was incomplete for the period considered in his sample, indicating that its effects would continue to be felt in the future. Finally, after obtaining corrected TFP measures, he concludes that TFP growth was quite low in Korean manufacturing for the period under consideration, a point to which we will subsequently return. 27 Citing many studies that uncover bi-directional causality between exports and growth, Krishna et al. (2003) examine the impact of trade on growth by using a model selection criterion based on the ex ante predictive ability to examine patterns of causation between income, export, import, and investment growth for 39 developing countries (excluding Turkey). In contrast to other studies, they find new evidence of unidirectional causality. Given at times the conflicting evidence on this issue, it appears to worthwhile to consider the application of Krishna et al. approach to Turkey in future research. 28 For a further discussion of endogenous growth models, see Aghion and Howitt (1998). 29 See OECD (2004a, p. 20). This value is obtained using year 2000 PPP (Purchasing Power Parity) prices. 30 We choose this period because data on capital stocks are only available for 1972–1999. For a further discussion of how capital stock data are derived, see the publication of the State Planning Institute (2002). 31 Hsieh derives the dual productivity residual using simple national income identities. Altux and Filiztekin (2002) provide a derivation based on the dual cost function, which yields an expression for the dual productivity residual under nonconstant returns to scale and imperfect competition as pit JK,L,M c Jitp Jit
(1 i) 1 i Yit iAit it
Productivity and growth, 1923–2003
32 33
34
35
36
37 38 39 40 41
42
43
44
57
where it denotes the time-varying markup and c are the cost shares of the inputs. Hsieh’s innovation is to make use of independent measures of the rental rate of capital that do not rely on national income accounts data to quantify TFP growth. They find that labor productivity growth for the period 1972–1979 is 3.16 percent but it falls to 1.94 percent for the period 1980–2000. These results are similar to the ones that we obtained from sectoral data. In their recent review, Oliner and Sichel (2000) argue that investment in ICT technologies is the reason behind enhanced TFP growth in the US during the 1990s. Other authors have provided evidence on the importance of ICT investment with firm-level studies. See, for example, Becchetti et al. (2003). Barro (1990) considers a simple endogenous growth model with infrastructure investment. Rebelo (1991) shows that differences in growth rates across countries may be explained by differences in government policy in a simple AK model that displays endogenous growth. The implications of models that link fiscal policy to growth have been studied by Easterly and Rebelo (1993), who find a positive impact of infrastructure investment on growth. Whereas public fixed core and non-core infra-structural investment grew at rates of 10.8 and 9 percent, respectively, during the period 1963–1979, these rates fell to 2.7 percent and 0.8 percent during 1980–1999, and the growth of real public fixed investment declined from 9.7 to 1.6 percent. Over the same periods, the growth in real private fixed investment declined from 7 to 6.1 percent. It has been suggested that incremental declines in military spending in developing countries typically lead to higher consumption or socially popular programs such as housing without increasing productive investments. The reason may be fixed-size investments or increasing returns/externalities in the generation of productive investment projects. A similar idea is used by Altux et al. (2002) to analyze the decisions of a sovereign borrower that seeks to finance a fixed-size investment project. Moreover, this type of behavior is more likely to arise in the absence of stable or durable political conditions and institutions. Crain and Oakley (1995) provide an empirical investigation of this hypothesis using inter-state data for the US. For a similar set of suggestions, see the OECD (2004b). See Inalcik and Quartert (1994), chapter 32, “Commerce.” See, for example, Barro and Sala-i-Martin (1995). Recently, various authors have argued that the speed of convergence across regions appears to be declining. Adjemian et al. (2000) present a model of integrated economies in which there is convergence during a transition to a steady state and persistence of initial conditions. In this model, the aggregate growth rate of the economy is correlated with the rate of spatial convergence. Carrington (2003) examines convergence across EU countries allowing for neighborhood spillover effects in standard growth regressions, and finds that convergence across regions is much lower than the 2 percent rule. Martin and Sunley (1998) argue that the 2 percent rule itself implies a slow rate of convergence, and examine the implications of endogenous growth theories in explaining regional convergence patterns. The data used in the study are Gross Provincial Product. The data for the years 1987–2000 are from the State Institute of Statistics (SIS), and the data for 1975–1986 are from Özütün (1998). The nominal GPP series are converted to real values using price deflators provided by SIS. Data on a total of 65 provinces are considered. Since the number of provinces has increased throughout the sample period, data on newer provinces are added to those for their former provinces. s s Our measure of specialization is given by: Ui s1(IKDi /IKDi s 1/( J 1)jiIKDj /IKDj), where s denote sector, i denotes province, and IKDsi denote value-added in region i in sector s. j it
58
Sumru Altux and Alpay Filiztekin
45 Another set of issues involves the role of external scale economies (agglomeration economies) on convergence and overall growth. Doxan (2001) uses cross sectional data on output per worker to detect the impact of agglomeration economies in Turkish manufacturing. Filiztekin (2002) examines local scale externalities in employment growth for the Turkish private manufacturing industry. 46 Turkish data are obtained from State Institute of Statistics of Turkey, and are based on national income and product account data. European data are from STAN Database provided by OECD. 47 The EU Accession countries such as the Czech Republic, Hungary, and Poland show high rates of productivity growth during 1991–2002. However, these countries simultaneously display negative rates of employment growth, no doubt as a result of the transition to a market economy. 48 This is accomplished by adding an additional constraint into the deterministic programming problems needed to calculate the various distance functions underlying the Malmquist index. See Färe et al. (1994, p. 75). 49 Notice that these scores do not control for capital deepening. This yields the anomalous situation that Australia and the United Kingdom have scores that are among the lowest in the sample. In our discussion, we focus on countries which have roughly similar labor–capital ratios. 50 Kumar and Russell (2002) and Maudos et al. (2000) construct world production frontiers using output and inputs of labor and capital under the assumption of constant returns to scale. Their results also indicate that technical change is non-neutral, leading to productivity growth at higher capital–labor ratios, and that capital deepening is the primary cause of convergence in the distribution of labor productivity. 51 Broadberry (1998) argues that Germany and the United States overtook Britain in the period 1870–1990 by moving resources out of agriculture and by improving comparative productivity in services. Caselli and Coleman (2002) present a model of structural transformation (the decline of agriculture as the dominating sector) to explain North–South and other regional convergence in the United States. 52 Mauro (1995) examines the relationship between corruption and growth for a cross-section of countries (including Turkey). Taking into account the endogeneity of various corruption indicators, he finds that corruption and lack of bureaucratic efficiency lower investment, thereby also lowering growth. 53 Çetinbav and Vergil (2004) estimate the size of the underground economy to be between 16–31 percent of the official economy for the period 1971–2000. They find that the underground economy causes the official economy but not the opposite, suggesting that measures to reduce the size of the underground economy would lead to growth in the official economy. See also Farrell et al. (2003).
References Ades, A.F. and E.L. Glaeser (1999) “Evidence on Growth, Increasing Returns, and the Extent of the Market,” Quarterly Journal of Economics 114, pp. 1025–1045. Adjemian, S., J. Glachant, and C. Vellutini (2000) “Regional Convergence and Aggregate Growth,” Manuscript. Aghion, P. and P. Howitt (1998) Endogenous Growth Theory. Cambridge, MA: MIT Press. Akdede, S.H. (2002) “Economic Liberalization, Markups, and Total Factor Productivity Growth in Turkey’s Manufacturing Industries,” Manuscript, Aydin Menderes University. Altux, S. (1989) “Time-to-Build and Aggregate Fluctuations: Some New Evidence,” International Economic Review 30, pp. 889–920. Altux, S. and A. Filiztekin (2002) “Scale Effects, Time-varying Markups, and the Cyclical Behavior of Primal and Dual Productivity,” Applied Economics 34, pp. 1687–1702.
Productivity and growth, 1923–2003
59
Altux, S., V. Özler, and M. Usman (2002) “The Role of Lender Behavior in International Project Finance,” Economic Theory 3, pp. 571–598. Atiyas, I. and H. Ersel (1992) “The Impact of Financial Reform: The Turkish Experience,” World Bank, Washington, DC. Bahmani-Oskooee, M. and I. Domac (1995) “Export Growth and Economic Growth in Turkey,” METU Studies in Development, 22, pp. 67–77. Balassa, B. (1978) “Exports and Economic Growth: Further Evidence,” Journal of Development Economics 5, pp. 181–189. Barro, R. (1990) “Government Spending in a Simple Model of Endogenous Growth,” Journal of Political Economy 98, pp. S103–S125. Barro, R. and X. Sala-i-Martin (1992) “Convergence,” Journal of Political Economy 100, pp. 223–251. Bayar, G. (2002) “Effects of Foreign Trade Liberalization on the Productivity of Industrial Sectors in Turkey,” Emerging Markets Finance and Trade 38, pp. 46–71. Becchetti, L., D.A.L. Bedoya and L. Paganetto (2003) “ICT Investment, Productivity, and Efficiency: Evidence at Firm Level using a Stochastic Frontier Approach,” Center for International Studies on Economic Growth Working Paper No. 29. Bernard, Andrew and Charles I. Jones (1996b) “Productivity and Convergence across US States and Industries,” Empirical Economics 21, pp. 113–135. Boratav, K. (1997) “Economic History (1908–1980),” In Sina Akvin (ed.) Modern Turkey, 1908–1980: History of Turkey Vol. 4, Istanbul: Cem Yayinevi. Broadberry, S. (1998) “How Did the United States and Germany Overtake Britain? A Sectoral Analysis of Comparative Productivity Levels, 1870–1990,” Journal of Economic History 58, pp. 375–407. Bulutay, T. (1995) “Employment, Unemployment and Wages in Turkey,” International Labor Office. Bulutay, T., Y. Tezel, and N. Yildirim (1974) Turkey’s National Income (1924–1974) 2 Ankara University Publications. Carrington, A. (2003) “A Divided Europe? Regional Convergence and Neighborhood Spillover Effects,” Kyklos 56, pp. 381–394. Caselli, F. and W.J. Coleman II (2002) “The U.S. Structural Transformation and Regional Convergence: A Reinterpretation,” Journal of Political Economy 109, pp. 584–616. Caves, D.W., L.R. Christensen, and W.E. Diewert (1982) “The Economic Theory of Index Numbers and the Measurement of Input, Output, and Productivity,” Econometrica 50, pp. 1393–1414. Çetinbav, H. and H. Vergil (2004) “Causality between Measured Economy and Underground Economy,” Yapi Kredi Economic Review 15, pp. 21–28. Crain, M.A. and L.K. Oakley (1995) “The Politics of Infrastructure,” Journal of Law and Economics 38, pp. 1–17. Derviv, K., J. De Melo, and S. Robinson (1982) General Equilibrium Models for Development Policy, Cambridge: Cambridge University Press. Diewert, W.E. (1976) “Exact and Superlative Index Numbers,” Journal of Econometrics 4, pp. 115–145. Doxan, E. (2001) “External Scale Economies in Turkish Manufacturing,” International Review of Applied Economics 15, pp. 429–446. Doxanlar, M. and M. Fisunoxlu (1999) “Causality between Exports and Economic Growth in Asian Countries,” Yapi Kredi Economic Review 10, pp. 3–12. Dönek, E. (1998) “Is the Turkish Economy Competitive in the New Technology Goods?,” Yapi Kredi Economic Review 9, pp. 35–54.
60
Sumru Altux and Alpay Filiztekin
Dunne, P., E. Nikolaidou, and D. Vougas (2001) “Defense Spending and Economic Growth: A Causal Analysis for Greece and Turkey,” Defense and Peace Economics 12, pp. 5–26. Easterly, W. and S. Rebelo (1993) “Fiscal Policy and Economic Growth: An Empirical Investigation,” Journal of Monetary Economics 32, pp. 458–493. Edwards, S. (1998) “Openness, Productivity, and Growth: What Do We Really Know?” Economic Journal 108, pp. 383–398. Erzan, R. and A. Fliztekin (1997) “Competitiveness of Turkish SME’s in the Customs Union,” European Economic Review 41, pp. 881–892. Erzan, R. and A. Filiztekin (2003) “Does Size Matter? (in Growth, Productivity, Employment and Vulnerability/Flexibility in Turkish Manufacturing),” Manuscript. Färe, R., S. Grosskopf, M. Norris, and Z. Zhang (1994) “Productivity Growth, Technical Progress, and Efficiency Change in Industrialized Countries,” American Economic Review 84, pp. 66–83. Farrell, D., D. Mean, and D. Baver (2003) “The Hidden Dangers of the Informal Economy,” McKinsey Quarterly (2004), pp. 27–35. Filiztekin, A. (1998) “Convergence across Regions and Provinces,” Koç University Working Paper Series No. 1998/08. Filiztekin, A. (2000) “Openness and Productivity Growth in Turkish Manufacturing,” Sabanci University Discussion Paper Series No. 0104. Filiztekin, A. (2002) “Agglomeration and Growth in Turkey, 1980–1995,” Manuscript, Sabanci University. Filiztekin, A. (2004) “A Multi-sectoral Co-feature Analysis of Fluctuations in Turkey,” Forthcoming, Emerging Markets Finance and Trade. Grossman, G.M. and E. Helpman (1991) Innovation and Growth in the Global Economy, Cambridge, MA: MIT Press. Hatemi-J, A. and M. Irandoust (2000) “Time-series Evidence for Balassa’a Export-led Growth Hypothesis,” Journal of International Trade and Economic Development 9, pp. 355–365. Hsieh, C. (2002) “What Explains the Industrial Revolution in East Asia? Evidence from the Factors Markets,” American Economic Review 92, pp. 502–526. Inalcik, H. and D. Quataert (1994) A Social and Economic History of the Ottoman Empire, 1300–1914, Cambridge: Cambridge University Press. Islam, N. (1995) “Growth Empirics: A Panel Data Approach,” Quarterly Journal of Economics 110, 1127–1170. Ismihan, M., K. Metin-Özcan, and A. Tansel (2002) “The Role of Macroeconomic Instability in Public and Private Capital Accumulation and Growth: The Case of Turkey, 1963–1999,” Applied Economics 37, pp. 239–251. Kim, E. (2000) “Trade Liberalization and Productivity Growth in Korean Manufacturing Industries: Price Protection, Market Power, and Scale Efficiency,” Journal of Development Economics 62, pp. 55–83. Krishna, K., A. Özyildirim, and N.J. Swanson (2003) “Trade, Investment, and Growth: Nexus, Analysis, and Prognosis,” Journal of Development Economics 70, pp. 479–499. Krueger, A.O. (1995) “Partial Adjustment and Growth in the 1980’s in Turkey,” In Reform, Recovery, and Growth: Latin America and the Middle East, National Bureau of Economic Research Project Report Series, Chicago, IL and London: University of Chicago Press, pp. 343–367. Krueger, A.O. and B. Tuncer (1982) “Growth of Total Factor Productivity in Turkish Manufacturing,” Journal of Development Economics 119, pp. 307–325.
Productivity and growth, 1923–2003
61
Krugman, P. (1994) “The Myth of Asia’s Miracle,” Foreign Affairs 73, pp. 62–78. Kumar, S. and R.R. Russell (2002) “Technological Change, Technological Catch-up, and Capital Deepening: Relative Contributions to Growth and Convergence,” American Economic Review 92, pp. 527–548. Kydland F. and E.C. Prescott (1982) “Time-to-Build and Aggregate Fluctuations,” Econometrica 50, pp. 1345–1370. Lains, P. (2002) “Southern European Backwardness Revisited: The Role of Open Economy Forces in Portugal and the Balkans,” Scandinavian Economic History Review 50, pp. 24–43. Lucas, R.E. Jr (1988) “On the Mechanics of Economic Development,” Journal of Monetary Economics 22, pp. 3–42. MacKinnon, R. (1964) “Foreign Exchange Constraints in Economic Development and Efficient Aid Allocation,” Economic Journal 74, pp. 388–409. Mankiw, N.G., D. Romer, and D. Weil (1992) “A Contribution to the Empirics of Economic Growth,” Quarterly Journal of Economics 107, pp. 407–437. Martin, R. and P. Sunley (1998) “Slow Convergence? The New Endogenous Growth Theory and Regional Development,” Economic Geography 74, pp. 201–274. Maudos, J., J.M. Pastor, and L. Serrano (2000) “Convergence in OECD Countries: Technical Change, Efficiency, and Productivity,” Applied Economics 32, pp. 757–765. Mauro, P. (1995) “Corruption and Growth,” Quarterly Journal of Economics 110, pp. 681–712. Miller, S.M. and M.P. Upadhyay (2000) “The Effect of Openness, Trade Orientation, and Human Capital on Total Factor Productivity,” Journal of Development Economics 63, pp. 399–423. Nishimizu, M. and S. Robinson (1984) “Productivity Growth in Manufacturing,” Journal of Development Economics 16, pp. 177–206. OECD (2004a) Economics Surveys: Turkey, October, Paris OECD (2004b) Small and Medium-sizes Enterprises in Turkey: Issues and Policies, October, Paris. Oliner, S.D. and D.E. Sichel (2000) “The Resurgence of Growth in the 1990’s: Is Information Technology the Story?” Journal of Economic Perspectives 14, pp. 3–22. Onaran, Ö. (2002) “Measuring Wage Flexibility: The Case of Turkey Before and After Structural Adjustment,” Applied Economics 34, pp. 767–781. Owen, R. and V. Pamuk (1998) A History of the Economies of the Middle East in the Twentieth Century. London and Cambridge: I.B. Taurus Publishers and Harvard University Press. Özötün, E. (1988) Provincial Distribution of the Gross Domestic Product of Turkey, 1979–1986, (Publication No. 1988/8). Istanbul: Istanbul Chamber of Commerce Research Department. Pamuk, V. (2000) “Intervention during the Great Depression: Another Look at Turkish Experience,” in Vevket Pamuk and Jeffrey Williamson (eds), The Mediterranean Response to Globalization Before 1950, Routledge Explorations in Economic History, London and New York: Routledge, 2000, pp. 321–339. Quah, D. (1996) “Empirics for Economic Growth and Convergence,” European Economic Review 40, pp. 1353–1375. Rebelo, S. (1991) “Long-run Policy Analysis and Long-run Growth,” Journal of Political Economy 99, pp. 500–521. Republic of Turkey Undersecretariat of the Treasury Publication (2003). Treasury Statistics. Romer, P.M. (1986) “Increasing Returns and Long-run Growth,” Journal of Political Economy 94, pp. 1004–1037.
62
Sumru Altux and Alpay Filiztekin
Rosenztein-Rodan, P.N. (1943) “Problems of Industrialization of Eastern and Southeastern Europe,” Economic Journal 53, pp. 202–211. Saygili, V., C. Cihan, and H. Yurtoxlu (2001) “Productivity and Growth in OECD Countries: An Assessment of the Determinants of Productivity,” Yapi Kredi Economic Review 12, pp. 49–64. Schmitz, J.A. (2001) “Government Production of Investment Goods and Aggregate Labor Productivity,” Journal of Monetary Economics 47, pp. 163–187. Sezgin, S. (2000) “Defense Expenditures and Economic Growth in Turkey: A Disaggregated Analysis,” Yapi Kredi Economic Review 11, pp. 67–78. Sezgin, S. (2001) “An Empirical Analysis of Turkey’s Defense–Growth Relationships with a Multi-Equation Model, (1956–1994),” Defense and Peace Economics 12, pp. 69–86. Solow, R. (1958) “Technical Change and the Aggregate Production Function,” Review of Economics and Statistics 39, pp. 312–320. State Institute of Statistics (1994) Gross National Product: Concepts, Methods, and Sources. State Planning Institute (2002) Capital Accumulation, Growth, and Productivity in the Turkish Economy, 1972–2000. Taymaz, E. (2005) “Are Small Firms Really Less Productive? An Analysis of Productivity Differentials and Firm Dynamics,” Small Business Economics 25(5), pp. 429–445. Temel, A. (1998) Sectoral Transformations during the Development Process: Manufacturing Industry and Employment, State Planning Institute. Temel, T., A. Tansel and P.J. Albersen (1999) “Convergence and Spatial Patterns of Labor Productivity: Nonparametric Estimations for Turkey,” Journal of Regional Analysis 29, pp. 3–19. Ventura, J. (1997) “Growth and Interdependence,” Quarterly Journal of Economics 112, pp. 57–84. Voyvoda, E. and I. Yeldan (2001) “Patterns of Productivity Growth and the Wage Cycle in Turkish Manufacturing,” International Review of Applied Economics 15, pp. 375–396. Yildirim, E. (1989) “Total Factor Productivity Growth in Turkish Manufacturing Industry between 1963–1983: An Analysis,” METU Studies in Development 16, pp. 65–96. Young, A. (1995) “The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian Growth Experience,” Quarterly Journal of Economics 110, pp. 641–680. Zaim, O. and F. Tavkin (1997) “The Comparative Performance of the Public Enterprise Sector in Turkey: A Malmquist Productivity Approach,” Journal of Comparative Economics 25, pp. 129–157.
3
Income and consumption inequality in Turkey What role does education play? Burcu Duygan and Nezih Güner
Introduction In recent years, Turkey’s economic preoccupation has shifted from inflation to inequality. As documented in Chapter 2, Turkey is finally enjoying a period of calm after years of macroeconomic instability. Accordingly, the media, the public, and the government are now able to step back from monthly inflation figures and take a longer-term view of economic growth. In doing so, many observers have been surprised and dismayed at the high level of income inequality, which is closer to that seen in the most unequal countries of Latin America and sub-Saharan Africa. It is also unfortunately far from the typical range observed among Turkey’s European neighbors. Nevertheless, it is important to note that over the last 40 years inequality has declined in Turkey, while it has risen in many countries. But does inequality really matter? After all, almost any economist would agree that some inequality is essential to a market economy. If a doctor or an engineer received the same wage as a high school dropout, why would anyone spend all those extra years in the classroom? Why should an entrepreneur invest in a risky project, if her risk is not compensated? Inequality is necessary to provide rewards to those who invest in their skills as well as to those who take risk and responsibility. Hence, it is arguable that inequality is simply the result of a well-functioning market economy. In addition, any attempt at redistribution must be judged against its effects on essential economic incentives that allocate talent and resources. Recent literature, however, shows that the effects of inequality on economic growth and development can be more complicated, and challenges the traditional view of the equity-efficiency trade-off. In their review of the inequality and growth literature, Aghion et al. (1999) conclude that inequality can indeed be detrimental for growth when it is coupled with imperfect credit markets and unequal access to education. The authors argue that inequalities can result in inefficiently low levels of physical and human capital investment and lower aggregate productivity and growth. Consider human capital investment. It is particularly prone to credit market imperfections since, unlike physical capital, it constitutes poor collateral. Credit opportunities for financing education are quite limited if not completely absent in many developing countries. If access to public education is limited, then family income becomes the sole determinant of education.
64
Burcu Duygan and Nezih Güner
This can result in a complete disassociation between children’s abilities and their education. Highly able but poor children may end up with little or no formal education. In such a world, talent – possibly the most important source of economic growth in a country – is misallocated, inequality is transmitted from one generation to the next, and aggregate economic performance worsens overall.1 The main goals of this chapter are threefold: First, we document the extent of income and consumption inequality in Turkey, using primarily the 1994 and 2002 Household Income and Consumption Expenditures Surveys. An analysis of consumption inequality is especially important for understanding whether changes in income distribution accurately capture changes in welfare. Permanent income or consumption is a better measure of the distribution of resources than current income. To our knowledge, this chapter is the first to apply either analysis to Turkey. Second, we look at who is rich and who is poor and document factors that separate rich and poor households. Finally, we try to go beyond the current levels of income inequality and discuss other factors that are critical to the intergenerational transmission of inequality, such as household formation, private and public spending on education, and women’s education. Our goal is to move beyond the current, short-term discussions on inequality, and to understand its persistence. In other words, we want to know what today’s distribution of resources and institutions imply for tomorrow’s inequality. The resulting picture is less than encouraging.
Income and consumption inequality in Turkey The last decade was a period of substantial change in Turkey, as instability of 1990s, which included two financial crises in 1994 and 2000/2001, ended with relative stability in the last couple of years. In this section, we take a look at the distributional dynamics during this period. Our goal is to explore the changes in the distribution of economic resources and household welfare. The analysis is based primarily on the 1994 and 2002 Household Income and Consumption Expenditure Surveys (HICES). A detailed description of these household surveys as well as the definitions of our variables and sample selection criteria are provided in the Appendix.
Income inequality Table 3.1 shows the distribution of income since 1963, looking at the share of income by quintiles and the Gini coefficient. The data for 1963–1987 are from Kasnakoxlu (1997), while the data for 1994–2003 period are based on the 1994 and 2002 HICES, and the 2003 Household Budget Survey. Similarly, Table 3.2 shows the distribution of income in the same period using the ratio of average income by each quintile with respect to the average income of the lowest quintile. In both tables, the measure of income is the total annual household income.
Income and consumption inequality
65
Table 3.1 Distribution of annual total household income: income shares of quintiles and Gini coefficients, 1963–2003 Quintiles
1963
1968
1973
1986
1987
1994
2002
2003
Lowest 20% Next 20% Next 20% Next 20% Highest 20% Gini Coefficient
4.50 8.50 11.50 18.50 57.00 n/a
3.00 7.00 10.00 20.00 60.00 0.56
3.50 8.00 12.50 19.50 56.50 0.51
3.90 8.40 12.60 19.20 55.90 n/a
5.24 9.61 14.08 21.15 49.94 0.44
4.86 8.63 12.91 19.03 54.88 0.49
5.3 9.8 14.0 20.8 50.1 0.44
6.0 10.3 14.5 20.9 48.3 0.42
Source: Data for 1963–1987 are taken from Kasnakoxlu (1997); for 1994, 2002, and 2003 are taken from State Statistical Institute and available at http://www.die.gov.tr/TURKISH/SONIST/ HHGELTUK/hhgeltuk.html
Table 3.2 Distribution of annual total household income: average income by quintiles, 1963–2003 Quintiles
1963
1968
1973
1986
1987
1994
2002
2003
Lowest 20% Next 20% Next 20% Next 20% Highest 20%
1 1.9 2.6 4.1 12.7
1 2.3 2.9 6.7 20
1 2.3 3.6 5.6 16.1
1 2.2 3.2 6.0 14.3
1 1.8 2.7 4.0 9.5
1 1.8 2.6 3.9 11.3
1 1.9 2.6 3.9 9.5
1 1.7 2.6 3.5 8.1
Source: Data for 1963–1987 are taken from Kasnakoxlu (1997) for 1994, 2002 and 2003 are taken from State Statistical Institute and available at http://www.die.gov.tr/TURKISH/SONIST/ HHGELTUK/hhgeltuk.html
These tables illustrate the current degree of income inequality in Turkey very clearly. The share of income accruing to the poorest quintile of the household income distribution was 5.3 percent in 2002 and 6 percent in 2003. In contrast, the share of the richest quintile was between 50.1 percent and 48.3 percent, respectively. The average household income of the households in the top quintile was 9.5 and 8.1 times higher than the average income of the households in the bottom quintile in 2002 and 2003, respectively. Moreover, the average household income rises quite slowly until the fourth income quintile. In 2003, the rise in the average household income was 70 percent from the first to the second income quintile, 53 percent from the second to the third, and 35 percent from the third to the fourth. The jump from the fourth to the last income quintile is, however, much more dramatic: average household income in the top quintile is more than twice the average income in the fourth quintile. These differences are all the more significant if we correct our income measure to capture differences in household sizes, and use annual per capita household income. As seen in Table 3.3, the share of the income accruing to the richest quintile is now 53 percent rather than 50.1 percent, while the share of the poorest groups declines from 5.3 to 4.6 percent. The richer households not only have
66
Burcu Duygan and Nezih Güner Table 3.3 Distribution of annual total household income per capita, 2002 Quintiles
Income share
Income ratio
Lowest 20% Next 20% Next 20% Next 20% Highest 20% Gini coefficient
4.56 9.01 13.55 19.68 53.20 0.49
1.00 1.98 3.05 4.67 12.64
Source: Authors’ calculations from the 2002 HICES.
more resources but also share it among a smaller number of household members. Another eye-opening exercise is analyzing the extreme ends of the distribution. An analysis at the percentile level shows that the annual average per capita incomes of the richest 1 percent of households was 250 times more than the poorest 1 percent in 2002: average annual income of the households at the very top end of the distribution was about $52,000 compared to $240 of the lowest 1 percent.2 A better measure to make individual welfare comparisons is to consider possible economies of scale within the household and use household income per adult equivalent person instead of household income per capita.3 For 2002, the Gini coefficients were not particularly sensitive to the choice of particular equivalence scale we used, ranging from 0.44 to 0.46. Furthermore, we believe that it is more important to consider their effect when studying consumption inequality and consequently provide a more detailed discussion on them in the next section. Moreover, note that experimenting with different equivalence scales changes the level of inequality measures but have very little effect on the trends. The numbers in Tables 3.1 and 3.2 also show that Turkey has had high income inequality for decades. Since 1963, income share of the households in the lowest income quintile has been about 3–6 percent, while the share of the top quintile was at least about 48 percent or higher. Yet, the distribution of household income improved significantly between 1963 and 1987. Gini coefficient declined from 0.56 in 1968 to 0.44 in 1987. The improvement in the income distribution has been mainly due to the decline in income shares of the top 20 percent. While their shares declined, income shares of the other quintiles increased. Between 1963 and 2003, the share of income accruing to the poorest household increased by 75 percent. The distribution since 1987, however, seems to have been fairly stable, with the only exception of a worsening in 1994 – the year of the first main financial crisis in the last decade. This is not surprising given the increased volatility of household incomes following layoffs and increased uncertainty in the economy in 1994. Since 1994, the distribution of income seems not to have changed much, and even improved some between 2002 and 2003 as income share of the poorest groups continued to increase over the last decade. The recent improvement in distribution of income should be considered, however, with some
Income and consumption inequality
67
Table 3.4 Distribution of household per capita income, 1990s Quintiles
Latin America and Caribbean
Sub-Saharan Africa
East Asia and Pacific
South Asia
Eastern Europe
Middle East and North Africa
Developed Turkey Countries 2002
1 and 2 3 and 4 5
13.2 33.8 52.9
14.1 33.5 52.4
18.1 37.5 44.3
21.7 38.4 39.9
22.2 40.0 37.8
17.8 39.8 45.4
18.4 41.8 39.8
15.1 34.8 50.1
Source: All numbers, except for Turkey, are from De Ferranti et al. (2004), Table 2.3, and are for the 1990s.
care. In particular the income data from 1994 and 2002 household surveys only capture 50–60 percent of the national income for these years: the annual per capita income based on the 2002 survey was just over $1,400 compared to a GNP per capita of about $2,600 for the same year.4 Furthermore, 1994 is a poor baseline because there was a financial crisis that year – a period of increased volatility and income inequality. To better interpret these findings, Tables 3.4 and 3.5 present similar statistics from other countries and place Turkey in a broader context. Table 3.4 shows per capita income distribution in Turkey and different parts of the World. In only two regions, Latin America and sub-Saharan Africa, the distribution of per capita income is more unequal than Turkey. A closer look at the income shares in Turkey and in a more equal region, such as within developed countries, also shows that the main difference is the larger share of income received by the top 20 percent. Of course, if the top 20 percent is getting a larger share in Turkey than they do in developed countries, some other groups must be getting less. Surprisingly, it is the middle-income group (households in the third and fourth income quintiles) that gets a lower share, 41.8 versus 34.8 percent. The income share of the bottom 40 percent is quite close at 18.4 versus 15.1 percent. At a more disaggregated level, Table 3.5 portrays the GDP per capita and inequality measures for individual countries. We include two types of countries: some that are widely considered as Turkey’s peers, such as Mexico and Argentina, and a group of current and potential European Union (E.U.) members. The latter group consists of some countries that entered E.U. in 1980s (Greece, Portugal, and Spain), some recent members (Czech Republic, Poland, and Hungary), and some incoming members (Bulgaria and Romania). Income inequality in Turkey is clearly very high when compared to the EU countries. Even when compared to Poland, the most unequal EU country reported in Table 3.5, Turkey fares poorly. If we compare Turkey with the other developing countries in Table 3.5, however, suddenly the level of inequality does not look excessive. With the exception of Ukraine, all other developing countries exhibit more unequal distributions than Turkey, especially those in Latin America.
68
Burcu Duygan and Nezih Güner Table 3.5 International comparisons Country
GDP per capita ($)
Gini Coefficient
Brazil Chile Mexico Argentina Malaysia Philippines Russian Federation Thailand Turkey Poland Portugal Greece Spain Bulgaria Romania Hungary Ukraine Czech Republic
7,480 9,810 8,950 10,920 8,940 4,640 8,920 7,450 6,690 11,450 17,980 19,920 22,020 7,610 7,140 13,780 5,410 15,650
0.59 0.57 0.55 0.52 0.49 0.46a 0.46a 0.43a 0.42 0.37 0.36 0.35 0.33 0.32 0.30a 0.29 0.29 0.26
Source: GDP per capita numbers are for 2003 and taken from the World Development Report (2005). They are measured in purchasing power parity (PPP) terms. Gini coefficients are from the World Development Report (2005) and OECD (2005) and show income inequality, unless otherwise noted. Most recent available numbers are reported. Note a Measure of consumption inequality.
Consumption inequality Although income inequality has remained somewhat stable in Turkey over the last decade, the distribution of current income may not capture the accurate distribution of economic well-being among Turkish households. First, systematic misreporting of income is more common than misreporting of consumption expenditures in household surveys. Consequently, self-reported incomes may not be the most reliable measure for analyzing the welfare distribution. Second, the same change in income inequality may lead to different welfare distributions under different credit market structures that allow households to smooth income fluctuations. Finally, and more importantly, economic theory suggests that if a significant fraction of income variations are caused by changes in the transitory component of income, current income may not be the right measure of lifetime economic resources. For instance, 2002 data suggest that the correlation between monthly household income and monthly consumption expenditures is about 0.52 in Turkey. This figures becomes 0.75 when we use the logarithmic values of these variables to limit the influence of outliers. In either case, the fact that this correlation is far less than 1.0 is another way to see why we look at consumption inequality. For these reasons, many authors have focused on measures of consumption to capture distribution of welfare, instead of measures of income and earnings as
Income and consumption inequality
69
Table 3.6 Distribution of consumption expenditures per person, 1994–2002 Nondurable
Consumption shares (percentages), by quintile Lowest 20% Next 20% Next 20% Next 20% Highest 20% Gini Std Dev. Of Log
Total expenditures
1994
2002
1994
2002
6.11 10.67 15.27 21.87 46.08 0.39 0.71
6.50 11.12 15.41 21.40 45.57 0.40 0.72
5.74 9.96 14.48 21.11 48.70 0.42 0.73
5.44 9.88 15.18 20.26 49.24 0.43 0.75
Source: Authors’ calculations based on the 1994 and 2002 HICES.
discussed in detail in Krueger and Perri (2002).5 In this section, we therefore present a brief overview of the evolution of consumption inequality in Turkey and compare it with the evolution of income distribution to see if there is any “hidden” prosperity or poverty. We continue to use the 1994 and 2002 HICES. Our main measure of consumption is nondurable consumption expenditures. This measure, following Attanasio and Davis (1996), includes spending on food, alcoholic beverages, tobacco, utilities, personal care, transportation services, gasoline and motor oil, apparel, education, health services, and miscellaneous expenditures. Each component of consumption expenditures in 1994 is deflated by its corresponding monthly Consumer Price Index (CPI), since 1994 was a year of significant month-to-month price changes due to the 1994 financial crisis. Results for total monthly consumption expenditures are also reported as an alternative measure. Table 3.6 presents the distribution of consumption expenditures between 1994 and 2002. In particular, it shows the share of consumption expenditures by each quintile along with the Gini coefficient and standard deviation of the logarithm of consumption expenditures – two common measures of inequality. As expected, consumption inequality is lower when compared to income inequality. What’s immediately more interesting from this table is the increase in consumption inequality, regardless of measure used, despite the decrease in income inequality in the same period discussed earlier. We believe this result is driven mainly by two factors, both related to the fact that 1994 data being collected during a crisis period. First, as shown in Duygan (2005), consumption inequality decreased in Turkey in 1994 during the crisis period as a result of falling nondurable spending mainly by the rich: people cut down on caviar and Swedish massages, possibly in response to increased uncertainty caused by the crisis and also taking advantage of high interest rates and saving instead. Second, income inequality was quite higher in 1994 compared to other years as shown in Table 3.1. This was again an impact of the crisis given the increased volatility of incomes during the period and unemployment risk. This suggests that consumption
70
Burcu Duygan and Nezih Güner
inequality was lower than usual in 1994, while income inequality was higher.6 Hence, though surprising, we believe it is mainly because of this crisis effect that we observe declining income inequality and increasing consumption inequality between 1994 and 2002. These results, however, are expressed in per capita terms. An important issue for analyzing the distribution of resources is to account for varying needs caused by differences in family structures. The per capita measure assumes that needs increase linearly with family size and that the needs of children and adults are identical. To capture economies of scale within the family and adjust for varying needs, we use four different equivalence scales: Eurostat, Oxford, and OECD scales discussed in Gürsel et al. (2000) and a scale based on the consumption literature.7 Table 3.7 shows the results from this exercise. The main effect of the choice of alternative equivalent scales seems to be on the level of consumption inequality, but not much on its trend. However, the increase in consumption inequality in the last decade is much less significant when it is computed using total expenditures as opposed to nondurable spending. This result holds for all equivalence scales except in the case of per capita consumption.
What is behind the inequality? In the previous section we documented the extent of income and consumption inequality in Turkey. Why is inequality high in Turkey? This is the question we turn to next. Household income distribution is determined by a complex process. First, people are born with different initial conditions. Some are raised in richer families who have the necessary resources to invest in their children’s human capital formation. Others are born in poorer families who have only limited means. Similarly, some have parents with high human capital and can therefore benefit from their formal education and their connections, while others may not be that fortunate. Second, institutions, such as the education system, play an important role. If high-quality education were freely available to all children, family background would not play such an important role. On the other hand, if education opportunities are limited by individuals’ economic or social background, education can in fact worsen the differences in initial conditions rather than reducing them. Finally, given their initial conditions and the institutional environment, people make educational, professional, and personal choices, all of which impact the distribution of resources in a society. These choices include whether to take a particular job as opposed to another, whom to get married to, how many children to have, and how much to save.
Who are the poor and the rich in Turkey? We start our analysis by taking a closer look at the personal characteristics of the rich and the poor. Table 3.8 compares households at the top and the bottom of the per capita income distribution along four dimensions: the educational attainment
Income and consumption inequality
71
Table 3.7 Distribution of consumption under alternative equivalence scales Quintile consumption shares (percentages)
I. Per capita consumption Bottom 20% Top 20% Gini Std. dev. of log II. Total household consumption Bottom 20% Top 20% Gini Std. dev. of log III. Per adult equivalent consumption a. Eurostat scale (1 0.5 NA 0.3 NC ) Bottom 20% Top 20% Gini Std. dev. of log b. Oxford scale (1 0.7 NA 0.5 NC ) Bottom 20% Top 20% Gini Std. dev. of log c. Cutler and Katz (1992) (NA 0.4 NC )0.5 Bottom 20% Top 20% Gini Std. dev. of log d. OECD scale (N)0.5 Bottom 20% Top 20% Gini Std. dev. of log
Nondurable
Total expenditures
1994
2002
1994
2002
6.11 46.08 0.39 0.71
6.50 45.57 0.40 0.72
5.74 48.70 0.42 0.73
5.44 49.24 0.43 0.75
7.04 42.56 0.35 0.64
6.53 43.12 0.36 0.68
7.01 44.68 0.37 0.64
6.12 45.17 0.37 0.65
7.05 43.01 0.36 0.64
7.21 43.50 0.36 0.65
6.78 45.57 0.38 0.65
6.11 45.77 0.38 0.65
6.89 43.46 0.36 0.65
7.08 43.65 0.37 0.66
6.59 46.05 0.39 0.67
6.01 46.41 0.39 0.67
7.23 42.45 0.35 0.63
7.34 43.19 0.35 0.64
7.03 44.95 0.38 0.64
6.33 45.03 0.37 0.64
7.03 43.04 0.36 0.64
7.34 43.50 0.36 0.66
6.79 45.54 0.38 0.65
6.29 45.78 0.38 0.66
Source: Authors’ calculations based on the 1994 and 2002 HICES expenditures.
of household heads and spouses, household size, labor force participation of spouses, and rural versus urban residency. Education is a major factor that separates rich and poor households. This is not surprising as one of the best-documented facts in economics is the strong link between education and earnings.8 Table 3.8 shows that more than 70 percent of the household heads who are at the bottom of the per capita income distribution have at most primary education, while only 6% of them have a high school degree. The situation for the spouses is even more dramatic: about 95% of them have at most primary education. In contrast, families at the top of the income distribution have more than 50 percent of household heads and 40 percent of their spouses have at least a high school degree. Table 3.8 also shows that
72
Burcu Duygan and Nezih Güner
Table 3.8 Household characteristics of rich and poor (income), 2002 Quintiles Education of household head Below Primary Primary Above Primary, but below High School High School More than High School Education of spouse Below Primary Primary Above Primary, but below High School High School More than High School Household size Number of children Number of own children Labour force participation of spouse Residence Rural Urban
Poorest 20%
Richest 20%
Overall population
30.05 58.24 7.03 4.20 0.47
5.39 35.89 8.54 23.33 26.84
14.92 53.26 9.57 14.21 8.05
48.43 47.70 2.04 1.71 0.13 6.7 3.39 2.57 0.31
10.09 42.81 9.32 22.69 15.09 3.7 1.0 0.9 0.32
28.83 53.16 5.26 9.34 3.41 5.1 2.1 1.6 0.30
45.10 54.90
25.41 74.59
39.98 60.02
Source: Authors’ calculations from the 2002 Turkish HICES.
educational attainments of husbands and their wives are quite similar to each other at both ends of the income distribution. The differences in terms of household size are also quite large. While the households at the bottom of the income distribution have on average more than three children in the household, the rich households have only about one. When we count only own children, the difference is again very significant. Hence, as it is well documented in the literature, the poor not only have lower incomes but also have to share their limited resources among more individuals. These two factors – number of children and education – are not necessarily independent of each other. This is particularly true for the relation between women’s education and fertility. Evidence from both developed and developing countries shows a strong and negative relation between women’s education and fertility.9 There is surprisingly little difference between rich and poor households with respect to female labor force participation – about 30 percent for both groups. This result is in contrast with observations from Latin America, a region with very high income inequality. There, female labor force participation constitutes an important dimension along which poor and rich households differ, together with fertility and education.10 This dissimilarity might be explained by the particularly low labor market participation of married women in Turkey. Bavlevent and Tunalı (Chapter 4 in this volume), however, argue that given the rise in educational attainment in recent years, a larger fraction of women born in the 1990s and beyond will enter the labor market in greater numbers and remain in the
Income and consumption inequality
73
labor market even after marriage. Therefore, it is reasonable to expect that the female labor force participation will be a more important determinant of household income in the future. Although higher labor force participation of married women will benefit all households, it is likely to increase inequality given the similarity between husbands’ and wives’ educational attainments as shown in Table 3.8. As also seen in Table 3.8, 75 percent of the rich households lives in urban areas while about half of the poor do so. We also use our consumption measure to analyze the characteristics of poor and rich households in Turkey. After all, the distribution of consumption is more concentrated than the income distribution and implies lower poverty rates such that the composition of the poor may be quite different under our consumption measure relative to the composition suggested by income. Table 3.9 presents the results of this analysis based on the per capita nondurable expenditures distribution. Results are very comparable to those based on the per capita income distribution, shown in Table 3.8: households tend to be larger, have more children, and be less educated in the bottom 20 percent than in the top. One striking difference, though, is observed for the labor force participation of the spouse. According to the nondurable consumption expenditures distribution, labor force participation is much lower among the spouses on the top 20 percent of the distribution compared to the bottom 20 percent. For the top Table 3.9 Household characteristics of rich and poor (consumption), 2002 Quintiles Education of household head Below Primary Primary Above Primary, but below High School High School More than High School Education of spouse Below Primary Primary Above Primary, but below High School High School More than High School Household size Number of children Number of own children Labour force participation of spouse Residence Rural Urban
Poorest 20%
Richest 20%
Overall population
30.01 57.94 6.28 5.33 0.44
7.18 36.36 8.22 22.61 25.64
14.92 53.26 9.57 14.21 8.05
50.08 45.94 2.50 1.45 0.03 6.91 3.43 2.52 0.34
10.92 43.18 8.68 22.12 15.09 3.54 0.88 0.79 0.25
28.83 53.16 5.26 9.34 3.41 5.1 2.1 1.6 0.30
50.14 49.86
20.83 79.17
39.98 60.02
Source: Authors’ calculations based on the 2002 HICES. Note For all but labor force participation (LFP) and education variables, reported numbers represent the means for the corresponding variables. For LFP and education, sample proportions are shown.
74
Burcu Duygan and Nezih Güner
20 percent of the consumption distribution, 25 percent of the households have spouses participating in the labor force, unlike 34 percent in the bottom 20 percent. On the other hand, the spouses in the households on the top 20 percent of the income distribution have slightly higher labor force participation rates than the bottom 20 percent, which is rather puzzling. The results from Tables 3.8 and 3.9 suggest that education is the key asset that is unequally distributed between the rich and the poor households. Next we look at how valuable this asset is and analyze the effect of education on earnings. If education has a large effect on earnings, then even small differences between the educational attainments can generate large income disparities. The standard model relating education and earnings is based on Mincer (1974). The standard Mincerian earnings equation is given by log(ei) 1si 2(age si 6) 2(age si 6) 2 i, where ei is earnings or wage of person i, si is years of schooling, and (age si 6) is potential experience. The basic idea is that if education and experience increase a worker’s productivity, then earnings should rise with these factors. The parameter 1, which is usually referred to as the returns to schooling or the Mincer coefficient, then represents the percentage increase in earnings associated with one extra year of schooling. The quadratic term in experience is usually added to capture a concave relation between experience and earnings, which would be the case if agents’ post-schooling investment in human capital declines by their age. The first column of Table 3.10 shows the results from this regression carried out using the 2002 HICES. The returns to schooling is quite high: one extra year of schooling increases earnings by on average 13 percent after controlling for experience.11 Another way to capture the link between education and earnings is to enter education into Mincer regression as a categorical variable using dummy variables. This is done in the second column of Table 3.10, with “below primary” as the omitted educational category. These results capture the benefit of each level of educational attainment in a clear way: people with primary education earn about 26 percent more than people whose education is less than primary school, and gains associated with higher levels are even more significant. We use total wage and salary income and not hourly or weekly wages as our dependent variable in Table 3.10. As a result, our estimates reflect the overall effect of education and experience on incomes and do not necessarily provide a good measure of the market price of skills and experience. Tunalı (2003) and Dayıoxlu and Tunalı (2004) provide much more detailed estimates of male and female wage equations for 1988, 1994 and 2002, with hourly wages for urban workers between ages 20 and 54 as dependent variable. They also control for selection bias in their analysis, which we do not attempt here. In general our estimates of the returns to education and experience are higher than their estimates. For 2002, for example, using a specification similar to column (2) in Table 3.10, Tunalı (2003) finds 8.7 percent returns to one extra year of experience, which is about 2 percent points lower than our estimate of 11 percent. Tunalı (2003) and Dayıoxlu and Tunalı (2004) also show that returns to education for males have increased significantly in Turkey during recent decades. Liberalization of Turkish
Income and consumption inequality
75
Table 3.10 Relationship between education and earnings (1) Years of schooling
0.1257*** (0.0047)
Primary Above Primary, but below High School High School More than High School Years of experience Years of experience squared Constant R2 No. of observations
(2)
0.1099*** (0.0092) 0.0018*** (0.0002) 19.3560*** (0.1170) 0.2432 5230
0.2573*** (0.0738) 0.5754*** (0.0829) 0.9263*** (0.0777) 1.5638*** (0.0782) 0.1108*** (0.0092) 0.0018*** (0.0002) 19.7390*** (0.1214) 0.2416 5230
Source: Authors’ calculations based on the 2002 HICES. Notes *** denotes statistical significance at the 1 percent level. Robust standard errors are presented in parentheses. Results are based on a sample of all males between the ages of 20–54 in the 2002 HICES. The age restriction is aimed at capturing the “prime age” male population, a group that is relatively homogeneous. The dependent variable is the logarithm of the annual wage earnings.
economy since the early 1980s has been the key contributor to this rise, as more educated, and in particular those with a university degree or more, benefited greatly from new opportunities that the liberalization has created.12 Few patterns emerge from the results presented in this section. First, education plays an important role explaining income inequality in Turkey, both because it is unequally distributed between the rich and the poor and because it has large effect on earnings. Second, fertility differences between the rich and the poor households are significant and contribute to the per-capita income differences. Finally, educational attainments of husbands and wives are quite similar at both ends of the household income distribution. The well-educated marry each other, while people with less education form households with other less-educated people, without much mixing across groups. In the next section, we turn into the problem of intergenerational persistence of income inequality and present a simple model in which these three factors, education, household formation and fertility determine not only the current level of household income distribution but also determine how inequality is transmitted from one generation to the next. Table 3.11 highlights another aspect of income distribution in Turkey: government transfers. It shows the sources of annual household income for the rich and the poor households. For all households in the urban areas and especially for those
76
Burcu Duygan and Nezih Güner
Table 3.11 Sources of annual household income by per capita income quintiles and geographic area (percentage shares) ALL
Wages and salaries Self employment income Agricultural Asset income Transfers From government Share of retirement benefits
URBAN
RURAL
Lowest 20%
Highest 20%
Lowest 20%
Highest 20%
Lowest 20%
Highest 20%
45.9 30.8 20.3 1.8 21.5 11.0 24.96
39.1 28.9 10.2 9.5 22.5 18.6 47.68
61.1 15.1 2.5 1.4 22.4 11.4 31.75
43.9 21.3 1.6 10.1 24.7 20.3 48.77
28.1 49.2 40.9 2.4 20.4 10.6 17.21
24.6 52.4 36.6 7.5 15.5 13.5 43.42
Source: Authors’ calculations based on the 2002 Turkish HICES.
who are at the bottom of the income distribution, wages and salaries comprise the main source of household income. In rural areas, on the other hand, selfemployment income from agricultural is the main income source. Not surprisingly, asset income is much more important for the rich households than for the poor ones. What is rather surprising about Table 3.11 is the distribution of government transfers. In both urban and rural areas, richer households receive a larger share of their incomes from the government than the poor households do. As we show in Table 3.11, one contributing factor to this pattern is the retirement benefits, which account for almost half of the government transfers accruing to the rich, but only one-quarter to the poor. Although providing a detailed account of how government transfers affect income inequality in Turkey is beyond the scope of this paper, it is worthwhile to point out that government transfers, as it was also documented by Gürsel et al. (2000) for 1987 and 1994, do not play much of a redistributive role in Turkey in contrast to many developed countries. In fact, according to World Bank (2002), the pre-transfer (or market-determined) income inequality in Turkey is quite similar to the pre-transfer income inequality in some other more equal OECD countries. The main difference is that while income inequality is reduced by transfers and progressive taxation in these countries, it is largely unaffected in Turkey. A model of household formation, education, and persistence of income equality In this section we focus on certain aspects of the complex process that lie behind income inequality. In particular, we analyze how formation of households, fertility choices, and educational opportunities of children can interact and affect both the level and the persistence of inequality. As a structural background we first present a model of household formation and inequality by Fernández et al. (2001 and 2005).13 The environment: Consider an economy where there are two types of individuals, either skilled or unskilled. Each individual lives for three periods, during which he
Income and consumption inequality
77
is young, middle-aged, and old. At the start of the first period, young agents make a decision whether to become skilled or not. Becoming skilled is costly. Young agents borrow in order to finance their education, but how much they can borrow is limited by their parents’ resources. After their education decisions are made, young agents meet with others and form households. In the second period, middle-aged agents who have now formed households pay back what they had borrowed for their education, decide how much to consume, and how many children to have. Agents consume only in the last period and the interest rate is set to zero. When middle-aged, agents have children who make their own educations decisions, and when their parents are old they themselves form households and pay back their debts (if any), and so on. Household’s problem: Consider a middle-aged household with income I, and match quality, q, both to be specified later. The household chooses how much to consume, c, and how many children to have, n, to solve the following problem max[c log n q], c,n 0
subject to
c I(1 tn),
where , t 0, and is a constant. Here t represents the fraction of resources that each child consumes. If I , an interior solution to this problem is given by n
tI
and
c I .
As long as the household income is above , an increase in income increases consumption and reduces the number of children. Without loss of generality one can set log t log , and write the value of being in a household with income level I and match quality q (again for the interior case) as V(I,q) I log(I ) q. The labor market: How is income (I) determined? Suppose there is a competitive labor market in which skilled and unskilled workers, Ls and Lu, are hired to produce an aggregate consumption good according to a constant returns to scale production technology: F(Ls, Lu) LF (, 1 ) L u f
1 ,
where Ls/(Ls Lu) is the fraction of skilled labor in the population. Then, the wages for skilled and unskilled labor are given by
1 ,
ws f
and
wu f
1 f 1 1 .
78
Burcu Duygan and Nezih Güner
A low , that is a small proportion of skilled people in the population, increases skilled wages and decreases unskilled wages and results in a higher skill premium. Since education is costly, what matters is the net wages for the skilled people. Let ˜ s w d be the net wage for a skilled person after he/she pays back the cost of w education d. Household formation: After they decide to become skilled or unskilled, agents enter into a marriage market in which they match with other agents. Suppose agents match in two rounds. In the first round, agents match randomly and draw a random match quality q 0. Suppose q is distributed according to cumulative distribution function Q. Since matching is random, if is the fraction of skilled agent, then the fraction of matches between two skilled agents is given by 2, between a skilled and unskilled agent by (1 ), and between two unskilled agents by (1 )2. In the second round of matching, skilled agents only meet with other skilled agents and unskilled agents meet only with other unskilled agents, and draw a new match quality. Consider a skilled agent who is matched with an unskilled agent in the first round, and draws match quality q. Then, this skilled person faces a trade-off: should he/she accept a match with a skilled person and enjoy indirect utility V(Isu,q) or wait for the next round and enjoy V(Iss,q). Given wages, household ˜ s wu and Iss 2w ˜ s.14 incomes in these two options are given by Isu w Obviously the choice of the skilled agent will depend on how the match quality q compares with the income loss due to a match with an unskilled person. Given ˜ s versus wu, there will be a threshold match quality, q*, that the skill premium, w makes a skilled agent indifferent between accepting an unskilled agent in the first round and waiting for a skilled agent in the second round. It is easy to show that this threshold match quality is given by ˜ s() 2w ˜s () wu () log q* w . w˜s ( wu()
Given q*, we can find the fraction of households formed by two skilled denoted by ss, one skilled and one unskilled denoted by su, and finally by two unskilled agents denoted by uu as
2 (1 )Q (q*(t)), if ij ss if ij su
ij () 2(1 )(1 Q (q*())), (1 )2 (1 )Q (q*()), if ij uu. Consider matches between skilled and unskilled agents, that is su. In the first round there are 2(1 ) matches between skilled and unskilled agents. However, only those with high enough q, a fraction (1 Q (q*())) of them, will end up in marriages. Since the threshold quality, q*, is increasing in the skill premium, a high skill premium will result in less mixing between skilled and unskilled agents. In fact, Fernández et al. (2005) show that Q (q*) is the measure of marital sorting, and corresponds to the correlation between skill types in households, and that a higher skill premium, ws/wu, results in more sorting.
Income and consumption inequality
79
Education decisions: How do children make their education decisions in this environment? Children will decide to get education if the value of getting education, V s, is higher than value of remaining unskilled, V u. The function V s represents the benefit of education, both in terms of wages and household matching opportunities, and therefore depends on the future values of . Suppose education has a non-monetary cost, represented by , and this cost is distributed according to a cumulative distribution function . If there are no borrowing constraints, then as long as V s V u, an agent will choose to receive education. In a world without borrowing constraints, a fraction (*) of children will get education (with * V s V u). But education is costly and there are borrowing constraints. Suppose a family with income level I can borrow at most Z (I ), so each child can borrow up to Z(I )/n(I ). If z(I )/n(I ), not all children will be able to get an ˆ of them will, where this fraction is defined by education. Only a fraction () Z(I ) ˆ d. n(I )()
(3.2)
ˆ (*), children will be constrained in Given a household income level I, if () their education decisions. This model provides a nice framework to link different forces that contribute to inequality in a society: First, in a society with high skill premiums there is more marital sorting, and skilled and unskilled agents do not mix. Second, the interaction between inequality and sorting has aggregate consequences due to differential fertility and borrowing constraints. A high degree of marital sorting results in more children being born to poor households. Furthermore, children who grow up in poor families face borrowing constraints that limit their educational opportunities. Finally, education decisions determine the level of skill premium. Indeed, Fernández et al. (2005) show that this model can generate multiple steady states: one steady state that has low skill premium, low sorting and high level of human capital and another one that has high skill premium, high sorting and low level of human capital. Fernández et al. (2005) examine this relationship between marital sorting and the skill premium using household surveys from 34 countries. They compute different measures of the skill premium, and find that all of these measures are positively and significantly correlated with marital sorting, where the latter is measured by the correlation of husband’s and wife’s education. Figure 3.1 is based on their results and shows the relation between the skill premium, measured by the Mincer coefficients, and marital sorting. Using data from the 2002 HICES we also include Turkey, which was not in their original sample.15 There is a positive and significant relation between marital sorting and the skill premium measured by the Mincer coefficients – the correlation is 0.63 across 35 countries and is significant at 1 percent level. Two things emerge from Figure 3.1. First, the skill premium in Turkey is quite high and is only surpassed by few countries, almost all in Latin America.16 Second, the correlation between spouses’ years of schooling is about 0.55 in Turkey, which is below the average of 35 countries in Figure 3.1.17 One interpretation is that there is somehow greater mixing across
80
Burcu Duygan and Nezih Güner
Brazil
Skill premium–Mincer coefficient
0.16
Paraguay Chile Panama Hungary Peru
Colombia
Mexico
0.12
Turkey Bolivia United States Uruguay Costa Rica
0.08 Britain Finland
0.04
Canada Ecuador Luxembourg Czech Republic France Italy Germany Taiwan Belgium Argentina Venezuela Norway Netherlands Spain Israel Poland Slovakia Sweden Denmark
Australia
0.3
0.4
0.5
0.6
0.7
0.8
Marital sorting
Figure 3.1 Marital sorting and skill premium. Source: Fernández et al. (2005) and authors’ calculations for Turkey using 2002 HICES.
educational groups in Turkey than in many Latin American countries, like Brazil, given the income gap between the skilled and the unskilled. Parental resources One implication of Equation (3.2) above is that children who grow up in poor families are constrained for two reasons. First, their families have limited resources. Second, these limited resources are split among a large number of siblings. This mechanism generates intergenerational persistence of income inequality, as children who grow up in poor households end up being poor when they grow up. This lack of mobility from one generation to the other is often viewed as an indication of lack of equal economic opportunity in a society. For the United States, for example, evidence is quite strong that economic status is quite persistent across generations. Persistence coefficient for relative economic status is at least 0.4 or higher.18 Furthermore, most of the transmission can be explained by parental investment in human capital.19 Table 3.12 shows per child expenditure on education by different income quintiles in Turkey. Children in the richest households receive almost 14 times more educational spending than children in the poorest households. A comparison between Tables 3.3 and 3.12 shows that the distribution of educational
Income and consumption inequality
81
Table 3.12 Education expenditure per child by per capita income quintile, 2002, households with children Quintiles
Spending ratio
Lowest 20% Next 20% Next 20% Next 20% Highest 20%
1.00 1.03 2.81 4.21 13.63
Source: Authors’ calculations from the 2002 Turkish HICES.
spending is indeed much more unequal than the distribution of income. While the average per capita income of households in the second quintile is about twice the average per capita income of households at the bottom quintile, perchild educational expenditure is almost identical between these two groups. Women’s education In the earlier analysis the only link between parents and their children is through parents’ economic resources. Although parental education is an important determinant of parental income, it can play an independent and crucial role in human capital formation of children. This might be especially true for early education, which as several recent papers show is fundamental in determining future outcomes.20 Women’s education plays a particularly critical role in the human capital formation of the next generation. According to Schultz (1993 and 2002), a growing body of empirical literature demonstrates that both health and schooling of children are much more closely related to their mother’s education than to their father’s. Recent analyses by Tunalı (1996) and Rankin and Aytaç (2005) on Turkey also document that maternal education is a better predictor of children’s school enrolment than paternal education. Rankin and Aytaç (2005) show that the probability of receiving no education is essentially zero for both girls and boys if their mothers have at least junior high school education.21 These positive externalities are one reason why total ( private plus social) returns to education can be larger for women’s education than men’s education and therefore justify larger spending on women’s schooling. According to Schultz (2002, p. 207): “Evidence from a growing number of countries in all regions of the world demonstrates that increasing investment in women’s human capital, especially education, should be a priority for countries seeking to increase both economic growth and human welfare.” Given this evidence, the status of women’s education in Turkey is quite troubling. According to the 2004 Human Development Report, in 2002 the literacy rate of women aged 15 and above was 78.5% in Turkey. Among countries with
82
Burcu Duygan and Nezih Güner
similar per capita incomes, Turkey has one of the lowest female literacy rates.22 Furthermore, the gap between male and female literacy is quite large in Turkey. This is shown in Figure 3.2. Among 88 countries in this figure, the average gender gap in the literacy rate is about 4 percent, whereas it is 16 percent in Turkey. Despite recent efforts to close this gap, it is still rather large.23 In fact, Turkey’s gender gap in literacy is surpassed by only two countries in this sample with similar per capita GDPs, Tunisia and Libya. Similar gender gaps are also observed in enrolment rates. According to the 2004 Human Development Report, the combined primary, secondary and tertiary enrolment rate was 74 percent for males in contrast to 62 percent for females in 2001/2002. These large disparities, combined with fertility differences in Table 3.8, imply that a large fraction of children grow up in households in which the mother (or the father) lacks the basic skills to help with their education. Public spending on education Parental resources are only one of the factors that determine the human capital investment that children receive. Almost everywhere in the world there is some form of public education. In fact, education policies are often viewed as a fundamental force to decrease existing income inequality from one generation to the next. To see this, one can imagine two worlds: one in which all, including basic education, is completely private and another one in which basic education is
MOR
Male literacy rate–female literacy rate %
EGYPT INDIA LIB TUN
20 ALG SYRIA
OMAN
TURKEY GUAT
SA IRAN
10
LEB PERU JORDAN IND CHINA
SIN
MAUR
BOSNIA
HK
MAL SRI ELSAL
POR
THAIL
0
MEX ECUPAR CRO SWAZ ROM TTSAF BUL NAM ALB VEN ARME CUBA RUSSIA CHIPOL LAT ARG COL DOMBRA CR HON PHI URG
KUW KOR HUN SLOVK
CZH
ISR GR SLOVN
SP NZ
ITL UKFRA GER
NET
UAE
BOTSW JAM
–10
10,000
20,000 GDP per capita, $, 2002
Figure 3.2 Male–female literacy rate difference, ages 15. Source: 2004 Human Development Report.
30,000
USA
Income and consumption inequality
83
publicly provided. In the first world, differences in educational spending, such as those documented in Table 3.12 would simply imply that one’s future is determined primarily by parental backgrounds, while in the other world this would be less true. Table 3.13 shows the share of education in total government expenditure in Turkey. While about 18 percent of all government expenditure was spent on education in 1990, this share plummeted to less than 9 percent in 2001 and recovered only slightly afterwards. As documented by Altux and Filiztekin (Chapter 2 in this volume), the 1990s was pretty much of a lost decade for Turkey, as successive governments struggled to cope with mounting deficits. Unfortunately, a major effect of this struggle has been a significant drop in public spending on education. Figure 3.3 shows some cross country evidence on the relationship between GDP per capita and public spending on education as a percentage of GDP.24 Turkey’s spending on education, which is about 3.7 percent of GDP, is quite low from this cross country perspective as well. Obviously, what matters for inequality is not only the level of public spending on education, but also how it is spent. In a recent study, Restuccia and Urrutia (2004) argue that public spending on early education is much more effective than public spending on college education in reducing intergenerational persistence of income inequality. If children from poor families do not receive adequate levels of early education, they are not able to build the necessary human capital to attend colleges. As a result, even if the college education is provided for free, as long as parental background determines who can go to college, free college education is likely to be ineffective in reducing intergenerational persistence of income inequality. Table 3.14 shows the share of private spending in total education expenditure across countries. The importance of private spending differs significantly between countries. In some countries, such as Malaysia and Norway, public spending plays the dominant role in both early as well as in more advanced stages. In others, such as Chile, private spending is the major component at both levels. In most countries, however, the share of private spending in more advanced education is higher than its share in early education. Average share of private spending is 10.5 percent on early education, while it is 27.6 percent on advanced education. In this table, Turkey stands out as a country in which private spending plays an important role in early education and a limited role in advanced education. Indeed, Turkey has one of the highest shares of private expenditures for early education and one of the lowest for more advanced education. One interpretation of the numbers in Tables 3.13 and 3.14 is that public spending on education has been declining. Private spending has filled the gap in Table 3.13 Share of education in total government expenditure
% share of education
1990
1995
2000
2001
2002
2003
17.94
12.22
10.10
8.71
9.64
10.30
Source: Population and Development Indicators, table 5: Population and Social Development, State Statistical Institute, available at http://nkg.die.gov.tr
84
Burcu Duygan and Nezih Güner
CUBA DEN
8
NAM
MAL SWE
As a percentage of GDP, %
EST
ISR NOR
TUN NZ FIN
JAM
6
BELARUS
LAT SAF
SWAZ MOR
4
POR
BEL FRA
SWZ
USA
POL IRAN
MEX
HUN
CR PAR ARG JORDAN COL PAN OMAN CRO UKR SLOVK MAC INDIA BRA TT SYRIA CHI GABON TURKEY ROM AZER MAUR PERU ARMEPHI RF LEB LIB URG ELSAL DOM
ITL
CAN NET
UKGERAUS CZH
SP
IRE HK
GR KOR
JAP
BOTSW
2 GUAT IND ECU
0 10,000
20,000
30,000
GDP per capita, $, 2002
Figure 3.3 Public expenditure on education as a percentage of GDP. Source: 2004 Human Development Report.
primary and secondary education, while government spending still constitutes the bulk of spending in higher education. To be sure, the share of private spending in primary and secondary education was not always this high; it was just 6 percent in 1995. The transmission of inequality from one generation to the next can be quite high if private spending continues to play a greater role in primary and secondary education. If high quality public education is not available for children from poor households, their early education will be determined by their family’s limited resources (financial as well as non-financial, such as their parents’ education) and ability to pay for education as documented in Table 3.12. Even if higher education is provided free of charge by the government, this will only benefit children who could afford to go to a private primary or secondary school to begin with. It will not significantly improve the equality of opportunities between children of different backgrounds.
Conclusions In this chapter we documented different dimensions of inequality in Turkey. Our main aim was to look beyond aggregate inequality statistics, to highlight the factors that separate the rich from the poor and better understand the intergenerational transmission of income inequality.
Income and consumption inequality
85
Table 3.14 Private education expenditure as a percentage of total education expenditure, 1996–1999 Country
Primary and secondary
Country
Tertiary
Philippines Peru Chile Germany Turkey Korea, Republic of Indonesia Australia Mexico Czech Republic Switzerland Spain Argentina United States Japan Canada Hungary France Israel Uruguay Belgium Netherlands Austria Ireland Denmark Jordan Italy Malaysia Norway Sweden Portugal
40.3 38.2 31.3 24.1 21.8 20.7 18.2 15.9 13.8 12.5 11.9 10.8 10.6 9.2 8.3 8.3 8.0 7.3 7.2 6.9 6.0 5.7 5.2 3.1 2.1 2.0 2.0 2.0 1.0 0.9 0.2
Korea, Rep. of Chile Thailand Japan Indonesia Philippines Peru Vietnam United States Australia Canada Israel United Kingdom Spain Ireland Argentina Italy Jordan Hungary France Czech Republic Netherlands Mexico Croatia Sweden Germany Portugal Norway Turkey Malaysia Denmark Iceland Switzerland Austria
83.3 75.8 67.5 58.3 56.4 55.5 55.4 54.0 53.2 43.9 43.4 40.6 37.3 27.9 27.4 25.7 25.3 24.0 23.4 14.5 14.1 12.5 12.1 12.0 10.7 7.9 7.7 6.0 5.8 4.0 2.8 2.3 1.5 1.1
Source: World Bank Education Statistics, available at http:// devdata.worldbank.org/edustats/td.asp
Any study of inequality is a description of different lives. On the one hand there are households that are at the top quintile of the per capita income distribution. These households are well educated. More than 50 percent of the household heads and about 37 percent of their spouses have at least a high school degree among them. Given the large effects of education on earnings, these high levels of human capital command high labor earnings. These households also have few children, just about one. Quite a different picture emerges when we look
86 Burcu Duygan and Nezih Güner at the households in the bottom quintile of the per capita income distribution. They lack human capital. About 88 percent of such households’ heads and more than 95 percent of their spouses have at most primary education. Furthermore, they have a large number of children, about 2.5. As a result, children who grow up in rich households receive more than 13 times the per capita educational expenditures received by the children of poor households. Furthermore, the children in poor households do not benefit from their parents’, especially from their mothers’, human capital. These differences would not be a big concern if education were available to all children independent of their parental resources. The recent decade, however, witnessed a significant decline in the share of education in total government spending in Turkey. Part of this gap has been filled by private spending in education. Especially, private spending on primary and secondary education has become significant, adding to the initial differences between the children from the poor and the rich households. Turkey is entering into the twenty-first century with a great potential, its young population. Whether and how this potential will materialize and bring a reduction in existing inequalities at the same time will depend critically on education. In particular, it will depend on whether the best and brightest of the country have access to high quality education independent of their parental resources.
Appendix: data description The majority of our analyses and the statistics reported in this chapter are based on the 1994 and 2002 HICES, provided by the SIS of Turkey.25 Both surveys provide detailed information on household consumption expenditures as well as household incomes. As discussed in Gürsel et al. (2000), the survey data quality is “ambiguous” according to the OECD standards – the total disposable income data gathered through the surveys were about 50–60 percent of the national income for the corresponding years. The application methods of both surveys are relatively similar, despite some differences in sampling strategies.26 The 1994 survey was applied monthly to about 2,600 households, changing every month between January 1 and December 31 in 1994, yielding a total sample size of about 26,000 households. The primary income component of the 1994 survey – also known as the Household Income Distribution Survey – was then applied to these same households in the beginning of 1995. In this second stage, however, new households were substituted following attrition by 20 percent of the households from the first round. In other words, the samples in two parts of the survey were not perfectly identical. Although this issue does not affect our results given our data aggregation level, it is worthwhile to note that the numbers reported for income distribution are based on the second round of the survey while the consumption distribution data come from the first round. The 2002 survey, though much smaller in terms of its sample size with about 9,600 households, was also collected in a similar way. It was applied to about 800 households (650 urban, 150 rural), again changing every month, between January 1 and December 31 in 2002.27 In both surveys, households report consumption expenditures for the month of data
Income and consumption inequality
87
collection – the interview month, and income data for the year preceding the interview month. In all our calculations for 2002 we use household weights provided by the SIS. For our overall analysis, we include only those households that report positive income and positive total consumption expenditures for the interview period, yielding a total sample size of over 21,000 households in 1994 and 9,555 households in 2002. The Mincer regression is carried out using only the sample of all men between the ages of 20 and 54, representing the prime age male population. When computing the Mincer coefficient for Figure 3.1, following the sample selection restrictions by Fernández et al. (2005), we restrict our sample to married men between ages 30 and 60. The income definition used in our analysis is the total annual disposable household income, as well as its per capita adjusted version. For the analysis on Mincer regressions, we use the 2002 survey data on total annual labor income, that is only wage earnings. For the analysis on consumption expenditures, two definitions are implemented. The first one, called “nondurable” spending, includes expenditures on nondurable goods and services – food, alcoholic beverages, tobacco, utilities, personal care, transportation services, gasoline and motor oil, apparel, education, health services, and miscellaneous expenditures. Each component of consumption is deflated by its corresponding monthly Consumer Price Index (CPI) for both the 1994 data and 2002. This is especially crucial for 1994, which was a year of significant month-to-month price changes as a result of the financial crisis. The second definition of consumption is the total consumption expenditures, which includes all direct out-of-pocket expenses of the household including durable goods. It is a variable measured on a monthly basis and directly reported in the surveys. Finally, one important note as mentioned briefly earlier is that the 1994 survey was collected around the 1994 financial crisis in Turkey. The crisis, which began as a currency crisis, hit unexpectedly in April 1994. In retrospect, the recovery was quick and the crisis lasted less than one quarter. Thus, this short survey encapsulates data from before, during, and after the entire shock. Consequently, the data are not representative of “regular” times and reflect a period of crisis-adjustment. The implications of this fact considered when we discussed our results earlier.
Acknowledgment We would like to thank Özlem Sarıca for her help with the 1994 and 2002 Turkish Household Income and Expenditures Surveys, and Jesse Bump, Remzi Kaygusuz, and Wnsan Tunalı for helpful comments and suggestions.
Notes 1 Becker and Tomes (1986) and Galor and Zeira (1993) are two early papers that develop the basic ideas discussed earlier. 2 These findings are based on authors’ calculations using the 2002 HICES. Though similar, they are still significantly different from the findings in Sönmez (2001). Using
88
3
4 5 6 7 8 9 10 11
12
13 14 15 16
Burcu Duygan and Nezih Güner data from 2000 income tax returns, he reports that the average monthly income of the households in the top 1 percent in Turkey was as much as $13,000 as opposed to $54 for the poorest 1 percent. If a household with total household income I has N members, then per capita household income is simply I/N. However, when household size increases, from one to two, total income does not need to double in order to achieve the same level of per-capita consumption. As a result, household income per adult equivalent person is calculated as I/E, where E represent the number of adult equivalent persons in a household. Three commonly used scales calculate E as 1 0.5NA 0.3NC (Eurostat scale), 1 0.7NA 0.5NC (Oxford Scale), and N 0.5 (the OECD scale), where N is the total household size, NA is the number of adults in the household and NC is the number of children in the household. GNP per capita data are provided by the State Institute of Statistics (SIS) and are available at: http://nkg.die.gov.tr/en/goster.asp?aile2 (accessed on May 1, 2005). Contributors to this literature include, among others, are Cutler and Katz (1991, 1992), Mayer and Jencks (1993), Deaton and Paxson (1994), Blundell and Preston (1998), and Slesnick (2001). This pattern of decreasing consumption inequality and increasing income inequality during crises is also reported for Korea during the East Asian Crisis, and Indonesia as discussed by Duygan (2005). See discussion in Cutler and Katz (1992). See Ashenfelter and Rouse (2000) for a recent review of empirical literature on the link between schooling and income. See Hotz, Klerman and Willis (1997) and Schultz (1997) for recent reviews of fertility behavior in developed and developing countries, respectively. See Hausman and Székely (1999) for a study of Latin American inequality along these three dimensions. The 2002 HICES data provide only categorical education levels and not actual years of schooling. To compute these, we follow Akçixit and Tunalı (2002). In particular, we assume that it takes 2 years to attain literacy, and respectively 5, 8, and 11 years to complete primary school, middle school, and high school. At the university level, there are 2-year and 4-year programs, so years of schooling for these categories are computed as 13 and 15, respectively. For those who do report a post-graduate degree, we assume years of schooling to be 17. Note that this computation method will underestimate years of schooling for some individuals, such as those who went to college but did not finish it. Years of experience is then computed as age minus years of schooling minus 6 for those with at least a primary school degree, and as age minus 12 for those without any schooling. This method does lead to only a few of the sample individuals having negative experience. In these cases, we set years of experience to zero. Dayıoxlu and Tunalı (2004) also document the gender gap in wages, an important aspect of wage inequality that we do not analyze in this chapter. Despite recent improvements in relative educational attainment of women in Turkey, their results show that gender gap in wages increased between 1988 and 1994. See also Fernández and Rogerson (2001) and Fernández (2002) for models in which family formation affects inequality and macroeconomy, and Fernández (2003) for a recent review of the literature. Note that an interior solution for household problems is guaranteed 2wu . Marital sorting is measured as the correlation between husband’s and wife’s years of education across couples. The sample consists of married couples in which husband is between ages 36 and 45. To be consistent with Fernández et al. (2005), the Mincer coefficient for Turkey is calculated on males between ages 30 and 60. For this age group it is estimated to be 0.12 rather than 0.13 that was reported in Table 3.10. In Figure 3.1, average Mincer coefficient across 35 countries is 0.092 with a standard deviation of 0.034.
Income and consumption inequality
89
17 Average correlation between husbands’ and wives’ years of education across 35 countries is 0.60 with a standard deviation of 0.109. 18 The persistence is usually measured as the coefficient obtained by regressing logarithm of children’s earnings on logarithm of parents’ earnings, with controls for parents’ and children’s ages. See Mulligan (1997) and Solon (2002) for recent reviews of empirical literature. 19 For recent models of intergenerational persistence of income inequality, see Mulligan (1997), Aiyagari et al. (2000), and Restuccia and Urrutia (2004). 20 See, among others, Neal and Johnson (1997), Currie and Thomas (1999), and Carneiro and Heckman (2002). 21 Such positive effects are also present in more advanced countries. See, for example, Currie and Moretti (2003) who found large effects of maternal education on infant health. 22 Among 39 countries with per capita incomes greater than $4,000 and less than $10,000, Turkey (with a per capita income of $6,390) had the lowest female literacy rate (being surpassed by El Salvador, Libya, Iran, Tunisia, Guatemala and Algeria). 23 The gap was 22.4 percent in 1990 with 89.9 percent literacy rate for men versus 67.4 percent for women (see Population and Development indicators, table 3.3: Population and Education, State Statistical institute available at http://nkg.die.gov.tvr (accessed on September 1, 2003). 24 Among 75 countries in Figure 3.3, the average share of public spending in GDP is 4.62 percent with a standard deviation of 1.63. 25 The 2002 Survey is also called the 2002 Household Budget survey by the SIS. 26 See State Institute of Statistics (1997a,b) for a description of the survey methodologies, as well as the SIS News Bulletin (11/6/2003) available at: http://www.die.gov.tr/ TURKISH/SONIST/HHGELTUK/071103.htm 27 One main reason for the reduction in sample size is that, SIS have actually started collecting these Household Budget Surveys at an annual basis starting with 2002. At the time of the writing of this chapter, 2003 data were recently released and 2004 data were still at the cleaning stage. As of 2004, SIS plans to provide the “Household Income and Living Standards Survey” for studying distribution of income, and the household budget surveys to include mainly consumption expenditure data.
References Aghion, P., E. Caroli, and C. García-Peñalosa (1999) “Inequality and Economic Growth: The Perspective of the New Growth Theories,” Journal of Economic Literature, 37: 1615–1660. Aiyagari, R., J. Greenwood, and N. Guner (2000) “On the State of the Union,” Journal of Political Economy, 108: 213–244. Akçixit, U. and W. Tunalı (2002) “Average Years of Schooling in Turkey: Late 80’s vs. Late 90’s: An Examination Based on Household Labor Force Survey Data.” Manuscript. Koç University. Ashenfelter, O. and C. Rouse (2000) “Schooling, Intelligence, and Income in America,” in Meritocracy and Economic Inequality, K. Arrow, S. Bowles, and S. Durlauf (eds). Princeton, NJ: Princeton University Press. Attanasio, O.P. and S. Davis (1996) “Relative Wage Movements and the Distribution of Consumption,” Journal of Political Economy, 104: 1227–1262. Becker, G.S. and N. Tomes (1986) “Human Capital and the Rise and Fall of Families,” Journal of Labor Economics, 4: S1–S39. Blundell, R. and I. Preston (1998) “Consumption Inequality and Income Uncertainty,” Quarterly Journal of Economics, 113: 603–640. Carneiro, P. and J. Heckman (2002) “The Evidence on Credit Constraints in PostSecondary Schooling,” Economic Journal, 112: 705034.
90
Burcu Duygan and Nezih Güner
Currie, J. and D. Thomas (1999) “Early Test Scores, Socioeconomic Status and Future Outcomes,” National Bureaus of Economic Research Working Paper #6943. Currie, J. and E. Moretti (2003) “Mother’s Education and the Intergenerational Transmission of Human Capital: Evidence from College Openings,” Quarterly Journal of Economics, 118: 1495–1532. Cutler, D. and L. Katz (1991) “Macroeconomic Performance and the Disadvantaged,” Brookings Papers on Economic Activity, 1–74. Cutler, D. and L. Katz (1992) “Rising Inequality? Changes in the Distribution of Income and Consumption in the 1980’s,” American Economic Review, 82: 546–551. Dayıoxlu, M. and W. Tunalı (2004) “Falling Behind While Catching Up: Changes in the Female-Male Wage Differential in Urban Turkey, 1988 to 1994,” Manuscript. Koç University, Istanbul, Turkey. Deaton, A. and C. Paxson (1994) “Intertemporal Choice and Inequality,” Journal of Political Economy, 102: 437–467. De Ferranti, D., G.E. Perry, F. Ferreira, and M. Walton (2004) Inequality in Latin America: Breaking with History? Washington, DC: World Bank. Duygan, B. (2005) “What is the Welfare Cost of Financial Crises When Risk-Sharing is Imperfect?,” EUI, Finance and Consumption Working Paper. Fernández, R. (2002) “Education, Segregation, and Marital Sorting: Theory and an Application to UK,” European Economic Review, 46: 993–1022. Fernández, R. (2003) “Household Formation, Inequality and Macroeconomy,” Journal of European Economic Association, 1: 663–697. Fernández, R. and R. Rogerson (2002) “Sorting and Long-Run Inequality,” Quarterly Journal of Economics, 116: 1305–1341. Fernández, R., N. Guner and J. Knowles (2001) “Love and Money: A Theoretical and Empirical Analysis of Household Sorting and Inequality,” NBER Working Paper no. 8580. Fernández, R., N. Guner and J. Knowles (2005) “Love and Money: A Theoretical and Empirical Analysis of Household Sorting and Inequality,” Quarterly Journal of Economics, 120: 273–344. Galor, O. and J. Zeira (1993) “Income Distribution and Macroeconomics,” Review of Economic Studies, 60: 35–52. Gürsel, S., H. Levent, R. Selim, and Ö. Sarıca (2000) Individual Income Distribution in Turkey. Istanbul, Turkey: Turkish Industrialists’ and Businessmen’s Association (TUSIAD). Hausmann, R. and M. Székely (1999) “Inequality and the Family in Latin America.” Inter-American Development Bank, Working Paper no. 393. Hotz, V.J., A. Klerman and R. Willis (1997) “The Economics of Fertility in Developed Countries” in Handbook of Population and Family Economics, M.R. Rosenzweig and O. Stark (eds). Amsterdam, Netherlands: Elsevier. Kasnakoxlu, Z. (1997) “Income Distribution in Turkey: Who Gets What?” Private View, Autum. Istanbul, Turkey: Turkish Industrialists’ and Businessmen’s Association (TUSIAD). Krueger, D. and F. Perri (2002) “Does Income Inequality Lead to Consumption Inequality” NBER Working Paper no. 9202. Mayer, S. and C. Jencks (1993) “Recent Trends in Economic Inequality in the United States: Income versus Expenditures versus Well-Being,” in Poverty and Prosperity in the USA in the Late Twentieth Century, D. Papadimitriou and E. Wol (eds). New York: St. Martin’s Press. Mincer, J. (1974) Schooling, Experience, and Earnings. New York: National Bureau of Economic Research. Mulligan, C. (1997) Parental Priorities and Economic Inequality. Chicago, IL: University of Chicago Press.
Income and consumption inequality
91
Neal, D. and W.R. Johnson (1997) “The Role of Pre-Market Factors in Black–White Wage Differences,” Journal of Political Economy, 104: 869–95. OECD (2005) Society at a Glance: OECD Social Indicators. Paris, France: OECD Publications. Rankin, B.H. and I.A. Aytaç (2005) “Gender Inequality in Schooling: The Case of Turkey,” Manuscript. Koç University, Istanbul, Turkey. Restuccia, D. and C. Urrutia (2004) “Intergenerational Persistence of Earnings The role of Early and college Education,” American Economic Review, 94: 1354–1378. Schultz, T.P. (1993) “Returns to Women’s Education,” in Women’s Education in Developing Countries: Barriers, benefits, and Policies, E.M. King and M.A. Hill (eds). Baltimore, MD: Johns Hopkins University Press. Schultz, T.P. (1997) “The Economics of Fertility in Low Income Countries,” Handbook of Population and Family Economics, M.R. Rosenzweig and O. Stark (eds). Amsterdam, Netherlands: Elsevier. Schultz, T.P. (2002) “Why Governments Should Invest in More to Educate Girls,” World Development, 30: 207–225. Slesnick, D. (2001) Consumption and Social Welfare: Living Standards and Their Distribution in the United States. London, UK: Cambridge University Press. Solon, G. (2002) “Cross-Country Differences in Intergenerational Earnings Mobility,” Journal of Economic Perspectives, 16: 59–66. Sönmez, M. (2001) Gelir Uçurumu: Türkiye’de Gelirin Adaletsiz Bölüvümü. Istanbul: Om Yayinevi, Ekonomi Politik Dizisi. State Institute of Statistics (1997a) Results of The Household Income Distribution Survey 1994. Ankara, Turkey: State Institute of Statistics. State Institute of Statistics (1997b) Household Consumption Expenditure Results 1994. Ankara, Turkey: State Institute of Statistics. Tunalı, W. (1996) “Education and Work: Experiences of 6–14 Year Old Children in Turkey,” Education and the labor Market in Turkey: Proceedings of a Seminar Held in Ankara, T. Bulutay (ed.). Ankara, Turkey: State Institute of Statistics. Tunalı, W. (2003) “Comparison of Wage Equation Estimates: 1988 HLFS, 1994 IDS, 2002 HBS.” Mimeo. Koç University, Istanbul, Turkey. United Nations Development Programme (2005) Human Development Report 2004. New York: Oxford University Press for the United Nations Development Programme. World Bank (2002) Turkey: Poverty and Coping After Crises. World Bank Report no. 24185, Human Development Unit, Europe and Central Asia Region. Washington, DC: World Bank. World Bank (2005) 2005 World Development Report. Washington, DC: World Bank.
4
Female labor supply in Turkey Wnsan Tunalı and Cem Bavlevent
Introduction Examination of the historical record from western industrialized countries, as well as cross-country data, shows that female labor force participation rates initially decrease, and subsequently increase as a function of indicators of the level of development, such as average levels of education and per capita income (Psacharopoulos and Tzannatos, 1989; Goldin, 1995; Mammen and Paxson, 2000; Tansel, 2002). Taken at face value, the secular decline in female labor force participation observed in Turkey since the 1950s fits this model. However why female labor force participation rates in urban areas have remained under 20 percent for the last 15 years, despite favorable initial conditions, and favorable trends in education and fertility, is harder to explain. In particular marriage continues to bring about separation from the labor market. Starting with the early days of the Republic, legislation in Turkey paved the way for inclusion of women in the public sphere and women’s representation in white collar jobs has been underscored as one of the early achievements of the new regime (Kazgan, 1981; Öncü, 1981). Three generations later, the fact that only a small fraction of the women get to collect the market returns on their investments at a time when the demand side of the urban labor market seems to have become more receptive to women, is only one reason why one may be interested to know more about female labor supply in Turkey. The aim of this chapter is twofold. One purpose is to document the relevant developments in the labor market between 1988 and 2003, when significant structural changes took place in the Turkish economy. The second is a thorough empirical examination of the labor supply decision of married women. In fulfilling the first objective, we discuss in turn, developments on the demand side, the relevant demographic changes, and patterns on the supply side. In our empirical examination, we investigate both the extensive and intensive margins of the labor supply decision. Since women’s work can assume different forms over the course of development and the life cycle of a given individual, we estimate a multi-category model of the labor force participation. We find that the distinction between self-employment and wage labor is an important one. We then report the first set of labor supply elasticities estimated for married women in Turkey. Finally, we
Female labor supply in Turkey 93 estimate a structural version of the participation model and establish that small changes in wages elicit very large participation responses from married women. There are practical reasons for focusing on the 15-year period between 1988 and 2003. Since our econometric analysis is based on the first nationwide Household Labor Force Survey (HLFS) conducted in 1988, it makes sense to start with that year. At the time of writing, 2003 was the last year for which HLFS data were available, hence the end point.1 In establishing labor market trends, we mostly focus on the decade between 1988 and 1998, during which time Turkey became increasingly more integrated with the World economy. We choose 1998 as an alternate end point of our analysis mainly, because the two major earthquakes that struck the economic heartland of the country in 1999 and the financial crises in 2000 and 2001 left an indelible mark on the labor market. We contend that the patterns that emerged after the 2001 crisis are better understood when contrasted with those over the 1988–1998 period.
Labor market trends, 1988–2003 This section is devoted to an examination of the stylized facts that characterize the period under examination. We devote special attention to the course of compositional changes in employment. We also highlight demographic trends that have important labor market implications. Finally we turn to labor supply and examine in turn participation patterns and weekly hours of work. Although females constitute our subject, we often talk about males as well, in an effort to underscore the tell-tale signs of the changes taking place. Employment trends The “jobless growth” feature of the post 2001 recovery has received much attention. However, examination of the 15-year record reveals that Turkey’s employment growth has been weak even over the 1988–1998 period, when the economy grew at an average annual rate of 4.5 percent. This moderately strong expansion record is muted by sizable fluctuations around the mean. Although the ups and downs cannot be picked up by the commonly used labor market indicators, the variability in the growth rate is likely to have dampened employment growth. In the 10 years between 1988 and 1998, total employment (age 12 and over) in Turkey grew at the rate of 1.5 percent per annum, from about 19 million to about 22 million. In urban areas (locations with population of 20,000 or more) the growth rate was more than double this figure, approaching 3.2 percent per annum. Consequently the share of the urban component of the total workforce increased from 38 to 45 percent. During this time the number of individuals between the ages of 20 and 54 grew by more than 3 percent per annum. The faster employment growth recorded in urban areas was barely sufficient for absorbing the prime-age population into the workforce. Female employment also grew but at half the total annual rate. Consequently the share of females in total employment declined from 31 to 29 percent. This
94
Wnsan Tunalı and Cem Bavlevent
drop conceals significant compositional changes that accompanied urbanization. The share of females working in urban areas increased from 19 to 27 percent of total female employment, and the share of females among workers located in urban areas increased from 15 to 18 percent. The urban female workforce registered an average annual growth rate of 4.8 percent per annum, while the figure for the male workforce was 2.9 percent. The employment figures given in Table 4.1 for 1988, 1998, and 2003 are useful for drawing comparisons between the sub-period 1998–2003 and the sub-period 1988–1998 we have been focusing on. Note that the data in Table 4.1 are for individuals aged 15 and over.2 The toll of the 2001 crisis on the labor market is reflected in the negative figures that show up under the columns for 1998–2003. Remarkably average aggregate employment growth between 1998 and 2003 has been negative 0.6 percent per annum. The data in Table 4.1 also capture the compositional changes. The dramatic reduction in the role of Agriculture (including hunting, fishings, and forestry) is a prominent trend. Over the 15-year period under examination, there was a net loss of 1 million jobs in Agriculture.3 Between 1988 and 2003, the share of Agriculture in male employment declined from about 34 to 24 percent. In the case of females, the reduction was more dramatic, from around 77 to 59 percent. Note that in 1988, 51.1 percent of the (civilian non-institutionalized) population aged 15 and over was residing in urban areas. By 2003 this figure was 61.3 percent.4 Urbanization was accompanied by rural– urban migration. Since the Construction sector relies on unskilled labor, it serves an important function by introducing young unskilled male migrants to the urban labor market. Remarkably the significant employment growth in Construction that took place between 1988 and 1998 was wiped out between 1998 and 2003. Predictably urbanization was accompanied by above-average employment growth in Services and Manufacturing (including mining and utilities). Between 1988 and 2003, the share of Services in employment went up from 32 to 43 percent. The share of Manufacturing also went up, from 16 to 18 percent. In 15 years 3.5 million new jobs were created in Services (3.2 percent growth per annum), and 1 million in Manufacturing (2.1 percent growth per annum). In both instances, growth rates of employment were substantially higher for females (5.4 versus 2.8 percent in Services, 3.6 versus 1.8 percent in Manufacturing). It has been noted that worldwide women employed in the Manufacturing sector typically hold low-skill, low-paying jobs, in labor-intensive, small-scale plants, and they tend to be concentrated in export-oriented sectors. The evolution in Turkey appears to fit the pattern which has been dubbed “feminization of employment” in the literature that examines labor market effects of structural adjustment (Çaxatay and Berik, 1990; Çaxatay and Özler, 1995). Going beyond the stylized facts, an important question is whether trade liberalization and export orientation had an impact on employment growth. Recent research sheds some light on this issue. Based on evidence from aggregate time series (Erlat, 2000), as well as 4-digit (Taymaz, 1999) and plant level data (Krishna et al., 2001) from the Manufacturing sector, it has not been possible to detect a link between trade growth and aggregate employment growth. The next
Source: HLFS database, SIS web site.
Male Agriculture, forestry, hunting, and fishing Manufacturing, mining, and utilities (electricity, gas, water) Construction Services Total Female Agriculture, forestry, hunting, and fishing Manufacturing, mining, and utilities (electricity, gas, water) Construction Services Total All Agriculture, forestry, hunting, and fishing Manufacturing, mining, and utilities (electricity, gas, water) Construction Services Total 4,909 3,082 1,296 6,395 15,682 4,131 638 29 1,297 6,095 9,040 3,720 1,325 7,692 21,777
4,231 2,361 998 4,934 12,524 4,020 449 15 750 5,234 8,251 2,810 1,013 5,684 17,758
965 9,168 21,147
7,166 3,848
28 1,652 5,888
3,446 762
937 7,516 15,259
3,720 3,086
30.8 35.3 22.6
9.6 32.4
93.3 72.9 16.5
2.8 42.1
29.9 29.6 25.2
16.0 30.5
12.1 30.7 6.1 52.3 21.8 14.3 69.7 86.7 120.3 12.5 13.1 36.9 4.7 61.3 19.1
27.7 17.5 2.7 16.6 19.4 3.4 27.4 3.4 20.7 3.4 27.2 19.2 2.9
1988– 2003
24.2 0.1
1998– 2003
1988– 1998
2003
1988
1998
Change (percent)
Employment (thousands)
Table 4.1 Total employment (age 15 and over) by sector and sex, 1988, 1998, and 2003
2.7 3.1 2.1
0.9 2.8
6.8 5.6 1.5
0.3 3.6
2.6 2.6 2.3
1.5 2.7
1988– 1998
6.1 3.6 0.6
4.5 0.7
0.7 5.0 0.7
3.6 3.6
6.3 3.3 0.5
5.4 0.0
1998– 2003
Growth rate (percent)
0.3 3.2 1.2
0.9 2.1
4.2 5.4 0.8
1.0 3.6
0.4 2.8 1.3
0.9 1.8
1988– 2003
96
Wnsan Tunalı and Cem Bavlevent
question is whether export-orientation brought about composition effects, in the form of “feminization of manufacturing employment.” Using plant level data from the period 1983–1985, Özler (2000b) finds evidence in favor of the feminization phenemenon: female share in total employment increases as export orientation increases (as measured by the share of exports in total output of the manufacturing sub-sector the firm operates in). Using micro data from two rounds (October 1988 and 1994) of the HLFS and province level measures of employment shares of export oriented sectors, Bavlevent and Onaran (2004) detect a small favorable effect of export orientation on female employment in the case of single and/or younger women but not for married women. It is well known that the productive roles that women shoulder change during the course of development (Boserup, 1970; Schultz, 1990). In the period under examination, predictable changes were also seen in the composition of the female workforce in terms of employment status. It is useful to start with the 1988–1998 subperiod and focus on urban areas. During this decade among employed females aged 15 and over residing in urban areas, the share of regular and casual wage and salary workers increased from 68.4 to 77.3 percent. Share of own-account workers decreased from 10.8 to 6.3 percent; share of unpaid work in a family enterprise decreased from 13.4 to 8.6 percent. In terms of absolute numbers, the number of women working on their own account remained the same (117,000), while unpaid family workers increased (from 145,000 in 1988, to 160,000 in 1998). During this time the number of wage and salary workers nearly doubled, from 740,000 to more than 1.4 million. Number of female employers increased fourfold and accounted for 2.3 percent of the urban female workforce in 1998. Number of casual (seasonal, occasional, or temporary, as defined by the SIS) female workers increased from 70,000 to 160,000, but their share in the urban female workforce decreased. In 1998 the share of wage labor (regular and casual wage workers combined) among urban female employment reached its historical peak of 83 percent. The two earthquakes that hit the industrial heartland of Turkey in August and November 1999 appear to have ushered in a shift from regular wage and salary employment to other forms of employment. Predictably the crises in 2000 and 2001 reduced their share further. The data from 2003 suggest that the trends which were broken in 1999 could be reemerging. In 2003 the share of regular wage and salary workers among urban females was 70.7 percent and the share of casual workers was 9.4 percent, for a combined total of 80.1 percent. Privatization has been an important agenda item in the structural reforms introduced back in 1980. As Ökten (2005) argues, the coalition governments have been less than eager in pursuing policy measures that reduce the role of the State. Nonetheless, a decrease in public sector employment opportunities can be observed during the period under examination. The reduction of the share of the public workforce has been implemented partly by the privatization of the Stateowned Economic Enterprises (SEE) and partly by limiting the hiring of new government employees and encouraging the retirement of existing ones at an early age. HLFS data provide an accurate picture of the public sector starting with 1996.5
Female labor supply in Turkey 97 According to these data, 24.7 percent of the workforce engaged outside private agriculture was employed by the public sector in 1996. This figure was 23.7 in 1998 and 22.8 in 2003. Historically females have especially been visible in the government bureaucracy and public services (such as health and education). According to the HLFS, over the 1996–2003 period the share of females in the public sector fluctuated between 20.6–22.7 percent, while the share of females in total non-agricultural employment fluctuated between 14–18 percent. During this time highest female shares (in the 23.5–25.5 range) were observed among public sector workers in Services. Examination of the 2000–2001 rounds of the HLFS reveals that the public sector continued to hire younger workers (ages 15–39 in the case of males, 15–34 in the case of females) during the crisis period and that females were favored (Tunalı et al., 2003: 28–30). Although public sector employment opportunities continue to be important for females, there is a clear trend for privatization of female employment. In 1998, 34.8 percent of the female workforce engaged outside private agriculture was employed by the public sector. In 2003, the public share was 29 percent. To recapitulate, over the 15-year period under examination, the labor market in Turkey has been transforming in predictable ways, as a consequence of the liberalization and privatization of the economy. As such, the labor market has become more accommodating of women. In the next subsection, we argue that demographic changes have also been favorable, in the sense of removing some of the constraints which have been responsible for low labor market attachment of women. Demographic trends Population According to the 2000 General Census, Turkey’s population stood at near 68 million. Nearly two-thirds of the population was residing in urban locations with 20,000 or more inhabitants. Between 1990 and 2000, the population growth rate was 2.68 percent in urban areas and only 0.42 percent in rural areas.6 The primary factor behind the large difference between the two rates is rural-tourban migration. The figures from the 2000 census confirm the conclusions drawn in Shorter (1995) that Turkey is in the last stage of the demographic transition, a period characterized by a definite and irreversible decline in the rate of population growth. Based on the assumption that replacement levels of fertility will be reached around 2005–2010, the study predicted that the population would stabilize somewhere between 95 and 98 million by the middle of the twenty-first century. Examination of the likely evolution of the age structure of the population shows a dramatic rise (from 44 to 52 percent by 2010) in the share of individuals between the ages of 20 and 54. The equally dramatic fall (from 35 to 26 percent) in the share of the youth (0–14), combined with the increasing (from 4 to 6 percent) but still small share of the elderly (65 and over), point at a favorable dependency profile.7 Education Up until 1997, the Turkish educational system hinged on 5 years of publicly funded elementary education, after which significant attrition set in.
98
Wnsan Tunalı and Cem Bavlevent
Taking the non-institutional civilian population aged 12 and over as our base, in 1988, 1 out of 3 females, and 1 out of 10 males were illiterate. Ten years later, the ratios were 1 in 5, and 1 in 20. Among females residing in urban areas, the illiteracy rate went down from 24.2 to 16.6 percent. In 1988 urban males on average had 6.24 years of education, while urban females had 4.64 years. In 1998, the averages were respectively 7.36 and 5.84 years.8 During this time, the share among urban males of those with high school education, or higher, went up from 19.3 to 29.3 percent. Among urban females this share increased from 12.1 to 20.9 percent. Thus the population of females we study in our microeconometric investigation has low levels of human capital (as measured by formal schooling), but significant improvements have taken place in 10 years. In 1997 publicly funded compulsory basic education was extended to 8 years of schooling. The education reform has had a major impact on the course of human capital accumulation in Turkey.9 In particular enrolment rates increased and the gender gap decreased. The ratio of basic education graduates who proceed with secondary education increased to over 90 percent in 2001. While the ratio of girls among those who graduated in 2000 was 40.7 percent, their share increased to 47 percent among new enrolments (Akpınar and Ercan, 2002). Although girls remain far from parity in secondary schools and above, trends signal the mitigation of the societal obstacles detrimental to the education of girls. As we discuss later, each level of completed schooling brings women closer to the labor market. All indications are that women from the post 1980s generation will be entering the labor markets in significantly larger numbers and with better human capital endowments. Marriage and fertility Marriage is almost universal in Turkey. Data from the 1993 Turkish Demographic and Health Survey (TDHS) reveal that only 1 percent of the women in the 1944–1949 birth cohort remained single. There is evidence that the distribution of age at first marriage is shifting to the right. The median age at marriage went up from 18 for the 1944–1949 and 1950–1959 cohorts to 19 in the 1960–1969 cohort and 22 in the 1970–1981 cohort. By age 30, 97.5 percent of the women in the 1944–1949 cohort, 96 percent of those in the 1950–1959 cohort, and 91 percent of those in the 1960–1969 cohort were married (own calculations, based on age-corrected data from 1993 TDHS). The delayed entry into marriage has been accompanied by reductions in fertility. Based on surveys done 10 years apart (1988 and 1998 TDHS), the total fertility rate (TFR, for ages 15–49) declined from 3.0 to 2.6. In 1998 TFR was 3.1 in rural areas and 2.4 in urban areas. The Western region had the lowest average (2.03), and the Eastern region had the highest (4.19). When broken down by education, the figures are 3.89 for women without any formal schooling, 2.55 for those with primary ( 8 years) education, and 1.61 for those with higher (8 years or more) education. Marriage and fertility are known to have significant labor supply implications. There is reason to believe that for years to come, a very large majority of women in Turkey will spend a considerable chunk of their potential working lives in the married state. However these women will be having fewer children.
Female labor supply in Turkey 99 Participation trends Data compiled by Bulutay (1995) show that the secular decline in the labor force participation can be traced back to the 1950s when Turkey began its transition from an agricultural economy to an industrial one. According to data from the General Census, between 1955 and 1990, the female labor force participation rate (for individuals aged 12 or over) declined from about 70 percent to about 40 percent. During the same period, the percentage of the population residing in urban areas went up from around 25 to 60 percent. Figure 4.1 shows the recent trends in participation, based on the HLFS. Participation rates for males residing in rural areas have been the highest, and are followed closely by males in urban areas. Female participation rates are quite a bit lower, and the rural–urban gap is strikingly large. Gender differences are attributable to the domination of the traditional sexual division of labor, which assigns males the responsibility for providing income for the entire household and females the responsibility of home production. Rural–urban differences are attributable to the primacy of agricultural production in rural areas. As Anker (1990) argues, it is a lot easier for members to qualify as a participant when households engage in agriculture. The fact that Turkey’s population is predominantly Muslim, and that traditional values trust women with the responsibility of getting married, bearing children, and engaging in home production, does not by itself provide an explanation for the historical decline of the female participation rate. Arguably a more fundamental force has been rural–urban migration, which breaks the continuity between home production and market production present in the agricultural household setting (Tunalı, 1997). Migration to cities where paid employment 90.0% 80.0% 70.0%
LFPR
60.0% 50.0% 40.0% 30.0% 20.0% 10.0% 0.0% 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 Urban-M
Urban-F
Rural-M
Rural-F
Total
Figure 4.1 Labor force participation rates by location and sex, 1988–2003. Source: HLFS database, SIS website.
100
Wnsan Tunalı and Cem Bavlevent
Table 4.2 Participation by marital status, females aged 15 and over residing in urban Turkey, 1988, 1998, and 2003 1988
1998
2003
Share (%) in population
Share (%) in labor force
LFPR (%)
Share (%) in population
Share (%) in labor force
LFPR (%)
Share (%) in population
Share (%) in labor force
LFPR (%)
Single Married Divorced Widowed All
22.3 68.4 1.1 8.2 100
43.1 50.4 2.3 4.2 100
34.3 13.1 36.1 9.9 17.7
24.4 66.4 1.4 7.8 100
48.3 45.7 3.6 2.4 100
33.8 11.8 41.0 5.9 17.1
23.8 65.8 2.2 8.2 100
41.4 51.2 5.1 2.3 100
31.1 13.9 42.0 5.7 17.9
Total number
8.5 million
1.5 million
—
13.2 million
2.3 million
—
15.2 million
2.7 million
—
Source: HLFS database, October 1988 and 1998, Fourth Quarter 2003.
opportunities have to be sought outside the home altered women’s economic role in the household and resulted in their withdrawal from the labor force. During the period under examination, the female labor participation rate (for those 15 years old or over) did not exceed 20 percent in urban areas. As seen in Table 4.2, single women, as a whole, were about 2–3 times more likely to participate as married women. Divorced women have the highest participation rate but account for a small (yet rising) fraction of the population. Married women comprise the lion’s share in the population (66–68 percent of the female population at risk) but no more than 50 percent of the labor force. The discouraging effect of marriage on female labor supply is well documented. If children are present, child care considerations have an impact on the outcome. Since market-based child care services are utilized by only a handful of families, and not all families live in extended households where help from older adult females might be available, presence of young children in the household is likely to emerge as a major factor limiting participation. Investigations of the extensive margin of the labor supply decision of females on data from Turkey lend credence to this explanation and on the whole corroborate research findings from other countries. Using micro data from urban Turkey, Dayıoxlu (2000), Kasnakoxlu and Dayıoxlu (1996a,b, 1997), Kawar (1998), Tansel (1994, 1996), and Tunalı (1997) estimated participation equations. These studies confirmed that educated women are more likely to engage in market work, in the form of wage employment, holding constant male education (or wages), and non-labor income where possible. They also revealed that participation probabilities as a function of age peaked early, because marriage typically brought about withdrawal from the labor market. Some of these papers included controls for household demographics and established the tension women faced in dealing with the burden of reproduction and other domestic responsibilities. By way of underscoring the untapped potential of women, it is helpful to compare participation data from Turkey and other countries. In Table 4.3 labor force
Female labor supply in Turkey 101 Table 4.3 LFPR for persons aged 15–64 by sex and age group, Turkey and other OECD countries, 2003
LFPR, all persons aged 15–64 Men, 15–64 Women, 15–64 LFPR, ages 15–24 Men, 15–24 Women, 15–24 LFPR, ages 25–54 Men, 25–54 Women, 25–54 LFPR, ages 55–64 Men, 55–64 Women, 55–64
Turkey
OECD-Europe
Total OECD
51.1 74.0 28.1 38.4 50.6 26.8 59.1 87.7 29.8 34.0 47.1 21.4
67.0 77.6 56.4 45.6 51.0 40.2 79.4 91.1 67.6 43.1 54.8 32.0
69.8 80.2 59.6 50.3 55.1 45.5 79.9 92.0 68.0 53.4 65.4 42.0
Source: OECD (2004, Annex Tables B and C).
participation rates (LFPRs) for 2003 are given for the 15–64 age group and three phases in the life cycle, for Turkey and two groupings of OECD member countries: European, and all. Turkey’s aggregate LFPR is strikingly low. In fact Hungary, the country that is just above Turkey in the rankings, has a LFPR which is almost 10 points higher (51.1 versus 60.6 percent). When the data are broken by sex, it becomes clear that extremely low female LFPRs bring the average down. Notably prime-aged women (ages 25–54) have less than half the LFPR of their counterparts in OECD. Note that prime-aged males are also behind their OECD counterparts and the gap expands during late career (ages 55–64). This is attributable to the generosity of the retirement stipulations of the Social Security System in Turkey (see Sayan, 2005). Why the urban female labor force participation rate remained stable and below 20 percent during a period that harbored many favorable trends is a puzzle. Although a firm answer to the puzzle is yet to be found, some clues are available (Tunalı et al., 2003, Chapters 4 and 5). The lenient retirement stipulations have surely been a factor in low participation rates of women over 40. In the case of younger women, urban unemployment rates which were considerably higher for women than men up until the end of the 1990s may have discouraged female participation (Tansel, 2002). On a more fundamental level, the volatility in real income experienced by many households during the 1990s is likely to have dampened the demand for private services which employ women. A similar argument applies to export oriented manufacturing jobs which are selective of women (especially in textiles and apparel): these are at the mercy of global exchange rate fluctuations. In fact, there is evidence from private manufacturing data that women are the last to be hired, and the first to be fired when firms lay off workers (Özler, 2000a). As we argue later, the data on hours of work suggests that the labor market in Turkey does not offer the flexible working arrangements that have been credited for the dramatic rise in the fraction of females in the workforces of
102
Wnsan Tunalı and Cem Bavlevent
the OECD countries. Last but certainly not the least, the inadequacy of labor market legislation in addressing the needs of working women can be offered as a potential explanation.10 Hours of work We end this section with a brief look at the intensive margin. We focus on workers aged 15 or above, residing in urban areas. The average weekly hours of work has shown some variation over time around an increasing trend for both male and female workers since the initiation of the HLFS.11 Notably, length of the work week in Manufacturing has gone up from 47.8 in 1988 to 53 hours in 2003. The corresponding figures are 49.4 to 54.2 hours for males and 40.1 to 48.2 hours for female workers. Closer inspection reveals that hours have been especially high 2000 onwards. Figure 4.2 shows the evolution of weekly hours for females residing in urban areas broken down by employment status. An increasing trend since 1988 is visible in all the categories except for casual wage and salary workers. Over the 1988–1998 period, casual workers, who do not have employment protection, typically worked longer hours than regular wage and salary workers. This ordering reversed after 1999. In 2003 regular workers clocked in 47 hours on average while casual workers toiled for 38.5 hours. Evidently major changes have taken place since 1988, the year in which the micro dataset we employ in our empirical work was collected. The surges in the length of the work week appear to have come at a time when employment growth has been absent, hinting at a substitution relation. 65 60
Hours
55 50 45 40 35 30 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Regular W&S
Casual W&S
Employer
Self-employed
2003
Unpaid f.w
Figure 4.2 Hours of work per week by employment status, females residing in urban areas, 1988–2003. Source: HLFS database, SIS web site.
Female labor supply in Turkey 103 The share of part-time workers in Turkey is very low compared to OECD countries. In 1988, barely 3.5 percent of employed individuals in Turkey reported less than 30 hours of work. Women accounted for nearly 45 percent of this segment. In 1990 the share of part-time workers in OECD countries was close to 15 percent, and women accounted for more than 70 percent of the total.
Scope and theoretical framework Our micro data source is the October 1988 HLFS conducted by the State Institute of Statistics (SIS). This is the first nationwide labor force survey that applied the standards proposed by the International Labour Organization (ILO). We use this dataset rather than a more recent one because questions concerning income were removed after the two rounds (administered in smaller samples) fielded in 1989. Relevant questions have recently been reintroduced in the revised HLFS (2002 onwards) but wage and income data have not been released. Our analysis is confined to married women between the ages of 20 and 54, who reside in urban areas (population 20,000) and are not in school. We chose to restrict the sample to married women for several reasons. First and foremost is the fact that this enables us to exploit a well-understood theoretical framework which can be supported with the data. Since marriage is almost universal in Turkey, this choice should not influence the generality of our findings. Our operational sample is limited to women who are not in school, and are identified in the survey data as the sole wife of the household head. Since employment in agriculture typically reflects joint market and home production, we also excluded 179 women who were engaged in agriculture. In order to exploit the available information in the data fully, we excluded 1,535 women whose husbands did not report income despite being employed. The variety of ways in which women in Turkey get involved in the labor market is another factor which complicates the investigation of female labor supply. Besides wage labor (the activity that is conventionally understood as work), other forms of work include informal employment or income generating opportunities that can be carried out in the home as well as work in a family enterprise without pay, or the production and/or sale of a good or service on one’s own account. Since the literature on labor supply evolved around the stylized facts found in industrialized country labor markets, forms of employment other than wage labor have not received much attention. Yet the studies by Tansel (1996, 2000), Tiefenthaler (1994), Tunalı and Ercan (1998) establish that there is much to be learned from treating alternative employment categories as distinct alternatives. In our operational sample of 8,962 married women, 7,770 non-participants constitute a sizable majority (86.7 percent). Among 1,192 participants, 266 are unemployed. Employed women are observed in 1 of 5 categories: regular wage and salary employment (710), casual work for wages (35), employer (11), ownaccount work (120), and unpaid work in a family enterprise (50). Although distinctions between all seven groups are potentially relevant, small cell sizes and computational concerns preclude us from exploiting the full richness of the data. As a compromise we combine the first two employment categories as “wage
104
Wnsan Tunalı and Cem Bavlevent
labor,” and the remaining three as “self-employment.” We recognize the fact that non-participants devote a significant portion of their time to home production, and refer to that option as “home work.” Consequently we are able to capture the variety of ways in which women can realize their productive potential with some degree of realism. We introduce further realism by allowing unemployment to be a distinct state. Thus, the model we utilize distinguishes between four distinct labor force participation (LFP) categories: non-participation (or home work), selfemployment, wage labor, and unemployment. The participation model described in the section Participation equations captures the sorting mechanism and allows for the examination of the intensive margin of the labor supply decision after adjusting for selectivity. In the theoretical framework we envision, the male household head is the main bread winner.12 His labor supply decision is assumed to be exogenous in our analysis (in fact we choose our working sample so that all the husbands are gainfully employed). His wage rate is included among the determinants of the wife’s labor supply decision. The wife (our unit of analysis) views other household members’ incomes as non-labor income, which is assumed to be exogenous to her decision. We make two additional assumptions that concern hours. First, we assume that the hours of work decision is made freely – that is, there are no restrictions on hours imposed by the employer.13 Second, we assume that wages do not depend on hours – that is, they are fixed over the entire range of hours of work.14 Given these assumptions, we proceed to obtain wage and income elasticities of labor supply by regressing hours of work on own wage, husband’s wage, and non-labor income. It is possible to use the parameters of the wage equation estimated on the wage labor subsample to predict “market” wages for everyone. This allows us to estimate a “structural” version of the participation model by including predicted own “market” wage, non-labor income and husband’s wage as covariates. The identification issues are addressed later.
Econometric model Participation equations In our model, there are three latent variable equations which express the utility in a given state as the sum of a systematic and a random component, where the systematic component is a linear function of observed variables: Reservation utility:
U *0i 0zi v0i,
(4.1)
Self- employment utility:
U *1i 1zi v1i,
(4.2)
Wage labor utility:
U *2i 2zi v2i.
(4.3)
As the reference state, non-participation provides the reservation utility. Loosely speaking, equations (4.1)–(4.3) constitute the “structural form” of the model. We
Female labor supply in Turkey 105 assume that the individual chooses the state in which her utility is highest; thus it is sufficient to work with utility differences. Towards that end we define two new latent variables: y*1 U *1 U *0 ( 11 01)z (v1 v0) 1z u1;
(4.4)
y*2 U *2 U *1 ( 21 11)z (v2 v1) 2z u2
(4.5)
We refer to s as “reduced form” parameter vectors, and the us are the reduced form disturbances. Here and later the individual subscript i has been dropped to avoid notational clutter. The latent variables may be interpreted as follows: y*1 propensity to be self-employed or the excess of the utility under the self-employment option over that in the home work option (reservation utility). y*2 incremental propensity to engage in wage labor rather than self-employment so that y*1 y*2 is the excess utility from wage labor over home work. The unemployed need to be handled differently: Although they choose to participate, their preferences between self-employment and wage labor is typically not known. We follow Magnac (1991) and operationally define an unemployed woman as someone who envisions higher utility in at least one of the wage labor and self-employment sectors, relative to her reservation utility. With this assumption, it is possible to keep track of all four observed participation states by focusing on sign combinations of the two latent variables:
0 home work, 1 self-employment, LFP 2 wage labor, 3 unemployed,
if y*1 0 and y*2 y*1 0, if y*1 0 and y*2 0, if y*1 y*2 0 and y*2 0, if y*1 0 or y*1 y*2 0.
(4.6)
Let Pj denote the probability that an individual will be observed in state j, j LFP 0, 1, 2, 3. We have: P0 P ( y*1 0 and y*1 y*2 0), P1 P ( y*1 0 and y*2 0), P2 P ( y*1 y*2 0 and y*2 0), P3 P ( y*1 0 or y*1 y*2 0) 1 P0.
(4.7)
We need to parameterize these probabilities in terms of the observables (z) so that the selective nature of participation can be studied. We assume that the reduced form disturbances u1 ( v1 v0) and u2 ( v2 v1) are bivariate Normally
Wnsan Tunalı and Cem Bavlevent
106
distributed (independently of z) with mean vector zero and a positive definite covariance matrix , where
V
uu 1
11
2
12
12 . 22
(4.8)
The expressions for the membership probabilities are given in the Appendix. Wage and hours equations Presently we express the wage (w) and hours (h) equations as: w w xw vw,
(4.9)
h hxh vh,
(4.10)
where xs are vectors of observed covariates, s are the parameter vectors to be estimated, and vs denote random disturbances which have zero means and constant variances independently of x. Wage–hours pairs are observed only in the subsample of women who work for wages. In estimating (4.9) and (4.10), we need to take into account the (potentially) non-random selection mechanism described in the section on Participation equations. To achieve this, we focus on systems of three equations which in turn link (4.9) and (4.10) with the pair of selection equations (4.4)–(4.5) via (4.7). For m w, h, we define zero-mean vectors of disturbances (u1, u2, vm), with positive definite covariance matrices:
m
u1 11 V u2 12 vm 1m
12 22 2m
1m 2m . mm
(4.11)
Estimation With stronger distributional assumptions (namely multivariate normality), full information maximum likelihood estimation may be employed. However we follow the established practice and pursue the computationally simpler Heckman–Lee two-step estimation procedure introduced by Heckman (1979) and Lee (1979). First step Reduced form parameters of the participation choice are estimated using maximum likelihood. The relevant likelihood function is L4
LFP0
P0i ·
P1i ·
LFP1
LFP2
P2i ·
P3i,
(4.12)
LFP3
where the Pji s are defined in the Appendix for j 0, 1, 2, and P3i 1 P0i. Under the maintained assumptions, maximization of (4.12) will yield maximum likelihood
Female labor supply in Turkey 107 estimates of the reduced form parameters subject to identification. The conditions for identification can be established following Terza (1985). In our empirical work we apply the normalization 11 1, and estimate 1, 2, 22, and 12. We use the estimates for 22 and 12 to compute the correlation coefficient between the two reduced form participation equations. The correlation coefficient,
12/(12), is a key element of the estimated model that reveals the role played by unobserved heterogeneity in the participation decision. Since this version turns out to be computationally fragile, we also estimate versions with the restriction 22 k11.15 Second step With the first step estimates in hand, we form (maximum likelihood) estimates of the selectivity adjustment terms (s) using the expressions given in the Appendix. Abusing the notation in (4.9)–(4.10) in the interest of continuity, we express the estimating equations as: ˆ 21 w2 ˆ 22 w, log(w) wxw w1
(4.13)
ˆ ˆ h hxh VV w log(w) wh log(wh) h1 21 h2 22 h.
(4.14)
Here V denotes non-labor income, and log(wh) denotes the natural logarithm of the respondent’s husband’s wage rate. The hats (“ˆ”) indicate that the estimated versions of the s are being employed, and s are zero-mean, heteroskedastic disturbance terms which are uncorrelated with the right hand side variables. In establishing our theoretical framework, we mentioned the possibility that wages and hours could be jointly determined. Since (4.13) may be viewed as a reduced form equation, the problem can be side-stepped in estimating the wage equation, but it has to be confronted in (4.14). Therefore, we turn to IV estimation and use the predicted log-wage from (4.13) as a regressor in the hours equation.16 Technically the parameters of (4.13) and (4.14) are identified even when xw, xh, and z completely overlap, because s are non-linear functions of the linear index functions. We follow the preferred practice and impose additional restrictions by exploiting the economic content of our model and test the over-identifying restrictions. Linear regression of log(w) (or h) on the covariates and the selectivity adjustment ˆ 21 and ˆ 22) yields consistent estimates of the regression parameters w, terms ( w1, w2 ( h, V, w, wh, h1, h2). Since m1 (1m 2m)/a, m2 2m/2 (for m w, h), and first step estimates of 2 and a are available (subject to the normalization 1 1, and the restriction 2 k1 if applicable), consistent estimates of 1m and 2m may be formed. In light of the heteroskedasticity of m, the standard errors reported by the typical regression package would be incorrect. However the reported standard errors are correct under the null hypothesis of “no selection” (H0: m1 m2 0). Thus conventional procedures can be used to test this null hypothesis against the alternative that selection is present.17 Third step The aim is re-estimation of the participation equation with a set of covariates that afford structural interpretations: own market wage, non-labor income, and husband’s wage. Since market wages are observed only in the wage–labor subsample, for the others we rely on counterfactual predictions.
108
Wnsan Tunalı and Cem Bavlevent
Reintroducing the subscript for the individual and conditioning on work status (and the covariate vector xw) equation (4.9) yields: E(wi LFPi j ) wxwi E(vwi LFPi j ) wxwi j1j1i j2j2i,
j 0, 1, 2;
(4.15)
where 21 ( w1) (1w 2w)/a, 22 ( w2) 2w/2, 11 1w/1, 12 22, 01 11, 02 21. In the case of unemployed women, the situation is more complicated. We know they prefer the market options over the home work option, but we do not know the type of work they will engage in. A possible solution is to modify the prediction equation as follows: E(wi LFPi 3) xwi (P1i/P3i)(1111i 1212i) (P2i/P3i)(2121i 2222i).
(4.16)
With the first step estimates in hand, consistent estimates of the probabilities Pjis and s can be formed using the expressions given in the Appendix. Estimates of a and 2 are also available from the first step. These can be used alongside second step estimates of w, w1, w2, and 1w to predict market wages for everyone.
Measurement issues Estimates of weekly, monthly, and annual labor supply equations have all been reported in the literature. To maximize the variation in the dependent variable, we chose to focus on annual labor supply. In our dataset, we have information on actual hours of work during the reference week as well as “usual” hours worked per week. Our judgment call was that usual hours would be more appropriate. We therefore computed our annual hours of work measure as “actual days worked per year” “usual hours worked per week”/7. The kernel density estimate of annual hours of work is given in Figure 4.3. We used monthly earnings to arrive at a measure of the hourly wage for women who work for wages or salaries. To be specific, we divided “monthly income” by “monthly days of work” “usual hours worked per week”/7.18 This formulation works because, according to the accounting system employed in the HLFS, an unbroken job spell corresponds to 31 days per month, and 366 days per year. The hourly wage of the husband is computed in analogous fashion, regardless of status. Kernel density estimates of the resulting wage distributions are given in Figures 4.4 and 4.5. Summary statistics for the variables that appear in the estimating equations are reported in Table 4.4. Subsample means reveal that average annual hours of work of wage (and salary) workers is substantially higher than that for the self-employed (means were respectively 1,929 and 1,372 while medians were 2,091 and 1,098). In fact while two-thirds of the self-employed report weekly hours of less than 40 hours, the corresponding figure for wage workers is only 20 percent. Notably 53 percent of wage workers have a usual work week of 40 hours, and the standard
Female labor supply in Turkey 109
0.003
0.002
0.001
0 0
1,000
2,000
3,000
4,000
Yearly hours of work (W&S workers)
Figure 4.3 Distribution of wife’s annual hours of work. Source: 1988 HLFS October.
1
0.5
0 0
2
4 Own wage (W&S workers)
6
8
Figure 4.4 Distribution of wife’s wages. Source: 1988 HLFS October.
deviation in hours is smaller in the wage labor subsample (664 versus 1,015 for the self-employed). Evidently the wage labor option does not offer as much flexibility as the self-employment option. The HLFS provides no information on categories of income other than “earnings from all jobs.” We used the sum of the annual earnings reported by family members
110
Wnsan Tunalı and Cem Bavlevent
1
0.5
0 0
2
4
6
8
Husband's wage (W&S workers)
Figure 4.5 Distribution of husband’s wages. Source: 1988 HLFS October.
other than the respondent as our measure of non-labor income. This approach hinges on two assumptions: (i) the labor supply decisions of the other family members are exogenous; (ii) the total income of other family members is an endowment that our female decision maker can freely draw on. Both assumptions are fiercely contested in the literature. Unfortunately our dataset is conspicuously poor in providing the instruments we need to entertain more realistic models of female labor supply. The remaining explanatory variables fall into 1 of 3 categories, which may be termed personal, household, and labor market characteristics. Personal characteristics include age (or experience), and education dummies which identify the highest level of completed schooling.19 Illiterates constitute the reference education category. Elementary school graduates make up nearly half of the full sample. The share of high school and university graduates is by far the highest in the wage labor subsample. Also among personal characteristics is a dummy which identifies women married to self-employed husbands. In Turkey, most selfemployed females work in family owned businesses. By contrast, engagement in wage labor generally means that there are no family members in the work place. Therefore, a self-employed husband would imply in many cases that the wife also finds herself in the same business. Household variables capture demographic composition effects. The list includes a dummy for extended households, the number of children in various age groups distinguished by sex, as well as interactions between household type and child, counts. Labor market characteristics were obtained from secondary
Table 4.4 Sample means (std. devs.) of variables by labor force participation status Variable
Full sample
Nonparticipant
Selfemployed
Own wage
—
—
—
Husband’s wage (1,000 TL)
1.86 (4.67) Non-labor income (1 mil. TL) 4.58 (11.5) Annual hours of work — Age 33.3 Age squared/100 11.8 Experience 20.5 Experience squared/100 4.92 Illiterate (reference) 0.25 Literate without a diploma 0.090 Elementary school 0.49 Middle school 0.058 High school 0.087 University 0.030 Husband self-employed 0.33 Children aged 0–2 0.26 Children aged 3–5 0.34 Female children aged 6–14 0.41 Male children aged 6–14 0.44 Extended household 0.12 Ext. hh Children aged 0–2 0.029 Ext. hh Children aged 3–5 0.036 Ext. hh Female ch. aged 6–14 0.044 Ext. hh Male ch. aged 6–14 0.047 Share of textiles 0.31 Share of agriculture 0.17 Share of finance 0.060 Migration rate 0.024 Marmara (reference) 0.35 Aegean 0.11 South 0.12 Central 0.19 North West 0.040 East 0.078 South East 0.081 North East 0.033 Population 200,000 or more 0.67 Population 1 million or more 0.51 Share of Welfare Party 0.073 Share of Left of Center 0.35 Sample size 8,962
Wage worker
Unemployed
1.39 — (3.17) 1.87 1.61 1.95 1.48 (4.83) (1.68) (3.43) (4.33) 4.65 4.45 4.28 3.54 (11.8) (6.83) (8.00) (13.5) — 1,372 1,929 — (1015) (664) 33.4 34.3 32.8 30.7 11.9 12.4 11.2 9.84 20.9 21.6 16.4 17.2 5.12 5.35 3.29 3.48 0.27 0.22 0.055 0.13 0.095 0.12 0.034 0.071 0.52 0.46 0.22 0.48 0.055 0.11 0.060 0.094 0.059 0.072 0.35 0.20 0.0069 0.022 0.28 0.023 0.35 0.39 0.16 0.18 0.26 0.18 0.22 0.22 0.35 0.28 0.27 0.34 0.42 0.48 0.32 0.36 0.45 0.50 0.34 0.43 0.13 0.14 0.11 0.079 0.029 0.028 0.031 0.019 0.036 0.033 0.039 0.019 0.046 0.039 0.032 0.030 0.049 0.050 0.039 0.030 0.31 0.40 0.30 0.31 0.18 0.20 0.15 0.19 0.059 0.053 0.064 0.056 0.023 0.021 0.035 0.018 0.35 0.32 0.37 0.28 0.11 0.10 0.14 0.21 0.12 0.27 0.099 0.13 0.18 0.10 0.23 0.17 0.037 0.028 0.055 0.075 0.080 0.088 0.044 0.086 0.087 0.088 0.039 0.015 0.034 0.0055 0.027 0.041 0.66 0.69 0.74 0.60 0.50 0.42 0.61 0.46 0.074 0.059 0.062 0.058 0.35 0.33 0.36 0.36 7,770 181 745 266
112
Wnsan Tunalı and Cem Bavlevent
data sources and are defined at the province level. The list includes the share of textiles in the total manufacturing employment of the province in 1988, the GDP shares of finance and agriculture, and the net migration rate measured as the change in a province’s population between 1985 and 1990, divided by its population in 1990. We also include seven regional dummies (with Marmara region as the reference category) plus two indicators for the size of the location of residence. As in Kawar (1998), we use province level information on the share of votes received (in the 1987 general elections) by the pro-Islamist Welfare Party (WP) to capture attitudinal influences on, and the composition of, the labor market. Kawar’s argument is that votes for WP would demonstrate stronger attachment to a traditional view of women’s role in the household and would therefore make labor market orientation less likely. Following her line of thinking, we also include the share of votes received by the two major left of center parties (Democratic Left Party and Republican People’s Party). Provinces in which this share is higher are likely to feature a more liberal attitude towards female labor. Our specification allows for the possibility that the society might view the two sectors of employment differently. It remains to be seen whether stronger traditional values implies a stronger opposition to participation in the wage labor, rather than self-employment.
Empirical results Estimation was implemented using STATA 8.0. In the first and third steps, the derivative-free maximum likelihood option (lf) was used. We relied on non-linear transforms to confine estimates of correlation and variance parameters to their theoretical ranges. Since the sampling frame of the HLFS relies on stratification, we utilized the expansion factors provided in the HLFS data to construct appropriate probability weights. To help offset the presence of generated regressors and clustering, robust (Huber–White) standard errors are reported throughout. Reduced form participation equations Participation decisions are assumed to be utility driven. Therefore we used all personal, household, and labor market characteristics as explanatory variables in the first step. Age was included in quadratic form to keep track of life cycle effects. The maximum likelihood estimates of the “normalized” model (11 1) yielded
ˆ 0.971, a magnitude which is dangerously close to 1.20 Therefore, we estimated “restricted” versions of the model to examine the sensitivity of the results to changes in the preset values of k 22/11 (with k 0.35, 0.5, 1, and 2). As it turns out, the restricted versions lead to similar inferences. In particular, estimated values of are all negative and large in absolute magnitude. Note that k is estimated to be 0.358 in the normalized version. Consequently it is reassuring to discover that the restricted version results for k 0.35 (i.e. 22 0.3511) reported in Table 4.5 are practically identical to the results from the normalized model (not reported).
Table 4.5 Maximum likelihood estimates of reduced form participation equations (restricted version) Variable
Age Age squared/100 Literate without a diploma Elementary school Middle school High school University Husband self-employed Children aged 0–2 Children aged 3–5 Female children aged 6–14 Male children aged 6–14 Extended household Extended household *Children aged 0–2 Extended household *Children aged 3–5 Extended household *Female children aged 6–14 Extended household *Male children aged 6–14 Share of textiles Share of agriculture Share of finance Migration rate Aegean South Central North West East South East North East Population 200,000 or more Population 1 million or more Share of Welfare Party Share of Left of Center Constant 11 22
No. of observations Log-likelihood
First selection
Second selection
Coef.
Std. error
Coef.
Std. error
0.078* 0.125** 0.256**
0.032 0.047 0.098
0.01 0.024 0.183*
0.032 0.047 0.084
0.124 0.473** 0.543** 0.948** 0.084 0.178* 0.078 0.054
0.076 0.108 0.101 0.153 0.060 0.070 0.061 0.060
0.047 0.192* 0.09 0.148 0.152 0.076 0.012 0.074
0.071 0.097 0.118 0.146 0.078 0.062 0.053 0.056
0.03
0.058
0.072
0.054
0.188
0.123
0.165
0.104
0.095
0.175
0.046
0.150
0.057
0.160
0.09
0.136
0.251
0.157
0.219
0.136
0.082
0.155
0.102
0.134
0.06 0.755* 3.661* 3.623** 0.226* 0.038 0.505** 0.389* 0.378* 0.291 0.843** 0.114
0.176 0.383 1.455 1.175 0.103 0.099 0.129 0.168 0.173 0.149 0.280 0.089
0.173 0.950** 2.881* 4.197** 0.324** 0.016 0.544** 0.523** 0.422** 0.146 0.963** 0.027
0.158 0.341 1.249 1.042 0.092 0.085 0.127 0.157 0.156 0.125 0.267 0.079
0.16
0.086
0.138
0.074
5.943** 0.689 1.704**
0.997 5.588** 0.444 1.170** 0.582 0.251 1 [normalized] 0.35 [restricted] 0.970** (0.039) 8962 2443.1
0.974 0.437 0.671
Notes Robust standard errors are reported. Asterisks denote statistical significance at the 5 percent (*) and 1 percent (**) levels. Standard errors (reported in parentheses) for and 22 were computed using the “Delta” method.
114
Wnsan Tunalı and Cem Bavlevent
Having established confidence in them, we proceed to discuss the results from the normalized version in some detail. Based on the likelihood ratio statistic, the 4-way model fits the data well. Furthermore, tests of joint significance on the slopes (all 32 of them) provide very strong evidence that self-employment and non-participation are distinct states (p-value 0.0001), but weaker evidence that wage labor and self-employment are distinct states (p-value 0.12). When the second test is repeated under the restriction k 0.35, the second p-value is also near zero ( 0.0001). Note that the main difference between the estimates reported in Table 4.5 and those obtained under the normalized model is that the standard errors of the parameter estimates in the normalized version are 2 to 3 times as large as those in the restricted version. Once the scale parameter is fixed (identified exogenously), it is possible to get more precise estimates of the slopes. We chose to use the restricted results for hypothesis testing purposes to facilitate a fair comparison of the two selection equations. The estimate of 0.35 for k means that there is considerably more unobservable variation in equation (4.4) than in (4.5). Note that equation (4.4) deals with the choice between non-participation (home work) and self-employment while equation (4.5) deals with the choice between two types of employment. Evidently there is relatively more heterogeneity attributable to tastes and preferences in the former, whereas, in the latter, observable factors do a better job of accounting for perceived differences in the two sectors of employment. The negative sign for implies that a woman whose unobserved characteristics yield an above average propensity to engage in self-employment rather than home work, also has unobserved characteristics that yield a below average propensity to choose wage labor over self-employment. Put differently, unobserved traits of two women who are observationally identical except for their work status are very different. It could be that self-employment and wage-labor call for different skills (and those endowed with desirable skills in one segment have less of the skills demanded in the other segment), or time constraints operate differently, or tastes are different. The estimated slopes reflect pairwise differences in the contributions of the explanatory variables to the propensities. A positive (negative) coefficient in equation (4.4) (henceforth the “first” selection, or equation) means that the associated regressor increases (decreases) the propensity to choose the self-employment option over non-participation. Similarly, coefficients in equation (4.5) (henceforth the “second” selection, or equation) reveal how the regressors influence the propensity to engage in wage labor over self-employment. Finally, the coefficients on the same regressor in the two equations can be summed to measure the effect it has on the propensity to become a wage worker rather than remain a non-participant. If the sum is positive (negative) we conclude that the regressor makes wage employment more (less) likely. The way the model is formulated, the probability of unemployment is inversely related with the probability of non-participation. The probability of labor force participation displays a quadratic profile with respect to age. Maximum probability is reached at 31 years of age, regardless of
Female labor supply in Turkey 115 work status. With illiterates as the reference category, gaining literacy (without earning a diploma) increases the probability of participation, and schooling beyond the compulsory level (elementary school) boosts that probability. The contribution of a middle or high school degree is two times, and a university degree is nearly four times that associated with gaining literacy. Compared to illiterates, literates and middle school graduates are less likely to be engaged in wage labor; for the others self-employment and wage work are equally likely. Note the concentration of women with high school education or more among wage workers in Table 4.4. Holding other variables constant, completion of a high school or university does not favor wage labor over self-employment. Women who have self-employed husbands are just as likely to engage in self-employment as in home work, and least likely to engage in wage labor. Thus the presence of a self-employed husband does not increase the likelihood of employment. It could be that self-employment status of the husband captures an income effect or the force of traditions that orient married women to home work. Results from the “structural” versions we estimated support the second interpretation. Surprisingly the demographic composition of the household does not emerge as a discriminating factor in the reduced form model. In joint tests, the null that the household variables all have zero slopes cannot be rejected (p-value 0.14 in the first equation, 0.48 in the second). There is some evidence that the presence of small children (0–2-year-old) renders non-participation more likely, but the impact is small. As we show later, sharper results emerge from “structural” estimation. In what follows we take a p-value of 0.01 or less (in a two-sided test) as our standard for statistical significance and report the p-value when marginal significance can be claimed. There is evidence that location matters. Regional dummies are jointly statistically significant but city size dummies are not (p-value 0.12 in the first equation, 0.17 in the second). Variables that capture labor market conditions at the province level are jointly statistically significant. Higher shares of agriculture and higher net rates of migration increase the likelihood of engaging in wage labor over selfemployment. The first pattern is counterintuitive. Increases in the share of finance render wage–labor a more likely outcome than self-employment, but surprisingly non-participation becomes an even more probable outcome. Since these variables yield straightforward demand side interpretations in wage equations, we refrain from elaboration at this point. The last piece of evidence we have comes from variables that capture attitudinal influences on the labor market. As the share of the votes received by the WP increases, (i) the likelihood of engaging in self-employment rather than home work decreases considerably, while (ii) the likelihood of engaging in wage labor rather than self-employment increases by about the same magnitude. The second result is counterintuitive (but it is dampened in the third step). The explanation needs to be sought in the claim that the Welfare Party receives considerable support from the lower rungs of the working class. Predictably as the share of the votes received by the left of center parties increases, (iii) wage labor orientation increases relative to self-employment. The magnitude of the reduced form effect
116
Wnsan Tunalı and Cem Bavlevent
via (ii) pales by comparison with (iii). The magnitudes that emerge from our “structural” estimates are closer to one another. Wage equation Our wage equation specification is in the Mincer tradition. Covariates include human capital variables, demand side factors and regional dummies. Experience, rather than age (as in the first step), is included in quadratic form to keep track of life cycle effects. We exclude the household dummies, city size indicators, migration rate and vote shares of political parties. Based on Sargan’s (1984) test for overidentifying restrictions, the exclusions are warranted.21 The least squares estimates with and without correction for selectivity are ˆ 22 has the statistically reported in Table 4.6. Of the two selection terms, significant coefficient. A joint test also yields strong evidence in favor of selectivity, meaning that selection into the wage labor subsample as a result of the participation decision process is non-random.22 Upon combining the second step estimates with the first step estimates, we conclude ˆ 1w 0.350 and Table 4.6 Least squares estimates of the wage equation Variable
With selection Coef.
Experience Experience squared/100 Literate without a diploma Elementary school Middle school High school University Share of textiles Share of agriculture Share of finance Aegean South Central North West East South East North East 21 22 Constant Number of observations R2
Without selection Std. error
0.034** 0.013 0.072 0.037 0.080 0.181 0.028 0.120 0.256 0.144 0.437* 0.184 1.083** 0.250 0.282* 0.143 0.085 0.238 2.557* 1.225 0.169* 0.073 0.172* 0.080 0.007 0.073 0.170 0.099 0.096 0.115 0.086 0.114 0.210 0.113 0.171 0.146 0.458* 0.182 0.927** 0.345 735 0.379
Coef.
Std. error
0.031* 0.013 0.064 0.037 0.112 0.186 0.011 0.121 0.231 0.125 0.465** 0.119 1.038** 0.128 0.407** 0.122 0.150 0.238 4.241** 1.158 0.131 0.069 0.159 0.082 0.029 0.073 0.166 0.096 0.142 0.113 0.191 0.100 0.129 0.115 — — — — 0.923** 0.194 735 0.368
Notes Robust standard errors are reported. Asterisks denote statistical significance at the 5 percent (*) and 1 percent (**) levels.
Female labor supply in Turkey 117 ˆ 2w 0.274. Evidently unobserved traits that make self-employment more likely than home work imply above average market wages, while unobserved traits that render wage labor more likely than self-employment imply below average market wages. The wage equation fits the data well (R2 0.38). Experience profile is concave. Maximum wage is reached at about 24 years of potential labor market experience. Statistically speaking, individuals who have elementary school education or less earn the same wage. The returns to higher levels of schooling are sizable: 29 percent for middle school (p-value 0.066), 55 percent for high school and 195 percent for university. Average incremental returns are 8.8 percent per additional year of middle school (8 years versus 5 years of elementary school), 8 percent per additional year of high school (11 years versus 8 years of middle school), and a whopping 24.5 percent per additional year of university education (15 years versus 11 years of high school). We find that statistically significant regional wage differences are present and labor market conditions captured at the province level are statistically and economically significant. A one percent increase in the employment share of textiles depresses average wages by 0.28 percent. A one percent increase in the employment share of finance augments average wages by 2.6 percent. Hours equation Since there is no standard model of hours, we experimented with the prototypical specifications found in the literature and examined the sensitivity of the elasticity estimates. The results were surprisingly robust but disappointing. Own wage elasticity was consistently negative, and the income elasticity was almost always positive. The only theoretically sound elasticity was the cross wage elasticity. To frame the discussion, we rely on the results from a specification which uses age, age squared and education dummies as controls.23 Least squares estimates with, and without, correction for selectivity are reported in Table 4.7. Based on a joint test, selectivity is ruled out: the unobservable factors that determine selection into the wage labor sector do not influence annual hours of work. With an R2 of 0.08, we find that very little of the variation in hours has been explained by the model. Although poor fits are the rule rather than the exception in labor supply studies that rely on instrumented wages, R2 values close to 0.20 are not uncommon. Yet even a “kitchen sink” reduced form hours regression with the full list of covariates we used in the participation step yields an R2 of around 0.11. This finding is attributable to the fact that rigidities on hours of work are quite strong in Turkey. The results reveal that hours are negatively related to own wages and positively related to income. The non-labor income elasticity is estimated to be 0.019 and is marginally significant (p-value 0.08). The magnitude is small, but it counters the conventional assumption that leisure is a normal good. The combination of a negative gross wage elasticity and a positive income elasticity produce a marginally significant (p-value 0.05) small negative (0.182) compensated own wage
118
Wnsan Tunalı and Cem Bavlevent
Table 4.7 Least squares estimates of the annual hours equation Variable
With selection Coef.
Without selection Std. error
Log of predicted own wage 379.6 218.5 Annual non-labor income 8.663 5.025 Log of husband’s wage 167.3** 57.27 Age 13.34 42.98 Age squared/100 13.43 62.09 Literate without a diploma 30.44 234.2 Elementary school 185.7 189.0 Middle school 492.6* 236.1 High school 580.1* 282.6 University 456.3 431.8 21 21.25 198.4 45.89 275.4 22 Constant 1,295.9 988.9 Number of observations 735 R2 0.0804 Elasticities Gross wage (w/h) 0.196 [0.083] Non-labor income 0.019 [0.085] [V(V /h)] Total income (wV) 0.012 [0.085] Compensated wage 0.208 [0.066] ( w/h-wV) Cross (husband’s) 0.086 [0.0036] wage (wh/h)
Coef.
Std. error
328.8 8.707 173.4** 13.97 14.96 42.31 201.6 508.0* 624.5** 490.6* — — 1,226.1*
177.8 5.018 54.49 35.06 50.80 237.6 191.4 197.3 192.5 233.2 — — 598.3 735 0.0797
0.170 [0.065] 0.019 [0.083] 0.012 [0.083] 0.182 [0.048] 0.090 [0.0015]
Notes Robust standard errors are reported. Asterisks denote statistical significance at the 5 percent (*) and 1 percent (**) levels. Wage labor subsample averages reported in Table 2 were used in elasticity calculations. The figures reported in brackets next to the elasticities are p-values for the null hypothesis that the elasticity is zero against the two-sided alternative.
elasticity, something that runs counter to the basic tenets of consumer theory. The husband’s wage variable, on the other hand, has the correct sign. The cross-wage elasticity is calculated to be 0.09, and is statistically significant. Recall that our non-labor income measure includes husband’s labor income. It could be argued that by estimating the income elasticity alongside the cross-wage elasticity, we are asking for more than what the data can deliver. We therefore re-estimated shorter versions of the model, by excluding in turn husband’s wage and non-labor income. The elasticities calculated from the restricted models were not appreciably different. Observe that annual hours increase with education (actually the peak occurs at high school level). Going beyond the implausible explanation that married females in Turkey are “workaholics” who view leisure as an inferior good, one
Female labor supply in Turkey 119 could look deeper into what “work” and “leisure” capture. The standard model of labor supply does not draw a distinction between leisure and home work and labels all work as a “bad.” At higher levels of non-labor income, married women may find it easier to substitute relatively more desirable (say high status) market work for less desirable home work. To pursue this line of thinking, we excluded professionals and re-estimated our models, but the findings were equally disappointing. Given the robustness of our results, we conclude that estimation of married female supply elasticities should be pursued with a better theoretical framework and richer data. Structural participation equations We estimated structural versions of our participation model on the full sample by including predicted own market wage, non-labor income, husband’s wage alongside a shorter list of controls. Predictions were done using equations (4.15)–(4.16) and the wage equation estimates reported in Table 4.6.24 We excluded the education, region, and labor market dummies which were employed as determinants of wages. The maximum likelihood estimates from the “normalized” model (11 1) are reported in Table 4.8. The structural version converged faster, and yielded smaller absolute magnitudes as estimates of the correlation ˆ 22 0.242). Based on the ( ˆ 0.290) and the unrestricted variance parameter ( AIC, the fit in the structural version was substantially better. The estimates reveal a huge own wage effect on participation. Interestingly, an increase in earnings potential is more likely to orient married women towards self-employment than towards wage labor. However this is a second order effect compared to the participation enhancing effect of own wage. As predicted by the standard model, increases in both non-labor income and husband’s wage dampen the participation probability. Note that (annual) non-labor income is measured in 1,000,000 TL, while own and husband’s (hourly) wages are denominated in 1,000 TL. Slope comparisons indicate that the income effect is considerably smaller than the cross wage effect, and both pale by comparison with the own wage effect. Exclusion in turn, of non-labor income and husband’s wage, did not alter the picture. Taken together, these results serve to vindicate the theoretical framework adopted here. Results from the structural version of the participation model sharpen the role of household composition. Own children effects predominate, suggesting that time constraints that women face figure in the participation decision. Evidently presence of very young (0–2-year-old) children is not a factor. However 3–5-year-old children deter the likelihood of engaging in wage labor relative to non-participation and self-employment. Presence of 6–14-year-old children increases the likelihood of self-employment relative to participation but does not help in the case of wage labor. Female children are more important than male children in relaxing the time constraint. Residence in extended households favors participation. The effect is considerable in the case of self-employment but smaller in the case of wage labor.
120
Wnsan Tunalı and Cem Bavlevent
Table 4.8 Maximum likelihood estimates of structural participation equations Variable
First selection Coef.
Predicted own wage Annual non-labor income Husband’s wage Age Age squared/100 Literate without a diploma Elementary school Middle school High school University Husband self-employed Children aged 0–2 Children aged 3–5 Female children aged 6–14 Male children aged 6–14 Extended household Extended household *Children aged 0–2 Extended household *Children aged 3–5 Extended household *Female children aged 6–14 Extended household *Male children aged 6–14 Migration rate Population 200,000 or more Population 1 million or more Share of Welfare Party Share of Left of Center Constant 11 22
No. of observations Log-likelihood
Second selection Std. error
Coef.
Std. error
5.638** 0.026 0.100* 0.307** 0.404** 0.256** 0.124 0.471** 0.533 0.931 0.298 0.208 0.006 0.410** 0.264 0.642**
0.317 0.017 0.042 0.103 0.150 0.099 0.085 0.138 0.580 1.049 0.241 0.109 0.117 0.151 0.136 0.194
0.574* 0.009 0.048 0.325** 0.464** 0.185 0.048 0.194 0.090 0.149 0.660* 0.094 0.232* 0.444** 0.365* 0.470*
0.290 0.013 0.035 0.123 0.179 0.121 0.075 0.187 0.117 0.154 0.264 0.096 0.111 0.163 0.142 0.196
0.178
0.272
0.023
0.265
0.247
0.287
0.386
0.285
0.497
0.257
0.350
0.244
0.183 2.918** 0.108 0.829** 11.573** 1.116 0.884
0.243 0.871 0.096 0.134 1.154 0.831 1.532
0.211 1.485 0.111 0.357* 3.648* 1.995* 4.739** 1 [normalized] 0.242 (0.222) 0.290 (0.255) 8962 1914.2
0.220 0.881 0.089 0.154 1.607 0.854 1.618
Notes Robust standard errors are reported. Asterisks denote statistical significance at the 5 percent (*) and 1 percent (**) levels. Standard errors (reported in parentheses) for and 22 were computed using the “Delta” method.
Conclusion Thanks to dramatic reductions in fertility, the share of Turkey’s working age population is rising and will remain high during the first half of the twenty-first century. Recent and pending reforms of the education system ensure steady improvements in the human capital endowments of the new entrants to the labor
Female labor supply in Turkey 121 force. Although these factors augur well from an endogenous growth perspective, labor force participation rates below 50 percent in 2000 and beyond, and unemployment rates that remain in double digits four years after the economic crisis in 2001, emerge as signs for concern. In this chapter we undertook a systematic study aimed at providing clues for the likely course of events. Towards that end, we enlisted two types evidence: descriptive, and econometric. Although females were our primary objects, we also devoted attention to males – who have traditionally been the primary bread winners of their households – in our descriptive analysis. In similar vein, although our focus was on labor supply, we also examined the evidence from the demand side in an effort to understand the supply patterns. Based on our evaluation of the big picture, it is clear that the structural reforms undertaken in the 1980s have helped to transform the labor market in predictable ways: Agricultural employment is declining, the share of wage and salary workers in the urban work force is increasing, and employment gains have been primarily in Services. Manufacturing employment also grew but at a much slower rate. The increased export orientation of this sector appears to have worked in favor of females. Low labor market attachment of women emerges as a major differentiating feature of the Turkish labor market when compared with those in the EU and others in the OECD. Yet there are many reasons for thinking that this situation will change. There is strong evidence that labor force participation rates rise sharply with education and increased schooling delays entry to marriage. Given the dramatic improvements in educational attainment observed after 1998, a larger fraction of women born in the 1990s and beyond will enter the labor market, delay marriage, and remain attached after getting married. The reduction in fertility combined with changes in home production technology should reinforce this orientation. Finally, the increase in the minimum legal age for retirement is likely to prolong careers (for males, as well as females). Our econometric investigation was directed at prime age (20–54) married women in urban areas, who stand to be most affected by the expected changes. We relied on a conventional static labor supply framework to draw behavioral inferences that can help us to go beyond the observed patterns. Our ultimate aim was to uncover the labor supply responses to primitives such as market wages and non-labor income. We studied the factors responsible for the observed female labor force participation patterns using a four-way choice model that distinguishes between the states of non-participation, wage (and salary) work, self-employment, and unemployment. Since unemployment was construed as a composite participation state, our results shed light on the distinctions between non-participation (home work), wage work, and self-employment. We found that there were both observed and unobserved factors at work. Therefore in our examination of the intensive margin of work, we took into account two forms of self-selection: (i) selection into the labor force, and (ii) selection into wage work rather than self-employment. The double-selection emerged as an important factor in the estimation of the wage equation but not in the hours of work equation. In fact, despite our best efforts, the annual hours equation failed to yield meaningful elasticity estimates.
122
Wnsan Tunalı and Cem Bavlevent
Our pursuit of the extensive margin was more rewarding. We found (a very strong) positive own-wage response and (small) negative responses to non-labor income and husband’s wage, all consistent with theory. Based on the differences in the estimated magnitudes, the behavioral trigger needed for a dramatic increase in the female labor force rate appears to be present. The question, of course, is whether the relative wage gains (demand shifts) that brought about the steady rise in female participation in high-income OECD countries can be realized anytime soon. Legislation that renders flexible working arrangements attractive has also been credited for the dramatic rise in the fraction of females in the workforces of OECD countries. Since social security legislation in Turkey does not align payroll taxes directly with hours of work, firms have been less than eager to hire part-time workers. At the same time, the ubiquity of informal employment arrangements has been pointed out as a potential source of flexibility. Although the data on hours of work of female wage and salary workers displays some variation, considerable concentration around the legally stipulated normal work week is present and incidence of part-time work is extremely low. Our inability to find theoretically plausible labor supply elasticites lends credence to the view that restrictions are important and need to be incorporated into the model.
Acknowledgments The empirical part of this chapter draws from Tunalı and Bavlevent (2001, 2004). Tunalı would like to acknowledge the hospitality of the California Center for Population Research at UCLA where the empirical work on the latter paper was completed. We are grateful to Richard Blundell, Andrew Foster, Mark Rosenzweig, Paul Schultz and Aysıt Tansel for their comments, and to Ali Çarkoxlu and Alpay Filiztekin for supplying province level data. Bengi Yanık provided research assistance on this version. Funding for this research was provided under grant no: ERF99-TK1004 from the Economic Research Forum through the World Bank DGF Program.
Appendix Handling selectivity Let G[t1, t2; r] denote the standard bivariate Normal cumulative distribution function with upper thresholds t1, t2, correlation coefficient r. With g(.) the associated density function, we have:
g(u ,u ;r) du du . t1 t2
G(t1,t2;r)
1
2
2
1
(4A.1)
We let Pj Pr(LFP j) and express the cell probabilities as: P0 G(C01,C02;C03), P1 G(C11,C12;C13), P2 G(C21,C22;C23),
(4A.2)
Female labor supply in Turkey 123 and P3 1P0. In terms of the parameters of the reduced form equations (4.4)–(4.5) and the reduced form covariance matrix (8), the arguments of the cell probabilities are: C01 1z/1, C02 (1 2)z/a, C03 b/a1; C11 C01, C12 2z/2, C13 12/12; C21 C02, C22 C12, C23 c/a2;
(4A.3)
with a 11 22 212, b 11 12, c 22 12. To account for selectivity in estimation and prediction, we need expressions for the truncated means in the outcome equations, E(vm | LFP), m w, h. These can be obtained following Rosenbaum (1961). For j = 0, 1, 2, we make use of the following new terms: C*j1
Cj1 Cj3Cj2 , Dj
C*j2
Cj2 Cj3Cj1 , Dj
Dj
(1 C2j3) . 1/2
(4A.4)
The truncated means can now be expressed as: * ) (C01)(C 02 , P0
01
11
* (C11)(C 12 ) , P1
21
(C21)(C *22) , P2
02 12
22
(C02)(C *01) ; P0
* (C12)(C 11 ) ; P1
(4A.5)
(C22)(C *21) ; P2
where (.) and (.) respectively denote the univariate standard normal pdf and cdf. In the expressions the subscript for the individual has been suppressed. Heterogeneity enters the picture via the Cs defined in (3A.3).
Notes 1 Household Labor Force Surveys conducted nationwide by the State Institute of Statistics (SIS) constitute the primary data source for the purposes of labor market analyses. After its initiation in October 1988, HLFS data have been collected biannually (in April and October) between 1989–1999 and quarterly since 2000. Detailed information on the HLFS is available on the SIS web page, http://www.die.gov.tr. 2 Starting with 2000, HLFS modified its labor force definition and started publishing data for the 15 and over age group. In 1998 workers in the 12–14 age group numbered around 850,000 nationwide and constituted 3.9 percent of the total. 3 There appears to be a break between the old (1988–1999) and new (2000 and beyond) HLFS series when the size of the agricultural workforce is examined (see Tunalı et al., 2003: 33). Thus this figure should be regarded as a rough estimate. 4 Since the 1988 data were only collected in October, in an effort to maintain comparability we relied on the statistics from the October 1998 and third quarter of 2003. 5 Public sector includes employees of the central government administration, local administrations (municipalities), and SEEs. A detailed examination of employment
124
6 7 8 9 10
11
12 13
14 15 16 17 18
19
Wnsan Tunalı and Cem Bavlevent
patterns up until 1998 based on government records, broken down by the three components may be found in Tansel (2001). This and other information from the 2000 General Census reported below were obtained from the web page of SIS. The projections have been provided by Frederic Shorter. The projection methodology is described in SIS (1995). To arrive at the measure of average schooling, we assumed that literates without a diploma had 2 years of schooling, and all those included in the higher education category had 4 years of additional education. This has huge implications for growth. As Altux and Filiztekin (2005) point out, endogenous growth models predict permanent changes in the growth path when initial conditions change. The new Labor Law (no. 4857 enacted in June 10, 2003) was designed, in part, to address employer concerns about lack of flexibility in the nature of the employment relationship, and intensity of work during the day/week/month. Since the regulations for implementing the law were not completed until mid-2004, it will take some time before its impact can be assessed. In the HLFS database accessible from the web page of the SIS, actual hours worked during the reference week reported for the following subdivisions: 1–16, 17–35, 36–39, 40, 41–49, 50–59, 60–71, 72. Assuming that workers are uniformly distributed within the hours of work categories, it is possible to calculate average hours of work from grouped data. We excluded workers who had a job but did not work during the reference week from the analysis and assumed that average hours of work in the 72 category was 80. This model is referred to as the “male chauvinist” model in Killingsworth (1983: 30). The label is not appealing, but arguably it describes the situation fairly well. This assumption is overly restrictive, but as Blundell and MaCurdy (1998: 1588) point out, it should not be taken literally. Since individuals can choose among employers that offer different wage–hours combinations, the labor supply function captures an “average” relationship that individuals have over hours of work and hourly earnings. More realistic models allow simultaneity between wages and hours (Moffitt, 1984; Tummers and Woittiez, 1991) or discrete choices over the range of hours (Hoynes, 1996; Van Soest, 1995). Dickens and Lundberg (1993) and Tummers and Woittiez (1991) offer formal treatments of the hours restrictions. We ignore any non-linearity introduced to the budget line because of the cap on payroll tax, progressive income taxation and payment of overtime wages. Moffitt (1986) and MaCurdy et al. (1990) address these issues. Throughout, we use the convention that double subscripts denote variances and covariances, and single subscripts denote standard deviations (e.g. j jj). Based on a Durbin–Hausman–Wu test, exogeneity of wages in the hours equation was rejected: the residuals from the log–wage regression were statistically significant in the hours equation (p-value 0.01) when included along with the predicted log-wages. Consistent estimates of mm and the standard errors can be obtained by modifying the expressions in Tunalı (1986). Note that the expression we used for obtaining hourly wage contains a calculated value of hours of work in its denominator, while a similarly calculated value serves as the dependent variable in the hours equation. As Borjas (1980) shows, measurement errors in the hours variable will lead to a spurious negative correlation in the hours equation (so-called “division bias”). In fact, we estimated a version of the hours equation with the actual wage and obtained a significantly negative coefficient. Zabel (1993) shows that the problem is mitigated if the predicted wage rather than the actual is used in the hours equation. This is yet another reason for pursuing the IV approach we adopted. Experience refers to potential labor market experience. It is defined as age minus years of schooling minus 7 if the individual has elementary school education or above, and as age minus 12 if the individual has not completed elementary school.
Female labor supply in Turkey 125 20 In some alternative model specifications we experimented with, the estimate of was still close to its lower bound. For example, in the version with the household variables excluded, we got ˆ 0.967, and ˆ 22 0.217. The general patterns observed in the first and second equations were similar to those we report here. 21 Regression of the residuals from the wage equation (n 735 observations) on the excluded variables (q 15) yielded an R2 of around 0.0045, a test statistic [nR2 ~ Chi-squared(q)] equaling 3.34, and a p-value of 0.99. We had to exclude 10 observations because of missing wage information. 22 In Tunalı and Bavlevent (2004) we examine the distributions of the actual and counterfactual wages derived from this model and conclude that individuals choose the sector in which they are the most productive. 23 The alternate specifications we tried were: (ii) household dummies in place of education dummies, (iii) both sets of dummies, and (iv) neither set of dummies. The exogeneity of wage was rejected in all models. Using (i) to denote the specification reported here, Sargan (1984) tests of the overidentifying restrictions yielded the following p-values (in models with the selection terms): (i) 0.066, (ii) 0.0023, (iii) 0.11, (iv) 0.0015. We chose (i) rather than (iii) because (iii) is open to the criticism that household composition is endogeneous. 24 To avoid outlier problems, before proceeding with ML estimation we capped predicted hourly wage at 7.23, which corresponds to the ninety-ninth percentile value in the observed value in the wage labor sample. This affected 27 observations.
References Akpınar, A. and H. Ercan (2002) Vocational Education and Training Report: Turkey 2001. Italy: European Training Foundation. Altux, S. and A. Filiztekin (2005) “Productivity and Growth, 1923–2003,” Chapter 1 in this volume, The Turkish Economy. Anker, R. (1990) “Methodological Considerations in Measuring Women’s Labor Force Activity in Developing Countries: The Case of Egypt,” Research in Human Capital and Development. Greenwich: JAI Press, 27–58. Bavlevent, C. and Ö. Onaran (2004) “The Effect of Export-Oriented Growth on Female Labor Market Outcomes in Turkey,” World Development, 32(8): 1375–1393. Blundell, R. and T. MaCurdy (1998) “Labor Supply: A Review of Alternative Approaches,” in Handbook of Labor Economics, Volume 3A, O.C. Ashenfelter and D. Card (eds), Amsterdam: Elsevier, 1559–1695. Borjas, G. (1980) “The Relationship Between Wages and Weekly Hours of Work: The Role of Division Bias,” Journal of Human Resources, 15: 409–424. Boserup, E. (1970) Women’s Role in Economic Development. London: Allen & Unwin. Bulutay, T. (1995) Employment, Unemployment and Wages in Turkey. Ankara: ILO/SIS. Çaxatay, N. and G. Berik (1990) “Transition to Export-Led Growth in Turkey: Is There a Feminization of Employment?” Review of Radical Political Economics, 22: 115–134. Çaxatay, N. and V. Özler (1995) “Feminization of the Labor Force: The Effects of LongTerm Development and Structural Adjustment,” World Development, 23: 1883–1894. Dayıoxlu, M. (2000), “Labor Market Participation of Women in Turkey,” in Women’s Identities and Roles in the Course of Change: Central Asia, East and Central Europe and Turkey, F. Acar and A. Ayata (eds), Köln: Brill, 44–73. Dickens, W.T. and S.J. Lundberg (1993) “Hours Restrictions and Labor Supply,” International Economic Review, 34(1): 169–192. Erlat, G. (2000) “Measuring the Impact of Trade Flows on Employment in the Turkish Manufacturing Sector,” Applied Economics, 32(9): 1169–1180.
126
Wnsan Tunalı and Cem Bavlevent
Goldin, C. (1995) “The U-Shaped Female Labor Force Function in Economic Development and Economic History,” in Investment in Women’s Human Capital, T.P. Schultz (ed.), Chicago, IL: The University of Chicago Press, 61–90. Heckman, J.J. (1979) “Sample Selection Bias as a Specification Error,” Econometrica, 47(1): 153–161. Hoynes, H.W. (1996) “Welfare Transfers in Two-Parent Families: Labor Supply and Welfare Participation under AFDC-UP,” Econometrica, 64(2): 295–332. Kasnakoxlu, Z. and M. Dayıoxlu (1996) “Education and Labor Market Participation of Women in Turkey,” in T. Bulutay (ed.), Education and the Labor Market in Turkey, Ankara: State Institute of Statistics Press, 145–173. Kasnakoxlu, Z. and M. Dayıoxlu (1997) “Earnings Inequality between Genders in Turkey,” in Economic Dimensions of Gender Inequality: A Global Perspective, J.M. Rives and M. Yousefi (eds), Westport Co: Praeger, 95–117. Kawar, L. (1998) “Defining Women’s Economic Participation in Turkey.” Honors Thesis, Department of Economics, Harvard University. Kazgan, G. (1981) “Labor Force Participation, Occupational Distribution, Educaitonal Attainment and the Socio-Economic Status of Women in the Turkish Economy” in, Women in Turkish Society, N. Abadan-Unat (ed.), Leiden: E.J. Brill, 131–159. Killingsworth, M.R. (1983) Labor Supply. Cambridge: Cambridge University Press. Krishna, P., M. Devasish and S. Chinoy (2001) “Trade Liberalization and Labor Demand Elasticities: Evidence from Turkey,” Journal of International Economics, 55(2): 391–409. Lee, L., (1979) “Identification and Estimation in Binary Choice Models with Limited (Censored) Dependent Variables,” Econometrica, 47(4): 977–996. MaCurdy, T., D. Green and H. Paarch (1990) “Assessing Empirical Approaches for Analyzing Taxes and Labor Supply,” Journal of Human Resources, 25: 415–490. Magnac, T. (1991) “Segmented or Competitive Labor Markets,” Econometrica, 59(1), 165–187. Mammen, K. and C. Paxon (2000) “Women’s Work and Economic Development,” Journal of Economic Perspectives, 14(4): 141–164. Moffitt, R. (1984) “The Estimation of a Joint Wage-Hours Labor Supply Model,” Journal of Labor Economics, 2(4): 550–566. Moffitt, R. (1986) “The Econometrics of Piecewise-Linear Budget Constraints: Survey and Exposition to the Maximum Likelihood Method,” Journal of Business and Economic Statistics, 4: 317–327. OECD (2004) Employment Outlook. Paris: OECD. Ökten, Ç. (2205) “Privatization: What has been Achieved?” Chapter 10 in this volume. Öncü, A. (1981) “Turkish Women in the Professions: Why So Many?” in Women in Turkish Society, N. Abadan-Unat (Ed.), Leiden: Brill, 181–193. Özler, V. (2000a) “Gender Differences in Job Creation and Destruction Over Business Cycles,” in Wages, Income and Wage Distributions, T. Bulutay (ed.), Ankara: State Institute of Statistics Printing Division, 141–162. Özler, V. (2000b) “Export Orientation and Female Share of Employment: Evidence from Turkey,” World Development, 28(7): 1239–1248. Psacharopoulos, G. and Z. Tzannatos (1989) “Female Labor Force Participation: An International Perspective,” World Bank Research Observer, 4(2): 187–202. Rosenbaum, S. (1961) “Moments of the Truncated Bivariate Normal Distribution,” Journal of the Royal Statistical Society, Series B, 2: 405–408. Sargan, J. D. (1984) “Wages and Prices in the U.K.,” in Econometrics and Quantitative Economics, D. Hendry and K. Wallis (eds). Oxford: Blackwell. Sayan, S. (2005) “The Political Economy of Pension Reform,” Chapter 10 in this volume.
Female labor supply in Turkey 127 Schultz, T.P. (1990) “Women’s Changing Perspective in the Labor Force: A World Perspective,” Economic Development and Cultural Change, 38: 451–488. Shorter, F. (1995) “The Crisis of Population Knowledge in Turkey,” New Perspectives on Turkey, 12: 1–32. SIS (1995) The Population of Turkey, 1923–1994. Ankara: State Institute of Statistics Press. Tansel, A. (1994) “Wage-employment, Earnings and Returns to Schooling for Men and Women in Turkey,” Economics of Education Review, 13(4): 305–320. Tansel, A. (1996) “Self-employment, Wage-employment, and Returns to Education in Turkey,” in Education and the Labor Market in Turkey, T. Bulutay (ed.), Ankara: State Institute of Statistics Printing Division, 175–208. Tansel, A. (2000) “Formal versus Informal Wage Sector Choice of Wage Earners and Their Wages in Turkey,” in Informal Sector (I), T. Bulutay (ed.), Ankara: State Institute of Statistics, Printing Division, 25–150. Tansel, A. (2001) “Public Sector Employment as a Social Protection Mechanism.” Economic Research Forum Working Paper no. 0104. Cairo, Egypt. Tansel, A. (2002) “Economic Development and Female Labor Force Participation in Turkey: Time Series Evidence and Cross-Province Estimates,” in Employment of Women, T. Bulutay (Ed.), Ankara: State Institute of Statistics, Printing Division, 111–151. Taymaz, E. (1999) “Trade Liberalization and Employment Generation: The Experience of Turkey in the 1980s,” ERC Working Paper Series, no.99/12, Middle East Technical University. Tiefenthaler, J. (1994) “A Multisector Model of Female Labor Force Participation: Empirical Evidence from Cebu Island, Peru,” Economic Development and Cultural Change, 43: 719–742. Tunalı, W. (1986) “A General Structure for Models of Double-Selection and an Application to a Joint Migration/Earnings Process with Re-migration,” in Research in Labor Economics, Volume 8, Part B, Greenwich: JAI Press, 235–282. Tunalı, W. (1996) “Education and Work: Experiences of 6–14 Year Old Children in Turkey,” in Education and the Labor Market in Turkey, T. Bulutay (ed.), Ankara: State Institute of Statistics Printing Division, 105–142. Tunalı, W. (1997) “To Work or not to Work: An Examination of Female Labor Force Participation Rates in Urban Turkey.” Proceedings of the 5th Annual Conference of the Economic Research Forum, Cairo. Tunalı, W. and C. Bavlevent (2001) “Estimation of Female Labor Supply Parameters When Self-Employment is an Option.” Mimeo. Tunalı, W. and C. Bavlevent (2004) “Married Women’s Participation Choices and Productivity Differentials: Evidence From Urban Turkey.” Mimeo. Tunalı, I. and H. Ercan (1998) “Labor Market Segmentation in Turkey,” in Main Characteristics and Trends of the Turkish Labor Market, T. Bulutay (ed.), Ankara: State Institute of Statistics Printing Division, 89–127. Tunalı, W., with H. Ercan, C. Bavlevent, O.D. Öztürk and U. Akçixit (2003) Background Study on Labour Market and Employment in Turkey. European Training Foundation, Torino, Italy. Van Soest, A. (1995) “Structural Models of Family Labor Supply: A Discrete Choice Approach,” Journal of Human Resources, 30: 63–88. Zabel, J.E. (1993) “The Relationship Between Hours of Work and Labor Force Participation in Four Models of Labor Supply Behavior,” Journal of Labor Economics, 11(2): 387–416.
5
A disaggregated analysis of price dynamics Mustafa Çaxlayan and Alpay Filiztekin
Introduction Turkey has been experiencing high inflation since mid-1970s. According to statistics published by the State Institute of Statistics (SIS), the annual inflation rate has been well above 20 percent since 1976 until the implementation of recent structural reforms. Interestingly, researchers have only recently begun to investigate the Turkish data in relation to linkages between inflation and the behavior of prices although high inflation rates observed throughout the world in the 1970s, and the 1980s have prompted many economists to propose macroeconomic models to explain the phenomena and to strive to find support for these models using aggregated or disaggregated data. Even so, given the unique experience of Turkey with prolonged inflation, the number of studies regarding the relationship between prices and inflation in Turkey are still too few and almost no research has been conducted to describe the behavior of prices. There is a vast theoretical literature investigating the linkages between variability of prices and inflation. In the main, researchers have used menu cost models, signal extraction models, and the information investment model. Menu cost models assume that it is costly to change the nominal price (label price) of a product. This basic idea explored by Sheshinski and Weiss (1977) is that the actual nominal price of the product is temporarily held constant as its real or relative price gradually moves in between the two optimally set upper (S) and lower (s) bands. This approach suggests that the width of the band is determined by the rate of inflation, and as the rate of anticipated inflation increases, firms widen the (S,s) bands in order to conserve on menu costs, causing higher dispersion. Alternatively, search models developed by Benabou (1998), Benabou and Gertner (1993), and Dana (1994), point out that if it is costly to search, an increase in the absolute value of unexpected inflation induces consumers to search less. Therefore, as “sellers can ‘hide’ behind aggregate or inflationary noise to charge higher prices taking advantage of consumers’ reduced information to increase their markups” (Benabou and Gertner, 1993, p. 70). Finally, Van Hoomissen (1988) proposes that price dispersion could be a result of customers with different stock of information on the type of product they purchase. She claims that in an inflationary environment information will become obsolete quickly leading to higher price dispersion in the current
Disaggregated analysis of price dynamics 129 and the future periods. Overall, empirical research suggests that although the mechanisms may differ, all three sets of models receive support from the data regarding a positive association between inflation and price volatility and that firms will not change product prices synchronously. In this context, to our knowledge, the first study to investigate the relationship between prices and inflation for Turkey is carried out by Karasulu (1998). Using the SIS data for food items collected from the largest three cities in Turkey (Istanbul, Ankara, and Izmir) over the period 1991–1996, she showed that there is a positive relation between inflation and relative price variability providing support for the menu cost models. Subsequently, Çaxlayan and Filiztekin (2003), using a half-century long disaggregated price data for 22 food products collected from 19 cities over the period between 1947 and 1997, pointed out that the association between inflation and price variability is nonlinear with respect to structural shifts in the rate of inflation.1 More recently Çaxlayan et al. (2004) proposed a new empirical model to explain the relationship between price dispersion and inflation while searching for the role of vendors’ pricing behavior within the framework of menu cost, signal extraction, and information investment models using a disaggregated data collected from three different types of vendors across 15 neighborhoods in Istanbul. There are many potential sources of inflation. For the case of prolonged inflationary episodes, economists unilaterally agree that inflation is a monetary phenomenon. Nevertheless, neither policymakers nor the public agree on the true costs of inflation. Actually, inflation’s costs are not well understood. An important aspect of high inflation is that it is simply a tax and as such it distorts the functioning of markets and distribution of income.2 Another easily identifiable cost associated with inflation is the so-called shoe leather cost, transaction costs of not holding cash. However, there are more subtle and more damaging costs associated with inflation. For instance, during inflationary episodes relative prices will be severely distorted due to existence of rigidities, which arise from firms’ reluctance to change their prices because either there are administrative costs having to do with reprinting price lists or informing the dealer, or implicit costs associated with adverse reaction of customers (Rotemberg, 1982). Consequently, significant welfare losses could emerge as a result of externalities to price adjustment while price stickiness may prevent the economy to reach full employment level. Furthermore, in inflationary environments the prices of products are not adjusted across vendors synchronously, even in a city. Naturally, in an economy where relative prices do not reflect the relative scarcity of goods, resources will not be allocated efficiently. In other words, inflation deteriorating the information content of prices will lead to inefficient allocation of resources as investors will not be able to determine accurately which of the potential projects yield higher returns as opportunities will not be easily identified. Moreover, inefficient firms may be able to survive longer than they should leading to further welfare losses.3 As inflation gets higher (additional costs associated with higher inflation is in dispute among economists and there seems to be less incentive among politicians
130
Mustafa Çaxlayan and Alpay Filiztekin
to reduce inflation as it exceeds a certain threshold), it becomes more variable and less predictable. That is higher inflation leads to higher uncertainty. Obviously, in such an economic environment investors would be more reluctant to undertake long term projects.4 Furthermore, as managers of existing companies could not even predict the actual demand for their products, they behave conservatively, giving up on possibly lucrative investment opportunities while many potential investors delay their entry into such a market. Naturally all these lead to welfare losses and it should not be too hard to conjecture that losses will be even higher for countries that experience high and prolonged inflationary episodes such as Turkey. In this chapter, we examine in detail the nature of relative price adjustments, namely the duration of the prices, the synchronization of price changes, and the relative price dispersion using a unique data set which is constructed by The Istanbul Chamber of Commerce. The data set consists of monthly price observations for 242 distinct products from 15 different neighborhoods of Istanbul over the period between July 1994 and June 2000. Generally, the price of food items are collected from three different type of vendors – supermarkets; bakkals (convenient stores); and pazars (open air markets) – for each neighborhood. For several other items such as dwelling expenses, household expenses, clothing, etc. there is a single price entry for each neighborhood. We can summarize our findings as follows. We observe that the average price duration is 3.1 months, while for some goods it could be as low as 0.23 or as high as 23.5 months. If we take the average price duration literally as the norm, it turns out that a vendor loses in real terms around 14.5 percent of the price due to ongoing inflation by not changing the price tag of the product for an average of 3 months. Furthermore, we observe that price changes across vendors are not synchronized; vendors do not act simultaneously in response to changes in the economy. This behavior implies that the price of a good in relative terms could be substantially different across vendors. Indeed, we find that the average price dispersion across vendors for food items is around 10 percent and that for clothing is around 25 percent. The average price dispersion is 15 percent when all goods are considered. We also show that the unconditional distribution of real prices is symmetric. Allowing heterogeneity across stores and neighborhoods reduces variability of real prices by a small margin it does not affect the shape of the distribution. The findings also raise some questions about the effectiveness of monetary policies pursued by the government. Particularly, nominal prices stickiness suggests that it will take a longer time for the monetary policies to have the desired effect on the economy. The rest of the chapter is organized as follows. In the section on The data, we give a brief survey of the theoretical literature on menu cost models, signal extraction models, and the information investment model. In the section on The behavior of prices in Istanbul, we describe our data and in the section The synchronization of price adjustment we analyze the duration of prices, the synchronization of the price changes and the dispersion of the relative prices. The final section concludes the analysis.
Disaggregated analysis of price dynamics 131
The data The data set we use is constructed by the Istanbul Chamber of Commerce to compute a broad-based Cost of Living (COL) index for wage earners in Istanbul which is considered the most dynamic city in Turkey generating about 22 percent of the GDP as of 2000 (a quarter of industrial output and services) and housing about 15 percent of the total population. In particular, our data is collected from 15 different neighborhoods (boroughs) of Istanbul accommodating more than 50 percent of the population of Istanbul according to 1996 Census Bureau report. Furthermore more than 60 percent of the value added generated in Istanbul is produced within the boundaries of the same 15 neighborhoods. Our data set spans the period 1994:2007 to 2000:2006 and consists of average monthly price observations (average of four weekly quotations) for 242 distinct products including food; dwelling expenses; household expenses; clothing; health and personal care; transportation and communication; culture, education, and entertainment, and miscellaneous. We had to drop some of the items (38 in total) since prices of these products are available only for certain months of the year.5 The data collectors visited the same vendor to record price observations on the same product (same brand, quantity/weight, and other characteristics) for the purpose of consistency. For food items, price of each product is collected from three different types of vendors, supermarkets, bakkals, and pazars, for each neighborhood.6 For some products, prices are obtained from specialized stores. Prices for services can be set by the state (electricity, petroleum, taxi fares per mile etc.) and private landlords (house rental). For several other items, mostly for durable good prices or tolls for highways, product prices across neighborhoods are identical. Therefore, the data set provides us the price of each product, Pijkt, indexed by the product i, the neighborhood (borough) j, the store type k, and the month t. The three distinct store types mentioned earlier have different characteristics. Bakkals are small convenience stores which are almost always family-owned and operated. Each neighborhood has at least a dozen bakkals and usually many more. Bakkals are also local institutions with an important social dimension, as consumers tend to drop in to buy one or two items and exchange news and gossip with the owner. Pazars are open-air markets for fresh produce and small consumer items. These markets approach the perfectly competitive ideal, since sellers operate small stalls with one to a handful of products each, and each product generally has several sellers. There is one main pazar in each neighborhood, open one day a week. Turkish supermarkets are similar to their Western counterparts. They are relatively large, corporate-owned, and stock a wide variety of distinct products and brands. There are typically 1 to 2 centrally located supermarkets per neighborhood. Table 5.1 provides information regarding the products listed under each category heading and the distribution of these products across different vendor types.7 Although we appreciate the possibility of categorizing products in our data set differently, we choose to follow the approach taken by the Chamber to eliminate potential discrepancies that may arise due to personal choices. Given that our
132
Mustafa Çaxlayan and Alpay Filiztekin
Table 5.1 Distribution of products across vendor types and categories Main expenditure categories
N
Foodstuff Dwelling Household Clothing Health/personal care Transportation/communication Culture/education Miscellaneous Total
71 10 33 45 16 11 13 5 204
Bakkal
Market
Pazar
Other
Total
70 0 9 0 4 0 0 0 83
33 0 9 0 0 0 0 0 42
12 0 0 0 0 0 0 0 12
1 10 25 45 12 11 13 5 122
116 10 43 45 16 11 13 5 259
0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00
2000,03
1999,11
1999,07
1999,03
1998,11
1998,07
1998,03
1997,11
1997,07
1997,03
1996,11
1996,07
1996,03
1995,11
1995,07
1995,03
1994,11
1994,07
–0.02
Figure 5.1 Cost of living index inflation.
data set is constructed using observations collected from different types of vendors, the total number of price series for each neighborhood is 259 while the total number of products is 204. Note that foodstuff and clothing categories contribute most of the observations, 40 and 24 percent, respectively. For foodstuff bakkals contribute information on all items, whereas supermarkets and pazars provide information only on 33 and 12 products, respectively. Also note that bakkals contribute information on 9 and 4 items under the household, and the health and personal care expense categories, as well. Before we start investigating the behavior of prices in Istanbul, we should indicate that the average inflation rate over the period of interest was high but relatively stable at about 60 percent per annum.8 While, the average monthly inflation rate was 4.7 percent according to COL Index, the minimum and the maximum inflation rate reported are 2.0 percent in August 1994 and 11.8 percent in October 1995. In Figure 5.1, we plot monthly inflation rate in Istanbul during the observation period.
Disaggregated analysis of price dynamics 133
The behavior of prices in Istanbul Using Turkish data, studies by Karasulu (1998), Çaxlayan and Filiztekin (2003) and Çaxlayan et al. (2004) investigate the role of inflation on the behavior of prices in great detail. While Karasulu reports a positive relation between inflation and real prices, Çaxlayan and Filiztekin emphasize the presence of nonlinearity in that relationship. Çaxlayan et al. further show that the basic predictions of menu and search cost and information models on price rigidities hold in Istanbul. They also show that the market structure could play an important role in defining the relationship between inflation and relative prices. In this section we examine, a different aspect, namely, the behavior of nominal prices in Istanbul. In particular we emphasize three aspects of the behavior of price changes: the duration of price quotations and the size of price adjustments; the synchronization of price adjustments; the heterogeneity of prices across vendors and neighborhoods.9
The duration of price quotations and the size of price adjustments Initially, we compute the duration of the price quotations for all products sold in all neighborhoods considering all vendors. Details of our results, sorted with respect to average duration of product prices within specific categories are presented in Table 5.2. In column 1 we list the number of observations for each product and in columns 2, 3, and 4 we present the direction of price movements in terms of percentages – negative (Down), no change (No) and upward changes (Up). Observe that despite a high inflation rate (around 4.7 percent per month), for 140 items in more than 50 percent of the instances, product prices remain unchanged. Considering high inflation in Turkey these numbers indicate a high level of rigidity. It must also be noted that 6.3 percent of prices declined throughout the sample period. Nonetheless, these are either fresh produces or clothing that are subject to seasonal sales.10 The items that exhibited the highest rate of positive changes are meat and dairy products and housing rental and electricity. In Turkey, rental contracts are often based on some measure of past inflation, thus observing frequent positive changes (76 percent) is not unexpected. Electricity, however, is state provided. Such a high rate of positive change implies governments’ reliance on state provided services to raise revenues. In our analysis, we assume that prices may adjust at any moment, not just at the monthly interval. Hence, the instantaneous probability of price change is ln(1 x) and the mean time between price changes can be computed as 1/ln(1 x) months, where x is the rate of change.11 In column 5 of Table 5.2, we report the duration for each item in our list. The duration of price quotations vary between 0.30 (green peppers) and 23.50 (funeral expenses) months. The distribution of duration over products is plotted in Figure 5.2. The mean and the median duration of price quotations are 3.09 and 2.22 months, respectively. The distribution is skewed to the left, almost two-thirds of products have mean
Table 5.2 The duration of price quotations and the size of adjustments
Foodstuff Eggs Chicken Mutton Veal Margarine Feta cheese Onion Butter Olives Sucuk Vegetable oil Rice Salami Potato Yogurt Honey Olive oil Sausage Kaseri cheese Lentils Cracked wheat Offal Cookies Flour Baklava Jam Chick peas Dried beans Chocolate Hazelnuts Granulated sugar Halvah Kesme Seker Candy Pasta Biscuit Flodough Walnuts Tomato paste Cigarette (imported) Roasted chickpeas Bread Peanuts Milk Ready soups Raisins Coffee Tea
N
Down
No
Up
3,096 2,160 2,160 2,160 2,160 3,096 2,016 2,160 3,096 2,160 3,096 3,096 2,160 2,016 2,160 3,096 3,096 2,160 2,160 3,096 3,096 1,080 1,080 3,096 1,080 2,160 3,096 3,096 2,160 1,080 2,160 2,160 2,160 2,160 3,096 2,160 2,160 1,080 3,096 72 1,080 1,080 1,080 1,080 2,160 1,080 2,160 144
0.28 0.21 0.07 0.07 0.03 0.03 0.28 0.02 0.03 0.02 0.03 0.05 0.02 0.24 0.03 0.02 0.07 0.03 0.06 0.06 0.02 0.01 0.01 0.02 0.00 0.03 0.04 0.05 0.01 0.04 0.04 0.01 0.04 0.00 0.05 0.01 0.01 0.03 0.05 0.00 0.01 0.02 0.00 0.02 0.02 0.01 0.00 0.00
0.09 0.16 0.17 0.18 0.21 0.24 0.25 0.26 0.26 0.27 0.28 0.29 0.3 0.3 0.31 0.31 0.31 0.32 0.33 0.33 0.34 0.36 0.36 0.36 0.37 0.38 0.38 0.38 0.38 0.39 0.4 0.41 0.41 0.41 0.42 0.43 0.44 0.44 0.44 0.46 0.47 0.47 0.48 0.5 0.51 0.51 0.53 0.56
0.63 0.63 0.76 0.75 0.76 0.73 0.47 0.72 0.72 0.71 0.69 0.66 0.68 0.46 0.67 0.67 0.62 0.65 0.61 0.61 0.64 0.63 0.63 0.61 0.63 0.59 0.58 0.56 0.6 0.57 0.56 0.57 0.55 0.59 0.53 0.56 0.55 0.53 0.51 0.54 0.52 0.51 0.51 0.48 0.47 0.48 0.47 0.44
Duration 0.42 0.55 0.56 0.58 0.64 0.69 0.72 0.73 0.73 0.76 0.78 0.8 0.82 0.83 0.84 0.84 0.86 0.88 0.89 0.91 0.94 0.98 0.98 0.99 1.00 1.03 1.03 1.05 1.05 1.06 1.1 1.14 1.14 1.14 1.16 1.18 1.22 1.23 1.23 1.28 1.32 1.32 1.38 1.45 1.48 1.49 1.59 1.74
(Table 5.2 continued )
Table 5.2 Continued
Baby food Sunflower seeds Bottled water Cola drinks Carbonated drinks Juices Beer Table salt Restaurant bill Filter cigarette (short) Raki Bafra cigarette Filter cigarette (long) Tadelle candy bar Fresh fruits and vegetables Peppers Tomatoes Cucumber Courgette Scallions Apple Lemons Curly lettuce Parsley Dwelling expenses Electricity services bill House rental Fuel oil Kerosene LPG Firewood Coal House decoration Doorkeeper fee Water services bill Household expenses Washing liquid Soap Cleaning powder Bleach Paper towel Oven Fridge Hob burner Washing machine TV set Radio-casette player Vacuum cleaner Iron
N
Down
No
Up
Duration
2,160 1,080 1,080 72 1,080 2,160 1,080 1,080 1,080 72 72 72 72 1,080
0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
0.57 0.57 0.58 0.61 0.61 0.61 0.62 0.62 0.62 0.63 0.63 0.63 0.63 0.69
0.43 0.43 0.41 0.39 0.39 0.39 0.38 0.37 0.38 0.38 0.38 0.38 0.38 0.31
1.78 1.78 1.86 2.03 2.03 2.04 2.09 2.11 2.11 2.13 2.13 2.13 2.13 2.65
2,016 2,016 2,016 2,016 2,016 2,016 2,016 2,016 2,016
0.35 0.40 0.37 0.41 0.30 0.26 0.29 0.27 0.06
0.04 0.05 0.06 0.08 0.15 0.16 0.21 0.28 0.65
0.62 0.55 0.57 0.51 0.55 0.57 0.5 0.46 0.29
0.30 0.33 0.35 0.40 0.53 0.55 0.64 0.78 2.31
72 432 72 72 1,080 72 72 72 72 72
0.00 0.00 0.01 0.01 0.00 0.00 0.00 0.00 0.00 0.00
0.14 0.24 0.36 0.36 0.39 0.54 0.58 0.83 0.88 0.89
0.86 0.76 0.63 0.63 0.61 0.46 0.42 0.17 0.13 0.11
0.51 0.69 0.98 0.98 1.06 1.63 1.86 5.48 7.49 8.49
2,160 2,160 2,160 2,160 2,160 72 72 72 72 72 72 72 72
0.04 0.03 0.02 0.04 0.05 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00
0.22 0.28 0.30 0.31 0.33 0.35 0.36 0.36 0.38 0.39 0.39 0.39 0.43
0.74 0.69 0.67 0.65 0.63 0.65 0.64 0.64 0.61 0.61 0.61 0.61 0.57
0.66 0.79 0.84 0.86 0.89 0.95 0.98 0.98 1.02 1.06 1.06 1.06 1.19
Table 5.2 Continued
Video player Plastic housewares Broom Replacement bulbs Steel kitchenware Dining set Bedroom set Matress Settee set Carpet Glass kitchenware Sheets Duvet Aliminum kitchenware Towels Blanket Heater Curtains Porcelain kitchenware Box of matches Clothing Men’s tailoring Men’s shoes Women’s shoes Men’s suit Men’s jacket Men’s pants Women’s suit Women’s skirt Men’s shirt Children’s pants Women’s dress Children’s jacket Children’s shoes Women’s blouse Children’s shirt Children’s jumper Women’s tailoring Cotton flannel Men’s shoe repair Women’s sweater Batiste Men’s pullover Poplin Woolen cloth (a) Women’s handbag Women’s casual dress Children’s sweater
N
Down
No
Up
Duration
72 2,160 2,160 2,160 1,080 72 72 72 72 72 144 1,080 1,080 1,080
0.00 0.00 0.00 0.03 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.02 0.01
0.44 0.46 0.53 0.57 0.63 0.64 0.64 0.64 0.65 0.67 0.69 0.69 0.7 0.7
0.56 0.54 0.47 0.4 0.36 0.36 0.36 0.36 0.33 0.33 0.31 0.31 0.28 0.29
1.23 1.29 1.55 1.76 2.19 2.23 2.23 2.23 2.34 2.47 2.67 2.68 2.81 2.85
1,080 72 72 1,080 1,080
0.00 0.00 0.00 0.00 0.00
0.71 0.72 0.72 0.77 0.79
0.29 0.28 0.28 0.23 0.21
2.92 3.07 3.07 3.8 4.28
2,160
0.00
0.88
0.12
7.7
936 936 937 864 864 864 864 864 864 864 864 864 936 864 864 864 937 72 936 864 72 864 72 1,008 865 864 864
0.00 0.13 0.14 0.16 0.16 0.15 0.16 0.16 0.14 0.13 0.16 0.14 0.11 0.14 0.12 0.11 0.00 0.00 0.00 0.12 0.00 0.11 0.00 0.00 0.03 0.04 0.08
0.64 0.65 0.65 0.65 0.66 0.66 0.66 0.66 0.66 0.66 0.67 0.67 0.68 0.68 0.69 0.7 0.72 0.74 0.74 0.75 0.75 0.76 0.76 0.77 0.77 0.77 0.77
0.36 0.22 0.21 0.19 0.18 0.19 0.18 0.18 0.19 0.21 0.18 0.19 0.21 0.19 0.19 0.19 0.28 0.26 0.26 0.13 0.25 0.13 0.24 0.23 0.2 0.19 0.15
2.22 2.29 2.31 2.33 2.38 2.4 2.41 2.42 2.45 2.45 2.46 2.5 2.56 2.56 2.67 2.85 3.05 3.26 3.33 3.46 3.48 3.63 3.71 3.75 3.76 3.78 3.89
(Table 5.2 continued )
Disaggregated analysis of price dynamics 137 Table 5.2 Continued
Linen Men’s sweatsuit Children’s casual clothing Women’s overcoat Terylene Men’s socks Cotton cloth Women’s nightgown Women’s slippers Woolen cloth (b) Men’s underwear Women’s socks Children’s socks Children’s pyjama Women’s lingerie Children’s slippers Men’s pyjama Children’s underwear Cultural and educational expenses Movie ticket Book (fiction) Notebook Pen Private lectures Museum fee School book Theatre ticket Football ticket Miscellaneous expenses Child allowances Gifts Religious expences Lawyer fee Funeral costs
N
Down
No
Up
Duration
1,008 864 864 864 2,088 864 72 864 937 1,008 864 865 864 864 864 936 864 864
0.00 0.04 0.03 0.10 0.00 0.01 0.00 0.02 0.02 0.00 0.01 0.00 0.01 0.02 0.01 0.01 0.02 0.00
0.77 0.78 0.78 0.78 0.79 0.79 0.79 0.79 0.8 0.8 0.8 0.81 0.81 0.81 0.81 0.82 0.82 0.82
0.22 0.19 0.19 0.12 0.21 0.2 0.21 0.18 0.18 0.2 0.19 0.19 0.18 0.17 0.17 0.17 0.16 0.18
3.92 3.93 3.96 4.05 4.18 4.28 4.28 4.31 4.39 4.4 4.48 4.63 4.72 4.82 4.85 4.89 5.02 5.02
72 72 1,080 1,080 72 72 72 72 72
0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.00
0.81 0.85 0.85 0.87 0.89 0.89 0.90 0.90 0.92
0.19 0.15 0.14 0.12 0.11 0.11 0.10 0.10 0.08
4.62 6.03 6.37 7.31 8.49 8.49 9.78 9.78 11.49
1,080 1,080 72 72 72
0.00 0.00 0.00 0.00 0.00
0.48 0.67 0.90 0.92 0.96
0.52 0.32 0.10 0.08 0.04
1.37 2.54 9.78 11.49 23.50
duration less than the mean of 3 months. Yet, a quarter of products’ prices last more than 4 months despite a high level of inflation.12 Our findings regarding prices stickiness in Turkey can be compared with findings from other countries. Most of the earlier empirical research reported less frequent price changes. For example, Blinder et al. (1998) surveying 200 firms found that median price change is about once a year. Cecchetti (1986) studied newsstand prices and found that the duration of prices is varying between 1.8 and 14 years and Kashyap (1995) reports mail-order catalog prices do last less than 15 months. Noting that these studies have employed US data where inflation is considerably lower than in Turkey, and longer duration of prices is expected. The only study where duration of nominal prices is investigated in a high inflationary
138
Mustafa Çaxlayan and Alpay Filiztekin 50 45
46
44
41
40 35 30
25
25 20
17
15 10
6
5
8 4
4
4
3 0
2 (12, ]
(11,12]
(10,11]
(9,10]
(8,9]
(7,8]
(6,7]
(5,6]
(4,5]
(3,4]
(2,3]
(1,2]
( ,1]
0
Figure 5.2 The distribution of price durations.
environment is by Lach and Tsiddon (1992). They have reported mean duration of 2.2 months in Israel between 1978 and 1979 when monthly inflation was 3.9 percent per month. The mean duration in Israel, however, dropped to 1.5 months during 1982 when inflation soared to 6.6 percent. A recent paper by Bils and Klenow (2002), that uses a broad range of consumer goods as we do here, reported much more frequent price changes in the United States than early research as well. They found that the median duration of prices is around 4.3 months. Given our data set, we find significant differences in duration of prices across broad categories. In Table 5.3, we depict summary of our findings for several subgroups within each category. While prices of food items change most frequently; around 60–80 percent of food item prices do change from one month to the other; over 70 percent of service prices remain unchanged. Interestingly, the prices of food products that are regulated by the state (cigarettes strictly, bread market relatively more loosely) have relatively longer price duration than other food products. Given high average monthly inflation rate in our sample period, long duration of price quotations implies significant real cost due to keeping the prices unchanged. For an average good, the real price of a product erodes around 14.5 percent before the vendor changes its price tag. It is also generally predicted that relative prices of goods are not constant. Therefore, in columns 5 and 6 of Table 5.3, we provide average inflation and cost of holding the nominal prices constant for each category, respectively. The cost varies around 5.5 percemt (for foodstuff) to 30 percent (for services). The figure is around 25 percent for housing which constitutes the largest item in total household expenditure. Notice that there is no significant linear relationship between product specific average inflation and average duration of price quotations. The best fit obtained with a quadratic formulation can only explain a quarter of variation in duration times.
Disaggregated analysis of price dynamics 139 Table 5.3 Within group duration of prices Educational expenditure Foodstuff Bread and cereals Meat products Milk, egg and oil Dried beans Nuts Fresh fruits Fresh vegetables Precooked food Cigarettes and drinks Restaurant bill Dwelling Expenses House rental Enerji Household expenses Household goods Household electronics Cleaning products Kitchenware Clothing Male Female Children Cloths Health and personal care Health Personal care Transport and communication Transportation Communication Cultural and educational Educational expenditure Cultural activities Miscellaneous Miscellaneous expenditure
Down
No
Up
Avgerage duration
Avgerage inflation
Cost
1.88 6.14 6.39 13.31 1.60 27.50 30.88 1.84 0.09 0.09
40.54 25.10 27.93 32.97 47.72 18.65 18.59 45.19 59.96 62.22
57.58 68.76 65.69 53.72 50.68 53.84 50.53 52.97 39.95 37.69
1.14 0.73 0.81 0.91 1.37 0.60 0.71 1.35 1.98 2.11
4.54 4.66 4.74 4.46 4.85 4.74 4.25 4.50 4.85 4.73
5.15 3.41 3.87 4.06 6.58 2.84 3.15 6.04 9.68 9.96
0.00 0.40
64.81 46.63
35.19 52.98
4.56 2.22
5.54 4.49
24.08 9.61
0.43 0.14 2.65 0.22
68.27 42.08 42.70 65.56
31.31 57.78 54.65 34.22
2.68 1.26 1.88 2.66
3.90 4.22 4.51 4.26
10.24 5.20 8.51 11.22
7.43 7.86 5.72 0.04
72.37 73.62 75.93 76.85
20.20 18.52 18.34 23.11
3.34 3.46 3.88 3.83
4.62 4.65 4.48 3.90
15.27 15.93 17.31 14.88
0.00 0.50
85.80 60.54
14.20 38.96
8.60 2.36
4.69 4.23
40.41 10.06
0.33 0.00
72.26 86.81
27.41 13.19
5.07 7.09
4.83 4.71
23.60 33.79
0.28 0.53
87.96 80.83
11.76 18.63
7.99 5.93
4.85 4.57
39.52 27.53
0.11
78.67
21.22
9.73
4.34
34.90
Certainly these figures are quite large. Moreover, our calculations are based on both increases and decreases of the prices of the products. If we were to restrict ourselves to upwards movements to calculate the duration of price changes, these figures would have been even larger. Yet, we are not alone in observing high real price erosion before any price adjustments are made. These findings suggest existence of nominal price rigidities despite high inflation. Hence, given that the
140
Mustafa Çaxlayan and Alpay Filiztekin Table 5.4 The duration of prices across vendors
Table 4A All goods Foodstuff Bakkal Supermarket Pazar Table 4B All common goods Bakkal Supermarket Pazar
Down
No
Up
Average duration
3.90 6.95 5.36 5.34 16.18
58.23 38.67 41.15 34.77 23.86
37.87 54.38 53.49 59.89 59.95
3.09 1.18 1.30 1.12 0.74
5.67 4.29 8.23 4.31
31.70 36.44 28.82 29.53
62.63 59.27 62.94 66.16
0.89 1.03 0.81 0.84
Notes Common goods across vendors are: rice; pasta; flour; cracked wheat; dried beans; lentils; chic peas; feta cheese; olives; vegetable oil; eggs; tomato paste; honey.
products are priced far from their competitive values, we can conclude that there will be significant real societal losses as a whole. We also investigate the behavior of price duration across different vendor types to see if there are any systematic differences across vendors in their price adjustments. In particular we would like to see if bakkals or supermarkets are slower in changing the nominal prices. Sellers at pazars are expected to adjust their prices most frequently as menu cost for these vendors is the least. Also, one may suggest that Pazar vendors act closest to those operating in competitive markets requiring prices in pazars to be very close to marginal cost of production. Whereas, bakkals and supermarkets may charge prices above their marginal cost for they would have some monopoly power. Hence, we compute the ratios of price changes and average duration of products prices for each type of vendors where predominantly food items are sold. The results are given in Table 5.4. The table confirms our expectations regarding the duration of prices in pazars. The frequency of price changes is higher in pazars. It must also be noted that price reductions are most frequently observed for this vendor type. The longest duration is observed for bakkals. Interestingly price reductions are less frequent in supermarkets, as one would expect supermarkets to have more involved in promotional activities. While we cannot verify actual market power of these different vendors, it is clear that there are significant differences.
The synchronization of price adjustments In the earlier subsection, we showed that on average prices in Istanbul change every 2–3 months despite an average monthly inflation of 4.7 percent implying
Disaggregated analysis of price dynamics 141 a significant erosion of prices in real terms; the average vendor experiences an average loss of 14.5 percent due to holding prices unchanged on average by 3 months. In such an environment, it becomes important to determine the timing as to when a vendor should change the price tag of a particular product. Since many of the vendors are selling identical products, that is, vendors are competing with each other, synchronization of price adjustments have additional implications on the profitability of each vendor. In particular, assume that an increase in money supply leads to a rise in demand for products sold. In response, one would expect that all product prices would increase so that overall there would not be any changes in the relative prices. However, not all vendors change their prices simultaneously for various reasons including informational asymmetries and search costs. Thus, the impatient seller may loose a portion of his customer base and may incur losses. Moreover, lack of synchronization of pricing across vendors leads to a protracted response of prices to changes in monetary policies. No doubt, when there is a significant time lag between the implementation of the monetary policy and its effects on the economy as a whole it becomes more difficult to determine the effectiveness of monetary policies. In some models of optimal money supply rule, for example, Fisher (1977), the lag in the response of aggregate price level is longer than the duration of price quotes because of lack of synchronization across stores. We therefore investigate the degree of synchronization of price changes for our data set. Although previously we used a methodology allowing us to work on the assumption that prices can adjust instantaneously, in this case we are bound by the fact that our data set is on a monthly frequency. Hence, the discussion on synchronization of prices across vendors should be taken with a grain of salt. Our words of caution particularly applies to fresh vegetables and fruits whose prices change less than (or about) a month. We compute the monthly average price changes for all goods and all types of shops. Figure 5.3 plots the percentage of stores that did not change their prices. A quick glance at this figure shows that in any given month significant amount of stores did not change their prices indicating that price adjustments are staggered. On average, each month 46.8 percent of stores did not change their prices. Should there be synchronization, then we should have observed more ups and downs. Under full synchronization, all prices would increase at the same time or none would change, and that would imply higher standard deviation of price changes. The highest number of stores changing prices is observed in October 1997 when monthly inflation was over 11 percent. Even then, a quarter of the firms did not change their prices. The pattern seems to be (asymmetrically) cyclical. We observe relatively more stores change (increase) their prices in the month of October every year. These results indicate the existence of coordination problems about the timing of price changes across stores in an environment where the annual inflation rate is around 60 percent per annum. Similar findings have been reported by Lach and Tsiddon (1992) for Israel.
142
Mustafa Çaxlayan and Alpay Filiztekin
65.0
55.0
Average = 46.8
45.0
35.0
2000,01
1999,07
1999,01
1998,07
1998,01
1997,07
1997,01
1996,07
1996,01
1995,07
1995,01
1994,07
25.0
Figure 5.3 Monthly percentage of stores with no price changes.
The distribution of prices So far we have shown that vendors, on average, do not change prices very often, neither do they act synchronously. Next we discuss whether individual rigidities imply rigidities at the aggregate level. Using an (S,s) model where initial distribution of real prices are assumed to be uniform, monetary shocks are nonnegative and idiosyncratic shocks are absent, Caplin and Spulber (1987) show that at any point in time firms which adjust their real prices simply offset the deterioration of real prices of all other firms which do not change their prices. Consequently money becomes neutral in equilibrium. Later, Cabellero and Engel (1989), introducing idiosyncratic shocks to the basic Caplin and Spulber (1987) framework, show that the unique stable equilibrium distribution is indeed uniform. As these models suggest, the behavior of aggregate price level and its response to monetary shocks depend on the shape of the distribution of real prices. We define the real price for product i sold by vendor j in neighborhood k during month t as the deviation of log nominal price Pijkt ln Pijkt from the log mean price of product i across all vendors and neighborhoods during that month, Yijkt 1/Nit j,k pijkt . In this formulation, the relative price of a product can be interpreted as the value of the price of an item in a district and shop compared to the prices of the same item in different districts and shops each month. In Figure 5.4 we present the distribution of real prices for all goods over our sample period,13 and in Table 5.5 the corresponding statistics are given for major categories. The distribution of prices is symmetric when all goods are pooled together. The symmetry holds also for food and clothing items, however skewed to the left for most of the services. More specifically, the median of the distribution for each group
Disaggregated analysis of price dynamics 143
–1.5
–1
–0.5
0
0.5
1
1.5
Figure 5.4 The distribution of real prices. Table 5.5 Heterogeneity controlled real price distribution: food items
All goods Food Household Clothing Health Other
Min
5%
10%
25%
50% 75% 90% 95% Max Skew.
0.15 0.10 0.10 0.25 0.12 0.11
0.50 0.36 0.43 0.79 0.42 0.45
0.24 0.17 0.16 0.43 0.20 0.20
0.18 0.13 0.12 0.32 0.14 0.14
0.09 0.06 0.06 0.16 0.07 0.07
0.00 0.00 0.00 0.00 0.01 0.01
0.10 0.07 0.06 0.17 0.07 0.07
0.18 0.12 0.11 0.32 0.14 0.13
0.23 0.16 0.16 0.41 0.18 0.17
0.47 0.34 0.40 0.73 0.44 0.34
Kurt.
0.91 21.71 1.34 34.43 1.11 30.86 0.15 3.46 0.10 4.07 0.67 5.42
coincides with the mean of the distribution, which is equal to zero by definition and the standard deviation is 15 percent on average (it can be as low as 0.2 percent or as high as 35.8 percent for different categories). The table indicates that 50 percent of prices lie within 10 percent of the mean price and 80 percent of prices are within 18 percent of it. For clothing items, the distribution is wider, 50 percent of prices lie within 17 percent of the mean. On average there is a threefold difference between the maximum and minimum price. These are non-negligible differences. It is clear that price dispersion is prevalent and the dispersion differs across different types of product categories. Some of the differences may, however, arise from cross sectional heterogeneity of prices across vendors and neighborhoods. Therefore, the problem can be reformulated and investigated after removing some of the potential sources of heterogeneity. In that regards, we can remove the vendor, neighborhood, and time effects from prices. To achieve our goal, we use a simple error components model to control for observable time invariant vendor, k, and neighborhood, j, as well as for the aggregate month effects, t, as follows: Yijkt i k j t ijkt.
144
Mustafa Çaxlayan and Alpay Filiztekin
The residual of this process, ˆijkt, is the relative price of a product which we want to further investigate. This process controls for all observable sources of heterogeneity and the generated price series can be considered as the price of a homogeneous good. In other words, the generated relative price series yield the net price controlling for all of the characteristics bound with the type of product, the vendor, the neighborhood, and the time the product is sold. We estimate the error components model using ordinary least squares techniques for each product and compute ˆijkt. The regression results suggest that the vendor and the neighborhood effects are significant confirming the presence of cross sectional heterogeneity of prices. The distribution of heterogeneity controlled real prices, ˆjkt, for all products and statistics on this distribution are given in Figure 5.5 and Table 5.6, respectively. A comparison between Table 5.5 and 5.6 shows us that there is still a considerable degree of variation left in our series. Our control variables have explained some of the movements in the prices series, yet the standard deviation has fallen by 2 percent points to 13 percent for all goods only, and the distribution is still uni-modal. Hence, price dispersion and the shape of the distribution cannot be explained by the presence of heterogeneity.
–1.5
–1
–0.5
0
0.5
1
1.5
Figure 5.5 The distribution of heterogeneity controlled real prices. Table 5.6 Heterogeneity controlled real price distribution: all goods
All goods Food Household Clothing Health Other
Min
5%
10%
25%
50% 75% 90% 95% Max Skew.
0.13 0.09 0.09 0.23 0.10 0.10
0.48 0.34 0.41 0.77 0.38 0.40
0.22 0.15 0.15 0.39 0.17 0.17
0.16 0.11 0.11 0.29 0.12 0.13
0.08 0.06 0.06 0.14 0.06 0.06
0.00 0.00 0.00 0.00 0.00 0.00
0.09 0.06 0.05 0.15 0.06 0.07
0.16 0.11 0.11 0.28 0.12 0.13
0.22 0.15 0.15 0.37 0.16 0.17
0.45 0.34 0.38 0.70 0.39 0.33
0.07 0.03 0.06 0.12 0.11 0.22
Kurt. 3.86 3.69 5.00 3.68 3.98 3.92
Disaggregated analysis of price dynamics 145
Conclusion In this chapter, we concentrate on the duration of price quotations, the synchronization of price changes, and the dispersion of prices by employing a detailed disaggregated database. The data set provides us price observations for 242 distinct products on a monthly basis spanning the period July 1994 to June 2000. The data are collected from various types of vendors including pazars, bakkals and supermarkets operating in 15 different neighborhoods (boroughs) of Istanbul. Despite high inflation in Turkey, we find that the nominal prices remain unchanged for a long period of time, the mean duration of price quotations is around three months. This implies an average loss of 14.5 percent in real terms before the nominal price is adjusted. There is no synchronization of price changes either. Furthermore, the distribution of prices is not uniform and the basic assumption of some models of price stickiness is not confirmed. This fact remains true even after we control for heterogeneity across vendors. Taken together, these facts imply that there is price staggering, and monetary policy has to be formulated accordingly.
Notes 1 They also report similar observations for expected and unexpected inflation. 2 One of the main reasons why inflation is more commonly observed in developing countries rather than developed economies may be related to the lack of an effective tax system. Developing countries are more often indulged to use inflation as a means of generating revenue. In the case of Turkey, for example, seigniorage revenue has been close to 3 percent of the Gross National Product (GNP) through 1990s (Küçüker et al., 1994). 3 See for instance Tommasi (1992). 4 Beaudry et al. (2001) show that uncertainty, deteriorating the information content of prices lead to inefficient allocation of investment. 5 These are mostly fresh produce only available in certain months every year. 6 Although relatively few, the price of some of the food items are not available for all three types of vendors or all neighborhoods. 7 Since our database does not contain information on the weights of products in the COL Index calculated by the Chamber, we investigate un-weighted prices changes within each category over time. To our benefit, previous research investigating the behavior of prices on volatility of prices has noted that weighted and un-weighted indices yield similar conclusions. 8 The stability of inflation during the sample period may be significant, since Çaxlayan and Filiztekin (2003) have shown that the empirical link between inflation and dispersion can break down in the presence of large structural breaks. 9 Lach and Tsiddon (1992) and Lach (2002) investigate similarly the behavior of prices during inflationary episodes in Israel. 10 Lach and Tsiddon (1992) also report negative price changes in their sample and interpret their finding with the presence of idiosyncratic shocks affecting stores, which in our case may correspond to seasonal sales. 11 Bils and Klenow (2002) calculate average duration of price quotations in a similar way. The resulting duration for some items, thus, could be less than a month. For instance, duration for Eggs is calculated as 1/ln(1 0.69) 0.42 month. 12 Distribution details for the duration of price quotations are as follows: p10 0.78; p25 1.06; median 2.22; p75 3.92; p90 7.13. 13 In this section we use 135 distinct products whose prices vary across vendors and neighborhoods.
146
Mustafa Çaxlayan and Alpay Filiztekin
References Benabou, R. (1988) “Search, Price Setting and Inflation,” Review of Economic Studies, 55(3), 353–376. Benabou, R. and R. Gertner (1993) “Search with Learning from Prices: Does Increased Inflationary Uncertainty Lead to Higher Markups?,” Review of Economic Studies, 60(1), 69–94. Beaudry, P., Çaxlayan, M., and F. Schiantarelli (2003) “Monetary Instability, the Predictability of Prices, and the Allocation of Investment: An Empirical Investigation Using U.K. Panel Data,” American Economic Review, 79(2), 213–218. Bils, M. and P.J. Klenow (2002) “Some Evidence on the Importance of Sticky Prices,” NBER working paper 9069. Blinder, A., E.R.D. Canetti, D.E. Lebow, and J.B. Rudd (1998) Asking About Prices: A New Approach to Understanding Price Stickiness, New York: Russell Sage Foundation. Caballero, R.J. and E.M.R.A. Engel (1993) “Heterogeneity and Output Fluctuations in a Dynamic Menu-Cost Economy,” Review of Economic Studies, 60(1), 95–119. Çaxlayan, M. and A. Filiztekin (2001) “Nonlinear Impact of Inflation on Relative Price Variability,” Economics Letters, 79(2), 213–218. Çaxlayan, M., A. Filiztekin, and M.T. Rauh (2004) “A Re-examination of the Empirical Relationship between Inflation and Price Dispersion,” University of Leicester WP 04/2. Caplin, A.S. and D.F. Spulber (1987) “Menu Costs and the Neutrality of Money,” Quarterly Journal of Economics, 102, 703–725. Checchetti, S.G. (1986) “The Frequency of Price Adjustment: A Study of the Newsstand Prices of Magazines,” Journal of Econometrics, 31, 255–274. Dana, J.D. (1994) “Learning in an Equilibrium Search Model,” International Economic Review, 35(3), 745–771. Fischer, S. (1977) “Long-term Contracts, Rational Expectations, and the Optimal Money Supply Rule,” Journal of Political Economy, 85, 191–205. Karasulu, M. (1998) “Relative Price Variability and Inflation: Empirical Evidence from Turkey,” ISE Review 2(6), 1–26. Kashyap, A. (1998) “Sticky Prices: New Evidence from Retail Catalogs,” Quarterly Journal of Economics 110, 1–26. Küçüker, C., H. Kazdaxli, and C. Erdemir (1994) Seigniorage in Turkey: Concept and Measurement, Ankara: Hazine Müstesarligi Yayini. Lach, S. (2002) “Existence and Persistence of Price Dispersion: An Empirical Analysis,” Review of Economics and Statistics, 84(3), 433–444. Lach, S. and D. Tsiddon (1992) “The Behavior of Prices and Inflation: An Empirical Analysis of Disaggregated Price Data,” Journal of Political Economy, 100(2), 349–389. Rotemberg, J.J. (1982) “Monopolistic Price Adjustment and Aggregate Output,” Review of Economic Studies, 49, 517–531. Sheshinski, E. and Y. Weiss (1977) “Inflation and the Costs of Price Adjustment,” Review of Economic Studies, 54, 287–303. Tommasi, M. (1992) “Inflation and Relative Prices: Evidence from Argentina,” Optimal Pricing, Inflation, and the Cost of Price Adjustment, in E. Sheshinski and Y. Weiss. (eds) Cambridge and London: MIT Press, 485–511. Van Hoomissen, T. (1988) “Price Dispersion and Inflation: Evidence from Israel,” Journal of Political Economy, 96(6), 1303–1314.
Part 2
The corporate sector
6
Capital structure decisions of Turkish firms Yılmaz Güney, Aydın Özkan, and Kürvat Yalçıner
Introduction Since the seminal work of Modigliani and Miller (1958) the theory on capital structure has successfully investigated the conditions under which the capital structure decision of a firm would be relevant for its value. Among the factors identified in the theoretical literature as relevant for the firm’s capital structure decision are size, expected bankruptcy and agency costs, growth opportunities, profitability, asset structure and non-debt tax shields. Moreover, despite conflicting results regarding some of the firm-specific characteristics, much of the previous empirical work on the subject has confirmed most of the predictions of the theoretical literature.1 However, much of the empirical analysis has been based on companies in the major industrialised countries and mostly in the United States. The evidence provided by these studies suggests that despite the differences in accounting, legal and institutional structures, factors that affect the capital structure choice of firms are similar across developed countries. The main motivation of this study stems from this generally accepted view. More specifically, the approach taken in the present chapter attempts to provide insights into the following questions. First, would the factors which appear to influence the financing decisions of firms in industrialised countries exert similar impact on the capital structure decisions of firms operating in a different environment? Second, if similar factors exert different impact, can we trace why our results are different? Our main contribution to the empirical literature on capital structure lies in three main areas. First, we test the main capital structure hypotheses established in the theoretical literature in a different environment than that in which much of the previous empirical work has been carried out. Turkey, with significantly different legal and regulatory framework, and relatively young stock market, provides an interesting environment for this purpose. Second, it focuses on the dynamics of the capital structure decision, which is another aspect that is lacking in most empirical capital structure studies (notable exceptions, though in different contexts, include Marsh, 1982; Jalilvand and Harris, 1984; Fischer et al., 1989; Özkan, 2001). In so doing, our chapter captures one important aspect of borrowing behaviour of firms in that market imperfections
150
Yılmaz Güney et al.
such as adjustment costs and constraints may prevent firms from adapting to new circumstances and immediately offsetting the effects of random shocks (see, for example, Myers, 1984; Fischer et al., 1989). The inclusion of lagged values of leverage and independent variables in the empirical specification allows for the possibility of delays in response of firms in adjusting their capital structures to changes in explanatory variables. Third, the analysis of this chapter addresses the endogeneity problem in testing the hypothesised relations between leverage and firm-specific characteristics. The endogeneity problem is highly likely to arise in the capital structure framework, because observable as well as unobservable shocks affecting leverage decisions of firms are also likely to affect some of other firm-specific characteristics such as the market value of equity and profitability. It is also possible that observed relations between leverage and its potential determinants reflect the effects of leverage on the latter rather than vice versa. The use of panel data and the Generalised Method of Moments (GMM) estimation procedure allows us to efficiently control the endogeneity problem. It enables us to include firmspecific effects (which accounts for the cross-sectional components of the unobservable shocks) and the time dummies (which control for macroeconomic shocks common to all firms) and to choose more efficient instruments to control for endogeneity. The findings of this chapter reveal important characteristics of firms in Turkey regarding their capital structure behaviour. First, in contrast to previous studies, there is strong evidence that company size exerts a negative impact on borrowing decisions of firms. Second, firms do not seem to adjust to their desired leverage ratios immediately, which might suggest that the costs of adjustment are significant. Finally, in line with the prediction of theory, profitability and asset structure of firms appear to influence leverage decisions. The finding that there is a negative relation between growth opportunities and leverage also confirms the empirical literature. We argue that the distinct features of the Turkish corporate sector, such as a high degree of ownership concentration and the absence of an institutional framework to facilitate sufficient protection for investors, may provide some explanation for these findings. In the following, we first briefly discuss the ownership structures of firms, protection of investors and the main characteristics of the financial system in Turkey. In the section on Firm-specific characteristics impacting leverage decision, we present the testable hypotheses and motivate the choice of firmspecific characteristics which are identified by the previous studies as significant in determining the capital structure decisions of firms. In doing so, in order to facilitate a comparison, we follow Rajan and Zingales (1995) and Booth et al. (2001) which provide international evidence on capital structure for developed and developing countries, respectively. In the section on Data and empirical model, the empirical model is presented and a brief description of our data is given. We then present estimation results for both the dynamic and static leverage models. The final section provides concluding remarks.
Capital structure decisions of Turkish firms
151
An overview of the financial system and ownership structure in Turkey Turkish financial and banking system Since the early 1980s the Turkish economy has gone through a significant transformation from a highly protected state-directed system to a market-oriented free enterprise system. The reform process in the financial sector was initiated as part of this liberalisation programme. The Istanbul Stock Exchange (ISE) was activated in 1986 and it has grown rapidly since then. The number of companies traded on the ISE increased from 80 in 1986 to 288 in 2002. Stock market capitalisation has increased from US$938 to US$33,773 million in the same period. Clearly, the importance of the equity market has increased significantly. There are both private sector and public sector instruments traded in the Turkish Securities Market, the public sector securities having the dominant share. Table 6.1 shows, as of 2002, the share of private sector securities in the value of outstanding securities is only 8 per cent. Similarly, the value of outstanding shares is 9 per cent of the value of outstanding government bonds and bills in 2002. The total amount of outstanding corporate bonds is negligible compared with that of shares, where the value of outstanding corporate bonds was less than 0.02 per cent of the value of outstanding shares in 2000. In light of this evidence, one might argue that the financing decision of firms in Turkey is more reflected in the choice of public (equity) and private (bank loans) finance. The Turkish banking system is based on the universal banking (bank-based) framework that enables commercial banks to carry out traditional depository and lending services as well as investment banking activities. As of 2002, there were 54 banks operating in Turkey, 14 of which were investment and development banks, and the rest are commercial banks. The total value of banking assets was US$130 billion, comprising significant amount of Turkish GNP in 2002. Another major characteristic of the banking sector is the high degree of concentration. At the end of 2002, the total assets of the 5 (10) largest banks amounted to 58 per cent (81 per cent) of the total assets of the banking system. Banks in Turkey have large investment positions in government securities, where they placed 10.5 per cent of their assets in government securities and 22 per cent of their total interest income came from these investments.2 Despite the limited role played by banks in the ownership of firms, one should note at this point that this characteristic of the banking system in Turkey may have important implications for the borrowing behaviour of firms. First, the ability of banks to provide both types of finance, debt and equity, to firms reduces the degree of asymmetric information between firms and banks, and as a result, increases the willingness of banks to engage in borrowing. Second, shareholding by one or more banks in a firm may provide signal to other potential lenders. Moreover, the relatively underdeveloped nature of the long-term market for corporate bonds can reinforce the effectiveness of these factors.
134 15,564
6,888 5.5 6,366 322
8,808 7 4,863 3,598 15,696 12.5
5,514 4.1 4,946 475
8,667 6.4 6,423 1,868 14,181 10.5
1991
110 18,737
1990
15,764 12.2 10,097 4,938 22,858 17.7
7,094 5.5 5,743 195
145 9,922
1992
18,680 13.5 13,122 4,460 26,429 19.1
7,749 5.6 4,931 115
160 37,824
1993
20,207 15.4 6,060 7,919 23,614 18.8
3,407 3.4 2,843 37
176 21,875
1994
26,501 15.3 8,601 10,609 31,467 19.1
4,966 3.8 3,761 40
205 20,565
1995
29,759 19 11,628 14,212 33,865 21.9
4,106 2.9 3,950 25
228 30,329
1996
34,425 20.7 17,440 11,599 38,976 23.9
4,551 3.2 4,441 22
258 61,348
1997
45,358 22 21,997 22,473 52,663 25.5
7,305 3.5 7,256 21
277 33,473
1998
54,563 29.8 46,088 7,579 63,451 34.6
8,888 4.8 8,882 6
285 112,276
1999
60,702 29.3 56,678 3,394 72,030 34.8
11,328 5.5 11,326 2
315 68,635
2000
8,555 4.8 8,555 —
288 33,773
2002
116,294 97,996 68.5 55.2 96,647 73,267 18,955 24,035 126,246 106,551 74.4 60
9,952 5.9 9,952 —
310 47,189
2001
Notes In US$ million except the number of firms. Other private securities that are not in the table are commercial papers, asset-backed securities, bank bills and bank-guaranteed bills, real estate certificates and profit and loss sharing certificates. Other public securities that are not in the table are revenue sharing certificates, forexindexed bills, privatisation bond. The figures of public and private sector securities between 1998 and 2002 were converted from TL to US$ using average exchange rate parity of the respective period.
Source: Capital Markets Board of Turkey 2002 and Istanbul Stock Exchange.
Stock market No. of firms Total market capitalisation Securities Private sector Total % of GNP Shares Corp. bonds Public sector Total % of GNP Govt. bonds T-bills Total % of GNP
Indicators
Table 6.1 Selected indicators
Capital structure decisions of Turkish firms
153
The ownership structure of companies In the following, we examine the ownership structure of companies in Turkey using individual firm data on ownership available for 125 firms in our sample for the year 1997. The main source of ownership data is the ISE publication, Year Book of Companies. We analyse data only for the immediate ownership in the sense that we do not trace backwards to find out who the ultimate controllers of companies are through the network of indirect ownership via corporations, which is beyond the scope of this study. Ownership is classified by families, banks, the State and other companies. The corporation is defined as ‘widely held’ if no shareholders hold at least 10 per cent of the company’s shares. Table 6.2 summarises the findings of our analysis. The striking feature of our analysis is that in 1997 for 94 per cent of the companies there is at least one large shareholder owning more than 10 per cent of shares. Put another way, only 6 per cent of the companies are widely held. Moreover, other firms are the predominant controlling shareholders in Turkey at the 10 per cent cut-off level controlling 57 per cent of corporations. Families account for 18 per cent of dominant shareholdings. The role of the State and banks is relatively minor compared with that of other companies and families. The State and bank holdings account for only 12 and 7 per cent, respectively, which suggests only a limited role played by them in the ownership of Turkish companies. The analysis of the immediate ownership structure of companies in Turkey possibly suggests that the corporate governance mechanism in Turkey is more like the one that is characterised by ownership by large investors, that is, concentrated ownership. Given that an active takeover market is absent in Turkey, concentrated ownership also seems to be the alternative for the control over management. However, we are unable to analyse further the effect of concentrated ownership on the capital structure choices of firms due to limited data availability on the structure of corporate ownership. Table 6.2 Ownership structure of firms in Turkey
Family Banks State Widely held Other firms
Controlling shareholder (%)
Major shareholder (%)
18 7 12 6 57
26 10 16 n.a. 72
Notes A firm is said to have major shareholders if they hold at least 10 per cent of the company’s shares. A firm is defined as widely held if there is no controlling shareholder. There may be more than one major shareholder in firms, in which case the shareholder with the largest share is assigned as the controlling shareholder.
154
Yılmaz Güney et al.
The fact that relatively few firms are widely held in Turkey is consistent with the view that companies in economies with poor shareholder protection are less likely to be widely held (La Porta et al., 1999). La Porta et al. (1998) who study the quality of legal protection for shareholders characterise Turkey as a country with relatively poor shareholder protection. They identify several legal provisions concerning shareholders rights and give countries a score of 1 or 0 for each measure to achieve an aggregate score which ranges from 0 for Belgium to 5 for Canada and the United States. Turkey’s aggregate score is 2 where the sample average is 3. In particular, Turkey fails to impose one-share-one-vote rule, does not allow proxy voting by mail, does not allow cumulative voting for directors and does not have mechanisms of proportional representation on the board, by which minority shareholders would put their representatives on boards of directors. Finally, companies in Turkey are not mandated by law to pay out a certain proportion of their earnings as dividends. La Porta et al. (1998) argue that the mandatory right may be a remedial legal protection for minority shareholders who have fewer legal rights. The earlier analysis suggests that the type of agency conflicts in the Turkish environment with weak legal protection of minority shareholders is more likely to be one of those between the controlling and the minority shareholders. This raises the question of how the controlling shareholders are monitored. We attempt to shed some light on this question by analysing the ownership data further to see if there are other large shareholders in firms that have at least one large shareholder owning more than 10 per cent of shares. As suggested by La Porta et al. (1999), the large shareholders are likely to monitor each other. The examination of the data reported in Table 6.2 reveals that 56 per cent of the companies with controlling shareholders have at least another large shareholder. In 6 out of the 23 family controlled firms there is at least one other large shareholder. It is interesting to note that none of these other large shareholders are other families but mostly other companies. The situation for companies in which other companies are the controlling shareholders is rather different. There is at least another major shareholder in 69 per cent of these companies, and in 17 per cent of the cases the families are the other major shareholders. Banks and the State comprise of 7 and 6 per cent of other large shareholders, respectively, and the remaining are the other companies. In 56 and 33 per cent of companies controlled by banks and the State, respectively, there is at least another large shareholder. These findings suggest that the monitoring of controlling shareholders may not be very effective for companies in Turkey with 44 per cent of controlling shareholders having no potential monitor. This is particularly striking for family controlling shareholders who do not have another large shareholder in the same firm in 74 per cent of the cases. The analysis of creditor rights for Turkey is less conclusive in La Porta et al. (1998). Turkey scores 2 out of 4 where the sample average is 2.3. In particular, the reorganisation procedure imposes an automatic stay on the assets, which prevents secured creditors from gaining possession of their security. Also, managers are allowed to keep their jobs in reorganisation, which further undermines creditors’
Capital structure decisions of Turkish firms
155
power. However, secured creditors are ranked first in the distribution of the proceeds from the liquidation of the bankrupt firm’s assets. Also, there are restrictions to file for reorganisation, where creditor consent is needed. To summarise, the analysis of this section suggests that the corporate governance mechanism in Turkey is characterised by concentrated ownership where other firms and families are the dominant shareholders in companies and the role played by the State and banks is limited; the legal protection of shareholders is insufficient; and more importantly the type of agency conflicts in Turkey is more likely to be one of those between the controlling and the minority shareholders and/or the incumbent and potential shareholders.
Firm-specific characteristics impacting leverage decision Similar to Rajan and Zingales (1995), we focus on four firm-specific characteristics: profitability, tangibility of assets, firm size and market-to-book ratio. This enables us to compare our results to those of studies using similar variables on capital structure in both developed and developing countries. Also, the panel nature of the data prevents us from including variables such as, for example, the return variability in our analysis. Now we discuss the hypothesised impact of each of these four variables on capital structure. Profitability. Myers’ (1984) and Myers and Majluf ’s (1984) pecking order model suggest that firms’ financing decisions follow a hierarchy. Firms prefer retained earnings as their main source of financing investment; the second preference is debt financing, and last comes new equity issuing. In their model, observed capital structure of firms reflects the cumulative requirement for external financing. According to this view, profitability of a firm gives the ability of the firm to use retained earnings over external finance, and one should expect a negative association between the profitability of a firm and its debt ratio. Tangible assets. Tangible assets might fetch higher value than intangible assets in liquidation (Shleifer and Vishny, 1992; Alderson and Betker, 1995). This is because they have alternative uses and hence are good candidates for debt finance, since creditors can take the assets away and redeploy them when the firm defaults (Williamson, 1988). Also, tangible assets can easily be used as collateral when firms raise debt (Titman and Wessels, 1988; Harris and Raviv, 1991). Firms with proportionately more tangible assets also find it difficult to shift to riskier projects when specific assets secure their debt ( Detragiache, 1994; Johnson, 1997). All these arguments suggest a positive relation between asset tangibility and financial leverage. Size. The impact of the size of a firm on its leverage decision is not clear-cut. To the extent that size is an inverse proxy for the degree of informational asymmetry between insiders in a firm and the outside investors, a negative relation should be expected between size and leverage. This is because a lower degree of informational asymmetry makes large firms more capable of issuing equity
156
Yılmaz Güney et al.
which is more informationally sensitive compared to debt. However, a number of authors have suggested that the size of a firm is positively related to its debt ratio. It is argued that larger firms are better diversified (Titman and Wessels, 1988), have easier access to the capital markets, and borrow at more favourable rates (Ferri and Jones, 1979). There are also studies which use size as an inverse proxy for the probability of bankruptcy (Rajan and Zingales, 1995). A further reason for smaller firms to have potentially lower leverage ratios might be that smaller firms are more likely to be liquidated when they are in financial distress (Özkan, 1996). Growth opportunities. Myers (1977) suggests that the amount of debt issued by a firm is lower when its investment opportunity set consists of growth options which would increase the firm’s value when undertaken. It is argued that debt financing may give shareholders sub-optimal incentives to pass up some of these valuable investment opportunities in some states of nature. Moreover, growth opportunities cannot be collateralised and do not generate current income. They are intangible in nature and more valuable when the firm is alive. Their value will fall precipitously if the firm faces bankruptcy, which suggests that expected bankruptcy costs for firms with greater growth opportunities will be higher (Myers, 1984; Williamson, 1988; Harris and Raviv, 1990). Larger expected bankruptcy costs would in turn imply lower financial leverage.
Data and empirical model To test the hypotheses of capital structure we use data for a sample of 227 publicly traded firms in Turkey during the period 1990–2002. Our primary source is company balance sheets and income statements data collected from the Year Book of Companies ( published by the ISE Listing Department).3 The initial sample consists of all firms for which data are available on this database which covers all the publicly traded companies. However, the panel nature of our data restricted us in using all the firms in the initial sample. The panel data of the analysis has been constructed as follows. First, firms which operate in the financial sector and the regulated industries were excluded. Second, missing firm-year observations for any variable in the model during the sample period were dropped. Finally, from these firms, only those with at least four continuous time series observations during the sample period have been chosen. These criteria have provided us with a total of 227 firms, which represents 2,462 firm-year observations. Leverage. We use three alternative measures for the dependent variable, leverage ratio (LEV). The first measure is defined as the ratio of long-term debt to book value of total assets. The second is the ratio of long-term debt to book value of total assets minus the book value of equity plus the market value of equity. Finally, the third measure is defined as the ratio of total debt to book value of total assets where total debt includes short-term as well as long-term liabilities of firms. Profitability. The firm’s profitability (PROFIT) is measured by the ratio of earnings before interest, tax and depreciation (EBITD) to total assets. This measure is in
Capital structure decisions of Turkish firms
157
line with others in the literature (see, for example, Titman and Wessels, 1988; Whited, 1992). Tangible assets. The fixed asset ratio (FIXAST) (net total of land and buildings, plant and machinery, construction in progress and any other fixed assets divided by total assets) is used as a proxy for the tangibility of the firm’s assets. Size. Firm size (SIZE) is measured by the natural logarithm of sales in constant 1990 prices. This measure is in line with similar studies in this area (see, for example, Titman and Wessels, 1988; Whited, 1992; Rajan and Zingales, 1995). Growth opportunities. Similar to other studies (Smith and Watts, 1992; Whited, 1992; Barclay and Smith, 1995a,b; Rajan and Zingales, 1995), we use the marketto-book ratio (MKTBOOK), defined as the ratio of book value of total assets minus the book value of equity plus the market value of equity to book value of total assets, as a proxy for the growth opportunities in a firm’s investment opportunity set. Table 6.3 presents the descriptive statistics for the sample firms. It reveals that, based on total debt, the average leverage ratio is 54.8 per cent. Also, this ratio remains relatively stable across the years, ranging from 49 to 61.6 per cent (not reported). The indebtedness of firms in Turkey is lower compared to those in developed countries. Rajan and Zingales (1995) report that the average ratio of total liabilities to total assets in G-7 countries is 65.5 per cent in 1991. Our finding is in line with that of Booth et al. (2001) who report a leverage ratio of 59.1 per cent for 45 firms for the period 1983–1990. This ratio is higher than the leverage ratio of firms in 6 out of the 9 developing countries in their study for a similar period. The long-term leverage ratio is substantially lower at book and market values, 12.9 and 7.7 per cent respectively. This is also consistent with the findings of the previous studies. For example, Demirgüç-Kunt and Maksimovic (1999)
Table 6.3 Descriptive statistics
LEV1 LEV2 LEV3 PROFIT FIXAST SIZE MKTBOOK No. of firms
Mean
Std. dev.
Min.
Median
Max.
0.1298 0.0770 0.5477 0.3159 0.3060 14.907 4.5808 227
0.1414 0.0984 0.2266 0.2239 0.2002 2.5002 10.131 227
0.0000 0.0000 0.0126 1.4353 0.0000 4.0099 0.3712 227
0.0893 0.0431 0.5471 0.2947 0.2751 15.059 1.8451 227
1.0000 0.8145 1.0000 1.9581 0.9973 22.503 89.749 227
Notes LEV1 is the ratio of long-term debt to total assets. LEV2 the ratio of long-term debt to total assets minus the book value of equity plus the market value of equity. LEV3 is the ratio of total debt to total assets. PROFIT is the ratio of earnings before interest, taxes and depreciation to total assets. FIXAST is the ratio of fixed assets to total assets. SIZE is the natural logarithm of firm’s sales (in TL million). MKTBOOK is the ratio of book value of total assets minus the book value of equity plus the market value of equity to book value of total assets.
158
Yılmaz Güney et al.
report that the difference between total debt and long-term debt ratios in developing countries are more pronounced than in the developed countries. Booth et al. (2001) also find similar results for developing economies. In the next section we report alternative estimates of the following specification: LEVit [LEVi,t1, PROFITit, FIXASTit, SIZEit, MKTBOOKit] i t it ,
(6.1)
where LEVit is the leverage ratio of firm i in year t, and allows for lags in regressors. The presence of lagged variables in the empirical specification allows for the possibility of delays in response of firms in adjusting to new circumstances. In particular, the presence of the lagged dependent variable makes allowance for the adjustment of the dependent variable to the desired capital structure. Our specification does not provide much insight into the precise nature of the adjustment pattern to the long-run equilibrium that is hidden in our specification. However, as discussed in the next section, it is possible to derive long-term multipliers for the explanatory variables, which in turn depend on the coefficient of the lagged dependent variable. In this sense this coefficient determines the rate at which the effect of a change in an explanatory variable will die out. Furthermore, the inclusion of the lagged dependent variable in the analysis can be justified by the existence of adjustment costs that makes it costly for firms to adjust to new circumstances causing the current capital structure not to be immediately adjusted to a new desired capital structure (for a discussion see, for example, Myers, 1984; Fischer et al., 1989). This aspect of the capital structure relation is also acknowledged in a number of empirical studies. For example, Rajan and Zingales (1995) attempt to account for the slow adjustment of leverage by averaging the explanatory variables over four years. i and t represent time-invariant firm-specific effects and firm-invariant time effects, respectively. It is assumed that firm-specific effects are not observable but have a significant impact on the firm’s capital structure decisions. They change across firms but remain fixed for a given firm through time. On the other hand, t varies through time but is the same for all firms at a given point in time. t captures mainly economy-wide factors that are outside the control of firms. There are several important estimation problems that arise in the dynamic panel data specification we employ in this chapter. First, in the presence of unobservable firm-specific effects, the OLS coefficients will be biased when regressors and unobservable firm-specific effects are correlated. Moreover, OLS will result in inconsistent parameter estimates since LEVi,t 1 will be correlated with i which does not vary through time. An obvious solution to these problems is to eliminate the firm-specific fixed effects i by taking first-differences. However, OLS still does not consistently estimate the parameters because the first-difference transformation introduces correlation between the lagged dependent variable and differenced errors, that is, LEVi,t 1 and it are correlated through terms LEVi,t 1 and i,t 1.
Capital structure decisions of Turkish firms
159
Second, it is unlikely that the firm-specific variables are strictly exogenous. That is, shocks affecting capital structure choices of firms are also likely to affect some of the regressors such as market value of equity, profitability and size. For example, if leverage of a firm increases one could then observe a negative relation between leverage and the market-to-book ratio assuming that leverage decreases the firm’s market value since it increases its financial risk. However, this would simply be due to the effect of leverage on the ratio under investigation, rather than vice versa.4 Third, the regressors in the analysis including the lagged dependent variable may be subject to random measurement errors, which also induce biases in the estimates. This problem may be particularly relevant for all indicators in our analysis that are calculated from company balance sheet data and market data (see Baltagi, 2001, for a detailed discussion of measurement error and panel data). The problems outlined here advocate the use of an Instrumental Variables (IV) estimation method, where the lagged dependent variable and endogenous regressors are instrumented using an appropriate set of instrumental variables. This chapter therefore employs the GMM method of estimation which provides consistent parameter estimates by utilising instruments that can be obtained from the orthogonality conditions that exist between the lagged values of the variables and disturbances (see Arellano and Bond, 1991). It allows both for an MA (1) error structure and the heteroscedasticity of the disturbances across firms in the sample. It is obviously subject to an optimal choice of instruments where the validity of instruments depends on the absence of higher-order serial correlation in the idiosyncratic component of the error term. It is, therefore, essential to test the assumption of the absence higher-order serial correlation. To this end, a test for the second-order serial correlation is reported. In this context, we also report the statistic for the Sargan test of over identifying restrictions, indicating whether the instruments and the residuals of the equation to be estimated are independent.
Results In the following, we first report the results of alternative estimates of the preferred dynamic leverage model. We then report the results for the static capital structure model for the purpose of comparison wherein all the lagged variables are dropped. Estimation of the dynamic leverage model Table 6.4 reports the results of alternative estimates of the dynamic leverage model. Model (1) gives OLS estimates of leverage equation in levels where the firm-specific effects remain and the lagged leverage is treated as exogenous. Models (2) and (3) present GMM estimates in levels and first differences, respectively. In both specifications, all variables except the lagged leverage are treated as exogenous where in the GMM-Levels specification fixed effects remain.
160
Yılmaz Güney et al.
Table 6.4 Alternative estimates of dynamic capital structure model Independent variables
LEVi,t 1 PROFITi,t PROFITi,t 1 FIXASTi,t FIXASTi,t 1 SIZEi,t SIZEi,t 1 MKTBOOKi,t MKTBOOKi,t 1 Correlation 1 Correlation 2 Wald test 1 (df ) Wald test 2 (df ) Wald test 3 (df ) Sargan test (df )
Dependent variable: LEVi,t (1) OLS
(2) GMM (Levels)
(3) GMM (Differences)
0.7668*** (0.0374) 0.0208* (0.0125) 0.0056 (0.0137) 0.0878*** (0.0184) 0.0653*** (0.0191) 0.0020 (0.0038) 0.0018 (0.0047) 0.0013 (0.0007) 0.0003* (0.0004) 2.024 0.7113 690.5 (9) 63.63 (11) 627.9 (30) —
0.8662*** (0.3192) 0.6012 (1.3520) 0.0358 (0.3790) 0.8250* (0.5008) 0.7650 (1.2150) 0.0121 (0.1576) 0.0775 (0.1489) 0.0216** (0.0104) 0.0165* (0.0100) 3.692 2.619 84.33 (9) 21.12 (11) 33.39 (30) 11.54 (12)
0.3347*** (0.1245) 0.0755 (0.1193) 0.0055 (0.1022) 0.0548* (0.0330) 0.1028 (0.1467) 0.0216 (0.0268) 0.0123 (0.0220) 0.0038 (0.0058) 0.0050* (0.0030) 3.377 0.4423 42.73 (9) 29.05 (11) — 56.15 (57)
Notes LEV is the ratio of long-term debt to total assets minus the book value of equity plus the market value of equity. PROFIT is the ratio of earnings before interest, taxes and depreciation to total assets. FIXAST is the ratio of fixed assets to total assets. SIZE is the natural logarithm of sales. MKTBOOK is the ratio of book value of total assets minus the book value of equity plus the market value of equity to book value of total assets. Time dummies are included in all models. Industry dummies are included only in models 1 and 2. Model 1 gives OLS estimates in levels. Models 2 and 3 give the GMM estimates in levels and first differences, respectively, where LEVi,t 2, and further lags are used as instruments. Asymptotic standard errors robust to heteroscedasticity and small sample bias are reported in parentheses. Correlation 1 and 2 are first and second order autocorrelation of residuals, respectively; which are asymptotically distributed as N(0,1) under the null of no serial correlation. Sargan Test is test of the overidentifying restrictions, asymptotically distributed as 2(df ) under the null of instruments’ validity. Wald tests 1, 2 and 3 test the joint significance of estimated coefficients, of time dummies, and of industry and time dummies, respectively; asymptotically distributed as 2(df ) under the null of no relationship. The coefficients of intercept terms are not reported. The estimation period is 1991–2002 in models 1 and 2, and 1992–2002 in model 3. ***, ** and * indicate coefficient is significant at the 1, 5 and 10 per cent level, respectively.
The estimation period is 1991–2002 for the estimates in levels and 1992–2002 in the first differenced form as a result of losing two cross-sections in constructing lags and taking first differences. In all models the sample contains 227 firms and 2,462 firm-year observations.
Capital structure decisions of Turkish firms
161
Turning to the estimation results, although the estimates show similarities in terms of the estimated coefficients and the test-statistics, under the first two forms in Table 6.4, some of the test statistics show evidence for the existence of misspecification. First, the assumption of serially uncorrelated errors is violated in both the OLS and GMM-Levels specifications as revealed by the serial correlation tests. This can be seen as an indication of the presence of unobservable firm-specific effects which are time-invariant. Second, there is an upward bias on the coefficient of the lagged dependent variable. The estimated coefficients of the lagged leverage variable are 0.767 and 0.866 respectively, both significant at the 1 per cent level, compared to 0.335 under the first-differenced form of GMM.5 This is not surprising since the lagged dependent variable is expected to be biased upwards due to correlation with the unobservable fixed effects. We accordingly conclude that the level specification is inappropriate to estimate the dynamic leverage model and, in the following GMM specifications, we control for fixed effects by first differencing. Moreover, comparing the estimates in levels and differences reveals significant differences, which may again indicate that correlated individual effects are omitted from the regression in levels. Under the first-differenced GMM specification, as expected there is evidence for negative first-order serial correlation, whereas Correlation 2 test suggests that second-order serial correlation is absent. Moreover, the Sargan test indicates that the instruments used in the GMM estimation are valid, that is, the instruments used are not correlated with the error term. The results generally seem to be in line with the predictions of the theory discussed in the previous section. However, as we discussed earlier, we are reluctant to treat regressors as strictly exogenous which would imply that the current values of the explanatory variables are not influenced by past values of the dependent variable and errors. We instead allow for the possibility that current and past values of the regressors may be uncorrelated with current idiosyncratic shocks, but there may be feedback effects from the lagged dependent variable or past shocks to current and future values of the explanatory variables. The latter would be a less strict assumption than assuming that leverage decisions of firms have no effect on explanatory variables at any point in the past. Therefore, in the following table we restrict our attention to GMM estimates where the dependent and explanatory variables are assumed to be endogenous and t 2 and further lags are used to instrument them. In Table 6.5 the results for the GMM estimates for the three different measures of leverage ratio are reported. In the second column of the table, leverage is defined as the ratio of long-term debt to total assets. In the third column, we replace the book value of equity by its market value the denominator which is given by total assets minus the book value of equity plus the market value of equity. Finally, in the fourth column, we define leverage as the ratio of total debt to total assets. Before discussing the results in detail, we draw two general conclusions. First, there is no evidence for misspecification for any of the three diagnostics. Second, the results show some similarities in terms of the estimated coefficients. This is particularly true for the first and third specifications where the impact of independent variables is very similar except that of the fixed assets ratio.
162 Yılmaz Güney et al. Table 6.5 Alternative GMM estimates of dynamic capital structure model Independent variables
LEVi,t 1 PROFITi,t PROFITi,t 1 FIXASTi,t FIXASTi,t 1 SIZEi,t SIZEi,t 1 MKTBOOKi,t MKTBOOKi,t 1 Correlation 1 Correlation 2 Wald test 1 (df ) Wald test 2 (df ) Sargan test (df ) Sargan test (df )
Dependent variable: LEVi,t Long-term book-debt
Long-term market-debt
Total book-debt
0.4518*** (0.0504) 0.0699* (0.0423) 0.0420** (0.0212) 0.1760*** (0.0679) 0.0474** (0.0241) 0.0191* (0.0112) 0.0075 (0.0167) 0.0024** (0.0012) 0.0012* (0.0007) 5.367 0.139 197.1 (9) 16.54 (11) 205.7 (216) —
0.5099*** (0.0553) 0.0015 (0.0318) 0.0036 (0.0169) 0.0913* (0.0523) 0.0963** (0.0465) 0.0027 (0.0067) 0.0109* (0.0066) 0.0009* (0.0005) 0.0002 (0.0006) 5.653 0.155 157.6 (9) 55.61 (11) 196.7 (216) 11.54 (12)
0.4884*** (0.0435) 0.2662*** (0.0482) 0.0644*** (0.0257) 0.0636 (0.1262) 0.0526 (0.0849) 0.0178* (0.0108) 0.0079 (0.0108) 0.0019** (0.0010) 0.0013* (0.0007) 6.712 0.032 244.5 (9) 96.8 (11) 208.8 (216) 56.15 (57)
Notes Long-term book-debt is the ratio of long-term debt to total assets. Long-term market debt is the ratio of long-term debt to total assets minus the book value of equity plus the market value of equity. Total book-debt is the ratio of total debt to total assets. All variables are treated as endogenous and LEVi,t 2, PROFITi,t 2, FIXASTi,t 2, SIZEi,t 2, MKTBOOKi,t 2, and further lags are used as instruments. The estimation period is 1992–2002. Also see notes to Table 6.4.
Turning to the GMM estimates, the coefficient of the lagged leverage is positive and significant at the 1 per cent level across the three specifications. This finding suggests that the dynamics implied by our model are not rejected and firms adjust their leverage ratios in an attempt to attain their desired debt ratios. However, this adjustment is possibly slow and costly given that the adjustment coefficient, which is one minus the lagged dependent variable’ coefficient, is significantly less than unity in three specifications. One possible explanation for this result would be that the costs of being far away from the desired debt ratio are not significant so that firms may not wish to reach their desired ratios as quickly as possible and/or firms’ leverage ratios are not persistent over time.6 The speed of adjustment coefficient is lower for the long-term market debt (0.490) than that for the total debt which includes both long-term and short-term debt
Capital structure decisions of Turkish firms
163
(0.512). Given that a significant proportion of total debt is short-term debt, this result would imply that issuing short-term debt is easier and possibly costs less than issuing long-term debt. This might be a reasonable explanation of the behaviour of firms with regard to the maturity structure of debt as the main source of corporate debt in Turkey is banks. Evidence emerges in Table 6.5 that profitability of firms exerts a negative influence on their borrowing decisions, that is, profitable firms use less debt. The estimated parameter of current profitability is significantly negative except for the long-term market-debt ratio. The negative impact of profitability is consistent with the view of the pecking order theory which predicts that internal finance is preferred over external finance. The negative relation between profitability and leverage is stronger than the finding of Rajan and Zingales (1995) who find significantly positive impact of profitability only in three countries (United States, Japan and Canada) based on total book-debt ratio. The stronger profitability effect in our study may be an indication of more significant agency and informational asymmetry problems associated with firms in Turkey. On the other hand, the association of leverage with past profitability is significantly positive except for the long-term market-debt ratio. This contradicts the view that the relation between past profitability and leverage should be negative because past profitability can be viewed as proxy for higher future growth opportunities, which are intangible assets whose value can substantially decrease in financial distress (Smith and Watts, 1992; Shyam-Sunder and Myers, 1999). One plausible explanation to the positive impact is that with a background of high profitability as a past performance it may be relatively easy for companies to raise external capital such as debt. This finding supports Jensen’s (1986) view that the disciplinary role of debt may lead to a positive relation. We interpret this direct relation as surprising given the absence of an active market for corporate control in Turkey. The results with respect to the fixed assets ratio are similar for the long-term book-debt ratio to those for the long-term market-debt ratio. The coefficient estimate on the current fixed assets ratio is significantly positive. This is in line with the view that the greater the proportion of tangible assets on the balance sheet, the more willing would lenders be to supply loans as the likelihood of suffering the agency costs of debt financing decreases, and leverage should be higher. However, to the extent that Turkey can be defined as a bank-based country, the positive impact does not provide support to the suggestion of Berger and Udell (1994) that firms whose relationships with creditors are close should need to provide less collateral, which in turn implies that tangibility may matter less in the bank-oriented countries. Again, the potentially higher costs associated with agency problem and asymmetric information for firms in Turkey would mean that the former effect dominates the latter if, of course, both are in place. Another noteworthy finding is due to the negative (although insignificant) estimated relationship between leverage and the tangibility of assets for the total book-debt ratio. This finding is in line with that of Booth et al. (2001) who find that tangibility tends to be associated with decreases in the total debt which is
164
Yılmaz Güney et al.
defined as the ratio of total liabilities to the sum of total liabilities and net worth. Their result implies that the relationship between the tangibility of the firm’s assets and its capital structure may depend on the maturity of debt. This is consistent with the view of Myers (1977) that the maturity of a firm’s debt should be matched to that of its assets because, by so doing, the so-called underinvestment problem can be curtailed, that is, the hypothesis is that firms with more tangible assets that have longer maturity should have more long-term debt in their capital structures. In order to investigate this further, similar to others in the literature (see, for example, Barclay and Smith, 1995a; Stohs and Mauer, 1996; Özkan, 2000), we estimated a debt maturity structure model where the dependent variable, debt maturity, is defined as the ratio of debt that matures in more than one year to total debt. In addition to the fixed assets ratio, other control variables included in the estimation are profitability, size, and growth opportunities of firms. The results, though not reported here, suggest that there is a significant positive relationship between the debt maturity structure of firms and the tangibility of their assets, which is consistent with the asset maturity argument and our results regarding the relation between the total debt ratio and asset structure. The sign on the coefficient of the lagged fixed assets ratio is negative and significant at 5 per cent for the long-term debt both in book and market values. However, the overall effect is positive for the long-term book debt.7 The estimated coefficient is positive but insignificant for the total debt. Due to the opposite signs of the estimated coefficients of current and past fixed assets ratio we investigated the possibility that the change in the fixed asset ratio rather than tangibility of the firm’s assets may influence leverage but we did not find such evidence. These conflicting signs may also be because of the econometric specification (common factor restricted form) of our dynamic model (see, Blundell and Bond, 1998). The impact of firm size on leverage is similar across different definitions of debt. The estimated coefficient of this variable is negative and significant at the 10 per cent level for the long-term book-debt and total book-debt regarding current size and for the long-term market-debt regarding lagged size. The results are similar when we estimate our model by using the logarithm of total assets as an alternative proxy for the size variable. The finding of a negative relation between size and leverage is not in line with the findings of the previous empirical studies.8 The negative impact is not consistent with the view that large firms might be more diversified and fail less often, so firm size may serve as an inverse proxy for the probability of bankruptcy. It is also not in line with the argument that, to the extent that the size of firms is an inverse proxy for the direct costs of bankruptcy, small firms are expected to borrow less than large firms. The negative sign of this variable is therefore worth discussing further. As also discussed in Rajan and Zingales (1995), size may be seen as an inverse proxy for the degree of asymmetric information between the firm and outside investors. Small firms are more likely to face the asymmetric information problem because their stocks may not be traded as often as the stock of larger firms, and are not followed by many analysts. To the extent that this is the case, one should expect the firm to increase
Capital structure decisions of Turkish firms
165
its preference for equity (which is more sensitive to informational asymmetry) relative to debt financing as its size increases, which implies a negative relation between size and leverage. Our analysis is not able to distinguish between these effects and inconclusive as to why we find a negative relation between size and leverage. The negative size impact is in line with that finding of Booth et al. (2001) for Turkey, who also find inconsistent size effects across countries in their sample. All these findings provide more support to Rajan and Zingales’ (1995) conclusion that why firm size is correlated with leverage is not really clear. The coefficient of growth opportunities (proxied by market-to-book ratio) is significantly positive for the long-term book debt and total debt regarding past market-to-book ratio. This finding is not in line with the traditional view that growth opportunities of firms have a negative impact on leverage. The lagged positive effect might reveal several features of borrowing behaviour of companies in Turkey. First, it might indicate that the type of agency-related costs to which firms in Turkey are exposed are such that growth firms prefer debt over equity financing when they need to raise new finance. Our analysis, in the section on An overview of the financial system and ownership structure in Turkey reveals that legal protection of small (minority) shareholders is weak and monitoring of large shareholders is not effective, both of which imply that equity is less attractive compared to debt when new finance is needed. Second, the universal banking system in Turkey, characterised by shareholding of banks in firms, is likely to reduce the asymmetric information, which would make the supply of debt financing to firms easier. Finally, since firms whose values are largely determined by growth opportunities face greater degree of informational asymmetry and more severe asymmetric information-induced financing constraints they may be expected to have higher leverage ratios in an attempt to signal higher quality (Myers and Majluf, 1984; Gul, 1999). On the other hand, the impact of the presence of growth opportunities on leverage is significantly negative for all dependent variables regarding current market-to-book ratio. This is a common finding in the literature, and it is explained by the argument that high growth companies tend to have high financial distress and agency-related costs. These costs might deter firm managers from raising loan capital, which eventually leads to lower debt ratios. Long-term multipliers As mentioned earlier, one could derive long-term multipliers (which can be interpreted as the cumulative effect of a change in explanatory variables) for the regressors. In Table 6.6, we report the values of such multipliers implied by our empirical specification.9 In general, the long-run effect of the independent variables is similar across different measures of leverage. For example, the effect of a unit increase in the profitability of firms is to reduce the long-term book-debt ratio of firms by 0.05 units. The reductions for the long-term market-debt and total debt are 0.04 and 0.39 units, respectively. Although the long-run effect of a change in fixed assets ratio for the long-term book debt is significantly positive, this effect is negative but insignificant for other leverage specifications.
166
Yılmaz Güney et al.
Table 6.6 Long-run multipliers Independent variables
PROFIT FIXAST SIZE MKTBOOK
Dependent variable: LEVi,t Long-term book-debt
Total market debt
Book debt
0.0508* (0.0318) 0.2346*** (0.0842) 0.0212 (0.0260) 0.0022* (0.0014)
0.0042 (0.0657) 0.0103 (0.0638) 0.0167* (0.0101) 0.0023** (0.0010)
0.3944*** (0.0961) 0.0217 (0.1553) 0.0504* (0.0304) 0.0013 (0.0011)
Notes Long-term book-debt is the ratio of long-term debt to total assets. Long-term market-debt is the ratio of long-term debt to total assets minus the book value of equity plus the market value of equity. Total book-debt is the ratio of total debt to total assets. Also see notes to Table 6.4.
Furthermore, the impact of size in the long-run equilibrium is negative for all leverage measures and significant for the long-term-market debt and total debt ratio. Similarly, the long-run effect of a change in market-to-book ratio on debt ratios is always negative and significant for both long-term debt ratios. Estimation of the static leverage model For the purpose of comparison of our results to those of the studies analysing the determinants of capital structure in a static framework, in Table 6.7, we report the results for the GMM estimates of the static leverage model. For the reasons given earlier in the chapter, we estimate the model in first differences and treat the regressors as endogenous. Confirming the results in Table 6.5, size and profitability exerts a negative impact on debt ratio regarding all leverage measures and the association is significant for long-term book debt and total debt ratios. As expected, the coefficient for fixed assets ratio is positive and significant for both long-term debt ratios. The coefficient of the fixed asset ratio is still positive but insignificant for the total debt. Finally, the negative and significant impact of growth opportunities on leverage still remains for both long-term debt ratios while this correlation becomes positive but insignificant for total debt ratio. Overall, it seems that the results of the static model are consistent with those of the dynamic model.
Conclusions This chapter has investigated the empirical determinants of leverage decisions of firms in Turkey. The empirical analysis we adopt in this chapter has several important distinct characteristics. First, a panel data set for 227 Turkish companies
Capital structure decisions of Turkish firms
167
Table 6.7 Static capital structure Independent variables
PROFITi,t FIXASTi,t SIZEi,t MKTBOOKi,t Correlation 1 Correlation 2 Wald test 1 (df ) Wald test 2 (df ) Sargan test (df )
Dependent variable: LEVi,t Long-term book-debt
Long-term market-debt
Total book-debt
0.0861** (0.0422) 0.3519*** (0.0957) 0.0570*** (0.0182) 0.0012** (0.0006) 3.327 1.695 45.43 (4) 47.71 (12) 161.2 (176)
0.0234 (0.0314) 0.0978** (0.0491) 0.0009 (0.0078) 0.0009*** (0.0003) 3.850 1.170 11.68 (4) 82.48 (12) 192.5 (176)
0.2764*** (0.0604) 0.0627 (0.1066) 0.0517** (0.0221) 0.0009 (0.0007) 3.931 3.462 45.79 (4) 131.7 (12) 190.4 (176)
Notes Time dummies are included in all models. All variables are treated as endogenous and PROFITi,t 2, FIXASTi,t 2, SIZEi,t 2, MKTBOOKi,t 2, and further lags are used as instruments. The estimation period is 1991–2002. ***, ** and * indicate that the coefficient is significant at the 1, 5 and 10 per cent level, respectively. See notes to Table 6.5.
was constructed, which allowed us to adopt a dynamic model. Moreover, the GMM estimation procedure together with panel data analysis enabled us to effectively control for firm-specific fixed effects which are unobservable yet important in affecting financial decisions of firms and to choose more efficient instruments to control for the endogeneity problem. Second, the chapter focuses on the dynamics of borrowing decisions of firms by allowing the slow adjustment of firms of their capital structure to changes in firm-specific characteristics. Last but not least, the analysis of this chapter has been carried out for firms in an environment which is different from that in which much of the previous empirical work has been conducted. We believe that the chapter provides new insights into our understanding of borrowing behaviour of firms in a number of ways. Our investigation has provided the following insights into the empirical determinants of financial behaviour of firms in Turkey. The findings demonstrate the importance of taking into account the dynamics of capital structure decisions of firms. We provide strong evidence of significant dynamic effects, which suggests that there are delays in response of firms in adjusting to new circumstances, which might imply that the costs of adjustment are significant enough to prevent firms from adjusting to their desired ratios completely in the current period. Alternatively, it may mean that the costs of being away from their desired ratios are not as significant. The results seem to provide support to the view that profitability of firms exerts a negative impact on their capital structure. Our results are also in line with the
168
Yılmaz Güney et al.
prediction that there is a positive relation between leverage and tangible assets. This is valid for the long-run equilibrium results since in the short-run the association turns out to be inverse through lagged fixed asset ratios. What is more, the empirical findings support the predictions in the literature that firms with greater growth opportunities have lower leverage ratios. Another salient and interesting finding is that size exerts uncommonly a negative impact on firms’ debt ratios. The distinct features of the Turkish financial system and corporate sector may help explain some of the findings in this chapter. Our analysis suggests that the Turkish model of corporate governance is characterised by a high degree of ownership concentration and the absence of a legal and institutional framework to facilitate sufficient protection for investors. This implies that the type and extent of agency-related costs to which firms are exposed in Turkey may be different. We suspect that the costs of debt financing associated with the agency relation between shareholders and debt holders may not be as significant as those of the agency conflict between the incumbent and potential (new) shareholders. This would in turn provide explanation as to why some of our empirical findings do not confirm what the previous empirical evidence suggests. The analysis of this chapter suggests two lines of future research. First, there are other important aspects of financing choice of firms to investigate in developing countries such as maturity and priority structures of debt and the choice between private and public debt. This line of research would also provide more insights into our understanding of the optimal capital structure decision of firms. Second, it is important to further analyse the precise nature of the adjustment of firms to the long-run target leverage ratio. In particular, the speed of adjustment and its determinants remain to be important issues to be investigated.
Notes 1 For example, Ferri and Jones (1979), Kim and Sorensen (1986), Chung (1993) and Özkan (2001) do not provide evidence for a significant relation between firm size and leverage. Johnson (1998) finds a negative relation between size and leverage. On the other hand, Homaifar et al. (1994) and Titman and Wessels (1988) report results which are consistent with the notion that larger firms have higher debt ratios. There is also conflicting empirical evidence on the relation between profitability and debt ratios. The findings of Kester (1986), Titman and Wessels (1988) and Rajan and Zingales (1995) lend strong support for the negative relationship, whereas Long and Malitz (1985) do not find such a relation between leverage and profitability. 2 Source: The Banks Association of Turkey and Central Bank of Turkey. 3 The source for the market value of equity is ISE Companies: Capital, Dividend and Monthly Price Data. 4 This potential problem is emphasised in several cross-sectional studies. For example, Johnson (1997) and Rajan and Zingales (1995) attempt to reduce this problem by lagging independent variables one period. 5 One should be cautious when comparing the results of alternative estimates at this stage due to the likely endogeneity of firm-specific characteristics. To investigate this possibility,
Capital structure decisions of Turkish firms
6
7
8 9
169
we treated all variables as endogenous in all models. The upward bias of the lagged leverage remained in both OLS and GMM levels specifications. Noting that the adjustment coefficient is restricted to a range between 0 and 1, firms can generally be expected to have two different types of costs: costs of making adjustment to their target ratios and costs of being in disequilibrium. The higher are the costs of being away from the target ratio the closer the adjustment coefficient will be to one or vice versa. The direct impact of the current FIXASTit on the current LEVit is given by the coefficient of FIXASTit, which is 0.176 for the long-term book-debt. The indirect effects of the lagged FIXASTit 1 on the current LEVit is the sum of the coefficient of FIXASTit 1 (say 0) and the product of the coefficients of LEVit 1 () and FIXASTit (1). That is, the indirect effects of FIXASTit 1 given by (0 1* ), is 0.0321 for the long-term book-debt. 1 can be interpreted as the impact via the current leverage and 1* the impact via the lagged leverage. One notable exception is Johnson (1998) who, using data for US firms to examine the relation between leverage and bank debt use, finds that size is negatively related to leverage. The long-run multiplier of a regressor can be derived from (0 1)/(1 ). That is, it is the ratio of the log polynomials (L) and (L) evaluated at L 1.
References Alderson, M.J. and B.L. Betker (1995) ‘Liquidation costs and capital structure’, Journal of Financial Economics, 39, 45–69. Arellano, M. and S. Bond (1991) ‘Some tests of specification for panel data: Monte Carlo Evidence and an application to employment equations’, Review of Economics Studies, 58, 277–297. Baltagi, B.H. (2001) Econometric Analysis of Panel Data, 2nd edition (Wiley, Chichester). Barclay, M.J. and C.W. Smith (1995a) ‘The maturity structure of corporate debt’, Journal of Finance, 50, 609–632. Barclay, M.J. and C.W. Smith (1995b) The priority structure of corporate liabilities, Journal of Finance, 50, 899–917. Berger, A. and G. Udell, (1994) ‘Relationship lending and lines of credit in small firm Finance’, Journal of Business, 68, 351–381. Blundell, R.W. and S.R. Bond (1998) ‘Initial conditions and moment restrictions in dynamic panel data models’, Journal of Econometrics, 87, 115–143. Booth, L., V. Aivazian, A. Demirgüç-Kunt and V. Maksimovic (2001) ‘Capital structure in developing countries’, Journal of Finance, 56, 87–130. Chung, K.H. (1993) ‘Asset characteristics and corporate debt policy: an empirical test’, Journal of Business Finance and Accounting, 20, 83–98. Demirgüç-Kunt, A. and V. Maksimovic (1999) ‘Institutions, financial markets and firm debt maturity’, Journal of Financial Economics, 54, 295–336. Detragiache, E. (1994) ‘Public versus private borrowing: a theory with implications for bankruptcy reform’, Journal of Financial Intermediation, 3, 327–354. Ferri, M.G. and W.H. Jones (1979) ‘Determinants of financial structure: a new methodological approach’, Journal of Finance, 34, 631–644. Fischer, R.H., R. Heinkel and J. Zechner (1989) ‘Dynamic capital structure choice: theory and tests’, Journal of Finance, 44, 19–40. Gul, F.A. (1999) ‘Growth opportunities, capital structure and dividend policies in Japan’, Journal of Corporate Finance, 5, 141–168.
170
Yılmaz Güney et al.
Harris, M. and A. Raviv (1990) ‘Capital structure and the informational role of debt’, Journal of Finance, 45, 321–349. Harris, M. and A. Raviv (1991) ‘The theory of capital structure’, Journal of Finance, 46, 297–355. Homaifar, G., J. Zietz and O. Benkato (1994) ‘An empirical model of capital structure: some new evidence’, Journal of Business Finance and Accounting, 21, 1–14. Jalilvand, A. and R.S. Harris (1984) Corporate behaviour in adjusting to capital structure and dividend targets: an econometric study, Journal of Finance, 39, 127–145. Jensen, M.C. (1986) ‘Agency costs of free cash flow, corporate finance and takeovers’, American Economic Review, 76, 323–339. Johnson, A.S. (1997) ‘An empirical analysis of the determinants of corporate debt ownership structure’, Journal of Financial and Quantitative Analysis, 32, 47–69. Johnson, A.S. (1998) The effect of bank debt on optimal capital structure. Financial Management, 27, 47–56. Kester, C.W. (1986) Capital and ownership structure: a comparison of United States and Japanese manufacturing corporations’, Financial Management, 15, 97–113. Kim, W.S. and E.H. Sorensen (1986) ‘Evidence on the impact of the agency costs of debt on corporate debt policy’, Journal of Financial and Qualitative Analysis, 21, 131–144. La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R.W. Vishny (1998) ‘Law and finance’, Journal of Political Economy, 106, 1113–1155. La Porta, R., F. Lopez-de-Silanes and A. Shleifer (1999) ‘Corporate ownership around the world’, Journal of Finance, 54, 471–518. Long, M. and I. Malitz (1985) ‘The investment financing nexus: some empirical evidence’, Midland Corporate Finance Journal, 3, 53–59. Marsh, P. (1982) ‘The choice between debt and equity: an empirical study’, Journal of Finance, 37, 121–144. Modigliani, F. and M.H. Miller (1958) ‘The cost of capital, corporation finance and the theory of investment’, American Economic Review, 48, 261–297. Myers, S.C. (1977) ‘Determinants of corporate borrowing’, Journal of Financial Economics, 5, 147–175. Myers, S.C. (1984) ‘The capital structure puzzle’, Journal of Finance, 39, 575–592. Myers, S.C. and N.S. Majluf (1984) ‘Corporate financing and investment decisions when firms have information that investors do not have’, Journal of Financial Economics, 13, 187–221. Özkan, A. (1996) ‘Corporate bankruptcies, liquidation costs and the role of banks’, The Manchester School, 64, 104–119. Özkan, A. (2000) ‘An empirical analysis of corporate debt maturity structure’, European Financial Management, 6, 197–212. Özkan, A. (2001) ‘Determinants of capital structure and adjustment to long run target: evidence from UK company panel data’, Journal of Business Finance and Accounting, 28, 175–198. Rajan R.G. and L. Zingales (1995) ‘What do we know about capital structure? Some evidence from international data’, Journal of Finance, 50, 1421–1460. Shleifer, A. and R.W. Vishny (1992) ‘Liquidation values and debt capacity: a market equilibrium approach’, Journal of Finance, 47, 1343–1366. Shyam-Sunder, L. and S.C. Myers (1999) ‘Testing static tradeoff against pecking order models of capital structure’, Journal of Financial Economics, 51, 219–244. Smith, C.W. and R.L. Watts (1992) ‘The investment opportunity set and corporate financing, dividend, and compensation policies’, Journal of Financial Economics, 32, 263–292.
Capital structure decisions of Turkish firms
171
Stohs, M.H. and D.C. Mauer (1996) ‘The determinants of corporate debt maturity structure’, Journal of Business, 69, 279–312. Titman, S. and R. Wessels (1988) ‘The determinants of capital structure choice’, Journal of Finance, 43, 1–19. Whited, T. (1992) ‘Debt, liquidity constraints, and corporate investment: evidence from panel data’, Journal of Finance, 47, 1425–1460. Williamson, O. (1988) ‘Corporate finance and corporate governance’, Journal of Finance, 43, 567–591.
7
Firm-level profitability, liquidity, and investment B. Burçin Yurtoxlu
Introduction Investments in capital equipment and R&D are important determinants of long-run economic performance. Since investment must be financed, issues related to the efficiency of financial markets, corporate finance, and corporate governance are critical determinants of investment behavior of firms. Despite a strong theoretical case established by Modigliani–Miller for the independence of investment and financing decision, a literature dating back some 40 years has found a consistent and often strong relationship between investment spending and liquidity at the firm level. Among many of the proposed explanations1 for this relationship, this chapter will focus on two theories – asymmetric information hypothesis (AIH) (Stiglitz and Weiss, 1981; Myers and Majluf, 1984) and the managerial discretion hypothesis (MDH) or empire building (Grabowski and Mueller, 1972; Jensen, 1986). It is also worthy of note that these two explanations have dramatically different predictions concerning the deviation of observed investment from levels that would maximize firm value. While the AIH explanation points to financing constraints and the resulting underinvestment, MDH results in over-investment. A large number of studies have shown the importance of these two factors by exploiting different characteristics of firms, such as dividend payout ratios, size or ownership structure (See surveys by Chirinko, 1993; Hubbard, 1998). Most of these studies, however, analyze firms from Anglo-Saxon countries, with relatively efficient capital markets and corporate governance regimes that provide the outside investors a strong protection against expropriation by insiders. The purpose of this chapter is to study the determinants of capital expenditures for a sample of publicly listed Turkish firms over the 1990–2002 period. In doing that, the chapter will focus on some specific corporate governance structures of firms that are likely to influence the degree of asymmetric information and the likelihood of pursuing goals other than shareholder wealth maximization. The next section gives a short overview of the trends in the macro level of investment of Turkey. The section on Models of firm level investment decisions, presents two theoretical models of firm-level investment and discusses the issues related to their estimation. In the section on Turkish corporate governance and
Profitability, liquidity, and investment 173 implications on investment, we provide a detailed account of the corporate governance structures in Turkey and relate these to the sensitivity of investment to the availability of internally generated funds. The section on Data and the empirical model, presents the data and the definitions of the variables used in this chapter. The section on The findings, presents the empirical findings and the final section is devoted to the implications of these findings on the growth process of Turkey’s corporate sector. Since Turkish listed firms exhibit most of the salient characteristics of weak corporate governance regimes such as highly concentrated family ownership, divergence of cash-flow and voting rights, pyramidal structures, and business groups, establishing a link from such factors to investment behavior has potential implications beyond the specific country study.
Investment in Turkey The trends in the level and growth rate of investment This section looks at the flow of investment in Turkey. At the risk of encountering data comparability problems Table 7.1 presents OECD figures on the share of gross capital formation in GDP2 from a large number of countries. Over the 1970–1980 and 1980–1990 periods this share is slightly higher than 19 percent for Turkey, which is lower than most of the remaining countries. In the 1990–2001 period the share of investment in GDP rises to almost 24 percent, a figure which is higher than most of the European countries, the group of developing countries and the USA. Do these figures suggest that Turkey invests too little or too much? Without having a proper benchmark it is hard to answer this question. As suggested by Bond (2001) there is a widespread misconception that investment is good and more investment is better. However, investment can be too high as well as too low, a point to which we will return later. Existing evidence suggests that investment leads to higher long-run growth (Levine and Renelt, 1992). However, it is less clear how important investment is in the growth process. For example, Young (1995) shows that capital accumulation played a huge role in the East Asian growth miracles, with a minor role for total factor productivity (TFP) growth. Nelson and Pack (1999) disagree and show that the roles of TFP growth and capital accumulation cannot be disentangled in a general production function without specifying the form of technical progress. Klenow and Rodriguez-Clare (1997), on the other hand, show that the cross-country variation in TFP growth rates accounts for 92 percent of the crosscountry variation in per capita growth. Similarly, Easterly and Levine (2000) find that many stylized facts about growth imply that most growth and income differences are explained by the “TFP residual” not by factor accumulation. Recent research on the causes of the Asian Crisis of 1997 adds a new dimension to this debate. One argument often found in the structuralist interpretations of the Asian crisis is that fragile financial systems resulted from features of weak corporate governance structures such as relationship banking and cronyistic
174
B. Burçin Yurtoxlu
Table 7.1 Share of gross investment in GDP and the growth rate of total gross fixed capital formation Country
Share of gross investment in GDP 1970–1980
Turkey Australia Austria Belgium Canada Denmark Finland France Germany Greece Iceland Ireland Italy Japan Korea Luxembourg Mexico Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States Developing countriesa East Asiaa
1980–1990
1990–2001
19.31 22.16 25.70 21.72 16.60 22.74 30.22 21.94 26.46 30.98 27.13 25.34 24.47 30.10 22.30 22.91 30.10 25.70 20.04 34.85 22.04 23.55 20.45 19.16 17.18 16.53
19.44 23.98 22.80 18.85 19.60 18.82 27.37 20.27 22.46 21.50 21.07 21.15 21.55 28.38 28.82 22.32 25.58 21.58 20.21 28.36 22.96 21.21 19.53 22.25 17.26 17.42
23.66 22.49 24.04 20.64 19.67 19.50 20.29 19.92 22.35 21.25 20.10 20.35 19.77 28.61 33.25 23.61 22.76 21.83 20.34 22.23 26.00 23.72 18.00 21.80 17.19 18.47
20.48 24.79
21.52 26.63
21.62 27.09
Standard deviation of the growth rate of total gross fixed capital formation 1985–1990
1990–2001
18.50 7.34 2.33 5.05 5.67 9.39 7.22 2.29 2.49 4.58 9.95 6.99 1.57 2.56 5.71 13.33 9.30 2.42 2.42 7.06 5.68 2.74 4.97 1.70 6.63 1.37
16.98 6.34 2.80 3.69 6.09 5.92 11.13 4.47 3.92 4.91 13.94 8.88 4.72 3.68 11.47 10.21 13.23 3.24 9.45 8.06 5.01 5.42 8.31 4.39 5.18 4.88
Source: Annual National Accounts provided by the OECD. Notes a From IFC data (Glen and Sumlinski, 1995). All periods are inclusive.
relations between businesses, financial institutions, and the government. According to this view such relations protected the Asian system from market discipline and encouraged over-investment that led to crisis. In support of this view, Johnson et al. (2000) show that measures of corporate governance and in particular the strength of protection offered to minority shareholders explain the extent of the exchange rate depreciation and stock market decline better than standard macroeconomic measures. While we share some of the concerns raised
Profitability, liquidity, and investment 175 3
on this point, the Leitmotiv of this argument that institutional features have an impact on investment performance is an important one, which is supported by recent research on the determinants of investment literature. Another dimension of the investment record of Turkey is reflected in its erratic movement over time. The last two columns in Table 7.1 report the standard deviations of the growth rates of the (real) total gross fixed capital formation over the last two decades. In terms of the volatility of growth rate of investment Turkey occupies the first place in both periods. Financing issues Table 7.2 reports evidence on the degree to which financial factors are perceived to influence the investment decision of firms in Turkey. These figures are taken from the World Business Environment Survey (WBES, 2000), which is a survey of over 10,000 firms in 80 countries that examines a wide range of interactions between firms and the state. Panel A in Table 7.2 reports the general constraints on the operation and growth of businesses in Turkey. For a total of 150 interviewed companies from Turkey the issue of financing is the third most important obstacle after inflation and political instability. In Panel B, high interest rates are reported to be the most important obstacle to financing for all size classes of firms. An interesting finding of the survey is the differential response to the questions by small and large firms. While the existence of special connections to banks or financial institutions are perceived to be relatively important for small firms, it is much less for large firms. Panel C reports the financing patterns of surveyed firms. Almost 50 percent of investment is financed out of internal funds. Interestingly, this ratio is slightly higher for large firms. Medium and large firms make much more use of public sector funds.
Models of firm level investment decisions The neoclassical q-theory of investment Starting with Kuh and Meyer (1957), early research4 emphasized the availability of internal funds as an important determinant of investment. This strand of literature was challenged by Jorgenson (1967) who modeled firm investment as a function of the cost of capital as defined by Modigliani and Miller (1958) instead of cash flow. According to Jorgenson, the cost of capital should be the sole determinant of investment and, in a world where managers maximize the value of their company, all investment projects should be undertaken until the expected return on investment equaled the cost of capital.5 The Modigliani–Miller theorem establishes the separability of firms’ investment and financing decisions implying the irrelevance of different sources of finance to corporate investment decisions. This proposition is based on the assumption of a perfect capital market where different sources of finance are perfect substitutes.
Table 7.2 Evidence from World Business Environment Survey: “Measuring Conditions for Business Operation and Growth” private enterprise questionnaire Question/response (A) General constraints Inflation Policy instability/uncertainty Financing Taxes and regulations Corruption Exchange rate Infrastructure Organized crime/mafia (B) Financing Issues High interest rates Bank paperwork/bureaucracy Collateral requirements of banks/financial institutions Banks lack money to lend Inadequate credit/financial information on customers Need special connections with banks/financial institutions Access to foreign banks Access to specialized export finance Access to non-bank equity/investors/partners Access to lease finance for equipment (C) Sources of finance Internal funds/retained earnings Local commercial banks Family/friends Public sector Equity, sale of stock Leasing arrangement Foreign banks Investment funds/special development finance Supplier credit Other Moneylenders, traditional or informal sources
Total
Small
Medium
Large
3.56 3.49 3.16 3.11 2.88 2.84 2.26 2.12
3.78 3.59 3.22 3.31 3.00 3.15 2.31 2.53
3.57 3.53 3.16 3.23 2.96 2.73 2.36 2.01
2.90 3.05 2.95 2.05 2.22 2.26 1.75 1.26
3.56 2.40
3.64 2.62
3.60 2.38
3.15 1.85
2.32 2.21
2.60 2.12
2.26 2.42
1.75 1.80
2.18
2.44
2.15
1.55
2.10 1.99
2.33 1.94
2.00 2.18
1.79 1.45
1.95
1.89
2.13
1.50
1.89
2.09
1.94
1.20
1.75
1.89
1.80
1.20
47.10 17.57 10.72 9.83 8.28 5.00 3.31
45.00 17.02 18.95 3.88 4.25 6.58 0.04
48.28 20.08 6.25 10.43 12.50 4.84 2.89
49.71 10.00 0.00 25.00 5.88 0.29 15.88
2.50 1.17 0.90
2.89 0.98 1.72
2.50 0.47 0.16
1.18 4.41 0.88
0.87
2.11
0.00
0.00
Source: The World Business Environment Survey (WBES) © 2000 The World Bank Group. Notes The numbers in panel A show the average response to the question “Please judge on a four point scale how problematic are the following factors for the operation and growth of your business factors are.” The numbers in panel B show the average response to the question “Please judge on a four point scale how problematic are these different financing issues for the operation and growth of your business factors are.”
Profitability, liquidity, and investment 177 On the other hand, all cash flow models of investment posit that a firm’s cost of capital rises significantly when the firm has to raise external capital. The most popular version of the neoclassical theory of investment is the q-theory of investment, which has been originally formulated by Grunfeld (1960). In this model, investment depends on a variable that captures expected growth in the demand for capital. Grunfeld (1960) proposed the firm’s current market value while most follow-up studies have replaced it with Tobin’s q to measure the expected growth. Hayashi (1982) demonstrates that marginal q is the theoretically correct measure of investment opportunities and establishes the conditions under which it equals average q. Osterberg (1989) proves that marginal q is a “sufficient statistic” for investment. In its simplest form, the q-theory posits that investment in capital equipment of a firm in period t, It, is a function of Tobin’s q It = + qit1 + it
(7.1)
Tobin’s q is the ratio of the market value of a firm to the replacement value of its assets. It is relatively straightforward to show that q is equal to the ratio of the firm’s return on these assets to its cost of capital. The asymmetric information theory of investment Despite its simplicity and elegance, the neoclassical theory of investment has received very poor empirical support. The coefficient on Tobin’s q is often statistically insignificant and other variables, which theoretically do not belong in the equation, are often highly significant. Most prominent among them is the cash flow. A literature which can be traced back to Kuh and Meyer (1957) including a countless number of studies reporting a positive and significant relationship between company cash flows and investment. Advances in the research on information economics in the last two decades suggest that asymmetric information problems in capital markets can have important effects on corporate investment. This literature has shown that problems of asymmetric information between borrowers and lenders lead to a gap between the cost of external financing and internal financing. Stiglitz and Weiss (1981) have applied this idea to loan markets and showed that if firms have private information about the quality of their investment projects, there is potential for adverse selection and rationing. Concerning equity markets Myers and Majluf (1984) demonstrate that managers who seek to maximize the wealth of their existing shareholders forego profitable investment projects since the external capital market underestimates the quality of the investment. In this context, managers know that the return on the investment project is greater than the opportunity cost of their shareholders’ capital. The capital market, however, pools the firm with other firms with less profitable investment opportunities. If the company issues shares to finance the investment, it would harm its current shareholders. The managers wish to maximize the wealth of these shareholders, thus foregoing the investment unless
178
B. Burçin Yurtoxlu
they have sufficient cash flows to undertake them.6 Under this AIH cash flow is again justified as a determinant of investment. Principal–agent problems and the managerial discretion theory of investment As stated earlier, many studies that analyze the determinants of investment found a positive relationship between cash flow and investment. Many of these studies have analyzed samples including the largest companies in the economy. It is unlikely that most of these large, mature companies suffer from asymmetric information problems. The most plausible theory and the one that is most often put forward to explain this relationship assumes the existence of agency problems. Managers have their own goals, as for example increasing the size or growth of their companies (Marris, 1964, 1998) and wish to pursue this even when they harm their shareholders. The discretion managers have to allocate their internal cash flows, as their choice leads them to favor this source of finance over bank loans or equity issues ( Jensen, 1986). Empirical evidence supporting the MDH of investment has been presented by Grabowski and Mueller (1972), Kathuria and Mueller (1995), and Gugler et al. (2004). Comparison of AIH and MDH and estimation issues Both the AIH and MDH predict a positive relationship between cash flow and investment, they make in all other respects opposite predictions. The AIH assumes that managers maximize their shareholders’ wealth, and they would like to reveal the nature of their investments to the capital market. On the other hand, under the MDH managers seek to maximize their own welfare and would, if anything, like to conceal the nature of their investments from the capital market. Under the AIH, the firm is cash constrained and should pay little or no dividends. Under the MDH the firm has excess cash and is likely to pay some dividends to keep its share price high enough to avoid takeover. The AIH predicts that the firm under-invests and thus has a return on investment greater than its cost of capital. Under the MDH the firm over-invests and has a return on investment less than its cost of capital. Starting with Fazzari et al. (1988) an important aspect of empirical research in investment has been to identify a priori criteria to differentiate between sets of firms for which one of the hypotheses is likely to apply. Studies testing the AIH have used a variety of criteria to differentiate between financially constrained and unconstrained firms including size, age, the level of dividend payments, leverage, and ownership structure. The basic methodology consists of estimating some variant of the following equation for samples of firms which should and should not fit the hypothesis: It qt1 CFt1 it,
(7.2)
Profitability, liquidity, and investment 179 with the prediction that the cash flow coefficient will be larger for the sample of firms for which the asymmetric information problems are most likely. On the other hand, some other studies, for example, Kaplan and Zingales (1997) argue that a high cash flow – investment sensitivity does not imply financing constraints. One source of this controversy has been the disagreement in identifying the firm characteristics that reflect financing constraints. Most of the firm characteristics (size, dividends, leverage) used by earlier studies are endogenous and hence jointly determined with investment and cash flow levels of firms. Such an endogeneity problem does not arise in studies that use firm characteristics that are exogenously determined, such as business group membership. In this chapter, we will follow Hoshi et al. (1991) given the striking similarity between the Japanese keiretsu and the business groups (BG) in Turkey. The idea put forward by Hoshi et. al. (1991) was that keiretsu members have access to external financing from the group’s “main bank,” which monitors the member firms closely and reduces the costs of information associated with external financing.7 These factors led Hoshi et al. (1991) to hypothesize a very weak correlation between the investment of group firms and their internally generated funds. On the other hand, for firms that were not part of a keiretsu, they hypothesized and found a strong link between cash flow and investment. A more recent study by Walker (2001) corroborates these results by presenting empirical evidence that both vertical and horizontal industrial groups in Japan have active capital markets. Other studies that use similar sample splitting criterion include Shin and Park (1999) who compare the investment-cash flow sensitivity of Korean chaebols and non-chaebol firms. They show that the investment- cash flow sensitivity is low and insignificant for chaebol firms but is high and significant for non-chaebol firms. Similary, Perotti and Gelfer (2001) report a negative correlation between cash flow and investment for members of hierarchical, bank-led groups in Russia, whereas they find for industry-led group firms and independent firms a significant investment-cash flow sensitivity. George et al. (2004) analyze the investment behavior of independent and group affiliated firms in India. They find strong investmentcash flow sensitivities for both types of firms but no significant difference between them. Other evidences are available for Italy (Schiantarelli and Sembenelli, 2000), Canada (Schaller, 1993), and Germany (Elston and Albach, 1995). The BGs in Turkey are similar to the keiretsu in that they are a collection of large industrial and financial companies, and are perhaps even more similar to the Korean chaebol, since the group as a whole, regardless of its size and the scope of its activities, is run as a family business where ownership remains in the hands of the owner family.
Turkish corporate governance and implications on investment Turkish companies exhibit highly concentrated and centralized ownership structures. Families, directly or indirectly, own about 80 percent of all companies and keep
180
B. Burçin Yurtoxlu
the majority control. The separation of ownership and control is mainly achieved through pyramidal or complex ownership structures and by using dual-class shares. In a substantial fraction of companies, the management coincides with the family of the controlling owner. Table 7.3 provides information on the identities and the percentage shareholdings of direct and ultimate shareholders in Turkey using data from year 2001. Panel A reports the concentration of direct ownership by the identity of the largest shareholder (C1). Holding companies are the most frequently observed direct shareholders having the largest stake in 121 companies. Their mean ownership is equal to 47 percent of the outstanding shares. Non-financial companies and financial companies have the largest stake in 57 and 39 companies, respectively. Their stakes are also close to 50 percent. Families are the largest direct shareholders of 54 companies and own on average about 35 percent of the shares. Direct foreign ownership is present in 21 cases with a mean stake of 65 percent. The Turkish state and some miscellaneous owners have the largest direct shareholdings in 6 and 7 companies, respectively. In the next column we present the mean and the standard deviation of the percentage of a company’s outstanding equity held by the largest five shareholders (C5). Overall, the 5 largest shareholders own about 64 percent of the equity. The final columns reports the frequency distribution of the outstanding equity held by a large number of small investors (dispersed). The mean of this fraction is equal to 33 percent of the equity. Ultimate owners are by definition individuals or the state. In pyramidal structures, such as depicted in Figure 7.1, industrial companies and financial institutions appear among the owners. Substantial numbers of share stakes are held by companies, which in turn are held by other shareholders. Therefore, to pinpoint who actually owns a company, pyramidal ownership structures and the impact of dual class shares must be taken into account. Figure 7.1 presents an example of separating the cash-flow rights from the control rights using pyramidal structures. Celik Halat, a medium-size industrial company has three direct owners: Dogan Holding with 48.44 percent, Disbank with 10 percent, and Dogan Dis Ticaret with 4 percent8 of the equity. The remaining 37.56 percent are distributed among a large number of dispersed shareholders. A closer inspection of the ownership chains of these three direct owners reveals that 49 percent of the shares of the largest direct owner, Dogan Holding belongs to Adilbey Holding, another holding company and to the Dogan Family with 16.7 percent of the shares. Dogan Family is the sole owner of Adilbey Holding with 100 percent of the shares outstanding. Dispersed shareholders held also 34.24 percent of the Dogan Holding. A similar chain of ownership relations is present in the second largest shareholder, Dıvbank, leading to the ultimate control of the Dogan Family. Since Dogan Family at each level has the majority control, the control rights amount to 62.44 (48.44 10 4) percent. Multiplying and summing over all relevant control chains, we come up with 39.82 percent of cash-flow rights, which is substantially lower than the control rights. The wedge in this example is equal to 1.56 (62.44/39.82), whereas the number of pyramidal layers (PYR) is equal to 4 (Figure 7.2).9
Panel B: Ultimate ownership and control Families 242 Oyak – T. IS Bank 26 Foreign 22 State 12 Miscellaneous 3 All 305
N
121 57 39 54 21 6 7 305
Panel A: Direct ownership Holding company Non-financial companies Financial companies Families Foreign companies State Miscellaneous All
Ultimate owner identity
N
Largest shareholder identity
67.03 65.66 66.58 66.90 66.58 66.62
67.01 64.25 69.28 57.17 69.28 66.23
50.56 37.74 64.68 62.26 41.10 50.85
53.58 35.09 66.78 54.56 24.59 53.00
Median
Mean
Mean
Median
Cash-flow rights
17.56 16.16 19.19 22.04 19.91 29.76 26.81 20.57
Control rights
46.39 47.83 41.18 34.71 65.75 72.56 51.13 45.88
5.29 1.00 1.03 1.10 2.91 4.57
Mean
Wedge
66.45 67.92 52.72 55.49 81.00 74.86 68.31 63.61
Mean
Mean
Standard deviation
C5
C1
Table 7.3 Ownership and control structure of Turkish listed companies
1.12 1.00 1.00 1.00 2.23 1.00
Median 1.86 1.17 1.32 1.33 1.88 1.79
PYR (mean)
15.71 13.40 20.47 24.40 12.98 26.32 18.28 20.23
Standard deviation
44 38 27 25 67 42
Dual (mean %)
31.74 30.53 44.66 38.92 18.6 19.31 28.13 33.2
Mean
15.22 13.69 22.95 24.94 13.48 20.26 18.88 19.24
Standard deviation
6 7 7 7 6 6
3 0 2 0 0 2
Family/ Board (median) board (median)
Dispersed
182
B. Burçin Yurtoxlu
Celik Halat A.S.
Dogan holding
Disbank
Dogan dis ticaret
Dispersed
48.44%
10%
4%
37.56%
Adilbey holding
Dogan family
Dispersed
Dogan holding
Milpa A.S.
Dispersed
49%
16.70%
34.24%
53%
12%
35%
Dogan family
Adilbey holding
Dogan family
Dispersed
Dogan holding
Dispersed
100%
49%
16.70%
34.24%
65%
35%
Dogan family
Adilbey holding
Dogan family
Dispersed
100%
49%
16.70%
34.24%
Dogan family 100%
Figure 7.1 An example of a pyramidal ownership structure (Celik Halat A.S. (2001)).
SELCUK GIDA
E. Alharal I. 3%, A 6.2%, B
E. Alharal V. 1%, A 1.6%, B
V. Franco 1%, A 1.5%, B
T. Berkan 0.8%, A 0.7%, B
Dispersed 84.2%, B
Figure 7.2 An example of dual class shares (Selcuk Gida A.S. (2001)).
The ownership structure of Celik Halat is by no means unique. The presence of a holding company and a bank as a direct or indirect owner is found in many other listed companies. In some cases these ownership relationships are organized within the legal form of a holding company so that the resulting conglomerate has a structure very similar to the Japanese keiretsu or South Korean chaebol. We refer to this form as a BG. The BGs in Turkey are the outcome of investments by a single family or a small number of allied families. The families keep these companies together as a coherent body among which resources and personnel may be shifted as needed (Bugra, 1994). Some of the BGs own a bank that serves as the “main-bank” of the group companies. The firms affiliated with a business group are linked to one another in two ways. The first may be described as a complicated web of inter-corporate shareholdings. The second way of organizing a group is to form a pyramid by spreading the voting rights of minority shareholders out over a large number of firms and concentrating those of the ultimate owner at the top of the pyramid as in Figure 7.1. In this way several companies are controlled by the same family at the top of the pyramid through a chain of
Profitability, liquidity, and investment 183 ownership relations. While pure forms of BGs can be found in Turkey, the typical BG incorporates both features. Families hold the majority control of a holding company which in turn has shareholdings in several other companies giving rise to a pyramidal structure. There are also cases where the companies hierarchically under the holding company own the shares of each other. We differentiate between two main sets of firms regarding their relationship to BGs. The first, which we call the independent or non-group firms, are not part of a BG. We determine whether a firm is independent or not by analyzing its ownership structure. Firms whose owners consist of only individuals and families are labeled as independent, if they have no subsidiaries that are part of a BG. In a few cases another company appears to be among the direct owners of an independent company. In this case we check whether this company belongs to a BG or not. The criterion that separates group and independent firms is the existence of a holding company in the ownership structure of the company. Group firms have as a major owner of their outstanding shares (a) a holding company, which also contains a bank or (b) at least one other company, which belongs to a BG. It is also possible to differentiate between groups that own a bank and those that do not own a bank.10 While BGs that also own a bank are very similar to the Japanese keiretsu, BGs, which do not own a bank, resemble a conglomerate with a multidivisional (M-form) organization with the holding company at the top or the center of the ownership structure serving as the central management team. One of the advantages of an M-form organization is that it serves as a mini-capital market internal to the organization (Williamson, 1975).11 A member company with a profitable investment project can raise money within the BG, which can transfer it from other group companies or raise the amount at lower rates due to its size and scope of operations in the external capital market. In addition to this, they also have increased monitoring incentives of such internal capital markets, because the owner of the BG has a residual control over the assets, and therefore gets more of the gains from monitoring (Gertner et al., 1994). These arguments imply that non-bank group firms are less dependent on internal funds to finance investment than independent firms. Nevertheless, we expect to see some impact of internal funds on investment, even for group firms, since the group as a whole could be liquidity constrained. Panel B reports the results of the exercise of tracing the ultimate ownership of companies. Families control 242 of the 305 listed companies with available ownership data. Twenty-six of the remaining companies are controlled by two legal entities which have rather unusual ownership structures. The first (Türkiye Is Bankasi ) is a quasi-private bank under managerial control and the second (OYAK Group) is a quasi-private group of companies.12 Foreign companies are the controlling owners of 22 Turkish companies and there are 12 companies under state ownership. The remaining three companies in the miscellaneous category are controlled by entities similar to foundations. At the direct ownership level, the average stake of the largest shareholder is about 45 percent. This number understates the true control potential of large shareholders. It is usual that the controlling owner has more than a single direct
184
B. Burçin Yurtoxlu
ownership stake. Combing all ownership stakes under the control of the ultimate owner, we observe that the true fraction of control rights of families is about 67 percent of the outstanding shares (column 3). The cash-flow rights are, however, substantially lower than this figure. The average cash-flow rights of families is 50 percent. In more than half of the companies controlled by families there are substantial deviations of cash-flow rights from control rights leading to an average wedge of 5.29 (wedge voting rights/cash-flow rights). It is worth noting that this figure is much higher than 1.32, which Yurtoglu (2000) reports for Turkish companies under ultimate family control using 1997 data. The main reason for this large difference is that 1997 data do not allow considering the role of dual class shares, whereas improved disclosure allows us to consider them using the latest data. In the next column, we present the average number of PYR used to sustain the wedge. The mean of PYR is very close to two. In only 73 companies we observe no pyramiding. The remaining 169 companies have two or more layers of ownership chains. Of course, an alternative way of creating a wedge is to use dual class shares. The next column reports the fraction of companies with dual class shares. In 44 percent of companies under family control, there is more than one class of shares. All of these features of the ownership structure of Turkish companies are consistent with the predictions of recent research in law and finance (LaPorta, Lopez-de-Silanes, Shleifer and Vishny, 1997, hereafter LLSV). LLSV have argued that differences in ownership concentration between Continental European and Anglo-Saxon countries are due to differences in their legal institutions. Anglo-Saxon legal systems offer shareholders with small stakes better protection from exploitation by managers than do the civil law systems found in Continental Europe. This weaker protection affects both the demand and supply schedules for new equity issues and results in both thinner equity markets and more concentrated ownership structures in civil law countries such as in Turkey.13
Data and the empirical model We analyze the link between internal funds and investment using equation (7.2) which is common to many earlier studies. The investment equations include cash flow (CF ) as the measure of liquidity14 and Tobin’s q. To eliminate problems associated with size effects, investment and CF are scaled by the firm’s capital stock at the beginning of the fiscal year. All of the independent variables are lagged one period to avoid their being partly endogenous. The general specification of the investment equation is: It/Kt1 i t 1(CFt1/Kt1) 2qit1 it,
(7.3)
investment (I ) outlays are reported in the footnotes to the annual statements of listed firms. CF is net profit after tax plus (accounting) depreciation.15 This variable is a measure of the net flow of cash into the firm. Tobin’s q is the ratio of the market value of equity plus the book value of total debt divided by total
Profitability, liquidity, and investment 185 assets. Time dummies ( t) are included for each year to filter out the effects of aggregate business cycle fluctuations and firm dummies ( i) account for unobserved and fixed firm-specific effects.16 While a significant fraction of the literature used the earlier specification, some scholars have criticized it due to endogeneity problems. The criteria used to split the sample can be correlated with both the firm-specific and time-invariant component of the error term, as well as with the idiosyncratic component. While the correlation with the time-invariant component can be eliminated by first differencing, correlation with the idiosyncratic component is best dealt within an IV or GMM context. In this case, consistent estimates can be obtained using appropriately lagged values of the explanatory variables as instruments. If the equation is first differenced and the idiosyncratic component of the error term in the level equation has the usual statistical properties, endogenous variables lagged twice would be legitimate instruments. Due to this reason, we will also report the estimated coefficients of equation (7.3) using the dynamic panel data estimator of Arellano and Bond (1991). Table 7.4 gives descriptive statistics on some characteristics of group and independent firms.17 The data are collected from the annual balance sheet and income statements provided by the Istanbul Stock Exchange (ISE). We follow 221 manufacturing firms traded at the ISE from 1990 to 2002. Our panel consists of 1,547 firm-year observations. Since not all of the 221 companies are traded over the full 13-year period and there are omissions in our data sources, the panel is unbalanced. The number of group firms in the sample is 76, while there are 145 non-group firms. We also report summary statistics for the full sample of firms. The summary statistics reported suggest some differences in the investment behavior and financial structure of group and independent firms. The mean and median values of investment are greater for group firms, the mean difference being statistically significant at the 5 percent level. Both the means and medians indicate that group firms have higher levels of cash flow than for non-group firms. The mean Tobin’s q of group firms is higher than the q of non-group firms, the difference being significant at the 1 percent level. This finding indicates that group firms have better investment opportunities than for non-group firms.18 The next set of summary statistics show that group firms have higher mean and median sales growth rates, the differences are, however, statistically insignificant at conventional levels. The use of debt by group and non-group firms is compared in the next set of descriptive statistics. Interestingly, non-group firms have higher total debt-total assets ratios than group firms. This difference comes, however, from the greater use of short term debt by non-group firms. The long-term debt–total assets ratios of the two types of firms are not statistically different. Group firms are larger as indicated by the level of employees and the level of total assets. They are also slightly older than non-group firms.19 The last two rows show that group firms have a substantially higher mean return on assets than non-group firms and that they pay a higher fraction of their profits in dividends.
186
B. Burçin Yurtoxlu
Table 7.4 Summary statistics
Number of firms Number of firm-years Mean I/K Median I/K Standard deviation I/K Mean CF/K Median CF/K Standard deviation CF/K Mean Tobin’s q Median Tobin’s q Standard deviation Tobin’s q Mean sales growth Median sales growth Standard deviation sales growth Mean debt/total assets Mean long-term debt/total assets Mean short-term debt/total assets Mean employees Mean total assets Mean age Mean return on assets Mean dividend pay-out ratio
Non-group firms
Group firms
All
145 934 0.099 0.061 0.114 0.114 0.109 0.154 1.648 1.380 0.885 0.035 0.032 0.300 0.605***
76 613 0.113** 0.071 0.118 0.161*** 0.157*** 0.136 1.853*** 1579*** 1.024 0.055 0.054 0.224 0.530
221 1,547 0.105 0.064 0.116 0.133 0.129 0.149 1.729 1.458 0.947 0.053 0.041 0.272 0.575
0.141
0.137
0.139
0.465*** 782 5.96e06 32 0.098 0.092
0.393 1074*** 7.57e06** 36*** 0.145*** 0.153***
0.436 892 6.60e06 34 0.117 0.116
Notes ** (***) indicate that the mean or the median is significantly greater the corresponding entry at the 5 (1) percent level, respectively. A two-sample Wilcoxon rank-sum (Mann–Whitney) test is used for the median test.
The findings The influence of business group membership on investment The basic results are presented in Table 7.5. The first part of the table reports parameter estimates of equation (7.3) using ordinary least squares (OLS). The equation includes a full set of time dummies and 19 two digit industry dummies. The first column reports the coefficients of CF and Tobin’s q for the pooled sample of group and non-group firms. The coefficient on CF is 0.13 and it is highly significant. Tobin’s q is marginally significant at the 10 percent level. The equation explains 16 percent of the variation in observed levels of investment. The next column presents the estimated coefficients of the same variables for the group firms. The coefficient of CF is equal to 0.10 and is significant at the 1 percent level. The coefficient of Tobin’s q is significant at the 5 percent level. On the
Profitability, liquidity, and investment 187 Table 7.5 OLS estimates of the impact of group membership on the investment-cash flow sensitivity of Turkish firms Variables
CF/K Q Constant N/NF Adj. R2
OLS
Firm fixed effects
All
Group firms Non-group firms
All
Group firms Non-group firms
0.133*** 6.63 0.006* 1.68 0.065*** 10.63 1547/221 0.16
0.102*** 2.63 0.010** 1.97 0.053*** 5.16 613/76 0.21
0.108*** 4.24 0.003 0.74 0.067*** 4.74 1547/221 0.31
0.081 1.59 0.006* 1.84 0.056*** 4.19 613/76 0.32
0.165*** 6.80 0.002 0.72 0.077*** 10.95 934/145 0.22
0.1301*** 4.23 0.002 0.46 0.1107*** 8.36 934/145 0.30
Notes All regressions include a full set of time and 19 two-digit industry dummies, which are constrained to sum to zero, so that the constant becomes an estimate of the average effect of time and industry dummies. Both time and industry dummies were jointly significant at the one percent level or better. The t-values are reported under the coefficients and they are computed using standard errors which are robust to heteroscedasticity. *, (**, ***) indicate that the coefficient is significantly different from zero at the 10, (5, 1) percent level, respectively.
other hand, the coefficient of CF is much higher for non-group firms (0.16). The difference between the CF coefficients for group and non-group samples (0.06) is significant at the 1 percent level. Noteworthy is also the insignificant coefficient on Tobin’s q for non-group firms. Both equations explain about 21 percent of the variation in investment. The next set of results in Table 7.5 is obtained after replacing the industry dummies by fixed firm effects.20 The first column reports the coefficients when the standard investment-cash flow regression is estimated for the pooled sample of 221 group and independent firms. The coefficient on Tobin’s q is not significant. The coefficient on CF is 0.108 and significant at the 1 percent level. The included variables along with time and firm effects explain 31 percent of the variation in investment. The next two columns contrast the difference between group and non-group firms. For the full sample of 76 group firms, CF has no statistically significant impact on investment using conventional confidence levels. Comparing the coefficients on Tobin’s q, we note that q has a much larger coefficient for group firms, whereas it has a small and insignificant impact on investment for non-group firms. The difference in the CF coefficients amounts to 0.05, which means that the impact of CF is almost 50 percent higher for non-group firms than for group firms. Since equation (7.3) contains no lagged dependent variables and past values of the independent variables are predetermined, the use of OLS method is consistent. One can relax these assumptions and estimate equation (7.3) by IV or GMM using lagged values of the included variables as instruments. If there is no measurement error, variables dated t 1 are potentially legitimate instruments
188
B. Burçin Yurtoxlu
Table 7.6 GMM estimates of the impact of group membership on the investment-cash flow sensitivity of Turkish firms Variables
All
Group firms
Non-group firms
(I/K )t1
0.0003 0.64 0.057*** 2.04 0.001 1.11 0.015*** 11.08 1196 / 208 0.18 0.01 0.68 0.06
0.0004 0.75 0.031 1.01 0.001 1.13 0.020*** 11.10 478 / 73 0.16 0.00 0.76 0.05
0.003 0.78 0.132*** 2.19 0.003 0.82 0.008*** 3.48 718 / 135 0.19 0.01 0.69 0.07
CF/K q Constant N/NF Sargan test AR(1) AR(2) R2
Notes All regressions include a full set of time dummies. *, (**, ***) indicate that the coefficient is significantly different from zero at the 10, (5, 1) percent level, respectively. Absolute values of the z-statistics are under the coefficients. Estimation method is one step GMM, which eliminates firm fixed effects by first differencing. Sargan test is the p-value of a Sargan-Hansen test of overidentifying restrictions. AR(p) is the p-value of a test that the average autocovariance in residuals of order p is zero. Instruments include lagged levels of the dependent and predetermined variables dated t-2 or earlier. The estimation procedure does not produce a measure of goodness of fit. Instead we report the correlation coefficient of the predicted values with the observed values of the dependent variable (Windmeijer, 1995).
after the elimination of fixed firm effects (Schiantarelli, 1996). Otherwise, variables lagged at least twice should be used as instruments. Therefore, we estimate a version of equation (7.3) by a systems GMM estimator developed by Arellano and Bond (1991), Arellano and Bover (1995) and Blundell and Bond (1998). This estimator eliminates firm effects by first-differencing and controls for potential endogeneity of explanatory variables. Endogenous variables lagged two or more periods will be valid instruments provided there is no second-order autocorrelation in the first-differenced idiosyncratic error terms. Sargan tests of over-identifying restrictions are also provided. Table 7.6 presents the estimation results for equation (7.3), which now includes a lagged dependent variable using the Arellano–Bond one-step GMM estimator. The estimates show that the lagged dependent variable is insignificant for the whole sample and for both group and non-group firms. The Sargan tests do not imply rejection of the overidentifying restrictions at conventional levels. There is some evidence of first order serial correlation in the residuals, however, AR(2) statistics suggest that there is no second order serial correlation confirming the validity of instruments. As in the OLS results, the CF coefficient for the whole sample is significant at the 5 percent level. Tobin’s q, on the other hand is not significant for the whole sample using the GMM method. The CF coefficient is
Profitability, liquidity, and investment 189 small and statistically insignificant for the group firms. The difference between the CF coefficients of the group and non-group firms is now much larger (about 0.10), since the group firms reveal more or less the same sensitivity of investment to CF in the GMM estimation. The results in Tables 7.5 and 7.6 suggest that BG membership has an important impact on the sensitivity of investment to cash flow. The sensitivity of investment to cash flow in non-group firms differs consistently from group firms suggesting that the availability of CF for group firms is not as important as for non-group firms. This finding is consistent with the idea that groups help to cope with asymmetric information problems in capital markets which may happen on three grounds. First, groups own their own banks which of course have an informational advantage. The group members are likely to rely on the financial resources of the group in case of financial distress which will ease their access to external finance. And third, even in the absence of group banks, groups may function as internal capital markets that complement the capital allocation function of the external capital markets. Whether these internal capital markets function efficiently is taken up in the next section. Efficiency of investment behavior in group firms All of three estimated coefficients of Tobin’s q in Tables 7.5 and 7.6 for non-group firms are statistically insignificant suggesting that their investment is not affected by their growth opportunities. On the other hand, the OLS results for group firms show that firms with higher Tobin’s q invest more (Table 7.5). If groups have internal capital markets, a finding supported by the smaller and in two out of three cases insignificant CF coefficient, then we should expect a group firm’s investment to be a function of the investment opportunities and internal funds of other group firms.21 To quantify the effect of other firms in the same business group on an individual group firm’s investment, we include the cash flows from other group firms within the same group after normalizing by the sum of their total assets (CFother). We also include a composite Tobin’s q for the group as a whole (qother).22 The results of this exercise are presented in Table 7.7. The estimated coefficients of CF and q are comparable to their counterparts in Tables 7.5 and 7.6. The CF coefficient is low and mostly insignificantly different from zero. Tobin’s q is marginally significant. More interesting for our purposes are the coefficients of the variables CFother and qother. CFother has a positive coefficient which is highly significant. This finding suggests that investment of individual group firms is affected by other group members’ liquidity. The three estimated coefficients of qother are, on the other hand, insignificantly positive. Therefore, we must reject the hypothesis that a group firm invests less when other firms in the same group have better investment opportunities. This finding casts doubt on the notion that groups function as efficient internal capital markets that allocate the internal funds to group companies with better investment opportunities.
190
B. Burçin Yurtoxlu
Table 7.7 Investment of group firms Variables
(1) OLS
(2) FE
(3) GMM
0.042 1.49 0.011* 1.94 0.134*** 2.42 0.003 0.25 —
0.030 1.41 0.008 1.89* 0.128 2.71*** 0.004 0.74 —
0.0002 0.90 0.025 0.89 0.002 1.12 0.054* 1.95 0.002 0.45 —
0.018 9.17 613/76 0.15
0.025 8.67 613/76 0.30
0.008 3.48 478/73 0.06
(I/K )t1 CF/K q (CF/K )other qother (CF/K ) * qother Constant N/NF Adj. R2
(4) FE
0.028 1.45 0.007 1.81 0.131*** 2.54 0.003 0.71 0.002 0.45 0.027 7.41 613/76 0.31
Notes All regressions include a full set of time dummies. *, (**, ***) indicate that the coefficient is significantly different from zero at the 10, (5, 1) percent level, respectively. The OLS equation includes 19 two-digit industry dummies. Absolute values of the t- (for GMM z-) statistics are under the coefficients. The R2 reported in the GMM equation is the correlation coefficient of the predicted values with the observed values of the dependent variable (Windmeijer, 1995).
As a further test of this point, we interact the CF variable of individual group firms with the Tobin’s q of the other group members. If internal capital markets work efficiently, we would see a negative and significant coefficient on this interaction term. The estimated coefficient, however, is insignificantly positive (column 4 in Table 7.7). This finding also suggests that there is no firm link between the investment opportunities of the group as a whole and the sensitivity of investment to internal funds. These findings suggest that even though there is a transfer of funds within the groups, these transfers do not seem to be consistent with the notion that the groups allocate internally generated funds to other group firms with better investment opportunities. Even though, the groups may help to alleviate the information problems in the capital markets, they may still be subject to agency problems. This final issue is also related to the fact that many of the group firms are unlisted and hence do not show up in the construction of CFother and qother variables. Whether unlisted group firms have better growth opportunities or not is an empirical question. These unlisted group firms, however, are surely under the sole control of the family managing the group, where they have the right incentives to maximize value. What the findings of Table 7.7 suggest is that in listed firms where the owners share the benefits, the efficiency of the internal capital markets should be questioned.
Profitability, liquidity, and investment 191 Further tests of the impact of ownership structure on investment So far we have analyzed the impact of the group membership on the determinants of investment. Other aspects of the ownership structure of companies can also have an impact on investment. For example, Gugler et al. (2005) argue that ownership identity and concentration do in fact have an impact on the investment of firms which are most likely to face information asymmetries. They estimate a smaller coefficient on CF for firms fitting the AIH and controlled by another company, financial institution or the state, than for firms that are family-controlled or with dispersed ownership, because the former are able to borrow more readily from the institution that controls them.23 As a final test of the efficiency of investment behavior by group and non-group firms we interact Tobin’s q with the cash-flow rights (CFR) and voting-rights (VR) of the ultimate owner.24 As the cash flow rights of the ultimate owner increase this leads to an alignment of interest with the minority shareholders. As voting rights increase the controlling owners can face incentives to deviate from value maximizing behavior. In fact, recent research shows that separating control rights from cashflow rights can create agency costs an order of magnitude larger than the costs associated with a controlling owner who also has a majority of the cash-flow rights in her corporation (Bebchuk et al., 1999). Consistently, Claessens et al. (2002) provide empirical evidence that such deviations are associated with a value discount.25 We expect that the sensitivity of investment to Tobin’s q to increase with higher cash-flow rights of the ultimate owner and to decrease with higher control (voting) rights. The results are presented in Table 7.8. In the first two columns we report the results for group firms. Both the OLS and fixed effects estimations show that the group firms’ investment becomes more sensitive to growth opportunities as cash-flow rights of the ultimate owners increases. On the other hand, investment becomes less sensitive to investment opportunities as the voting rights of the controlling owner increase. The results in Table 7.5 show that non-group firms’ investment is not influenced by their growth opportunities as measured by their Tobin’s q. The results in columns 3 and 4 of Table 7.8 indicate that as the cash-flow rights in these firms increase so that ultimate owners’ and minority shareholders’ interests become more aligned, the investment becomes responsive to investment opportunities. The q * CFR interaction term is positive and significant in both OLS and fixed effects regressions. The q * VR interaction has the right, positive sign, it is, however, only marginally significant. The results in Table 7.8 suggest that deviations of cash-flow rights from votingrights do have an impact on the investment behavior of companies, a finding that complements earlier empirical evidence on their value destroying impact. Robustness Although we believe that the regression evidence presented so far is fairly strong, we explore other specifications to examine the robustness of results with respect to specification and estimation technique.
192
B. Burçin Yurtoxlu
Table 7.8 Further tests of the impact of ownership structure as a determinant of investment Variables
(1) Group OLS
(2) Group FE
(3) Non-group OLS
(4) Non-group FE
CF/K
0.098*** 2.39 0.029*** 3.05 0.0001 0.93 0.0004*** 2.58 0.047*** 8.47 603/72 0.17
0.078 1.57 0.0002 1.72 0.0003*** 2.41 0.00021*** 2.09 0.008*** 3.48 603/72 0.31
0.171*** 6.61 0.0047 0.65 0.0003 3.31*** 0.0002 1.87 0.025*** 8.67 867/134 0.19
0.125*** 3.65 0.0257 1.68 0.00008 2.32*** 0.0003 1.40 0.027*** 7.41 867/134 0.33
q q * CFR q * VR Constant N/NF Adj. R2
Notes All regressions include a full set of time dummies. The OLS equation includes 19 two-digit industry dummies. *, (**, ***) indicate that the coefficient is significantly different from zero at the 10, (5, 1) percent level, respectively. Absolute values of the t-statistics are under the coefficients and they are robust to heteroscedasticity.
Some authors have explicitly analyzed the impact of firm size on the CF-investment sensitivity by analyzing the impact of cash flow and other financial variables for different size classes. Devereux and Schiantarelli (1990) report statistically insignificant differences in cash flow coefficient of small and large firms, while the impact of q increases with firm size. Kadapakkam et al. (1998) find larger firms make more use of internally generated funds in six OECD countries.26 While one can criticize such sample splits due to endogeneity reasons and due to the fact that firm size can proxy for other unobserved firm characteristics, many studies use firm size as an additional explanatory variable in investment equations such as our equation (7.3). As a robustness check we re-estimate all reported equations after including the natural logarithm of total assets in t 1. Size has almost uniformly a positive and significant impact on investment for all types of firms. The differences in the CF coefficients between group and non-group firms remain significant. Second, measurement errors or problems associated with omitted variables can bias panel data estimates if the firms in both group and non-group samples are heterogeneous. Although we have used fixed effects to correct for unobserved and non-varying determinants of investment, the use of fixed effects can exacerbate problems associated with heterogeneities in the effect of cash flow on investment (Pesaran and Smith, 1995). Pesaran and Smith (1995) suggest two procedures to correct for this. First, one can run a cross section regression for each year. The average value of the coefficient on cash flow across 12 cross sections is 0.11 for the pooled sample, 0.17 for non-group firms, and 0.07 for group firms. One can also average the variables over time for each firm and run a cross section regression.
Profitability, liquidity, and investment 193 This procedure yields a precisely estimated coefficient on cash flow of 0.15 for non-group firms and an insignificant coefficient of 0.031 for group firms. The results of these alternative specification and estimation procedures suggest that the predicted role of CF for group and independent firms is relatively stable. The argument for putting Tobin’s q in an investment equation rest on the assumptions of perfect competition, constant-returns-to-scale and that firms are price takers, which imply that the marginal and average returns on capital are equal, and equal a firm’s cost of capital.27 When firms are not price takers and markets are imperfectly competitive, however, marginal and average returns on capital do not coincide and equilibrium may exist in which a firm’s average return on capital differs from its marginal return. The same level of investment may be optimal for a monopolist as for a competitive firm even though the monopolist’s profits on existing assets, and hence its average q, are much larger than for the competitive firm. To predict the investments of these two companies more accurately, Gugler et al. (2004, 2005) have introduced a measure of marginal q, which they found to be a significant determinant of investment.28 We tried to use the same measure of marginal q instead of Tobin’s average q. However, this variable had in most of the regressions (not reported) an insignificant impact. Given the high volatility of share prices and hence market values measured at year end, it is possible that this variable is poorly measured and hence its coefficient is biased towards zero.
Conclusions Three conclusions stand out from this chapter. First, the availability of internal funds is an important determinant of fixed investment spending for Turkish manufacturing firms. This result is consistent with the findings of many studies on developed and developing countries. Second, the predicted relationship between CF and investment is different for group and independent firms. In particular, we find that CF is relatively more important for non-group firms compared to firms which are part of BG. For firms which are a member of a BG, the link between internal funds and investment ranges from weak to insignificant. We interpret these findings as evidence that group structure mitigates information asymmetries in raising external capital by its internal capital market function. However, additional tests suggest that such internal capital markets do not function in ways consistent with value maximization. Third, we show that ownership structure of firms and specifically some governance devices that produce a wedge between the interests of controlling families and minority shareholders do have an impact on investment. As the interests of these two conflicting parties become more aligned, investment becomes to be more responsive to growth opportunities. Both asymmetric information problems and agency costs caused by corporate governance mechanisms do have an impact on investment. Policy measures like enhanced transparency and better accounting standards as required by the
194
B. Burçin Yurtoxlu
new corporate governance code of Turkey can potentially reduce asymmetric information problems so that firms with good investment opportunities can have better access to external capital markets. Other measures such as one-share-one-vote or independent boards can potentially reduce the agency problems specific to civil law countries such as Turkey and hence they would increase the efficiency of corporate investment.
Acknowledgments I would like to thank Dennis C. Mueller, Thomas Stratmann, and participants at seminars at the 1999 EARIE conference, 1999 METU Conference in Economics and Corporate Governance conferences at the ESRC Center of the Cambridge University and Aarhus University for various comments and suggestions. An earlier version of this chapter was completed when I visited the Centre for Business Research at the Judge Institute of Management Studies, University of Cambridge, whose hospitality is gratefully acknowledged.
Notes 1 See, Mueller (2003, 119–155) and Blundell et al. (1996) for a review of different investment models. 2 This includes replacement investment as well as expansionary investment and covers all sectors in the economy. Since gross investment does not depend on any assumptions about asset lives and depreciation, it is relatively straightforward to make international comparisons using this measure. 3 See for example Singh (1999) and Singh et al. (2003). 4 See also, Donaldson (1961), Meyer and Glauber (1964), and Mueller (1967). 5 On the empirical front, however, cash flow models of investment were superior to the neoclassical model as confirmed by Elliott’s (1973) findings. 6 Myers and Majluf (1984) assume that the firm’s debt limit is reached so that the internal funds and funds from a new equity issue are the only alternatives. 7 The lower cash flow sensitivity for keiretsu members can also be explained by less managerial discretion through better monitoring within a BG. According to this view, the independent firms’ cash flow–investment sensitivity can be explained by the MDH. 8 To make the figure as simple as possible, we do not show the ownership structure of the third largest shareholder, Dogan Dis Ticaret. This company is also ultimately owned by Dogan family. 9 Pyramids are not the only possible way to separate cash flow rights from voting rights. By issuing shares with multiple voting rights, controlling owners can reduce their cash flow rights below their voting power. For an example of this type of ownership structure see the examples in figure 2 and in Yurtoxlu (2003). 10 An earlier version of this paper has separately analyzed groups with and without banks. In this version, we classify the sample firms as group firms, whenever they belong to a group, which also has a bank. 11 Williamson (1975, p. 148) argues that “assignment of cash flows to high yield uses is the most fundamental attribute of the M-form enterprise.” 12 This bank has an unusual ownership structure with 35 percent of its voting stock under the control of a state agency and 45 percent under the control of a foundation belonging to the employees of the bank. OYAK Group is founded in 1961 by a special law as a social security organization for the members of the Turkish army.
Profitability, liquidity, and investment 195 13 Alternatively, Roe (2003) suggests that differences in political ideologies between Common Law and Civil Law countries explains the differences in ownership concentration. 14 All equations that we report are also estimated including a stock measure of liquidity (cash stock cash on hand liquid assets) to check the robustness of the results to different measures of liquidity. In these regressions cash stock had a small, positive but statistically insignificant impact on investment mainly due to the high collinearity with cash flow. 15 Dividends are not subtracted from this figure. Doing so would imply that dividends are a higher priority use of cash flow and are thus decided first. When we estimate the model with this second measure including dividends the results do not change. 16 In Panel A of Table 7.5, we report parameter estimates of equation (7.1) by replacing the fixed-firm effects with industry effects and present them as baseline estimates. 17 For all variables that enter the equation (7.3), the extreme values defined as the top and bottom 1 percent of the sample observations are deleted. This allows for a uniform definition of outliers and leads to more robust results. 18 Such a difference can also indicate that group firms have more market power than nongroup firms, a difference that could be capitalized in the market value. See, Gugler et al. (2004, 2005) for such an interpretation and an alternative for the use of q in investment equations. 19 These numbers are relative to 2004. 20 A Hausman test shows that a random effects model would be inappropriate. The test for fixed effects is highly significant. 21 Shin and Park (1999) are the first to pursue this idea in the context of Korean chaebol members. 22 qother is constructed by summing over the market values of all group members other than the individual group firm and dividing by the total assets of all group firms other than the individual group firm. 23 See also, Hadlock (1998) and Oliner and Rudebusch (1992) for the impact of ownership structure on investment. 24 See the section on Turkish corporate governance and implications on investment for the construction of these two variables and Table 7.3 for summary statistics. 25 Yurtoglu (2000, 2003) reports similar value discounts for Turkey. 26 See also Carpenter and Petersen (2002). 27 See again, Hayashi (1982) for a formal development of the theory. 28 This derivation was first presented by Mueller and Reardon (1993). For other applications and additional discussion, see Mueller and Yurtoxlu (2000) and Gugler and Yurtoxlu (2003).
References Arellano, M. and S.R. Bond (1991) “Some tests of specification of panel data: Monte Carlo evidence and an application to employment equations,” Review of Economic Studies, 58: 277–297. Arellano, M. and O. Bover (1995) “Another look at the instrumental-variable estimation of error component models,” Journal of Econometrics, 68: 29–52. Bebchuk, L., R. Kraakman, and G. Triantis (1999) “Stock pyramids, cross-ownership, and dual class equity: the creation and agency costs of separating control from cash flow rights,” NBER WP 6951. Blundell, R. and S.R. Bond (1998) “Initial conditions and moment restrictions in dynamic panel data models,” Journal of Econometrics, 87: 115–143. Blundell, R., S. Bond, and C. Meghir (1996) “Econometric models of company investment,” in Matyas, L. and P. Sevestre (eds), The Econometrics of Panel Data: Handbook of Theory and Applications, Dordrecht: Kluwer Academic Publishers.
196
B. Burçin Yurtoxlu
Bond, S.B. (2001) “UK investment and the capital market,” Institute for Fiscal Studies. Bugra, A. (1994) State and Business in Modern Turkey, New York: State University of New York Press. Carpenter, R.E. and Bruce C. Petersen (2002) “Is the growth of small firms constrained by internal finance,” Review of Economics and Statistics, 84: 298–309. Chirinko, Robert S. (1993) “Business fixed investment spending: modeling strategies, empirical results, and policy implications,” Journal of Economic Literature, 31: 1875–1911. Claessens, S., S. Djankov, J.P.H. Fan, and L.H.P. Lang (2002) “Disentangling the incentive and entrenchment effects of large shareholders,” Journal of Finance, 57: 2741–2771. Devereux, M and F. Schiantarelli (1990) “Investment, financial factors, and cash flow: Evidence from U.K. panel data,” in Glenn Hubbard (ed.), Asymmetric Information, Corporate Finance, and Investment, Chicago, IL: Chicago University Press. Donaldson, G. (1961) Corporate Debt Capacity: A Study of Corporate Debt Policy and the Determination of Corporate Debt Capacity, Boston, MA: Harvard University Press. Easterly, William and Ross Levine (2000) “Its not factor accumulation: stylized facts and growth models,” manuscript, Minneapolis, MN: Carlson School Management. Elliot, J.W. (1973) “Theories of corporate investment behavior: revisited,” American Economic Review, 63: 195–207. Elston, J. and H. Albach (1995) “Firm ownership structure and investment: evidence from German manufacturing,” IFO Studien Zeitschrift für Empirische Wirtschaftsforschung, Heft 1. Fazzari, S.M., G. Hubbard, and B. Petersen (1988) “Financing constraints and corporate investment,” Brookings Papers on Economic Activity, 141–195. George, R., R. Kabir, and Jing Qian (2004) “Is investment – cash flow sensitivity a good measure of financing constraints?: new evidence from group-affiliated firms,” manuscript Tilburg University Gertner, R. H., David S. Scharfstein, and Jeremy C. Stein (1994) “Internal versus external capital markets,” Quarterly Journal of Economics, 109: 1211–1230. Glen, J.D. and M.A. Sumlinski (1995) “Trends in private investment in developing countries 1995,” International Finance Corporation, Discussion Paper 25. Grabowski, H. and Dennis. C. Mueller (1972) “Managerial and stockholder welfare models of firm expenditures,” Review of Economics and Statistics, 54: 9–24. Grunfels, Y. (1960) “The determinants of corporate investment,” in A.C. Harberger (ed.), The Demand for Durable Goods, Chicago, IL: Chicago University Press, 211–266. Gugler, K. and B.B. Yurtoxlu (2003) “Average Q , marginal Q , and the relation between ownership and performance,” Economics Letters, 78: 379–384. Gugler, K., D. Mueller, and B.B. Yurtoxlu (2004) “Marginal q, Tobin’s q, cash flow and investment,” Southern Economic Journal, 70: 512–531. Gugler, K., D. Mueller, and B.B. Yurtoxlu (2005) “Corporate governance and the determinants of investment,” manuscript, University of Vienna. Hadlock, C.J. (1998) “Ownership, liquidity, and investment,” RAND Journal of Economics, 29: 487–508. Hayashi, F. (1982) “Tobin’s Marginal q and average q: a neoclassical interpretation,” Econometrica, 50: 213–224. Hoshi, T., A. Kashyap, and D. Scharfstein (1991) “Corporate structure, liquidity, and investment: evidence from Japanese industrial groups,” Quarterly Journal of Economics, 106: 33–60. Hubbard, R.G. (1998) “Capital market imperfections and investment,” Journal of Economic Literature, 36: 193–225.
Profitability, liquidity, and investment 197 Jensen, M. (1986) “The agency costs of free cash flow: corporate finance and takeovers,” American Economic Review, 76: 323–329. Johnson, S., P. Boone, A. Breach, and E. Friedman (2000) “Corporate governance in the Asian financial crisis,” Journal of Financial Economics, 58: 141–186. Jorgenson, D.W. (1967) “The theory of investment behavior,” In R. Ferber (ed.), Determinants of Investment Behavior, New York: Columbia University Press. Kadapakkam P.-R., P. Kumar, and L. Riddick (1998) “The impact of cash flows and firm size on investment: the international evidence,” Journal of Banking and Finance, 22: 293–320. Kaplan, Steven N. and Luigi Zingales (1997) “Do investment-cash flow sensitivities provide useful measures of financing constraints?” Quarterly Journal of Economics, 112: 169–215. Kathuria, R. and Dennis C. Mueller (1995) “Investment and cash flow: Asymmetric information or managerial discretion,” Empirica, 22: 211–234. Klenow, P. J. and A. Rodríguez-Clare (1997) “The neoclassical revival in growth economics: has it gone too far?” in B. Bernanke and J. Rotemberg (eds), NBER Macroeconomics Annual, MIT Press, Cambridge, 73–103. Kuh E. and J.R. Meyer (1957) The Investment Decision: An Empirical Study, Cambridge, MA: Harvard University Press. La Porta, R., F. Lopez-de-Silanes, and A. Shleifer (1997) “Legal determinants of external finance,” The Journal of Finance, 52: 1131–1150. Levine, R. and D. Renelt (1992) “A sensitivity analysis of cross-country growth regressions,” American Economic Review, 82: 942–963. Marris, R. (1964) The Economic Theory of Managerial Capitalism, London: Macmillan. Marris, R. (1998) Managerial Capitalism in Retrospect, London: Macmillan. Meyer, J. R. and R.R. Glauber (1964) Investment Decisions, Economic Forecasting, and Public Policy, Cambridge, MA: Harvard University Press. Modigliani F. and M.H. Miller (1958) “The cost of capital, corporation finance and the theory of investment,” American Economic Review, 48: 261–297. Mueller, D.C. (1967) “The firm decision process: an econometric investigation,” Quarterly Journal of Economics, 81: 58–87. Mueller, D.C. (2003) The Corporation: Investment, Mergers, and Growth, Routledge, London and New York. Mueller, D.C. and E. Reardon (1993) “Rates of return on corporate investment,” Southern Economic Journal, 60: 430–453. Mueller, Dennis C. and B.B. Yurtoxlu (2000) “Country legal environments and corporate investment performance,” German Economic Review, 1: 187–220. Myers, S.C. and N.S. Majluf (1984) “Corporate financing and investment decisions when firms have information that investors do not,” Journal of Financial Economics, 13: 187–221. Nelson, Richard R., and H. Pack (1999) “The Asian miracle and modern growth theory,’ The Economic Journal, 109: 416–436. Oliner, S.D. and G.D. Rudebusch (1992) “Sources of financing hierarchy for business investment,” Review of Economics and Statistics, 74: 643–654. Osterberg, W.P. (1989) “Tobin’s q, investment, and the endogenous adjustment of financial structure,” Journal of Public Economics, 40: 293–318. Perotti, E. and S. Gelfer, (2001) “Red barons or robber barons ? Governance and financing in Russian FIGs,” European Economic Review, 45: 1601–1617. Pesaran, H.M. and R. Smith (1995) “Estimating long-run relationships from heterogeneous Panels,” Journal of Econometrics, 68: 79–113.
198
B. Burçin Yurtoxlu
Roe, M. (2003) “Delaware’s competition,” Harvard Law Review, 117: 591–645. Schaller, H. (1993) “Asymmetric information, liquidity constraints, and Canadian investment,” Canadian Journal of Economics, 26: 552–574. Schiantarelli, F. (1996) “Financial constraints and investment: Methodological issues and international evidence,” Oxford Review of Economic Policy, 12: 70–89. Schiantarelli, F. and A. Sembenelli (2000) “Form of ownership and financial constraints: panel data evidence from flow of funds and investment equations,” Empirica, 27: 175–192. Shin, H.H. and Y.S. Park (1999) “Financing constraints and internal capital markets: evidence from Korean ‘chaebols’,” Journal of Corporate Finance, 5: 169–191. Singh, A. (1999) “Asian capitalism and financial crisis,” in Global Instability and World Economic Governance by John Grieve Smith and Jonathan Michie (eds), London: Routledge. Singh, Ajit, Singh, Alaka, and Weisse, B. (2003) “Corporate governance, competition, the new international financial architecture and large corporations in emerging markets,” Management of Capital Flows, UNCTAD Geneva, 1–70. Stiglitz, J. and A. Weiss (1981) “Credit rationing in markets with imperfect information,” American Economic Review, 71: 393–410. Walker, M. (2001) “Internal capital markets and investment efficiency: evidence from Japanese groups,” Krannert School of Management Purdue University. WBES :World Business Environment Survey (2000) IFC, Worldbank Group. Williamson, O.E. (1975) Markets and Hierarchies, New York: The Free Press. Windmeijer, F. (1995) “A note on R2 in the instrumental variables model,” Journal of Quantitative Economics, 11: 257–261. Young, A. (1995) “The tyranny of numbers: Confronting the statistical reality of the East Asian growth experience,” Quarterly Journal of Economics, 110: 641–680. Yurtoxlu, B.B. (2000) “Ownership, control and performance of Turkish listed companies,” Empirica, 27: 193–222. Yurtoxlu, B.B. (2003) “Corporate governance and implications for minority shareholders in Turkey,” Journal of Corporate Ownership & Control, 1: 72–86.
8
Ownership concentration and corporate performance of Turkish companies Halit Gönenç
Introduction One of the most typical feature of emerging markets is poor functioning financial markets. Khanna and Palepu (2000) state that “financial markets in emerging markets are characterized by inadequate disclosure and weak corporate governance.” Moreover, “intermediaries such as financial analysts, mutual funds, investment bankers, venture capitalists, and the financial press are not fully evolved.” Since market imperfections make external financing very costly in emerging countries, firms finance their investments internally by creating business groups. Firms affiliated with business groups can benefit from instituting internal relations to mitigate external market failures.1 Similar institutional context is observed in Turkey. Most of the Turkish corporations listed in stock exchange are affiliated with each other within a business group and have highly concentrated equity ownership. Business groups are organized around a holding company and a bank serving as the main financial source of the business group, which creates well-organized internal capital market. In addition to internal financing opportunities of firms in business groups, Turkish stock market, the Istanbul Stock Exchange (ISE), shows a high development compared to the market capitalization with other emerging markets from 1990 to 1999 (Figure 8.1). The ratio of market capitalization to GDP increases from 12.6 percent with 110 listed firms in 1990 to 60.7 percent with 285 listed firms in 1999. In this chapter, we examine the effect of concentrated equity ownership measured by the total percentage of shares owned by the three largest shareholders on the firm performance of Turkish industrial firms operated in this institutional context. In fact, more than 85 percent of sample companies are controlled by large shareholders in Turkey. Therefore, our examination is related to the hypothesis stating that controlling shareholders creates an agency problem for the expropriation of minority shareholders. It is possible that concentrated ownership could potentially affect large shareholders to worry more about their own interests rather than outside shareholders. However, single or small groups of large shareholders can improve the degree of monitoring and thus lead to a better performance. Even though manager–owner conflict is less pronounced in highly concentrated equity ownership structure, we also aim to explore agency problems
200
Halit Gönenç
Market capitalization % of GDP
2.50 1990
2.25
1999
2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 US
UK
Japan
S. Africa
Singapure
Chile
Malaysia
Israel
S. Korea
Turkey
Philippines
Portugal
Thailand
India
Indonesia
Brazil
Chine
Peru
Argentina
Poland
Czech Rep.
Pakistan
Rus. Fed.
Sri Lanka
0.00
Figure 8.1 Market capitalization to GDP ratios for several selected countries: 1990 and 1999. Source: The World Bank Group, WDI Data Query. http://devdata.worldbank.org/data-query
between managers and outside shareholders. Our empirical analysis is based on the data of 185 industrial firms in the 1992–1998 period. We primarily focus on the belief that ownership structure is endogenous. The consideration is that the firm performance is at least likely to affect ownership structure as well as ownership structure is to affect performance. If the performance affects ownership structure, owners could easily increase or decrease their shareholdings. We accept that the endogeneity of ownership is also valid for the concentrated ownership. This is opposing the assumption in the La Porta et al. (2002) model. They assume that equity ownership of controlling shareholder is exogenously determined by history and life-cycle of the firm, and the sale of equity is not considered by the entrepreneur. We accept that there should be a market for share stakes to argue that the relationship between ownership structure and the firm performance should be examined simultaneously. Yurtoxlu (2000) observes, in the 1987–1997 period, that large shareholders in Turkish companies restructure their ownership within business groups. We also provide the evidence for changing percentage of shares owned by the three largest shareholders to support the argument that the largest shareholders vary their holdings of stock according to the future performance expectations. The evidence shows that concentrated ownership in about 75 percent of sample firms changes on average from the lowest 0.6 percent to the highest 13 percent in the entire sample period. In most countries large shareholders establish control over a firm through pyramid structure and cross-holdings among firms. The separation of ownership and control is also achieved through pyramidical and complex ownership structures by Turkish companies. This type of structure creates divergence between control rights and cash flow rights. Large shareholders may have control over a firm despite the little cash flow rights. Claessens et al. (2002) suggest that studying the
Ownership concentration of Turkish companies
201
separation of ownership and control requires data on both control and cash flow rights. Cash flow right of the largest shareholders is the share of the dividends. To argue this issue, for our sample, we examine dividend payout ratios against the percentage of shares owned by the three large shareholders. Moreover, we argue that the major ownership group that has a share in concentrated ownership becomes more important because business groups operate as well-organized internal capital markets with complex ownership structures in Turkey. Gürsoy and Aydoxan (2002) point out that identifying controlling owners keeps a vital role in a performance–ownership relationship. Therefore, we control the effect of major ownership groups in estimation models by using dummy variables for the state, holdings, financial institutions, family members, and foreign investors. We find some evidence with both OLS and 2SLS estimations that the presence of Turkish State has positive effect on market-to-book ratio of assets. The effect of financial institutions, on the other hand, is negative. Our results, however, do not reveal any information that the performance of firms affiliated with a business group is different from unaffiliated firms. Larger equity holdings by families or foreign investors do not also lead to a better performance. The chapter is organized as follows; in the section on Legal protection, ownership concentration, and corporate performance, we highlight the major points from the prior research on legal protection of outside investors, concentrated ownership, and the relationship between ownership structure and the firm performance. The section on sample, data, and model presents the data and definition of variables. The section on empirical results provides evidence regarding the relationship between concentrated ownership and the firm performance. The final section concludes.
Legal protection, ownership concentration, and corporate performance A recent research shows that differences in corporate governance mechanisms are important to understand the differences in the development of financial markets among countries. The literature has tended to categorize governance systems based on two classifications: (1) bank-centered versus market-centered and (2) the amount of legal protection afforded to outside investors (both shareholders and creditors). La Porta et al. (1997) show that better legal protection of outside investors leads to the expansion of financial markets with a higher number of listed firms and more valuable stock markets. Moreover, La Porta et al. (2000b) contend that the legal approach on protection of outside investors is superior to the understanding of governance systems than the conventional bank-centered and market-centered distinction. Corporate governance mechanisms According to La Porta et al. (2000b), “corporate governance is, to a large extent, a set of mechanisms through which outside investors protect themselves against expropriation by the insiders (both managers and controlling shareholders).”
202
Halit Gönenç
La Porta et al. (1998) show that countries adopt different law and its enforcement to protect outside investors. They argue that, in general, common law countries offer greater protection than civil law countries. Protection provided to both creditors and shareholders is the least in French-civil-law countries, in which Turkey is included. Their results show that in some countries laws are not sufficient to protect investors from expropriation of their funds. Shleifer and Vishny (1997) claim that if legal protection is poor, investors with a higher percentage of shares could have more effective control rights. In effect, concentration of ownership by one or more large equity blockholders leverages up legal protection. To support this argument, La Porta et al. (1998), with a sample of large firms from 49 countries, find a strong negative correlation between concentrated ownership, as measured by the combined stake of the three largest shareholders, and the quality of legal protection of investors. La Porta et al. (1999) also show that ownership structure of large corporations in 27 wealthy economies, except in economies with very good legal shareholder protection, is highly concentrated. Families or the state typically controls firms in countries with poor investor protection. Claessens et al. (2000) find that more than two-thirds of firms in nine East Asian countries are controlled by a single shareholder, and control is most pronounced among familycontrolled firms, except in Japan, which has fairly good shareholder protection. Lins (2003) examine 1,857 firms from 22 emerging markets. The results show that 58 percent of sample firms are controlled by one or more blockholders. Managers and their families are the dominant type of controlling blockholders in emerging markets. Implication of the findings for the earlier studies is that, share ownership is concentrated on most of both developed and developing countries. Empirical evidence on ownership and performance Shleifer and Vishny (1997) argue the benefits and costs associated with large shareholding that is an alternative corporate governance mechanism for legal protection. The correlation of concentrated ownership with the firm performance depends on the higher effect of either benefits or costs. In countries with poor legal rules protecting outside investors, owners with large ownership percentage can help reduce the agency problem between managers and outside shareholders described by Jensen and Meckling (1976) because they have power on managers to make decision against shareholders. Single or small group of large shareholders can also improve the degree of monitoring and thus lead to better performance. However, it is possible that concentrated ownership could potentially affect large shareholders to worry more about their own interests and not those of outside shareholders. In this case, as Shleifer and Vishny (1997) argue, the fundamental agency problem occurs between outside investors and controlling shareholders that have nearly full control over the managers. In countries with poor legal protection, firms are typically controlled by one or small group of owners. La Porta et al. (2000b) describe several ways for a major owner how to retain control of a firm. One of them is the pyramidical structure. In this structure, a holding company owned by the entrepreneur is at the top of
Ownership concentration of Turkish companies
203
the pyramid and controls several subsidiaries. Hence, a chain of ownership relations is created. The main feature of this type of control is to own small fraction of cash flows while having superior voting rights. Cross-shareholding is another way of keeping control. In this form, it is very hard for outsiders to gain the control of one group firm unless they buy all of them. The effect of the concentrated ownership on the firm performance in this context depends on the divergence between control and cash flow rights. La Porta et al. (2002) argue that the power of the controlling shareholders to expropriate outside investors is moderated by their financial incentives based on cash-flow ownership. La Porta et al. hypothesize that since expropriation is costly, higher cash flow ownership should lead to lower expropriation. By using large firm data from 27 of the wealthiest economies around the world, they find some evidence that higher incentives from cash flow ownership are associated with higher valuations. Claessens et al. (2002) also find that higher cash flow ownership in firms of several East Asian countries affects the firm value positively as opposed to stronger control rights that have adverse effects on it. However, Lins (2003) explore that control rights is 2.7 times bigger than cash flow rights due to having pyramid ownership structure and using non voting shares. Increasing the difference between control and cash flow rights decreases firm value. Moreover, this difference is large in countries where legal protection of outside investors is weak. Thus, the results for emerging markets are consistent with the hypothesis that monitoring of large investors is governing incentive for the expropriation of minority shareholders. Regarding the Turkish firms, Yurtoxlu (2000) finds a significant negative relationship between three firm performance measures and ownership concentration. Further, he shows that largest direct shareholdings have a small effect on performance, increasing the difference between control and cash flow rights which has a substantially negative impact on market valuation. On the other hand, Gürsoy and Aydoxan (2002) find that higher concentration leads to better market performance (price earning ratio and average returns), but lower accounting performance (return on assets and return on equity). They point out that identifying controlling owners seems to have an important role in a performance–ownership relationship. The other issue involves whether analyzing the effect of large shareholders on the firm performance should be tested in a simultaneous equation framework. In this sense, Demsetz and Villalonga (2001) argue that; “The possible divergence between insider and market-based expectations for the firm performance creates an incentive for management (or large shareholders) to vary its holdings of stock in accord with its expectation regarding future performance.” The consideration is that the firm performance is at least likely to affect ownership structure, as ownership structure is to affect performance. The main point in this issue is that there should be a market for share stakes. If performance affects ownership structure, owners could easily increase or decrease their shareholdings. Demsetz and Villalonga (2001) give leverage buyout and stock options held by managers as examples of how expected performance can cause ownership structure to change. They also point out that the market for corporate
204
Halit Gönenç
control involves the acquisition and sale of shares in ordinary exchange transactions in addition to hostile takeovers and management buyouts. Demsetz and Lehn (1985) treat ownership as endogenous and find that ownership fails to explain variations in the firm performance. The important issue here involves that analyzing the effect of equity ownership on the firm performance should best be tested in a simultaneous equation framework. Loderer and Martin (1997) take both Tobin’s q and managerial shareholding as endogenous for a simultaneous equation model and find that performance affects the size of managers’ stockholdings, but there is no evidence that larger stockholders lead to better performance. Cho (1998) also show by using a system of three equations that the firm performance affects ownership structure but not vice-versa. Himmelberg et al. (1999) do not find evidence that changes in managerial ownership affect the firm performance after controlling both for observed firm characteristics and firm fixed effects. In a recent study, Demsetz and Villalonga (2001) also find with the estimates of the two-equation model that variations across firms in observed ownership structures do not result in systematic variations in observed firm performances.2 Moreover, the 2SLS estimation results for the effect of the firm performance on ownership by large shareholders produce significant and negative coefficient for Tobin’s q. They interpret this finding that managers and outside largest shareholders prefer to hold fewer shares when firms seem to be doing well, because they expect that good performance will be followed by poorer performance.
Sample, data, and model Sample The analysis uses a dataset of Turkish industrial companies listed in the ISE for the period from 1992 to 1998. The data for this study basically comes from two sets of sources: (1) the data used for performance measures and control variables are obtained from a local source of the ISE database called FinNet, (2) the data on the equity ownership is gathered from a country specific source, the yearbook of ISE companies. The sample covers the period from 1992 to 1998 for 185 industrial firms that the data is available to calculate all variables. The issue of the Yearbook of Companies published by the Documentation Department of the ISE at the end of each year provides data on the first level of shareholding (direct ownership) in all Turkish publicly traded companies. We created ownership groups based on provided information in this book. The concentrated ownership is the sum of the percentage of three largest shareholders. Table 8.1 reports the mean values, standard deviation, minimum, and maximum values of variables for the 185 sample firms. All variables, except market risk of a firm’s stock (BETA) and the firm specific risk (SE-standard error), are the average of annual values for the 6 years from 1992 to 1998.3 Therefore, the symbolic representations of variables begin with a notation of “Av.” In this period, Turkish industrial firms, on average, operate with negative net working capital (AvNWC) and spend less for capital investments (AvINVEST). Mean
Ownership concentration of Turkish companies
205
Table 8.1 Sample descriptive statistics on ISE listed firms
AvMB AvROA AvDEBT AvASSET AvMAN AvCONT3 AvINVEST AvNWC AvDividend payout ratio BETA (market risk) SE (firm risk) STATE HOLDING FININST FAMILY FOREIGN
N
Mean
Std dev.
Min.
Max.
185 185 185 185 185 185 185 185 185 185 185 36 128 61 51 34
1.88 0.09 0.53 15.23 0.11 0.66 0.03 0.35 0.42 0.90 0.22 0.21 0.26 0.21 0.07 0.24
1.26 0.11 0.20 1.26 0.18 0.17 0.05 0.20 0.28 0.25 0.08 0.30 0.22 0.22 0.10 0.24
0.56 0.30 0.01 11.67 0.00 0.04 0.00 0.89 0.00 0.33 0.07 0.00 0.00 0.00 0.00 0.06
10.08 0.48 1.01 18.40 0.97 0.99 0.33 0.33 2.02 2.10 0.76 0.99 0.84 0.94 0.94 0.83
Notes Statistics on 185 Turkish industrial firms listed in Istanbul Stock Exchange (ISE). All variables given in this table, except market risk of a firm’s stock (BETA) and the firm specific risk (SE – standard error), are the average of annual values for the 6 years from 1992 to 1998.
values market-to-book ratio (AvMB) and return on assets (AvROA) are 1.88 and 0.09, respectively. Concentrated ownership (AvCONT3) has a mean value of 66 percent. The mean value of the percentage of shares owned by managers is 11 percent. Holding companies have shares in 128 companies with mean value of 26 percent. This implies that 57 out of 185 firms are not affiliated with a business group. The second largest ownership belongs to financial companies with 61 firms. The mean value of percentage of shares owned by financial institutions is 21 percent. The Turkish State has the same mean value of ownership shares with financial institutions, but only 36 firms are owned by the State. Family members own shares in 51 firms with mean value of 7 percent. The definitions of the variables are provided in the next section. Specification of simultaneous equation model and variables Our primary focus is to examine the potential endogeneity issue of ownership concentration. The endogeneity refers possible divergence between large shareholders that control the firm and market or accounting-based expectations for the firm performance. Large shareholders restructure their ownership within or without business group based on future expectations for the firm performance. To address this issue, our simultaneous equation system has two following equations. Our estimation depends on both OLS and 2SLS methods. To formulate the equations, we benefit from two past studies, Demsetz and Villalonga (2001), and Cho (1998).
206
Halit Gönenç
Performance f (investments, leverage, size, ownership) Ownership g (performance, size, liquidity, risk, industry) Main ownership variables are percentage of shares owned by management (AvMAN) and percentage of shares owned by the three largest shareholders (AvCONT3). The firm performance equation aims to measure if the concentrated ownership is systematically related to the firm performance. Conflict of interest between managers and outside shareholders is another version of agency problem that is less likely to be observed in highly concentrated equity ownership structure. To clear this issue out for Turkish firms, shareholdings by managers also appear as another explanatory variable in the firm performance equation. Our performance measures are market-to-book ratio (AvMB) and return on assets (AvROA). Market value is calculated as the sum of the market value of common stock and the book value of total debt. The denominator of the ratio is the book value of total assets. Return on assets is the ratio of net income to the book value of assets. Demsetz and Villalonga (2001) state that there are two major differences between market based (Tobin’s q) and accounting based performance measures (profit rate). The first difference is in time perspective, backward-looking for accounting profit rate and forward looking for q. The approach for who is actually measuring performance creates the second difference. Accounting profit is constrained by standards, but it is not affected by the psychology of investors. q, however, is constrained by investors’ acumen, optimism, or pessimism in estimation of future events. This discussion shows that each performance measure carries its own advantages and disadvantages. The consideration of reverse causations requires working with simultaneous equation models. Reverse causations means that variation in the firm performance affects variation in percentage of shares owned by three largest shareholders as well as is affected by it. If this is the case, the results obtained from single equation models are meaningless. As we will show in the section that we present as empirical results, OLS estimations provide evidence in favor of reverse causations. Therefore, AvMB or AvROA appear as an explanatory variable in ownership equation. Variable AvINVEST is capital investments as a fraction of the book value of assets and included in the firm performance equation to examine the issue that a higher level of investment may lead to a greater value. The argument here is that monitoring by the largest shareholders would influence firm management to be more diligent in finding positive NPV projects that will add value to the firm. There is another possibility that controlling shareholders would undertake many investments but not always for the right reasons due to the agency problem about expropriation of minority shareholders. In the firm performance equation, we control for leverage (AvDEBT). According to the pecking order theory, profitability (performance) should be negatively related to leverage. As far as Turkey is concerned, we could argue the following for the negative effect of leverage on the firm performance: Turkish industrial firms have very high short-term debt percentage in total debt ratio. The reason for a high percentage of short-term debt can be explained with a
Ownership concentration of Turkish companies
207
substantial amount of inflation, volatility, and political and economic uncertainty in the Turkish economy. This high degree of uncertainty in the economy makes investors to invest for very short-terms with high interest rates. AvASSET is a measure of firm size in both equations. Size is generally found negatively related to the firm performance and ownership structure. However, in Turkey, business groups create functional substitutes to external capital markets for mostly large firms. Therefore, we could observe positive relationship between size and the firm performance, and between size and concentrated ownership. Control variables for ownership are all dummy variables. The Turkish State, holding companies, financial institutions, family members and foreign investors are major types of controlling shareholders. The effects of all these type of owners are controlled for the influence of the presence of controlling shareholders on performance. Variable STATE is for state-controlled firms. State-controlled firms appeared in the sample of our study operate in highly regulated and oligopoly industries, which may give them to have superior performance. Variable HOLDING refers to a holding company participation in the ownership of a firm. Holding companies appear to be managerial center of business groups in Turkey. We add this variable to compare the performance of firms affiliated with a business group with unaffiliated firms. Khanna and Palepu (2000) discuss potential benefits and costs of diversified business groups in emerging markets and find that accounting and stock market measures of the firm performance of affiliates of diversified Indian business groups initially decline with group diversification and subsequently increase once group diversification exceeds a certain level. Ferris et al. (2003) show that chaebol-affiliated firms in South Korea suffer a value loss relative to nonaffiliated firms. Based on these results from these two studies, we would expect to find either positive or negative effects of holding ownership on performance. Financial institutions, especially banks, have a special feature of the Turkish business groups. Almost every private bank is under the control of a business group and serves as the main financial source of the business group. Past international evidence shows that bank ownership affects firm performance. Some authors (Aoki (1990), Prowse (1992), and Sheard (1989)) argue that banks are important monitors and help to reduce agency costs. Other authors (Hoshi et al. (1990) and Kaplan and Minton (1994)) contend that main banks are especially helpful during times of financial distress. Recently Gorton and Schmid (2000) find that performance of German firms improves to the extent that equity control rights of German universal banks are concentrated. On the other hand, Weinstein and Yafeh (1998) find that main banks’ client firms do not perform better than other Japanese firms do. These authors believe that main banks discourage risk taking by their client firms in part because the banks are also major debtholders of these firms. There could be several reasons for family members to provide good monitoring. Family members occupy the management of board in most of the family-controlled firms. They aim at keeping the control of business activity within the family. Family members can align interest with other shareholders because of their reputation. However, La Porta et al. (1999) argue that family members can implement policies that benefit themselves due to immense voting power and frequent involvement in
208
Halit Gönenç
management. All these reasons provide the information for a relationship between family holdings of shares (FAMILY) and the firm performance. Foreign owners (FOREIGN) seem as a strategic partner for a domestic corporation when, especially, they provide superior technology know-how. Thus, foreign ownership encourages greater monitoring. However, there could be a casual relationship indicating that foreign investors are simply attracted to better performing companies. Yurtoxlu (2000) argues that foreign institutional type of investors invest in Turkish companies for diversification not for control purposes. Variable AvNWC is used to examine the effect that liquidity may have on concentrated ownership. Firms with high liquidity are more subject to agency cost. Therefore, this variables controls how a small group of large shareholders use firm liquidity. If this group worries about their own interests and not those of outside minority shareholders, we should observe a positive relationship between liquidity and concentrated ownership. On the other hand, if owners with large controlling shares align their interest with those of minority shareholders, there should be a negative relationship between liquidity and concentrated ownership. A similar argument is also valid for equity ownership by managers. BETA and SE are related to the risks that affect stock ownership. BETA is for market risk and measured by a regression of the monthly return on a stock on the ISE100, which is a value weighted market portfolio, for the period 1992–1998 (BETA). Firm specific risk (SE), that is, the standard error of estimate from a regression of the monthly return on a stock on the ISE100, which is a value weighted market portfolio, for the period 1992–1998 (SE). BETA is used to measure the effect of market risk on the level of either concentrated ownership or managers. Higher market risk indicates higher return potential for shareholders. But, this potential requires a better monitoring by concentrating on outside investor shareholdings to make sure that they will not be adversely affected by rapidly changing economic conditions. In this case, the estimated sign of BETA should be positive. When concentrated owners or managers cannot make predictions effectively about future economic trends, as it is in Turkish economy, market risk makes higher ownership more costly. This implies that the higher the market risks, the lower the percentage of shares by the largest shareholders and managers. Firm specific risk, according to Demsetz and Villalonga (2001), is especially relevant measure of the risk of putting a large part of an investor wealth in the stock of a firm. Therefore, SE causes large shareholders to give more attention to the contribution of monitoring activities to the firm performance. Hence, higher firm’s specific risk should increase the value of concentrated ownership. When they are risk averse, however, they would like to reduce their ownership.
Empirical results The relationship between controlling shareholders and dividend payout ratio In the agency context where the fundamental problems occur between outside investors and controlling shareholders, dividends can play an important role.
Ownership concentration of Turkish companies
209
Dividend payment reduces incentives of controlling shareholders for expropriation of outside minority shareholders. In a firm with a highly concentrated ownership structure, when corporate earnings are distributed to investors by dividends, controlling shareholders are no longer capable of using these earnings to benefit themselves. This view, in fact, depends on the legal protection of outside shareholders. Minority shareholders use their legal powers to force companies to pay out cash. If shareholder protection is poor, as it is in Turkey, we should observe distribution of earnings in very low rates. This kind of finding also implies the possible existence of expropriation of minority shareholders. In an alternative agency view, dividends are a substitute for legal protection. Largest owners or entrepreneurs that control the firm need to establish a good reputation to raise external funds from outside shareholders who expect to be expropriated. La Porta et al. (2000a) state that “one way to establish such a reputation is by paying dividends, which reduces what is left for expropriation.” Moreover, in emerging markets including Turkey, business groups with pyramid ownership structures and cross-holdings among firms create internal financing opportunities. Even though paying dividend would not be a main consideration to establish a reputation, it would be a good way to ensure internal financing to the other firms in a business group. This view implies that dividend payout ratios should be higher with weak legal protection of shareholders. La Porta et al. (2002) show that an important source of substitution for legal protection is the cash flow rights by controlling shareholders. They claim that higher cash-flow ownership should lead to lower expropriation. Claessens et al. (2002) also state that the effect of the concentrated ownership on the firm performance depends on the divergence between control and cash flow rights, and they find evidence to support this statement. Cash flow right of the largest shareholders is the share of dividends. We compare, graphically, earnings per share and dividend to earnings ratio by looking at the association between these two and the percentage of ownership by three largest shareholders. With this kind of graphical presentation, we aim to show that the separation of control rights and cash flow rights for controlling shareholders should not be the main consideration for our sample firms in Turkey. Instead, we mainly focus on the major ownership group in controlling shareholders. In this sense, La Porta et al. (2002) examine the mean difference between cash flow rights and control rights and reach small differences for both civil and common law countries. Yurtoglu (2000) find small difference in mean value of the difference between cash flow and control rights for Turkish companies for the same period with our sample, 1992–1998. The difference does not exist for median level firms. It seems that larger differences occur at very high level of concentrated ownerships. Figure 8.2 presents two graphs; the first graph shows the relationship between mean value of earnings per share and concentrated ownership, and the second graph shows the relationship between mean value of dividend payout ratio, which is total cash dividends as a percentage of earnings, and concentrated ownership. Earnings refer to net income after taxes and interest. Ownership by the three largest shareholders is classified according to the different percentages of
210
Halit Gönenç
Mean earnings per share
(a) 3,000 2,500 2,000 1,500 1,000 500 0 0.00–0.50 (23)
0.50–0.60 (22)
0.60–0.65 (23)
0.65–0.70 (24)
0.70–0.75 (24)
0.75–0.80 (29)
0.80–0.85 (23)
0.85–1.00 (17)
Ownership by three largest dhareholders (No. of firms) (b) Mean dividend payout ratio
0.600 0.500 0.400 0.300 0.200 0.100 0.000 0.00–0.50 (23)
0.50–0.60 (22)
0.60–0.65 (23)
0.65–0.70 (24)
0.70–0.75 (24)
0.75–0.80 (29)
0.80–0.85 (23)
0.85–1.00 (17)
Ownership by three largest shareholders (No. of firms)
Figure 8.2 Association between both earnings and dividends and concentrated ownership for the period 1992–1998. Panel a: The relationship between earnings per share and ownership of largest shareholders. Panel b: The relationship between dividend payout ratio and ownership of largest shareholders. Notes Earnings refer net income after taxes and interest. Dividend payout ratio is total cash dividends as a percentage of earnings. Ownership by the three largest shareholders is classified based on the different percentage of shareholdings. All three variables are the average of annual values for the 6 years from 1992 to 1998.
shareholdings. Concentrated ownership percentage is less than 50 percent in 23 firms, between 50 and 60 percent in 22, between 60 and 65 percent in 23, between 65 and 70 percent in 24, between 70 and 75 percent in 24, between 70 and 80 percent in 29, between 80 and 85 percent in 23, between 85 and 100 percent in 17 firms. Earnings per share and dividend payout ratio, like other variables, are the average of annual values for the 6 years from 1992 to 1998. The comparison of the two graphs indicates that dividend payout ratios and earnings per share do not follow the same pattern. Firms with low level of earnings per share have paid dividends nearly as much as firms with high levels of earnings
Ownership concentration of Turkish companies
211
per share. In other words, dividend payout ratios are similar for different concentrated ownership groups, except for the highly concentrated ownership group with 85 percent or higher percentage of ownership by the three largest shareholders. We interpret these results that dividends are used to substitute for legal protection in Turkey. Firms pay dividends either to establish good reputation or to create internal financing among firms in the business group. The reason for this is that the difference between the percentage of control and cash flow rights is not large, except for the highly concentrated ownership group. If the difference between two rights were large, we would have seen lower dividend payments when earnings per share were low for the need of internal funds. The lowest dividend payout ratio of the highly concentrated ownership group increases the amount of internal funds, but it may also indicate expropriation of minority shareholders. Moreover, as we try to explain in the previous paragraph, the reason of lower dividend payout ratio is because the difference between control and cash flow rights is larger for this group. Based on the findings by Claessens et al. (2002), this should cause lower performance of firms in this group. To examine this issue, Figure 8.3 presents the relationship between market-to-book-ratio as a market performance measure and concentrated ownership. First of all, we observe that the highest average market performance is associated with the highest concentrated ownership level. This result is inconsistent with Claessens et al. (2002) result. Moreover, even though the firm market value increases with the controller shareholdings, there is no big difference, especially for ownership by the largest shareholder groups of between 50 and 85 percent. For 23 firms that have ownership by the largest shareholders of 0–50 percent show lower performance than other groups do. As a result, we can say that major owner categories in the concentrated ownership are more important than the type of rights.
Mean market-to-book ratio
2.5 2 1.5 1 0.5 0
0.00–0.50 0.50–0.60 0.60–0.65 0.65–0.70 0.70–0.75 0.75–0.80 0.80–0.85 0.85–1.00 (22) (23) (24) (24) (29) (23) (23) (17) Ownership by three largest shareholders (No. of firms)
Figure 8.3 Association between market-to-book ratio and concentrated ownership for the period 1992–1998.
212
Halit Gönenç
Changes in concentrated ownership and its categories In this section, we examine percentage changes in concentrated ownership and its categories over the period 1992–1998. This examination aims to show that the concentrated ownership pattern of Turkish companies is not stable. Large shareholders may change their ownership by selling their shares by using ordinary exchange transaction based on future performance expectations. In fact, large shareholders that worry about the future performance of firms may restructure their ownership within a business group to increase their monitoring capabilities. If we show that the percentage of ownership by large shareholders changes over the time, then we could say that equity ownership of controlling shareholder is endogenously determined. Table 8.2 gives the percentage of changes and breakdowns of these changes in both concentrated ownership, which includes the percentage of shares owned by the three largest shareholders, and shareholder categories appeared in concentrated ownership. Shareholder categories are managers, holding companies, the Turkish State, foreign investors, family members, financial institutions, and non-financial firms. The first two columns are summary statics including the number of firms and the mean value of the percentage of shares for concentrated ownership and its categories. For example, in 1998, there are 185 firms and mean value of concentrated ownership is 66.3 percent. Managers appear to be among the three largest owners in 84 out of total 185 firms with a mean value of 20.9 percent. Holding companies participate in concentrated ownership in 93 firms with mean value of 38.3 percent. In 1998, we observe the Turkish State in 16 firms with 37.9 percent ownership, foreign companies in 28 firms with 34 percent, family members in 21 firms with 11.4 percent, financial institutions in 40 firms with 28.5 percent and non-financial firms in 93 firms with 32.2 percent ownership. All these figures imply highly concentrated and complex ownership structure. The next two columns in Table 8.2 with the subtitle of “change” show the frequency and mean value of changes in percentage of shares owned by the three largest shareholders and by shareholder categories appeared in concentrated ownership. The highest percentage of change in concentrated ownership is experienced in 1993 with mean value of 13 percent. The ratio of the frequency of concentrated ownership changes to the total number of firms is about 75 percent for the entire sample period. The other remaining columns in the table give the breakdown of these changes into positive (increase in ownership) and negative (decrease in ownership) changes, respectively. For example, we observe positive (negative) changes of concentrated ownership in 79 (72) out of the total 151 firms in 1998. The mean value of positive (negative) changes from 0 to 10 percent occur for 47 (52) firms, from 11 to 50 percent for 27 (20) firms as examples in 1998. Positive changes with greater than 50 percent in concentrated ownership occur for 5 firms. We use one break point (25 percent) rather than two for changes in ownership categories. The values for changes in ownership percentages for the different
Ownership concentration of Turkish companies
213
categories and breakdowns of these changes show the following features: Managers, family members, and financial institutions, on average, reduce their ownerships. However, holding companies have tended to increase their shareholdings. These results imply that major owners that are the main participants of a business group have reduced their shareholdings and concentrated in the holding companies. We observe that non-financial firms increase and decrease their shareholdings, even though the mean value of the percentage of shares remains almost stable over the sample period. As we mentioned in the previous section, foreign owners invest in Turkish firms for diversification, not for control purposes. At the time of high volatility in economic condition, they prefer to reduce their stakes. The figures show that their shareholdings decreases after a serious financial crisis in January– March of 1994 resulted in a 75 percent devaluation of the TL against the US$. Changes in ownership by the Turkish State are due to the privatization program applied. The relationship between concentrated ownership and firm performance Table 8.3 presents OLS and 2SLS regression results from two sets of equations. The first set with two separate models has the firm performance as the dependent variable. The fraction of shares owned by management is the independent variable in the model 1, and in the model 2, fraction of shares owned by the three largest shareholders is the independent variable. The second set has ownership by managers and the three largest shareholders as the dependent variables in different two models in which the firm performance is the independent variable. In 2SLS estimations, performance and ownership variables are endogenous variables in the context of simultaneous equation framework. The firm performance is measured with market-to-book ratio, AvMB, in Panel A of Table 8.3. Focusing first on OLS estimates for the firm market performance equation (model 2), we find that concentrated ownership variable, AvCONT3, appears to be positively related with the firm market performance with the significant estimated coefficient of 1.55 at 1 percent significance level. This result shows that ownership concentration improves the degree of monitoring and reduces, therefore, conflict of interest between controlling shareholders and minority shareholders. The output of control by a small group of owners is better performance. The coefficient of ownership by managers, AvMAN, is not significant at any conventional level (model 1). This result is consistent with previous studies stating that agency problem between managers and outside shareholders is less likely observed in highly concentrated equity ownership structure. Whichever ownership variable is used as the independent variable in the firm performance equations (model 1 and 2), the Turkish State, STATE, has positive and financial institutions, FININST, has negative sign, and both are significant. The presence of financial institutions in equity ownership reduces the firm performance. This finding is inconsistent with the study by Gorton and Schmid (2000)
Mean
Mean
Concentrated ownership categories (a) Managers 1992 10 0.334 1993 17 0.233 1994 44 0.216 1995 60 0.261 1996 65 0.209 1997 62 0.239 1998 84 0.209
N
5 13 34 38 35 39
0.147 0.071 0.158 0.233 0.066 0.041
2 7 20 11 15 15
0.670 0.255 0.452 0.295 0.275 0.390
1 4 11 7 11 9
7 6 4 4 2 5
N
0.008 0.028 0.097 0.029 0.065 0.056
1 3 9 4 4 6
N
22 43 75 75 85 72
N
Total
1.332 0.557 0.887 0.761 0.850 0.890
17 24 53 54 63 52
N
3 6 14 27 20 24
N
Total
0.201 0.451 0.263 0.448 0.321 0.311
Mean
4 18 22 19 22 20
N
2 2 9 11 11 10
N
0.016 0.087 0.136 0.080 0.067 0.073
Mean
0.144 0.234 0.233 0.193 0.221 0.236
Mean
0.10–0.50
0–0.25
0.028 0.038 0.030 0.033 0.027 0.024
Mean
0–0.10
Negative changes
0.078 0.134 0.089 0.093 0.077 0.083
Mean
Negative changes
Mean
0.25
0.964 1.049 0.999 0.717 0.768 0.734
Mean
0.50
Mean
N
Mean
N
0.223 0.187 0.257 0.237 0.232 0.196
0–0.25
14 13 10 17 15 27
Total
Mean
0.025 0.028 0.037 0.025 0.019 0.024
N
24 39 28 35 29 47
Mean
Positive changes
0.233 0.169 0.181 0.139 0.121 0.128
N
0.10–0.50
Change
45 58 42 56 46 79
Mean
N
N
Mean
0–0.10
Total
N
Mean
Positive changes
Change
Change concentrated ownership 1992 91 0.636 1993 108 0.676 67 0.130 1994 134 0.672 101 0.040 1995 152 0.659 117 0.007 1996 171 0.642 131 0.006 1997 185 0.637 131 0.007 1998 185 0.663 151 0.027
N
Table 8.2 Changes in concentrated ownership and its categories
0.651 0.660 — 0.769 — —
Mean
1 4 5 16 9 14
N
0.571 0.633 0.492 0.701 0.632 0.480
Mean
0.25
1 1 0 2 0 0
N
0.50
(b) Holding companies 1992 29 0.307 1993 42 0.325 1994 55 0.308 1995 61 0.329 1996 72 0.330 1997 83 0.356 1998 93 0.383 (c)Turkish state 1992 15 0.461 1993 18 0.428 1994 13 0.461 1995 15 0.426 1996 17 0.436 1997 14 0.473 1998 16 0.379 (d) Foreign owners 1992 10 0.484 1993 11 0.414 1994 12 0.503 1995 17 0.388 1996 19 0.366 1997 22 0.384 1998 28 0.340 (e) Family members 1992 3 0.317 1993 4 0.574 1994 14 0.276 1995 13 0.165 1996 11 0.083 1997 13 0.130 1998 21 0.114 9 11 9 20 11 27
3 4 1 1 1 1
2 3 1 1 3 3
0 0 2 2 1 1
0.038 0.121 0.025 0.275 0.147 0.201 0.131 0.124 0.127 0.340 0.026 0.096
0.115 0.165 0.115 0.087 0.627 0.940 0.097 0.618 0.089 0.197 0.007 0.205
16 27 30 35 29 41
9 11 11 6 3 6
2 4 3 4 5 6
1 3 4 6 3 3
— — 0.204 0.147 0.116 0.289
0.115 0.224 0.442 0.016 1.060 2.219
0.233 0.089 0.010 0.166 0.112 0.000
0.136 0.526 0.377 0.650 0.584 0.426
0 0 1 1 1 0
2 2 0 1 1 1
2 3 1 1 1 1
7 8 5 12 3 15
— — 0.124 0.034 0.116 —
0.115 0.013 — 0.016 0.023 0.006
0.003 0.030 0.010 0.166 0.112 0.000
0.061 0.050 0.022 0.029 0.001 0.075
0 0 1 1 0 1
0 1 1 0 2 2
1 1 0 0 0 0
2 3 4 8 8 12
— — 0.284 0.261 — 0.289
— 0.645 0.442 — 1.578 3.326
0.694 0.266 — — — —
0.400 1.796 0.820 1.581 0.803 0.863
1 3 2 4 2 2
0 1 2 3 2 3
6 7 10 5 2 5
7 16 21 15 18 14
0.097 0.618 0.384 0.370 0.046 0.453
— 0.011 0.047 0.122 0.020 0.338
0.314 0.247 0.140 0.441 0.016 0.115
0.087 0.157 0.198 0.224 0.119 0.231
1 0 1 2 2 0
1 2 2 2 1
3 4 8 2 2 4
6 13 14 9 15 11
0.097 — 0.111 0.119 0.046 —
0 0.011 0.047 0.033 0.020 0.187
0.003 0.023 0.034 0.132 0.016 0.055
0.045 0.061 0.056 0.044 0.067 0.070
— 0.618 0.657 0.622 — 0.453
0 — — 0.300 — 0.413
0.626 0.546 0.563 0.647 — 0.356
0.338 0.571 0.484 0.492 0.378 0.822
(Table 8.2 continued )
0 3 1 2 0 2
— 0 0 1 0 2
3 3 2 3 0 1
1 3 7 6 3 3
Mean
0.127 0.179 0.099 1.599 0.249 0.134 0.211 0.053 0.018 0.934 0.147 0.044
16 24 23 26 22 24
22 31 34 41 38 48
15 19 17 27 11 25
9 12 10 10 10 8 0.384 0.196 0.214 1.462 0.143 0.302
0.299 0.620 0.103 4.547 0.954 0.347 7 15 15 18 8 16
7 8 9 7 7 6 0.029 0.064 0.068 0.046 0.071 0.045
0.036 0.068 0.072 0.130 0.029 0.095
Mean
N
N
Mean
0–0.25
Total
N
Mean
Positive changes
Change
8 4 2 9 3 9
2 4 1 3 3 2
N
0.694 0.689 1.303 4.295 0.334 0.759
1.217 1.723 0.376 14.854 3.112 1.101
Mean
0.25
7 12 17 14 27 23
7 12 13 16 12 16
N
Total
6 7 9 12 5 8 5 9 12 12 16 14
0.157 0.172 0.176 0.086 0.265 0.236
N
0.075 0.084 0.103 0.043 0.075 0.054
0.050 0.077 0.092 0.071 0.045 0.058
Mean
0–0.25
0.093 0.261 0.255 0.242 0.337 0.375
Mean
Negative changes
2 3 5 2 11 9
1 5 4 4 7 8
N
0.361 0.435 0.350 0.343 0.542 0.519
0.346 0.518 0.623 0.754 0.545 0.693
Mean
0.25
Note Concentrated ownership includes the percentage of shares owned by the three largest shareholders, and shareholder categories appeared in concentrated ownership.
Concentrated ownership categories (f) Financial institutions 1992 35 0.324 1993 36 0.345 1994 44 0.316 1995 43 0.307 1996 43 0.297 1997 54 0.285 1998 40 0.285 (g) Non-financial firms 1992 41 0.355 1993 45 0.386 1994 59 0.323 1995 66 0.319 1996 81 0.328 1997 84 0.311 1998 93 0.322
N
Table 8.2 Continued
Ownership concentration of Turkish companies
217
on German Banks. Financial institutions, especially banks, have a special feature of the Turkish business groups. Almost every private bank is under the control of a business group and serves as the main financial source of firms in the group. They are the main creditors as well as shareholders for firms because Turkish industrial firms don’t have a chance to borrow from the bond market. We can say that the negative estimated coefficient of FININST is consistent with the argument by Weinstein and Yafeh (1998) that main banks discourage risk taking by their client firms in part because the banks are also major debtholders of these firms. The estimated coefficients of variables for HOLDING, FAMILY, and FOREIGN are all positive, but our sample cannot detect any significant effect of these variables on the performance of firms. The control variables, AvDEBT has negative and significant, AvASSET negative, and AvINVEST positive, but insignificant coefficients. We see from OLS estimates for the ownership equation (model 4) that the firm market performance, AvMB, is related positively and significantly with percentage of ownership by the three largest shareholders, AvCONT3. This result shows that there is a reverse causation in which the firm performance affects concentrated ownership. We argue that it is not possible certainly to say that higher concentrated ownership leads to a better firm market performance. Consequently, the single equation model is the bias used to examine the relationship between concentrated ownership and the firm performance, at least on the sample of Turkish industrial firms. OLS estimate for the other control variables, except market risk, BETA, are significant determinants of the level of concentrated ownership. Concentrated ownership increases with firm size (AvASSET) and firm specific risk (SE), and decreases with liquidity (AvNWC). All these results are very consistent with the view that shareholders give more attention to the contribution of monitoring activities to the firm performance. 2SLS estimation results for both performance and ownership equations are also reported at Panel A of Table 8.3. The results raise considerable doubt about the relationship between concentrated ownership and the firm market performance. Neither concentrated ownership (model 2) nor the firm market performance (model 4) has significant estimated coefficient. On the other hand, variables AvDEBT, STATE, and FININST (model 2) explain the variation in the firm market performance. In Panel B of Table 8.3, we present OLS and 2SLS estimation results in which the firm performance is measured by an accounting base measure, Return on Assets (AvROA). AvROA is statistically insignificant in the OLS equations (models 3 and 4 in the first part of Panel B). In the 2SLS equations, however, it is significantly negative for concentrated ownership (model 4), at 1 percent significance level. This result implies that percentage of shares owned by the three largest shareholders decreases when firms seem to be doing well based on the accounting measures. Demsetz and Villalonga (2001) find similar result for the relationship between Tobin’s Q and ownership concentration. Their interpretation of this result is that “perhaps selling shares during good times in the expectation that today’s good performance will be followed by poorer performance.”
HOLDING
STATE
SE
BETA
AvNWC
AvASSET
AvDEBT
AvINVEST
AvMB
1.41 (2.91)*** 0.26 (1.29)
2.09 (1.06) 2.04 (4.30)*** 0.00 (0.01)
Panel A: firm market performance Constant 2.75 (2.29)** AvMAN 0.03 (0.06) AvCONT3
1.25 (2.66)*** 0.22 (1.13)
2.73 (1.41) 2.28 (4.88)*** 0.06 (0.69)
1.55 (2.76)***
2.68 (2.30)**
0.00 (0.18) 0.05 (0.68) 0.10 (1.65)* 0.42 (2.38)**
0.00 (0.20)
0.10 (0.55)
0.05 (4.45)*** 0.14 (2.03)** 0.04 (0.72) 0.31 (1.91)*
0.03 (2.49)***
0.19 (1.16)
AvCONT3 (4)
1.70 (2.93)*** 0.43 (1.60)
1.50 (0.70) 2.31 (4.11)*** 0.02 (0.27)
2.16 (1.56) 2.27 (1.00)
AvMB (1)
AvMAN (3)
AvMB (1)
AvMB (2)
2SLS results
OLS results
Table 8.3 Empirical results on the relationship between ownership and firm performance
1.12 (2.13)** 0.19 (0.91)
3.24 (1.51) 2.47 (4.31)*** 0.10 (0.90)
2.78 (1.29)
2.63 (2.23)**
AvMB (2)
0.00 (0.32) 0.08 (0.79) 0.09 (1.38)* 0.39 (1.99)**
0.02 (0.37)
0.08 (0.44)
AvMAN (3)
0.05 (4.27)*** 0.12 (1.21) 0.05 (0.78) 0.33 (1.87)*
0.02 (0.41)
0.18 (1.06)
AvCONT3 (4)
AvNWC
AvASSET
AvDEBT
AvINVEST
AvROA
0.21 (1.50) 0.30 (9.40)*** 0.01 (1.00)
0.21 (1.47) 0.30 (9.11)*** 0.01 (0.99)
0.00 (0.07)
0.17 (2.09)**
0.00 (0.34) 0.13 (1.83)*
0.17 (1.20)
0.05 (0.29)
AvMAN (3)
0.06 (6.00)*** 0.15 (2.22)**
0.10 (0.71)
0.39 (2.38)**
AvCONT3 (4)
0.11 (3.34)***
0.28 (1.52) 0.24 (4.64)*** 0.00 (0.55)
0.24 (2.19)** 0.45 (1.99)**
AvROA (1)
AvROA (2)
Yes 0.16 (4.40)***
AvROA (1)
Yes 0.04 (1.69)*
0.32 (1.03) 0.01 (0.04) 0.31 (1.09)
2SLS results
0.16 (4.59)***
0.47 (2.17)** 0.19 (0.88) 0.13 (0.57)
Ols results
0.12 (4.43)***
0.51 (2.25)** 0.13 (0.65) 0.18 (0.74)
Panel B: firm accounting performance Constant 0.17 (2.13)** AvMAN 0.02 (0.39) AvCONT3
INDUSTRY Adj R2 F
FOREIGN
FAMILY
FININST
0.15 (1.01) 0.25 (5.18)*** 0.01 (1.53)
0.17 (1.19)
0.15 (1.68)*
AvROA (2)
0.13 (3.82)***
0.44 (1.97)** 0.23 (1.01) 0.10 (0.39)
0.06 (4.66)*** 0.14 (0.99)
1.27 (2.56)***
0.02 (0.08)
AvCONT3 (4)
Yes 0.14 (3.76)***
(Table 8.3 continued )
0.01 (0.66) 0.43 (2.90)***
1.34 (2.65)***
0.42 (1.69)*
AvMAN (3)
Yes 0.04 (1.67)*
0.34 (12.23)***
0.04 (0.97) 0.01 (0.74) 0.02 (1.40) 0.02 (1.50) 0.01 (0.82) 0.34 (12.21)***
0.03 (–0.91) 0.01 (0.66) 0.02 (1.35) 0.02 (1.45) 0.02 (0.88) Yes 0.04 (1.79)*
Yes 0.25 (7.43)***
0.03 (0.60) 0.37 (2.42)**
0.07 (1.31) 0.38 (2.43)**
0.24 (7.84)***
0.10 (1.73)* 0.05 (1.74)* 0.06 (2.16)** 0.05 (1.96)** 0.01 (0.40)
0.32 (11.27)***
0.02 (0.38) 0.00 (0.14) 0.02 (1.48) 0.02 (1.16) 0.02 (1.15)
AvROA (2)
Yes 0.05 (1.91)**
0.06 (0.87) 0.64 (3.04)***
AvMAN (3)
Yes 0.21 (5.99)***
0.05 (0.75) 0.10 (0.50)
AvCONT3 (4)
Notes The definition of variables is given at Table 8.1. The t-values are given in parentheses. *, **, *** denote significance at 10 percent, 5 percent, and 1 percent level.
INDUSTRY Adj R2 F
FOREIGN
FAMILY
FININST
HOLDING
STATE
SE
BETA
AvROA (1)
AvCONT3 (4)
AvMAN (3)
AvROA (1)
AvROA (2)
2SLS results
OLS results
Table 8.3 Continued
Ownership concentration of Turkish companies
221
Conclusions This chapter investigates the relationship between ownership concentration and the firm performance for Turkish industrial firms as a laboratory for capital markets having this kind of institutional environment. Our analysis depends on 185 industrial firms for the period from 1992 to 1998. We measure ownership concentration with the percentage of shares owned by the three largest shareholders. The studies by La Porta et al. (2002) for a sample of large firms from 27 wealthy countries, Claessens et al. (2002) for East Asian countries, Lins (2003) for 22 emerging countries, Yurtoxlu (2000), and Gürsoy and Aydoxan (2002) for Turkish companies examine the effect of controlling shareholders on the firm performance with single equation model. On the other hand, our assumption is that ownership concentration can be determined endogenously as well as ownership structure suggested by Demsetz and Villalonga (2001), Cho (1998), and others. The endogeneity refers possible divergence between large shareholders that control the firm and market or accounting-based expectations for the firm performance. We show that large shareholders in Turkish industrial firms change and restructure their ownership. Shareholding adjustments can be made by ordinary sales or activities to restructure ownership within or without business groups. Thus, ownership by large shareholders should be affected by future expectations of the firm performance. As a result, there should be no systematic relationship between concentrated ownership and firm performance. Our regression analysis depends on a simultaneous equation system, and the results are presented for both OLS and 2SLS estimation methods. Based on OLS estimations, we find that concentrated ownership has positive and significant effect on the firm market performance. This result is consistent with the hypothesis that ownership concentration improves the degree of monitoring and reduces the conflict of interest between controlling shareholders and minority shareholders. Concentrated ownership is also positively and significantly affected by the firm market performance. This result shows that there is a reverse causation in which the firm performance affects concentrated ownership. On the other hand, 2SLS estimation results for both performance and ownership equations show that neither concentrated ownership nor the firm market performance has significant estimated coefficient. In the 2SLS equations by using return on assets, the firm accounting performance, however, has significantly negative effect on concentrated ownership. This result implies that the percentage of shares owned by the three largest shareholders decreases when firms seem to be doing well based on the accounting measure. We control the effect of major ownership groups in estimation models by using dummy variables for the state, holdings, financial institutions, family members, and foreign investors. We find some evidence with both OLS and 2SLS estimations that the presence of Turkish State has positive and significant effect on the firm performance. The effect of financial institutions, on the other hand, is significantly negative. Our results, however, do not reveal any information that the performance of firms affiliated with a business group is different from unaffiliated
222
Halit Gönenç
firms. Larger equity holdings by families or foreign investors do not also lead to better performance.
Notes 1 Khanna (2000) reviews existing theoretical and empirical work on the role of emerging market business groups. 2 Demsetz and Villalonga (2001) use two dimensions of ownership structure in their analyses: percentage of shares owned by management and by five largest shareholders referred as concentration to monitor and discipline management behaviour. Results on OLS estimation for the firm performance show that Tobin’s q is linked positively with ownership by managers and negatively linked concentration ownership. These findings are inconsistent with expectations that greater ownership concentration by outside investors does lead to superior firm performance and greater shareholdings by insiders does lead to a more entrenched management. 3 We don’t require for a firm to have available data for each sample year. For this reason, the number of firms changes year by year in the sample period. Therefore, the average of annual values covers less than 6 years for some number of firms.
References Aoki, M. (1990) “Toward an Economic Model of the Japanese Firm,” Journal of Economic Literature, 28, 1–27. Cho, M.-H. (1998) “Ownership Structure, Investment, and the Corporate Value: An Empirical Analysis,” Journal of Financial Economics, 47, 103–21. Claessens, S., S. Djankov, and L. Lang (2000) “The Separation of Ownership and Control in East Asian Corporations,” Journal of Financial Economics, 58, 81–112. Claessens, S., S. Djankov, J. Fan, and L. Lang (2002) “Disentangling the Incentive and Entrenchment Effects of Large Shareholdings,” Journal of Finance, 57, 2379–2408. Demsetz, H. and K. Lehn (1985) “The Structure of Corporate Ownership: Causes and Consequences,” Journal of Political Economy, 93, 1155–1177. Demsetz, H. and B. Villalonga (2001) “Ownership Structure and Corporate Performance,” Journal of Corporate Finance, 7, 209–33. Ferris, S.P., K.A. Kim, and P. Kidsabunnarat (2003) “The Costs (and benefits?) of Diversified Business Groups: The Case of Korean Chaebols,” Journal of Banking and Finance, 27, 251–273. Gorton, G. and F.A. Schmid (2000) “Universal Banking and the Performance of German firms,” Journal of Financial Economics, 58, 29–80. Gürsoy, G. and K. Aydoxan (2002) “Equity Ownership Structure, Risk Taking and Performance,” Emerging Markets Finance and Trade, 38, 6–25. Himmelberg, C.P., R.G. Hubbard, and D. Palia (1999) “Understanding the Determinants of Managerial Ownership and the Link between Ownership and Performance,” Journal of Financial Economics, 53, 353–384. Hoshi, T., A. Kashyap and D. Scharfstein (1990) “The Role of Banks in Reducing the Costs of Financial Distress in Japan,” Journal of Financial Economics, 27, 67–88. Jensen, M. and W. Meckling (1976) “Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure,” Journal of Financial Economics, 3, 305–360. Kaplan, S. and B. Minton (1994) “Appointments of Outsiders to Japanese Boards – Determinants and Implications for Managers,” Journal of Financial Economics, 36, 225–258.
Ownership concentration of Turkish companies
223
Khanna, T. (2000) “Business Groups and Social Welfare in Emerging Markets: Existing Evidence and Unanswered Questions,” European Economic Review, 44, 748–761. Khanna, T. and K. Palepu (2000) “Is Group Affiliation Profitable in Emerging Markets? An Analysis of Diversified Indian Business Groups,” Journal of Finance, 55, 867–891. La Porta, R., F. Lopez-De-Silanes, A. Shleifer, and R. Vishny (1997) “Legal Determinants of External Finance,” Journal of Finance, 52, 1131–1150. La Porta, R., F. Lopez-De-Silanes, A. Shleifer, and R. Vishny (1998) “Law and Finance,” Journal of Political Economy, 106, 1113–1156. La Porta, R., F. Lopez-De-Silanes, and A. Shleifer (1999) “Corporate Ownership Around the World,” Journal of Finance, 54, 471–517. La Porta, R., F. Lopez-De-Silanes, A. Shleifer, and R. Vishny (2000a) “Agency Problems and Dividend Policies Around the World,” Journal of Finance, 55, 1–33. La Porta, R., F. Lopez-De-Silanes, A. Shleifer, and R. Vishny (2000b) “Investor Protection and Corporate Governance,” Journal of Financial Economics, 58, 3–27. La Porta, R., F. Lopez-De-Silanes, A. Shleifer, and R. Vishny (2002) “Investor Protection and Corporate Valuation,” Journal of Finance, 57, 1147–1170. Lins K. (2003) “Equity Ownership and Firm Value in Emerging Markets,” Journal of Financial and Quantitative Analysis, 38, 159–184. Loderer, C. and K. Martin (1997) “Executive Stock Ownership and Performance, Tracking Faint traces,” Journal of Financial Economics, 45, 223–255. Prowse, S. (1992) “The structure of corporate ownership in Japan,” Journal of Finance, 47, 1121–1140. Sheard, P. (1989) “The Main Bank System of Corporate Monitoring and Control in Japan,” Journal of Economic Behavior and Organization, 11, 399–422. Shleifer, A. and R. Vishny (1997) “A Survey of Corporate Governance,” Journal of Finance, 52, 737–783. Weinstein, D. and Y. Yafeh (1998) “On the Costs of a Bank-centered Financial System: Evidence from the Changing Main Bank Relations in Japan,” Journal of Finance, 53, 635–672. Yurtoxlu, B. (2000) “Ownership, Control and Performance of Turkish Listed Firms,” Empirica, 27, 193–222.
Part 3
The government sector and reform
9
Privatization in Turkey What has been achieved? Çaxla Ökten
The best of all monopoly profits is a quiet life. (Sir John Hicks)
Introduction In the last two decades, many countries have launched extensive privatization programs. There is now a growing body of literature on the effects of privatization on efficiency. In this chapter, we first review the theoretical and empirical literature on privatization. We then focus on privatization efforts in Turkey and analyze the Turkish experience within the context of the literature. Privatization efforts in Turkey, fueled by the forces of globalization, started in 1985. These efforts should be considered as part of a larger plan which conceived that regulations, policies, and incentives should be readjusted to liberalize the Turkish economic environment for private and foreign direct investment (Öniv, 1991; Karatav, 2001). The primary objectives of the privatization program were to (1) minimize state involvement in economic activities; (2) promote competition, improve efficiency and increase the productivity of public enterprises; (3) relieve the state from the burdens of inefficient state industries; (4) facilitate a wider distribution of share-ownership and develop a viable capital market; and (5) create revenue for the government. Although the privatization process in Turkey had started earlier than in most developing countries, its progress – measured in terms of the size of divestiture – had been slower when compared with the principal Latin American and Eastern European cases (Ercan and Öniv, 2001). From its start in 1985 up to 2005, the total proceeds from privatization efforts have amounted to $9.4 billion. More than half of this has been realized in the 2000–2005 period after the 1999 IMF Stand By agreement that placed a particular emphasis on privatization. In the next section, we first review economic theory of privatization and identify arguments for and against public ownership. The economic theory of privatization is a subset of the vast body of literature on the economics of ownership and the role for government ownership of productive resources. There are two main branches in this literature: The Social View (Shapiro and Willig, 1990) and the Agency View (Vickers and Yarrow, 1988; Shleifer and Vishny, 1994).
228
Çaxla Ökten
We then turn our attention to empirical studies that examine the effects of privatization and identify common findings as well as conflicting results in the literature. In the section on Privatization process in Turkey, we describe the privatization environment and process in Turkey. As a case study, we focus on the privatization process of Türk Telekom to show the specific challenges that privatization efforts had to face. In the section on Objectives of Turkish privatization: what has been achieved?, we assess the achievements of the Turkish privatization experiment in terms of reducing state’s involvement in the economy, increasing efficiency, and generating revenue. The final section concludes.
Literature review Theoretical literature on privatization The economic theory of privatization is a subset of the vast body of literature examining the economics of ownership and the role of government ownership of productive resources. There are two main branches in this literature: the Social View and the Agency View. According to the Social View (Shapiro and Willig, 1990), state-owned enterprises (SOEs) are capable of curing market failures by implementing pricing policies that take into account of social marginal costs and benefits of production. A privately owned firm is expected to maximize profits whereas a state owned firm is expected to maximize social welfare. For example, in a natural monopoly market structure, efficiency calls for a single firm to exist. A profit maximizing monopoly will, however, charge too high a price and produce too low a quantity. This potential inefficiency can be solved by state ownership. The Agency View of firm ownership presents a strong critique of this theory. There are two complementary strands of the literature which differ as to whether the agency conflict is with the manager or the politician. Vickers and Yarrow (1988) argue that managers of SOEs may lack high-powered incentives or proper monitoring. Shleifer and Vishny (1994) stress that political interference in the firm results in excessive employment, poor choices of product and location, lack of investments and ill-defined incentives for managers. The Social View unequivocally predicts that efficient technology will be chosen by the state-owned firms. Models of the Agency View, on the other hand; while predicting that inefficient technologies will be chosen by politicians/managers, have different predictions for the direction of the distortion in the production process. They either predict that state-owned firms will have low investment levels (Shleifer and Vishny, 1994) or that they will use excess capital as well as excess labor (Vickers and Yarrow, 1988). The over-capitalization argument stems from bureaucratic inefficiency models. The founder of this line of literature, Niskanen (1975), proposed that bureaucrats are inclined to maximize their total budget rather than the utility of their sponsors. Vickers and Yarrow (1988) argue that the bureaucrats will subject the state-owned firm to over-investment and over-capitalization to justify high salaries and perks.
Privatization: what has been achieved? 229 On allocative efficiency, the Social View predicts that prices are likely to rise as a result of privatization. The Agency View, on the other hand; predicts that if a reasonable degree of competition ensues then allocative efficiency may actually increase as firms increase their productivity after privatization. Empirical literature on privatization Privatization and productive efficiency Firm performance has been the focus of the empirical literature on privatization. Studies cited in a survey of empirical studies of privatization almost unanimously report increases in firm performance associated with privatization (Megginson and Netter, 2001).1 Most of these studies compare post-privatization performance changes with either a comparison group of non-privatized firms or compare three year mean/median performance of privatized firms to their own mean/median performance during their last three years as state owned firms. Critics of these findings are quick to point out that most of the gains researchers have documented after privatization are due to selection bias. The argument is that better firms are privatized first and their comparison to more poorly performing firms which happen to remain public gives a spurious relationship between privatization and firm performance. Cross-sectional studies may not satisfactorily control for firm-specific effects and therefore address the selection problem for privatization. While comparing before and after three year averages of performance measures might be more promising, even that method may not entirely solve the selection problem. One could argue that, those firms would have improved at any rate even if they were not privatized (Omran, 2004). When Omran compares before and after privatization averages of performance measures of privatized firms from Egypt, he finds a significant increase in performance. When he carries out the same exercise for firms that remain state-owned, he finds that they also improve after the privatization period though they themselves are not privatized. Either the improvement of privatized firms had spillover effects on state-owned firms or privatization has nothing to do with the changes observed. The author suggests that other economic reforms that enhanced the competitive environment in which his sample of privatized and state-owned firms operate might have been responsible for his findings. Recent studies improve on methodology by using panel data methods. Using firm fixed effects, they control for unobserved firm heterogeneity a potential source for selection bias (Ehrlich et al., 1994; Frydman et al., 1999; Villalonga, 2000; Wallsten, 2001; Earle and Telegdy, 2002; Ökten and Arin, 2005). The results of these studies on privatization and firm performance are mixed. Ehrlich et al. use a sample of 23 comparable international airlines of different ownership categories over the period 1973–1983. Their results suggest that private ownership leads to higher rates of productivity growth and declining costs in the long-run, and that these differences are not affected by the regulatory environment. Their estimates suggest that the short-run effects of changes from state to private ownership on productivity and costs are ambiguous.
230
Çaxla Ökten
Villalonga examines 24 Spanish firms from different industries and finds that privatization does not increase firm efficiency – defined as rate of return on assets. He argues that political factors such as the business cycle during which the firm is privatized and foreign ownership are important determinants of firm efficiency. Wallsten (2001) finds that in the telecommunications sector, privatization by itself does not appear to generate many benefits and is negatively correlated with main line penetration. He points out the importance of regulatory framework ensuing privatization as he finds that privatization combined with the existence of a separate regulator, is correlated with increased connection capacity and labor efficiency as measured by employees per main line. Earle and Telegdy find that privatization increases labor productivity growth in their heterogeneous sample of Romanian firms. Frydman et al. find that privatization to outsider owners has significant effects on revenue performance, but not on cost reduction using data from the Czech Republic, Hungary, and Poland, on 218 state-owned firms of which 128 were privatized during the 1990–1994 period. We should note that testing the effects of privatization on firm performance is even more difficult in transition economies than in non-transition economies as privatization in these countries occurs at the same time as and is part of, other massive economy-wide changes (Johnson et al., 1994). Ökten and Arin (2005) find that privatization increases labor productivity of Turkish cement plants in a study which controls for firm fixed effects and time effects with yearly dummies. We will discuss this study in detail in the section on Empirical studies of the cement industry. Privatization and allocative efficiency Studies that examine the effect of privatization on allocative efficiency are rare (Megginson and Netter, 2001). These studies typically find that prices either increase or do not change after privatization. La Porta and Lopez-De-Silanes (1999) analyze Mexican firms from a variety of industries and find that consumer prices increase after privatization. In their analysis of the water and sewerage industry of England and Wales, Saal and Parker (2001) find that, output prices increase and furthermore, total price performance indices reveal that increases in output prices have outstripped increases in input costs. On the other hand, in a cross-country panel study of the telecommunications sector, Wallsten (2001) finds that prices are not correlated with privatization but are negatively correlated with competition measured by the number of mobile operators not owned by the incumbent. Ökten and Arin (2005) find that prices in the Turkish cement sector decrease following privatization. It is unrealistic to expect that the effects of privatization on prices will be the same in every industry. Market structure of an industry – market power of firms in the industry – as well as firms’ productivity will affect consumer prices. Studies should strive to differentiate the effects of private ownership from the changes in market structure and competitive environment induced by privatization. Privatization and input choice Empirical studies of privatization do not directly examine the changes in input choice as a result of privatization. Rather, they report changes in employment and capital investment, which may suggest a change in technology. In their survey article, Megginson and Netter (2001) report
Privatization: what has been achieved? 231 that almost all of the 22 studies from non-transition economies that they review find that capital investment spending increases significantly as firms are privatized. Perhaps surprisingly, they report that these studies are far less unanimous regarding the impact of privatization on employment levels in privatized firms. La Porta and Lopez-De-Silanes (1999), in their study of 233 privatized Mexican firms, find that ratios of investment to sales and investment to fixed assets significantly increase after privatization while employment significantly decreases. Bhaskar and Khan (1995) find that privatization has a large and significant negative effect on white-collar workers using employment data from Bangladesh, for 62 jute mills of which 31 were privatized. Privatization and market structure There are only a few studies that have sought to estimate the effects of market structure along with privatization. These studies typically include some type of measure for market concentration as an additional control when they analyze the effects of privatization on firm productivity. In general, they do not analyze how privatization affects market structure or how changes in market structure affect allocative efficiency. Angelucci et al. (2001) analyze the effects of competitive pressures – measured by Herfindahl index and share of imports in sales – and ownership changes on productivity in Bulgaria, Poland, and Romania. Anderson et al. (2000) analyze the effects of competition and ownership in the productivity of the newly privatized enterprises using data from Mongolia. Kattuman and Domanski (1997) analyze market concentration ensuing mass privatization in Poland and find that concentration increased promptly in several markets. Warzynski (2003) in his study of 300 Ukrainian firms finds that competition does not have a significant effect on firm performance measured by productivity and profitability while privatization has a marginal positive significant effect on profitability and an insignificant effect on productivity. However, he points out that competition and privatization might be complementary measures as he finds that competition increases the performance of privatized firms. Revenue generation Generating government revenue is a common objective in privatization. Governments use a variety of sales methods to achieve this objective including share issue privatizations (SIPs) and direct sales via auctions. According to the World Bank, over 12,000 companies were privatized during the period 1980–1993 (almost half were divested between 1991 and 1993), and less than 5 percent of these divestitures involved public share offerings. However, SIPs have accounted for the vast majority of the assets and employees moved from the state to the private sector through privatization. Perhaps, that is why a large number of studies focus on pricing in SIPs. Jones et al. examine 630 SIPs from 59 countries that raise over $446 billion during the period 1977–1997 and find that government consistently underprice SIP offers, tilt their share allocation patterns to favor domestic investors, impose control restrictions on privatized firms, and typically use fixed price offers rather than book building or competitive tender offers. The most likely explanation for underpricing is that it ensures a high demand for shares offered.
232
Çaxla Ökten
Few studies analyze the determinants of privatization prices that resulted from direct sales of assets and companies via auctions (Lopez-De-Silanes, 1997; Arin and Ökten, 2003). Lopez-De-Silanes (1997) examines 236 Mexican firms, which were privatized between the years 1983 and 1992. Arin and Ökten (2003) use a data set of 68 privatized firms from Turkey. Both studies find that the number of bidders increases privatization prices by increasing the level of competition in these auctions. However, there are notable differences on how firm characteristics affect privatization prices. While Lopez-De-Silanes finds that low profitability of state-owned prices explain the low price paid, Arin and Ökten find that revenues and market characteristics affect privatization prices, not profits. Arin and Ökten argue that what fundamentally determines the privatization price is the expected future profit of the firm. We will discuss Arin and Ökten study in more detail in the section on Revenue generation and share ownership.
Privatization process in Turkey Historically, Turkey has a long experience of relying heavily on SOEs. SOEs were established during the 1930s by the government to jump-start the economy which had collapsed with the end of the Ottoman era in 1923. Over the years SOEs grew enormously, leaving the control of a large section of the economy to bureaucrats and politicians. By 1960, share of public sector in total value added in manufacturing was 60 percent. Politicians exploited SOEs to provide jobs to their constituents at the expense of consumers, who were faced with higher prices. Consequently, in the 1980s, SOEs began to be perceived negatively due to poor financial performance, overstaffing, dependence on subsidies, protected markets, and corruption (Ertuna, 1998). Furthermore, public enterprises were able to operate with a “soft budget” constraint which constituted an additional source of relaxation in performance, resulting in greater inefficiency (Öniv, 1991).2 After a Military Regime (1980–1983), the first party that came to power was the Motherland Party (ANAP) under the leadership of Prime Minister Turgut Özal. Özal was a strong supporter of Thatcherism which promoted a reduction of the state’s role in the economy. Privatization first entered the political agenda with Özal’s trade and capital account liberalization program in 1984. Despite this initial enthusiasm, the privatization process has been slow. Since its start in 1985, total proceeds from privatization efforts amounted to $9.4 billion by 2004. More than half of this was realized in the 2000–2004 period after the 1999 IMF Stand By agreement that placed a particular emphasis on privatization. Figure 9.1 shows annual total proceeds from privatization during 1985–2004. The observed spike in year 2000 is largely due to the block sale of the 51 percent of POAS (Petroleum Products Distribution) to Dogan Holding-WvBank consortium group for $1.260 billion3 and public share offerings of the 31 percent of Tupras (Petroleum Refinery) which amounted to $1.194 billion. Since 1985, state shares in 244 companies, 22 incomplete plants, 6 real estates, 4 power generation plants, 6 toll motorways, 2 Bosphorus bridges, 29 plants, and 1 service unit have been taken into the privatization portfolio (OIB, 2005). Later, state shares in 23 companies, 4 power generation plants, and 4 real estates have
Privatization: what has been achieved? 233 3,000
2716.5
2,500
Million $
2,000 1,500
1265.7 1019.7
1,000 536.5
2005
2004
2003
3.7 2002
2001
2000
1999
171.6
119.8
38.3 1998
1997
1996
1995
1994
1993
1992
1991
1990
1989
0
565.5 514.6 486 465.5 422.9 411.8 292 243.8 131.2 28.6 1985–1988
500
Figure 9.1 Revenues from privatization (million dollars).† Note † As of February 1, 2005 while the manuscript was in press revenues in 2005 amounted to 1,315.1 million dollars.
been taken out of the portfolio for various reasons. Over half of those that remained in the privatization portfolio have been fully privatized. Most of the full privatizations have been sales of assets or sales of minority state shares. Currently, there are 34 companies, 29 plants, 1 real estate, 6 toll motorways, 2 Bosphorous bridges in the portfolio. State shares amount to more than 50 percent in 24 of the 34 companies. Block sales have been the most prevalent method of privatization. Forty-two percent of the total proceeds have been realized through block sales, 18 percent through public offerings, 16 percent through asset sales, and 12 percent through international offerings, 8 percent through sales in Istanbul Stock Exchange (ISE). Table 9.1 shows the list of the highest grossing (in total proceeds) 47 block sales. We notice that 19 of these are cement plants highlighting the importance of cement industry in the privatization process. Table 9.2 shows domestic public share offering privatizations, Table 9.3 shows international public share offering privatizations and Table 9.4 shows the list of companies that were privatized through a combination of block sales and public offerings. The privatization reforms have not been fully carried out as intended, due to the lack of a legal framework, conflicting laws, and a wavering political will. Privatization efforts faced strong opposition by entrenched vested interests, notably senior bureaucrats in government departments and SOEs, Workers’ Unions who have expressed serious concern about the possibility of mass lay offs, and leftist political parties (Karatav, 2001). Following many amendments in the laws governing privatization of SOEs and a mind boggling number of decree-laws that were frequently invalidated by the
100.00 96.64 100.00 100.00 99.74 100.00 100.00 100.00 100.00 51.00 100.00 99.99 100.00 100.00 100.00 99.73 100.00 45.00 100.00 100.00 100.00 100.00 60.00 99.97
100.00 96.64 100.00 100.00 99.74 100.00 100.00 100.00 100.00 49.33 100.00 99.99 100.00 100.00 100.00 99.73 100.00 45.00 100.00 100.00 100.00 100.00 100.00 99.97
Etibank (Bank) Asil Çelik (Steel) Lalapava Cement Bursagaz (Gas distribution) Kümav (Mining) Sümerbank (Bank) Denizli Cement Anadolubank (Bank) Denizbank (Bank) Wpragaz (Gas distribution) Wskenderun Cement Deniz Nakliyati T.A.V. (Transportation) Eti krom A.V. (Mining-chrome) Ladik Cement Vanliurfa Cement Gaziantep Cement Adiyaman Cement Çayeli Bakir Wvl. A.V. (Mining-copper) Ergani Cement Esgaz (Gas distribution) Eti Gümüv A.V. (Mining-silver) Konya Krom Man. A.V. (Mining-chrome) Havav (Airport services) Petlas (Tyre production)
Share sold (%)
State share (%)
Company
Table 9.1 Privatizations through block sales
Yazeks A.V. Kombassan Holding A.V.
Rumeli Çimento San. A.V. Kolin Wnv. Tur. San. Ve Tic. A.V. Kss Mad. Wnv. Tur. San. Ve Tic. A.V. Konya Selçuklu San. Tic. A.V.
Yildirim Div Ticaret Paz. A.V. Rumeli Çimento Rumeli Çimento Rumeli Holding Teksko Giyim San. A.V. Inmet Madencilik A.V.
Medya-Wpek holding A.V. Yazici-Güriv-Parsan Rumeli Çimento A.V. Çalik Enerji San. Tic. A.V. Zeytinoxlu Holding A.V. Wpeks Tekstil San. A.V. Modern Çimento Mehmet Rüvtü Bavaran Zorlu Holding Primagaz A.G. Oyak-H. Ömer Sabanci Armatörler Den. Ve Nak. A.V.
Buyer
17.04.95 06.05.97
03.04.97 10.03.04 13.08.04 05.06.98
14.09.04 21.04.93 21.04.93 03.12.92 16.08.95 23.09.04
02.03.98 29.08.00 14.06.96 19.04.04 28.09.95 17.10.95 04.12.92 07.05.97 29.05.97 27.01.92 02.12.92 24.03.00
Date
36,000,000 35,750,000
46,700,000 43,000,000 41,200,000 40,700,000
58,050,000 57,598,687 57,405,988 52,695,898 52,500,000 49,250,000
155,500,000 131,000,000 125,890,000 120,000,000 108,100,000 103,460,000 70,100,000 69,500,000 66,000,000 64,066,776 61,500,000 59,000,000
Sale price ($)
100.00 99.30 100.00 100.00 100.00 100.00 100.00 100.00 99.89 40.00 99.90 100.00 98.30 100.00 100.00 56.00 99.78 30.00 100.00 100.00 98.41 88.33 100.00
100.00 99.30 100.00 100.00 100.00 100.00 100.00 100.00 99.89 40.00 99.90 100.00 98.30 100.00 100.00 56.00
99.78 30.00 100.00 100.00 98.41 88.33 100.00
Rumeli Çimento Gan International Zonguldak Yatırım Makinaları A.V. Aksu Mad. San. Tic. A.V. K.M.M. Kayseri Maden Metal A.V. Atlantic Ind. Ltd Ercan Madencilik A.V
Yibitav Holding SCF Rumeli Holding Erçimsan Evyap San. Tic. A.V. Yibitav Holding Erdemir Canlar Oto Wnv. San. Ve Tic. A.V. Oyak/Gama A.V. Turgay Ciner SCF Rumeli Çimento A.V. SCF Çimentav Grubu Ce-Ka Wnv. Mak. Mad. San. Tic. A.V. Torunlar Gida San. Tic. A.V. 06.05.93 01.07.91 13.05.97 25.10.04 22.05.96 18.11.88 01.10.97
25.12.92 08.09.89 03.12.92 17.06.93 09.07.98 25.12.92 15.04.04 09.01.98 12.06.96 30.03.98 08.09.89 12.06.96 08.09.89 18.06.96 12.04.04 08.10.04
Notes This list does not include all the privatizations through block sales. Companies are ranked according to the total proceeds from their privatization.
Source: Compiled and rearranged from tables of Privatization Administration.
Çorum Cement Ankara Cement Trabzon Cement Avkale Cement Yarimca Porselen (Porcelain) Sivas Cement Div-Han A.V. Kurtalan Cement Elazix Cement Havav (Airport services) Pinarhisar Cement Van Cement Balikesir Cement Kars Cement Eti Bakir A.V. (Mining-copper) Bet Kütahya Veker (Sugarproduction) Bartin Cement Günev Sigorta (Insurance) Filyos Atev Tuxlasi Eti Elektrometalurji Çinkur (Mining) Ansan – Meda Bozüyük Seramik A.V. (Ceramics) 20,568,669 18,900,000 18,150,000 15,320,000 14,000,000 13,000,000 12,000,000
35,000,000 33,000,000 32,551,000 31,158,000 30,500,000 29,400,000 28,500,000 28,100,000 27,850,000 27,100,000 25,000,000 24,500,000 23,000,000 22,250,000 21,879,000 21,438,000
236
Çaxla Ökten
Table 9.2 Public share offering privatizations
Erdemir (steel) Arçelik (appliances) Bolu Cement Çelik Halat (steel) Petkim (petrochemicals) Konya Cement Ünye Cement Mardin Cement Thy (turkish airlines) Adana Cement Adana Cement Migros (grocery chain store) Ditav (spare parts) Tüprav (petroleum refinery) Tofav Türk (auto) Tofav Türk T. Wv Bankasi (bank) Tüprav (petroleum refinery) Thy (turkish airlines) Total
State share at date of offering (%)
State share offered (%)
Date of offering offered
Total proceeds ($)
Number of applicants
48.65
2.93
9–10.04.1990
53,105,711
33,953
13.32
5.83
30–4/1.5.1990
19,890,196
12,618
34.50 19.42
10.38 13.25
30–4/1.5.1990 30–4/1.5.1990
8,268,150 7,750,179
8,157 6,517
99.97
8.08
18–29.6.1990
150,617,183
76,119
39.87 49.21 46.23 100.00
31.13 2.86 25.46 1.55
24–25.10.1990 1–2.11.1990 22–23.11.1990 29–11.7.12.1990
17,663,979 927,162 9,161,501 4,976,165
6,396 281 1,280 2,488
23.86 23.42 42.22
17.16 17.16 36.40
18–20.2.1991 18–20.2.1991 25–26.2.1991
25,162,623 2,795,847 5,609,246
3,355 3,355 3,951
14.77
2.51
6–7.5.1991
219,411
1,263
100.00
1.66
27–29.5.1991
6,036,589
15,456
23.13 4.46 12.30
0.85 0.13 4.90
13–14.6.1991 4–7.3.1994 4–6.5.1998
6,119,572 2,824,239 240,702,529
3,147 801 80,978
96.42
23.88
5–7.4.2000
839,028,679
369,566
98.17
8.05
1–3.12.2004
65,326,696
29,280
1,466,185,657
658,961
Constitutional Court, the first stand alone Privatization Law was ratified by Parliament in November 1994. After a round of revisions, the Privatization Law took its final form in April 1997. A legislation making international arbitration in disputes over contracts involving provision of public services was passed in August 1999. This legislation opened the door for active foreign participation particularly in infrastructure and utility privatizations (Aybar et al., 2001). Stand By Agreement with the IMF in December 1999 and the ensuing stabilization program to establish macroeconomic stability placed a particular emphasis on privatization which in turn increased the speed of privatization. Currently, there appears to be both external pressure and internal political will by the one
Privatization: what has been achieved? 237 Table 9.3 International public offering privatizations
Tofav Türk (Auto) T. Wv Bankasi (Bank) Tüprav (Petroleum refinery) Poav (Petroleum distribution) Thy (Airline) Total
State share at date of offering (%)
State share offered (%)
Date of offering
21.13 12.30 96.42
16.67 7.39 6.77
03.03.1994 4–6.05.1998 5–7.04.2000
42.30 98.17
3.50 14.95
6–15.03.2002 1–3.12.2004
Total proceeds ($)
330,000,000 391,949,083 265,491,985 38,891,914 125,952,471 1,152,285,453
Source: Compiled and re-arranged from tables of Privatization Administration in Turkey.
party majority government to complete the remaining more controversial privatization projects. We will now review the privatization process of one of these, the privatization of Turk Telekom since this case is illustrative of the challenges that the privatization efforts had to face. Privatization of Turk Telekom The telecommunications industry was once considered a textbook example for a natural monopoly since the fixed landline structure of this industry resulted in large-scale economies and hence efficiency required one firm to exist. Recent technological advances in this sector, such as the spread of wireless technology and communications via internet have produced alternatives to the fixed landline structure and made a competitive environment more viable. Even if there is one dominant firm in this market, whether that firm should be state-owned or privately owned but supervised by a regulatory agency is open to debate. The trend in the world has been for the privatization of the telecommunications industry. During 1990–1999, 49 countries have privatized their telecommunications firms either partially or fully (Yılmaz, 2000). An analysis of the privatization process of Turk Telekom is useful in understanding the privatization environment and process in Turkey. The privatization of Turk Telekom was initiated in 1994, only months after the September 1993 communique establishing the Turk Telekom as a separate entity from PTT and has not been completed as of today. In May 1995, law 4107 authorized the sale of 49 percent of the company and opened the door for telecommunications license agreements. However, since then privatization of telecommunication services has been subject to political and legal squabbling. In February 1996, the Constitutional court overturned critical parts of the law. After several round of cancellations the law 4161 was enacted. In the context of this law, Turk Telekom privatization was linked to sector reform, and company valuation, which would be followed by the sale of the company (Aybar et al., 2001). In 1998, the Council of Ministers adopted a sales strategy which consisted of a block sale of 20 percent to a strategic partner followed by a 19 percent of initial
Afyon Cement Block sale Public offering Tofav Oto Tic. (Auto) Block sale Public offering Nixde Cement Public offering Block sale Çukurova Elekt. (Electricity) Public offering Block sale Kepez Elektrik (Electricity) Public offering Block sale 51.00 39.87 16.00 1.36 12.72 87.10 5.45 11.25 8.14 25.39
39.00 23.00 99.84 87.12 18.65 11.50 42.05 25.39
State share offered (%)
99.60 48.60
State share at date of offering (%)
Table 9.4 Privatizations through block sales and public offerings
Public offering Rumeli Elekt.
Public offering Rumeli Elekt.
Public offering Oyak-Sabanci
Fiat Auto S Public offering
SCF Public offering
Buyer
16–17.04.1990 16.02.1993
16–17.04.1990 16.02.1993
13–14.05.1991 23.03.1992
22.02.1991 13–14.06.1991
08.09.1989 21–26.03.1991
Date
9,390,359 33,158,988
38,829,409 81,096,791
2,647,286 22,500,000
13,203,441 966,248
13,000,000 8,422,698
Total proceeds ($)
8,320
22,184
1,125
3,147
12,591
Number of applicants
20.00 7.75 4.15 94.05 22.00 18.00 70.00 30.00 4.02 51.00 13.00
49.00 29.00 54.68 50.38 40.00 18.00 100.00 30.00 100.00 93.30 42.30
Public offering Wv Doxan A.V. Public offering
SAS Public offering
Public offering Alcatel B.V.
Public offering Bilfer-Dedeman
NTL Public offering
Source: Compiled and re-arranged from tables of Privatization Administration in Turkey.
Netav (Telecommunıcatıons) Block sale Public offering Gima (Grocery stores) Public offering Block sale Teletav (Telecommunıcatıons) Halka Arz Block sale Usav (Aırport servıces) Block sale Public offering Petrol Ofisi A.V. (Petroleum dıstrıbutıon) Public offering Block sale Public offering Total Public offering Block sale 27–29.5.1991 21.07.2000 27.02–15.03.2002
09.02.1989 20–22.10.1993
29.02–2.03.1988 19.08.1993
3–4.06.1991 02.03.1993
01.03.1993 3–5.11–12.3.1993
14,386,888 1,260,000,000 129,479,256 1,747,747,798 241,548,765 1,506,199,033
14,450,000 15,205,871
13,090,225 21,002,400
406,902 21,787,413
26,000,000 8,723,623
51,470 167,590
17,206
4,672
41,695
283
4,897
240
Çaxla Ökten
public offering of Turk Telekom’s shares. The block sale to a strategic investor requires participation of an international telecommunications operator that would bring expertise and accelerate the commercialization of Turk Telekom (Aybar et al., 2001). The steps involved in the privatization of this enterprise would include a financial review, valuation, a decision regarding an appropriate method of investment, invitation of bids and screening of prospective investors, negotiations with short-listed firms and final recommendation to cabinet for approval (Karatav, 2001). Unfortunately, sales of minority shares of Turk Telekom did not attract foreign investors as it was hoped due to obscure management rights (Financial Times, 18.09.2000). Reconsidering the percentage of ownership stake to be offered to strategic partners and the controversial management rights the government authorized the block sale of 33.5 percent of the company in December 2000. The tender committee was to expect offers till May 14, 2001. However the financial crisis triggered by a confrontation between the prime minister and the president led to a collapse of the December 1999 Stand By Agreement with the IMF and the tender was cancelled. Since the uncertainty about management rights would be an obstacle to a successful privatization in by then a depressed global telecom market, it was proposed that a minimum of 51 percent of Turk Telekom’s shares be privatized. The privatization process gained momentum following the November 2002 elections which drastically altered the political landscape of Turkey. None of the ruling coalition parties received enough votes to be represented in Parliament, while a brand new party was able to form a majority government by itself. The Council of Minister Decree – dated November 13, 2003 stipulated that a minimum 51 percent of Turk Telekom shares were to be offered block, while remaining shares would be offered to public following the block sale. With the enactment of law 5189, the foreign ownership restriction on the part of foreign investors was lifted, the scope of the golden share has been limited and the satellite business has been taken out of Turk Telekom to function as a separate public entity (OIB, 2005). The formal tender process for the block sale of 55 percent of Turk Telekom commenced with the tender announcements on November 25, 2004. Accordingly, to be able to submit bids, bidders would be required to satisfy the pre-qualification criteria determined by the Tender Committee. Applications for pre-qualification were delivered to the Privatization Administration on January 11, 2005 where 13 national and international bidders qualified. The bidding deadline for the tender is May 31, 2005 (OIB, 2005).
Objectives of Turkish privatization: what has been achieved? The primary objectives of the privatization program were to (1) minimize state involvement in economic activities; (2) promote competition, improve efficiency and increase the productivity of public enterprises; (3) relieve the state from the burdens of inefficient state industries; (4) facilitate a wider distribution of
Privatization: what has been achieved? 241 share-ownership and develop a viable capital market and; (5) create revenue for the government. In this section, we review evidence on to the achievements of Turkish privatization with respect to these objectives. State’s involvement in economic activities According to the Privatization Administration Statistics the following privatization implementations have been realized since the start of privatization process in 1985: ●
●
● ●
●
●
●
State completely withdrew from cement, animal feed production, milk-dairy products, forest products, civil handling and catering services, and petroleum distribution sectors. More than 50 percent of the state shares were privatized in tourism, iron and steel, textile, sea freight, and meat processing sectors. State has partially withdrawn from the ports and petroleum refinery sector. Privatization of public banks has commenced with Sümerbank and continued with Etibank, Denizbank, and Anadolu Bank. The international and domestic offering of the 12.3 percent state shares in WvBank in May 1998, had been the largest public offering in Turkey until that time and recorded as one of the largest privatization proceeds among the emerging European markets. Public shares in Netas (telecommunication firm) and Tofas (auto manufacturer) were issued to foreign investors through international public offering for the first time, which served as a driving force of the integration of ISE with foreign capital markets. Public shares in many companies were issued to the public, particularly in the beginning of this decade and this enhanced the institutionalization of ISE. 14.95 percent of the Turkish Airlines has been privatized in an International and Domestic Offering in December 2004, sales amounting to $126 million.
An appropriate measure to assess the degree of state involvement in economic activities is the share of public sector in total value-added manufacturing. Figure 9.2 shows the share of public sector and of private sector in total valueadded annually for the 1985–2000 period, every 5 years for the 1970–1985 period and every 10 years for the 1950–1970 period. Data is compiled from the State Statistics Institute’s Statistical Indicators (Istatistiki Gostergeler 1923–2002, DIE). The share of public sector is practically unchanged between 1950–1960 and is almost 60 percent of total value added. In the next 25 years, it falls by 34 percent to about 40 percent of total value added. The reduction in the share of public sector is significantly faster in the post-privatization period. The share of public sector is down from 40 percent in 1986 to 18.5 percent in 2000, which is a 54 percent reduction. Since the rate of reduction in the share of public sector in total value added is significantly faster in the post privatization period than in the decades prior to 1985, we can conclude that the privatization program appears to
242
Çaxla Ökten
90 80 70 60
%
50 40 30 20 10
Public share
2000
1999
1998
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
1987
1986
1985
1980
1975
1970
1960
1950
0
Private share
Figure 9.2 Share of public and private sector in total manufacturing value-added.
have been – at least partially – successful in achieving its objective of reducing the state’s involvement in economic activities. However the privatization process has been slower compared with the principal Latin American and Eastern European cases and is yet to be completed. Privatization and efficiency In the case of the cement industry, there are relatively large numbers of stateowned plants that were sold during the 1989–1998 period, and hence sufficient time has elapsed for proper empirical analysis. Since almost all empirical studies on the Turkish privatization experience have focused on this industry, we will examine the privatization process in this industry and then discuss the findings of empirical studies. Privatization in the cement industry The first cement plant of Turkey was established in 1911 by a private firm. By 1950, four more private plants had been built. Only after 1950 did the cement industry develop on a large scale by means of a government initiative. A public enterprise, CISAN (Turkish acronym for Turkish Cement Industry Co. later named CITOSAN), was established in 1953 to build 15 plants in various regions. Before the privatization of the cement plants began in 1989, the public share in the cement industry was nearly 40 percent (Saygili and Taymaz, 2001). It is believed that each company was able to exercise some monopoly power within its hinterland (Ertuna, 1998), most probably due to the distance between firms and the lack of proper transportation facilities in the public sector.
Privatization: what has been achieved? 243 In 1986 a French company, Sema-Metra Conceil was contracted by the Turkish government and the World Bank to prepare two reports, one on the structural regulation of the cement sector and privatization and the other on the plan for the reorganization of CITOSAN. In the latter report, Sema-Metra Conceil suggested that plants in the west be privatized first since they could be as profitable as private plants, and recommended that the eastern plants be restructured prior to privatization. The report also suggested privatization on a plant-by-plant basis, as the sale of the state firm as a single entity may have led to an unhealthy monopoly (Tallant, 1993). In 1986 there was a major change in the economic environment of the cement plants. Prior to 1986, the Turkish Cement Producers’ Association (TCPA) set prices and market areas for all cement companies, however after 1986 firms were encouraged to operate independently and maximize profits. Sema-Metra’s first report might have partially led to this change. Privatization in the cement industry started in 1989, with the initial sale of five factories to the French firm Cement Francais (SCF). By 1998, the sale of 22 cement plants had been completed.4 The recommendations of the Sema-Metra report were taken into consideration, and the western plants were privatized first.5 It may also be the case that the privatization of the eastern plants was delayed, as the eastern region suffered from unemployment and terrorism throughout the 1990s, and the public enterprises were used as means of employment. Privatization of the cement plants was carried out under the Privatization Administration of Turkey. Most of the privatizations were realized through block sales using closed-bid auctions and through a combination of block sales and public offerings in a few cases. Public sector employment was guaranteed to all workers that lost their jobs because of privatization. Hence there were no disposal costs of workers for the buyers of the privatized firms. Saygili and Taymaz (2001) point out that, holding companies had a tendency to acquire plants in specific regions. For instance, Rumeli Holding bought plants in the eastern region and along the Black Sea coast. The Turkish Armed Forces Pension Fund (OYAK) and Sabanci Holding; one of the biggest conglomerates in Turkey formed an alliance and purchased companies in the Central Anatolia, Southern Anatolia, and Marmara regions. Set Cement Holding (a subsidiary of Italcementi which merged with Ciment Francais) focused on Central and Western regions, and finally, Lafarge and Yibitas bought cement plants in neighboring provinces of Central Anatolia. Saygili and Taymaz (2001) argue that, privatization through block sales, instead of public offerings in the stock market, gave rise to bigger regional monopolies. According to the report of the Central Anatolian Board of Export, however, the privatization of public cement plants increased competition in the industry and decreased prices. Today, the Turkish cement industry consists of 39 private plants, some owned by giant industrial conglomerates and others by small one-plant companies. There are four foreign investors in the industry, namely, French Lafarge Coppee, Ciment Vicat, German Heidelberger Zement/CBR, and Italian Italcementi. Cement consumption continues to grow at sound levels and Turkey continues to be a major exporter of cement. According to the report of the Central Anatolian
244
Çaxla Ökten
Board of Export, in 1998 Turkey was the largest cement producer in Europe and seventh in the world (OAIB, Cimento Sektoru Raporu, 1998). Empirical studies of the cement industry There are several empirical studies that analyze the impact of privatization on the Turkish cement industry due to availability of pre- and post-privatization data for this sector (Çakmak and Zaim, 1992; Tallant, 1993; Karatav, 1995; Özmucur, 1998; Saygili and Taymaz, 2001; Ökten and Arin, 2005). Most of these studies only analyze how privatization affects firm efficiency and yield somewhat mixed results. Ökten and Arin (2005) also analyze how privatization affects allocative efficiency. We can classify these studies into two groups. In the first group, studies compare the performance of public versus private plants and yield mixed results (Çakmak and Zaim, 1992; Tallant, 1993; Saygili and Taymaz, 2001). In the second group, studies compare pre and post privatization performance of privatized plants and present a more clear picture (Karatav, 1995; Özmucur, 1998; Ökten and Arin, 2005). All of these studies report a significant improvement in labor productivity in the post privatization period. We now analyze the studies in the first group in more detail. Çakmak and Zaim (1992) compare the efficiency of private and public plants in 1985. They employ the stochastic production frontier approach to estimate the production frontier and technical efficiency at the plant level. Their input variables in the production process are value of expenditures for labor use, value of expenditures for energy use and depreciation on capital. Their findings suggest that private plants on average are no more efficient than public plants. Tallant (1993) analyzes the relative efficiency of the public sector with respect to the private sector in Turkish cement industry in a cross-sectional study. He finds that private plants are more efficient in terms of labor productivity and capacity utilization. However, he argues that the better showing in physical measures is closely related to geographic location as western plants perform better which indicates that the initial location decision has had more to do with firm performance than public ownership per se. Tallant also analyzes financial measures of efficiency and finds that ratio of operating profit to net sales revenues is less for state owned plants than in private ones. However by his own account he is skeptical of financial measures as he states that Turkish financial measures are not detailed, accounting standards tend to vary and financial disclosures do not provide a complete picture of the firms’ financial health. Saygili and Taymaz (2001) analyze the effects of ownership and privatization on technical efficiency using a panel data set of public and private cement plants for the years 1980–1995. They have a sample of 13 private plants, 14 privatized plants, and 6 publicly owned plants during their period of study. They use dummy variables for these ownership categories omitting the dummy for public plants and estimate the effects of ownership on technical efficiency relative to the omitted dummy variable. In other words, they measure the relative performance of private or privatized plants with respect to the plants that remain public during the
Privatization: what has been achieved? 245 study period (until 1995). They estimate a stochastic frontier production function and the efficiency effects model simultaneously. They find that private plants were clearly more efficient than state-owned plants privatized after 1995 but a comparison of the average technical efficiency of private plants and public plants privatized in 1989 reveals no statistically significant difference. Their conclusion is that privatization authorities started the privatization process with the most efficient state-owned plants and the postprivatization performance of these “better” plants did not show any significant improvement. Their conclusion has a similarity to that of Tallant study in that public plants located in the east and privatized later during the process perform more poorly than private plants. We now examine the studies in the second group. Karatav (1995) uses data collected by Istanbul Chamber of Industry and observes an improvement in labor productivity but consistently poor financial performance of the five plants privatized in 1989. Özmucur (1998) analyzes a panel of public and private cement plants, using the results of the Istanbul Chamber of Industry 500 largest firms of Turkey surveys. He estimates a separate equation for each plant to determine the year of structural change for employment and labor productivity for the 1981–1995 period. He finds that structural change coincided with time of privatization for public plants and reduction in employment which to a degree happened in all plants was significantly higher in the privatized plants. Ökten and Arin (2005) analyze the effects of privatization on firm productivity and allocative efficiency using a panel data set of 22 privatized cement plants during 1983–1999 period. The data set enables the authors to avoid the endogeneity problem associated with sample selection. All public cement plants in Turkey have been privatized by 1998 and the authors have pre- and post-privatization data for all. They use firm fixed effects and yearly dummies to control for period specific effects in their baseline specification. The results show that privatization increases labor productivity and decreases prices significantly indicating an improvement in both productive and allocative efficiency. Their results on productive efficiency are robust to controlling for changes in the competitive environment (market structure) using a regional Herfindahl Index (HHI) while privatization has no longer a significant effect on prices in the presence of this control. This highlights the importance of market structure and competitive environment following privatization. They also find evidence that production technology becomes more capital intensive as the capital endowment, investment and capital labor ratios increase following privatization. How can we reconcile the differences in results as to the effects of privatization on firm efficiency? One explanation can be the differences in the competitive environment in which public plants operate in. Before the start of privatization, there are many private plants in the western regions while few exist in the eastern regions. Hence public plants located in the east face little competition from private plants. Figure 9.3 presents the Herfindahl Index for Marmara (west) and Eastern Anatolia (east) regions (Ökten and Arin, 2005). HHI is calculated by
246
Çaxla Ökten
(a)
2,400 2,300 2,200 2,100 2,000 1,900 1,800 1,700 1,600 1,500 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 HHI1
(b)
HHI2
11,000 10,000 9,000 8,000 7,000 6,000 5,000 4,000 3,000 2,000 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 HHI1
HHI2
Figure 9.3 Market concentration measured by Herfindahl Index: (a) Marmara (west) region, (b) Eastern Anatolia (east) region.
summing the squares of market shares of plants in each region. If two or more plants are owned by the same parent company, then market share of the parent company in region is used in the calculation. The higher the HHI, the more concentrated the market.
Privatization: what has been achieved? 247 In the graphs, HHI(1) is the Herfindahl index including the publicly owned plants and HHI(2) is the Herfindahl index excluding the publicly owned plants. In calculating HHI(1), share of output sold by publicly owned plants is considered as the share of a single firm – the public enterprise in that region. In calculating HHI(2), only output sold by privately owned plants is considered. Vertical lines indicate the years of privatizations. From Figure 9.3, HHI(1) for Marmara Region has ranged between 1,600 and 2,300 during 1980–2000 period and actually increased after privatization. In contrast, public plants in the eastern region formed a monopoly prior to the start of the privatization process since no private plants existed in this region. The four public plants were privatized in 1993 and 1996 and HHI dropped significantly – from 10,000 to 3,000 – as each plant was sold to a different company. Hence the lack of competition from the private sector, might be responsible for poor performance of public plants located in the east and the relatively more competitive environment in the west might be responsible for the better performance of public plants located in the west. Another possible explanation for the differences in findings is that studies are measuring short-run effects of privatization and while there is a significant improvement in labor productivity in the short-run, there are few changes in total productivity. We should note that, according to the Ökten and Arin (2005) study, capital and investment are increasing after privatization. This might indicate that plants are switching to more capital intensive technologies. These technologies might be more cost efficient in the long-run than labor intensive technologies, although both types of technologies are on the production possibilities frontier. Furthermore, the effects of investment in capital and technology might be observed with a lag. Hence a study that spans a longer post-privatization period may help reconcile some of the differences in results. Ökten and Arin (2005) study is the only study that examines the effect of privatization on allocative efficiency of the cement industry. Their measures for allocative efficiency are plant specific cement prices (in log) and the relative inflation rate. The relative inflation rate is calculated by subtracting the wholesale price index inflation rate from the plant specific price inflation rate. Prior to the price de-regulation in 1986, the price of each publicly owned plant was set to the same amount by CITOSAN, the Public Enterprise. The authors lack data on these prices but have an industry-wide price index from the State Planning Institute of Turkey. Hence, they calculate the plant specific price inflation rate by using this industry-wide price index prior to 1986, and by using plant specific prices after 1986. Since this variable merely indicates rate of change, it is possible to construct it using two different price indices as long as we code the year for which we switch from one index to the next as missing. Their goal in constructing relative inflation rate is to achieve a longer series on price. Both regressions, plant specific cement prices and the relative inflation rate, control for firm specific and period specific effects by adopting a firm fixed effect specification and including year dummies. The authors find that privatization decreases both cement prices and the relative inflation rate. A switch from public ownership to private ownership decreases cement prices by 32 percent.
248
Çaxla Ökten
Revenue generation and share ownership Since its start in 1985, the total proceeds from privatization efforts have amounted to $9.4 billion by 2004. Block sales via auctions have been the most prevalent method of privatization. There is only one empirical study that analyzes the determinants of privatization prices in Turkey (Arin and Ökten, 2003). The authors use a data set of 68 privatized firms from Turkey gathered from official statistics of the Privatization Administration. The sample includes 24 cement plants/grinding facilities, 29 dairy product plants, 4 ports, 3 marinas, 2 airline service firms, and 6 heavy industry manufacturers like mining and metal firms, which were privatized in Turkey between the years 1989 and 1998. In that sample, 65 out of 68 firms are sold through block sales via auctions. The remaining 3 firms are sold through a combination of block sales and public offerings. Interestingly, the authors find that revenues affect privatization prices, not profits and hence argue that what fundamentally determines the privatization price is the expected future profit of the firm. Potential buyers would discount firms’ current cost information if they believe that these firms were inefficient. Current costs and hence profits do not affect privatization prices because they do not reflect expected future profits, whereas revenue and market characteristics are good indicators for future profitability. Firms’ profit margins have positive and significant effects on privatization prices when the whole heterogeneous sample is used similar to findings for Mexico by Lopez-De-Silanes (1997). However, when authors concentrate on a single industry (cement industry), this variable is no longer significant. Therefore, their interpretation for the effect of this variable is different from that of Lopez-De-Silanes. They argue that profit margins measure the differences in market power of firms in different industries (market structures) rather than differences in firm efficiency and this is what gets reflected in privatization prices. Arin and Ökten (2003) find that the number of bidders increases privatization prices consistent with the results for Mexico. Sales of SOEs through public offerings have been utilized on a limited scale. The limited size and depth of the ISE has undoubtedly restricted the sale of SOEs by public offerings. Hence one could argue that, the objectives of wider share ownership and developing a viable capital market have not been met. However, share issue privatizations include some of the largest SOEs as can be seen from Tables 9.2 and 9.3.
Concluding remarks Economic theory specifies a role for state ownership only in natural monopoly type of market structures where efficiency requires a single firm to exist. Even then, state ownership may not improve efficiency if SOEs pursue other objectives due to political or bureaucratic pressure (Vickers and Yarrow, 1988; Shleifer and Vishny, 1994). Historically, Turkey has had a long experience of relying heavily on SOEs even in markets that can not be characterized as a natural monopoly. SOEs were established during the 1930s by the government to jump-start the economy that collapsed with the end of the Ottoman era in 1923. SOEs grew enormously over the years and
Privatization: what has been achieved? 249 served as engines of economic activity in the absence of a strong private sector. By 1960, share of public sector in total value added in manufacturing was 60 percent. By the 1980s, SOEs began to be perceived negatively due to poor financial performance, overstaffing, dependence on subsidies, protected markets, and corruption. Hence, privatization came into the political agenda first with Prime Minister Özal’s trade and capital account liberalization program in 1984. Despite the initial enthusiasm, it is not surprising that privatization of SOEs was a slow process. The preferences that created and promoted SOEs were also reflected in the country’s laws and institutions that challenged many privatization attempts. Still, a large number of SOEs and state shares in companies have been divested between 1985 and 2005 and the total proceeds have amounted to $9.4 billion. Block sales have been the most prevalent method of privatization. One could argue that this policy contradicts with the aim of developing a viable capital market and facilitating a wider distribution of share ownership. Arin and Ökten (2003) analyze the determinants of privatization prices in block sales via auctions and find that revenue and market characteristics affect privatization prices, not profits and argue that what fundamentally determines the privatization price is the expected future profit of the firm. They also find that competition measured by the number of bidders increases privatization prices. There are several empirical studies that analyze the impact of privatization on the Turkish cement industry due to availability of pre- and post-privatization data for this sector (Çakmak and Zaim, 1992; Tallant, 1993; Karatav, 1995; Özmucur, 1998; Saygili and Taymaz, 2001; Ökten and Arin, 2005). Results of these studies are mixed on the effects of privatization on firm efficiency. Studies that span a longer post-privatization period and consider the links between ownership type and the competitive environment are needed to reconcile differences.
Notes 1 A survey by Djankov and Murrell (2002) examines the effects of privatization in transition economies. They conclude that in most countries, privately owned firms perform better than state-owned firms. 2 Bartero and Rondi (2000) show that consistent with theoretical predictions, Italian stateowned firms respond to financial pressure under a hard budget constraint by increasing total factor productivity and reducing employment whereas no such positive effect is observed under a soft budget constraint. 3 This is an interesting case. In July 1998, PA High Council reversed its previous decision that favored the offer of $1.6 billion by the highest bidder, the Akmaya-Orteks AS group and decided to sell 51 percent of POAS to the third highest bidder, the WvBank, Park Holding, Bayindir and PUIS consortium group. This group agreed to raise its initial bid to the level of the highest bid. Nevertheless, the privatization of POAS was suspended by the Ankara Supreme Court, as the privatization deal was deemed to be against the Constitution and the principle of wider share ownership. The partial sale of POAS took place in March 2000 after the legal framework has been prepared for its divestment (Karatav, 2001). 4 Also privatized were the two cement grinding facilities. 5 Two exceptions were Denizli and Lalapasa. These two public plants were established in 1987 and 1991 respectively, in order to meet the growing demand in the western regions.
250
Çaxla Ökten
References Anderson, J.H., Y. Lee, and P. Murrell (2000), “Competition and Privatization Amidst Weak Institutions: Evidence from Mongolia,” Economic Inquiry, 38(4): 527–549. Angelucci, M., S. Estrin, J. Konings, and Z. Zolkiewski (2001), “The Effect of Ownership and Competitive Pressure on Firm Performance in Transition Countries, Micro Evidence from Bulgaria, Romania and Poland,” CEPR Discussion Paper: 2985. Arin, K.P. and C. Ökten (2003), “The Determinants of Privatization Prices: Evidence from Turkey,” Applied Economics, 35(12): 1393–1404. Aybar, C.B., E.S. Guney, and S. Suel (2001), “Privatization and Regulation in Turkish Telecommunications: A Preliminary Assessment,” SNHU International Business Program Working Paper Series No. 2001–2002. Bartero, E. and L. Rondi (2000), “Financial Pressure and the Behaviour of Public Enterprises under Soft and Hard Budget Constraints: Evidence from Italian Panel Data,” Journal of Public Economics, 75: 73–98. Bhaskar, V. and M. Khan (1995), “Privatization and Employment: A Study of the Jute Industry in Bangladesh,” American Economic Review, 85: 267–273. Çakmak, E. and O. Zaim (1992), “Privatization and Comparative Efficiency of Public and Private Enterprises in Turkey: The Cement Industry,” Annals of Public and Cooperative Economics, 63: 271–284. Djankov, S. and P. Murrell (2002), “Enterprise Restructuring in Transition: A Quantitative Survey,” CEPR Discussion Papers: 3319. Earle, J.S. and A. Telegdy (2002), “Privatization Methods and Productivity Effects in Romanian Industrial Enterprises,” Journal of Comparative Economics, 30(4): 657–682. Ehrlich, I., G. Gallais-Hamonno, Z. Liu, and R. Lutter (1994), “Productivity Growth and Firm Ownership: An Analytical and Empirical Investigation,” Journal of Political Economy 102(5): 1006–1038. Ercan, M. and Z. Önis (2001), “Turkish Privatization: Institutions and Dilemmas,” Turkish Studies, 2(1). Ertuna, O. (1998), “Constraints of Privatization: The Turkish Case,” Bogazici University, Working Paper. Ficici, A. (2001), “Political Economy of Turkish Privatization: A Critical Assessment,” Working Paper, New Hampshire College. Financial Times (18.09.2000), a news article by Leyla Boulton. Frydman, R., C. Gray, M. Hessel, and A. Rapaczynski (1999), “When Does Privatization Work? The Impact of Private Ownership on Corporate Performance in the Transition Economies,” Quarterly Journal of Economics, 114(4): 1153–1191. Istatistiki Gostergeler (Statistical Indicators) (1923–2002), Devlet Istatistik Enstitusu, Ankara. Johnson, S., H. Kroll, and E. Santiago (1994), “Strategy, Structure and Spontaneous Privatization in Russia and Ukraine,” in Changing Political Economies: Privatization in PostCommunist and Reforming Communist States, Vedat Milor (ed.), Boulder: Lynne Rienner, pp. 147–173. Karatav, C. (1995), “Has Privatization Improved Profitability and Performance of the Public Enterprises in Turkey,” in Privatisation Policy and Performance: International Perspectives, P. Cook and C. Kirkpatrick (eds), London: Prentice Hall, pp. 244–262. Karatav, C. (2001), “Privatization in Turkey: Implementation, Politics of Privatization and Performance Results,” Journal of International Development, 13: 93–121. Kattuman, P. and R. Domanski (1997), “Industrial Concentration Under Shock Therapy: Poland in the Early Transition Years,” University of Cambridge, ESRC Centre for Business Research Working Paper: WP76.
Privatization: what has been achieved? 251 La Porta, R. and F. Lopez-de-Silanes (1999), “The Benefits of Privatization: Evidence from Mexico,” Quarterly Journal of Economics, 114(4): 1193–1242. Lopez-De-Silanes, F. (1997), “Determinants of Privatization Prices,” Quarterly Journal of Economics, 112: 965–1025. Megginson, W. and J. Netter (2001), “From State to Market: A Survey of Empirical Studies on Privatization,” Journal of Economic Literature, 39: 321–389. Niskanen, W. (1975), “Bureaucrats and Politicians,” Journal of Law and Economics, 18(3): 617–643. OIB (Ozellestirme Idaresi Baskanligi – Privatization Administration) (2005), “Turkiye’de Ozellestirme (Privatization in Turkey),” www.oib.gov.tr (accessed February 1, 2005). Ökten, C. and P. Arin (2005), “How does Privatization Affect Efficiency? Evidence from Turkey,” Bilkent University Working Paper Series, No. 05–04. Omran, M. (2004), “The Performance of State-Owned Enterprises and Newly Privatized Firms: Does Privatization Really Matter?,” World Development, 32(6): 1019–1041. Öniv, Z. (1991), “Evolution of Privatization in Turkey: The Institutional Context of Public Enterprise Reform,” International Journal of Middle East Studies, 23: 163–176. Orta Anadolu Ihracatcilar Birligi (1998), Cimento Sektoru Raporu. (Central Anatolian Board of Export, Cement Industry Report.) Özmucur, S. (1998), “Privatization and the Labor Market in Turkey,” in Main Characteristics and Trends of the Turkish Labour Market, Tuncer Bulutay (ed.), State Institute of Statistics: Ankara, pp. 171–209. Saal, D. and D. Parker (2001), “Productivity and Price Performance in the Privatized Water and Sewerage Companies of England and Wales,” Journal of Regulatory Economics, 20(1): 61–90. Saygili, S. and E. Taymaz (2001), “Privatization, Ownership and Technical Efficiency: A Study of the Turkish Cement Industry,” Annals of Public and Cooperative Economics, 74(4): 581–605. Shapiro, C. and R. Willig (1990), “Economic Rationales for the Scope of Privatization,” in The Political Economy of Public Sector Reform and Privatization, B.N. Suleiman and J. Waterbury (eds), London: Westview Press, pp. 55–87. Shleifer, A. and R. Vishny (1994), “Politicians and Firms,” Quarterly Journal of Economics, 109: 995–1025. Tallant, D. (1993), “Relative Efficiency of Public and Private Sector Ownership and Privatization of the Turkish Cement Industry,” Review of Social, Economic and Administrative Studies, 7(1–2): 73–103. Vickers, J. and G. Yarrow (1988), “Privatization: An Economic Analysis,” MIT Press Series on the Regulation of Economic Activity, 18. Cambridge, MA and London: MIT Press. Villalonga, B. (2000), “Privatization and Efficiency: Differentiating Ownership Effects from Political, Organizational, and Dynamic effects,” Journal of Economic Behavior & Organization, 42: 43–74. Wallsten, S.J. (2001), “An Econometric Analysis of Telecom Competition, Privatization, and Regulation in Africa and Latin America,” Journal of Industrial Economics, 49: 1–19. Warzynski, F. (2003), “Managerial Change, Competition and Privatization in Ukraine,” Journal of Comparative Economics, 31: 297–314. Yılmaz, K. (2000), “Turk Telekomunikasyon Sektorunde Reform: Ozellestirme, Duzenleme ve Serbestlesme (Reform, Privatization, Regulation, and Liberalization in the Turkish Telecommunications Sector),” in Devletin Duzenleyici Rolu, Tesev Yayinlari, 19.
10 Political economy of pension reform in Turkey Serdar Sayan
Introduction Making sure that retired workers will be able to maintain at least some minimum standard of living during later parts of their lifetime is an important social policy goal for all governments. For this reason, many countries legally require workers to have some old-age insurance coverage provided through publicly managed pension schemes that are typically run on a Pay-As-You-Go (PAYG) basis. Public PAYG schemes finance pensions and other old-age benefits to retired workers by using contributions collected from currently active workers and their employers in the form of payroll taxes. This makes revenue–expenditure balances of publicly managed pension funds sensitive to reductions in the relative sizes of retirees collecting benefits and active workers continuing to pay contributions. If such reductions are not matched with increases in contributions and/or reductions in old-age benefits over extended periods of time, the resulting deterioration in revenue–expenditure balances will eventually lead to sizable deficits that need to be funded by the Treasury.1 Certainly, different political economy issues shape up policy makers’ attitude towards growing pension deficits and the timing of policy responses. First of all, even though a gradual decline in the number of workers per retiree is certain to occur over time, incumbent governments tend to delay taking action to curb the growth in pension deficits as long as possible, since increasing contributions and/or lowering pension benefits are not likely to prove popular among voters. While not as commonly observed, a second and more extreme form of myopic policy action involves interventions that change the incentive structure driving individuals’ work/retirement decisions or the eligibility criteria for retirement in such a way to speed up the deterioration in pension balances for political rent-seeking purposes. In either case, avoiding timely action on account of the lack of electoral support often causes pension deficits to eventually reach alarming magnitudes, requiring a major overhaul of the pension system without further delay at some point. Such an overhaul typically involves a parametric pension reform (Chand and Jaeger, 1996) to change the existing values of public pension program parameters (contribution and replacement rates and statutory entitlement ages for pension benefits2) markedly so as to restore balances or shrink deficits to tolerable levels by
Political economy of pension reform
253
increasing contributions collected and/or reducing benefits paid out. In general, previous delays in parametric reform require adjusting values of pension program parameters more dramatically than it would have been necessary in the absence of such delays. Since increasing the contribution rate and entitlement age are likely to face opposition from younger voters, whereas reducing the replacement rate is likely to displease older voters, the implementation of reform itself is a process driven by political economy considerations: New values of parameters must be set in such a way to minimize the harm to re-election chances of the incumbent government (i.e. left no choice but) to introduce the reform.3 In addition to adjustments to pension parameters, the reform process may also call for a restructuring of the existing institutional setup of the system or establishment of new institutions, possibly giving rise to power struggles between supporters and opponents of institutional reform and hence creating additional political economy concerns. The purpose of this chapter is to discuss some of these political economy issues in the context of recent efforts to rehabilitate the Turkish pension system, and to investigate alternative parametric reform scenarios that could be adopted for that purpose. Discussion in the remainder of the chapter is organized as follows. The next section describes the structure and problems of the pension system in Turkey while the section on Pension reform process provides a brief account of the recent pension reform process. The section on Alternatives of the reform of 1999, looks into alternative reform options and discusses advantages and limitations of these alternatives. The final section concludes the chapter and summarizes likely developments ahead.
Structure and problems of the Turkish pension system The publicly managed social security system in Turkey was set up and began to provide old-age insurance coverage and other social security benefits in the 1940s. Notwithstanding the work under way to create a single public pension fund administration, the system is currently made up of three different agencies, each with its own PAYG pension operations to provide retirement benefits to different groups within the working population, and a number of pension funds that commercial banks have set up for their own employees. SSK, the largest of the three public agencies, provides social security coverage to blue-collar workers employed in both private and public sectors and white-collar workers in the private sector, whereas ES serves as the pension fund administration for white-collar workers employed by local and central governments. The third one, BK, covers farmers, artisans, and other self-employed people. Participation in retirement plans offered is compulsory for all wage and salary earners as well as self-employed individuals in principle but actual coverage is far less comprehensive due to serious compliance problems. Voluntary participation is possible for people who are active outside compulsory coverage categories.4 Prior to 2003, additional pension benefits beyond coverage provided by these three public agencies were only available through a number of pension funds
254
Serdar Sayan
privately set up by some companies, universities etc. to provide optional coverage to their own employees. After the legal and regulatory framework was drawn up to allow working individuals to voluntarily purchase additional coverage from private companies, many insurance companies began to offer individual retirement plans in 2003, essentially transforming the single-pillar pension system in Turkey into a two-pillar system. The introduction of individual retirement accounts to be managed by private companies was in fact part of a larger process for social security reform, started in 1999 following serious financial difficulties that social security agencies run by the state began to experience due to the imbalance between contributions collected and benefits paid out. The primary culprit behind the deficits that became persistent after the 1980s was pension/old-age insurance operations that generated significant losses primarily due to generosity of pension benefits to retirees and their survivors relative to the amount and duration of contributions. This relative generosity resulted from pension parameters introduced and maintained by different governments purely for the purposes of reaping political benefits or avoiding political risks. Among the three agencies, pension (and health) benefits provided by ES have traditionally been even more generous than the other two, making public service a particularly attractive choice of employment for many people. When the pension crisis hit Turkey, PAYG-based pension funds run by the state in several other countries were also encountering financial difficulties of their own due to declining worker-to-retiree (or active-passive) ratios. Differently than those countries where decreasing worker-to-retiree ratios resulted from aging of respective populations, however, the fall in this ratio in Turkey was mainly due to various reductions in pension benefit entitlement ages (Table 10.1). Pension regulations in effect until 1999 allowed female/male workers to begin collecting retirement benefits as early as 38/43 years of age. Although relatively few people actually retired before 40 years of age, about one-third of people who applied to SSK to get their retirement benefits started during 1996–1999 were younger than 45 years of age.5 Despite some variation across social security agencies, the median retirement age has been significantly lower than OECD averages (Gillion and Cichon, 1996). The reductions in entitlement ages financially squeezed public pension funds by lowering both average contributions net of benefits per worker6 and worker–retiree ratios (Table 10.2). Table 10.1 Statutory entitlement ages in Turkey before 1999 Years
Female workers
Male workers
1950 1965 1969 1986 1992
60 55 38 55 38
60 60 43 60 43
Source: Kenar et al. (1996).
Political economy of pension reform
255
The decline in worker-to-retiree ratios resulting from the tendency of workers to take advantage of regulations allowing for low retirement ages has indeed been the major contributor to the growth in pension deficits which, by the end of the 1990s, had reached unsustainable proportions. Interestingly, the entitlement ages were not pulled down to such preposterously low levels as 38 and 43 as a result of the demands by trade unions and other pressure groups. Instead, entitlement age cuts were initiatives of ruling parties aiming to increase their chances of getting re-elected by achieving quick reductions in unemployment through positions opened up by young retirees (Ivıklı, 1998). It soon became obvious, however, that cuts to minimum retirement ages were rather short-sighted and uncalculated moves by politicians, since most workers who opted to retire at such early ages were taking up employment elsewhere, thereby adding pension receipts to wage/salary income from their new jobs. Finding employment was relatively easy for these young retirees as they were usually more experienced and skilled than younger people seeking first time employment. Furthermore, they were willing to accept, often unregistered, employment at lower wages/salaries since they already had a pension income. Some did not even have to really quit their job: they would apply to get their retirement benefits started as soon as they become eligible, and would get “re-hired” for the very same position and continue working in a business-as-usual fashion. Many employers allowed and even encouraged their employees to go through retirement procedures without effectively quitting their jobs, since they had nothing to lose. On the contrary, they would at least save the employers’ share of contributions by re-hiring these workers after they started collecting their pensions.7 As such, Turkey’s experience has been a clear example to how badly designed schemes can encourage premature retirement (Barr, 2002). The lack of labor market regulation with respect to re-employment of young retirees, coupled with already low compliance rates with the existing regulations and the large size of unregistered workers, not only limited the job creation effect hoped of lower entitlement ages, but also contributed to the growth in pension losses and hence public sector borrowing requirements (PSBR) that were already running high (Table 10.3). Given the way high and growing PSBR had been driving interest rates up, the most probable net result of reductions in statutory entitlement ages was a hike in unemployment rates due to crowding out of private investment – effectively suppressing employment creation capacity of the economy below what would have been in the absence of such attempts at quick fixes for the unemployment problem. Even though not carefully contemplated reductions in entitlement ages for short-term political benefits were the primary reason underlying the untimely experience of Turkey with pension crisis, when the country still had a predominantly young population,8 it was by no means the only populist intervention of politicians to disturb actuarial balances of the pension system. Such interventions started back when the policy-induced decline in worker-to-retiree ratios had not reached critical levels. As publicly managed pension funds were still running surpluses thanks to the advantageous age composition of masses they covered,
Table 10.2 Pension coverage by status and agency: 1990–2002 (in thousands) Active insured/employment (%)
1990
2000
2002
SSK (Sosyal Sigortalar Kurumu) total 1 Insured (active contributors) 2 Voluntarily covered contributors 3 Active contributors in agriculture 4 Pension recipients (passive users) 5 Dependents 6 Active/passive ratio [(1 2 3)/4] 7 Dependency ratio [(5 4)/(3 2 1)] BK (Bax-Kur) total 1 Insured (active contributors) 2 Voluntarily covered contributors 3 Active contributors in agriculture 4 Pension recipients (passive users) 5 Dependents 6 Active/passive ratio [(1 2 3)/4] 7 Dependency ratio [(5 4)/(3 2 1)] ES (T.C. Emekli Sandixi) total 1 Insured (active contributors) 2 Pension recipients (passive users) 3 Dependents 4 Active/passive ratio [1/2] 5 Dependency ratio [(3 2)/1] Private pension funds total 1 Insured (active contributors) 2 Pension recipients (passive users) 3 Dependents 4 Active/passive ratio [1/2] 5 Dependency ratio [(3 2)/1] Overall total 1 Insured (active contributors) 2 Voluntarily covered contributors 3 Active contributors in agriculture
18,977
34,139
35,261
86
3,446 300
5,283 844
5,257 942
52 214
74
185
149
101
1,596
3,339
3,748
134
13,560 2.39
24,488 1.89
25,166 1.69
85 29
3.97
4.41
4.55
15
13,361 1,967 106
15,036 2,182 255
15,548 2,193 238
16 12 124
752
876
891
18
595
1,277
1,394
134
9,940 4.75
10,446 2.59
10,833 2.38
9 ↓
3.73
3.54
3.68
↓
8,586 1,560 843
9,766 2,164 1,297
10,699 2,373 1,409
24 52 67
6,182 1.85 4.50 312 84 32
6,305 1.67 3.51 324 78 71
6,917 1.68 3.51 324 72 78
11 ↓ ↓ 10 14 143
195 2.63 2.70 41,236 7,058 406
174 1.10 3.12 59,265 9,707 1,099
175 0.92 3.53 61,832 9,894 1,180
10 ↓ ↑ 50 40 190
826
1,061
1,040
26
% Change 1990–2002
Political economy of pension reform
257
Table 10.2 Continued 1990 4
Pension recipients (passive users) 5 Dependents 6 Active/passive ratio [(1 2 3)/4] 7 Dependency ratio [(5 4)/(3 2 1)] Total population Share of insured population (%) Active insured/employment (%)
2000
2002
% Change 1990–2002
3,068
5,985
6,628
116
29,878 2.70
41,413 1.98
43,090 1.83
44 ↓
3.97
3.99
4.10
↑
58,754 72.7 39.1
68,036 87.1 46.3
70,171 88.1 46.3
19 21 18
Source: State Planning Organization, The Eighth Five Year Development Plan: 2001–2005, and Annual Program Documents 2003 and 2004 (all in Turkish).
Table 10.3 Pre-1999 growth of social security deficit in Turkey Year
Share of total transfers (treasury funding) to social security agencies in total budget deficit ( %)
Share of total transfers (treasury funding) to social security agencies in GNP (%)
1993 1994 1995 1996 1997 1998
17 20 36 26 34 35
1.20 1.13 1.44 2.16 2.55 2.85
Source: Ayas (1998).
governments began to tap these surpluses to finance overspending in other budget items, thereby diverting investable resources of pension funds to unrelated areas such as agricultural support purchases (Kılıçdaroxlu, 1998; Sayan, 2001). This diversion, coupled with the mismanagement of assets that had been acquired by pension funds, helped pull back the crisis in time. Over the years, politicians continued to meddle with the pension system by enacting several laws to make previously uncovered voters eligible to receive pension benefits that are incomparably higher than the amounts they were expected to pay in back contributions, and by handing out pre-election bonuses to retirees in the form of arbitrary raises to pension incomes across the board and “one time” lump sums called social assistance payments. Kılıçdaroxlu (1998) calculated that these social assistance payments handed out without a matching increase in contributions amounted to US$ 13.8 billion over the 1977–1996 period. In addition to the burden they placed on pension balances, these payments and arbitrary
258
Serdar Sayan
additions to pension incomes have created uncertainties for all retirees, making it difficult for them to plan ahead (TUSIAD, 2004). These pre-election bonuses have broken the link between contributions and benefits which was already rather weak, particularly in the case of ES, the pension fund administration for civil servants employed by local and central governments. Due to the historically privileged status of this group with its roots in the strong and long-lasting bureaucratic tradition in Turkey, ES has always had a special place among the three agencies in terms of the generosity of pension (and health) benefits it provided relative to contributions collected. In summary, the introduction of imprudent legislation for short-term political gains – as in the case of entitlement age reductions and payments to retirees of pre-election bonuses not tied to contributions – and mismanagement of pension funds under political pressure – as in the case of the use of excess contributions in financing other public expenditures – have all contributed to the financial crisis of the pension system that Turkey prematurely experienced in the 1990s. While some governments were directly responsible for starting or speeding the deterioration in pension balances through politically motivated steps, others are to be blamed for their failure to take action to stop or even reverse this process in fear of voter reaction. In other words, the losses generated by the Turkish pension system proved unsustainable by the end of the 1990s as a result of the actions or the lack thereof by different governments despite the advantageous age profile of Turkish population. This end could still have been delayed, if the large number of unregistered workers could be reduced through appropriate labor market policies so as to generate additional contribution revenue. The number of unregistered workers in the private sector and self-employed individuals is very high indeed (currently estimated by the Ministry of Labor and Social Security to be around 11 million) due, to a large extent, to the high rates of payroll taxes – that are supposed to be collected from self-employed individuals themselves, or private sector workers and their employers as contributions. High payroll taxes have not only encouraged employers to employ unregistered workers but also reduced competitiveness of businesses that hire only registered workers, creating additional compliance problems and collection difficulties. Even for registered workers and the self-employed, compliance problems cause actual contribution receipts to fall significantly below potential levels (Topal, 1999; TUSIAD, 2004). Low compliance manifests itself particularly in the form of a widespread tendency among employers to underreport payments to workers or to withhold contributions collected rather than transferring them to pension fund administrations, as there exists economic incentives for such violations (such as low late transfer penalties relative to market interest rates). These violations often go unpunished due, in part, to the inability of governments to enforce full compliance with pension laws, and in part to the lack of their willingness to pursue small businesses run by a few employees for political and sometimes economic (as during the times of economic crises) reasons. Regardless of the reasons, governments’ tendency to look the other way causes large amounts of potential
Political economy of pension reform
259
contribution revenue to be lost. Prior to 1999, data published by BK itself indicated that only 2.8 percent out of 3 million registered individuals owed no back contributions, whereas 27 percent made almost no contribution after getting registered with the remaining 70 percent making irregular payments (Sayan, 1999).
Pension reform process Implementation of the reform of 1999 Growing share of budget deficits in GNP and the rapidly increasing contribution of funding needs of the social security system to this share had begun to threaten future solvency of the system by the second half of the 1990s. Furthermore, total deficit of the whole system was projected to increase to 16.8 percent of GNP by the year 2050 in the case of non-intervention. The divergence between expenditures and revenues in pension operations was largely responsible for the bleak outlook and the GNP share of pension losses alone was projected to exceed 5 percent by 2010, should the pre-1999 values of pension parameters be retained (Ayas, 1998; see also Gillion and Cichon, 1996 for similar projections). Having realized that there was almost no room left for delaying action any further any more, the coalition government that was in power prepared a proposal for a major pension reform in 1999. The obvious need to extend average contribution period (and shorten average benefit collection period) made the entitlement age the most critical parameter of reform. In the early drafts of the reform bill, the entitlement ages were set at the substantially higher levels of 63 and 65 respectively for female and male workers who would start working for the first time after the new law has gone into effect. For the existing workers, the bill called for gradual increases over time that were to be progressively higher for more junior workers. Even though past reductions in entitlement ages were not directly demanded by workers, the bill’s provisions to increase them faced a very strong resistance, with “say no to retirement in the grave” becoming the slogan of opposing groups. Under intensive pressure from trade unions and other interest groups as well as opposition parties in the parliament, the government had to modify the proposal to limit the entitlement age hike to 58 and 60 for female and male workers, respectively (Sayan and Kiraci, 2001a). After heated debates between the government and the opposition groups inside and outside the Parliament, the bill was legislated into law no. 4447 in September 1999. While it raised entitlement ages to 58/60 for first time employees and set a transition period of 8 years for gradual increases for existing employees, the values of average contribution and replacement rates were maintained.9 Besides the age increase, major modifications introduced included the extension of the indexation period for calculation of pension benefits to the entire duration of employment and the linking of increases in pension incomes to consumer inflation (TUSIAD, 2004) to curb the tendency of governments to make arbitrary adjustments to pensions received by retirees for political benefit.
260
Serdar Sayan
The parametric reform was a crucial step to control the rapid growth in pension and hence social security deficits. Yet, significant and lasting reductions in deficits would have required more radical measures that were not necessarily feasible politically. One of the major shortcomings of law no. 4447 was its failure to restore the missing link between contributions and pension benefits for ES on account of the government’s reluctance to deal with the reaction that civil servants including higher ranking bureaucrats, and the members of the judiciary, police forces, and the military – particularly outspoken pressure groups that most governments would rather not have a confrontation with for political reasons. So, the only significant provision of the reform act for personnel covered by ES has been the increase in entitlement ages. Even then, the effectiveness of this most notable provision of law no. 4447 was reduced by a decision that the Turkish Constitutional Court (the supreme court in the country) took in response to an appeal by some members of the Parliament. As an act of political solidarity with trade unions and other groups opposing the hike in entitlement ages, a number of deputies from the main opposition party challenged the conformity of age adjustment with the constitution by filing an appeal with the supreme court. Following its examination of the appeal, the Constitutional Court ruled in February 2001 that the law’s (non-linear) scheme for gradual increases in the entitlement age during the transition period violated the equality principle and hence constitutional rights of workers in certain age groups and ordered that this provision be reconsidered. The Constitutional Court required that the government make legal arrangements within six months to make sure that the transition to new entitlement ages will be smoother (Sayan, 2002). Consequently, the transition period was extended until 2020 to increase entitlement age roughly by one every year, seriously reducing the effectiveness of the initially planned age increase scheme as a tool for reducing pension deficits (Table 10.4). Notwithstanding the modifications to age adjustments, law no. 4447 had been intended to be just the first stage of a larger reform to overhaul the system. The additional steps envisaged for the completion of reform process included the introduction of individual retirement accounts managed by the private sector as an optional second pillar; improvement of information technology infrastructure for more accurate record keeping with less paperwork and red tape in such a way to increase compliance and improve efficiency of operations, and the initiation of a work program to merge different rules and regulations of coverage under each of the three pension institutions into a unified framework. Such merging of different rules and regulations to create a similar cost and benefit structure of coverage across social security agencies was rather significant since it would essentially put an end to the privileged status of ES. This required major institutional restructuring to achieve and hence was set as a goal that needs to be pursued during the later stages of reform. Despite serious political risks of such a move, the incumbent government in 1999 facing strong pressure from the IMF saw the inevitability of such an institutional reform. It soon began to take the initial steps in that direction but adopted a public relations strategy based on
Political economy of pension reform
261
Table 10.4 Minimum entitlement ages for female workers (who were already employed at the time of the passage of new regulations) Years
Prior to 1999
Law no. 4447 (1999–2002)a
Revision of law no. 4447 (after 2002)
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021–2060
38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38
38 41 43 45 47 48 49 50 51 58 58 58 58 58 58 58 58 58 58 58 58 58
38 38 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 58
Note a Minimum entitlement ages that could not be put into effect due to the Constitutional Court’s ruling.
the avoidance of emphasizing the ultimate goal of bringing costs and benefits of coverage under ES in line with the other two agencies. The first step for the elimination of differences between the way contributions are linked to benefits across agencies and establishment of common standards was the creation of a social security administration to coordinate the harmonization of rules and regulations and to serve as an umbrella organization for the existing agencies. The government issued decree no. 618 for this purpose in October 2000 but legal complications arising later dictated that a law be approved by the Parliament delaying the establishment of this administration until the enactment of law no. 4947 in July 2003 (TUSIAD, 2004). There has also been some progress in the achievement of other targets set for successive waves of pension reform after 1999. The most significant of these has been the creation of an optional second pillar through the entry of private companies into the pension insurance business (Teksöz and Sayan, 2002). Private insurance companies in Turkey had long been offering pension benefits to
262
Serdar Sayan
individuals who want additional coverage through life insurance policies even before, but the purchase of defined-contribution plans specifically designed for the purposes of providing additional pension income to individuals covered by one of the public pension schemes became possible only in 2003. The bill allowing for the opening of individual retirement accounts with private companies was enacted into law in 2001, and following the completion of the regulatory framework, a number of private companies began to operate in 2003, selling pension plans to individuals who would like to take advantage of market determined rates of return on their long-term savings and the tax incentives provided.10 Yet reduced disposable incomes of workers due to high contribution rates charged by public pension schemes and the recent economic crises have so far limited the growth in the demand for individual retirement plans. Evaluation of the reform of 1999 Despite a series of initiatives including the enactment of the Pension Reform Act of September 1999, the reform process had only limited success in curbing social security deficits. In fact, the passage of law turned out to be largely ineffective in stopping the growth in social security deficits except for a temporary improvement in the balances of SSK and to some extent BK. The balances of ES, on the other hand, continued to deteriorate since law no. 4447 largely kept conditions for ES coverage intact except for the entitlement age hike whose effectiveness was mitigated by the ruling of the Constitutional Court. As a result, the ratio of deficits to GNP even exceeded its pre-reform values (Table 10.5). While there is no doubt that the Constitutional Court decision requiring the gradual increases to entitlement age be slowed down has been a major contributor to the failure of reform, it was not the only factor. The costs of timidity of the incumbent government in extending the coverage of parametric reform to ES on account of political risks have also become evident by now, as ES, the lowest-ranking agency in terms of the size of membership, became the highest-ranking contributor to overall social security deficits. In addition, the growth in lump-sum social assistance payments not tied to contributions also continued especially during severe economic crises that Turkey experienced between 1999 and 2002. Perhaps more importantly, the size of unregistered worker population could not be Table 10.5 Transfers to publicly managed social security agencies funded by the treasury (as percentage of GNP) Total
1999 (%)
2000 (%)
2001 (%)
2002 (%)
2003 (%)
2004a (%)
SSK BK ES Total
1.41 1.02 1.32 3.75
0.32 0.84 1.41 2.57
0.69 0.99 1.49 3.17
0.87 0.96 1.71 3.54
1.35 1.38 1.72 4.45
1.39 1.27 1.86 4.52
Source: Turkish Social Security Administration. Note a Provisional.
Political economy of pension reform
263
reduced. The reform of 1999 has not served this purpose at all indeed, since the cost of employing registered workers remained high due to the the maintenance of high contribution rates at their pre-1999 values. The evolution of GNP shares of transfers in Table 10.5 clearly indicates the need for a more radical and comprehensive reform which would not leave any of the existing agencies out of its scope, including new adjustments to pension parameters. It is interesting to note that in the absence of such a reform, Turkey’s aspirations to become a full member of the European Union could be killed by the deficits of its publicly managed social security system alone, since the upper limit of public deficits to GNP ratio allowable for members has been set at 3 percent by the Maastricht criteria. Furthermore, the upcoming demographic pressure increases the urgency of such a comprehensive reform. Even though Turkey still has a young population relative to the most of its OECD partners, population aging gradually sets in (Kenc and Sayan, 2001). The country is projected to get out of the demographic window of opportunity characterized by the increasing share of working age people in total population during the next two decades, signalling the need to take additional measures to avoid even larger deficits (TUSIAD, 1999). The next section discusses some alternatives that could be considered to face the challenges ahead of the Turkish pension system.
Alternatives of the reform of 1999 As Table 10.5 indicated, law no. 4447 has achieved little in curbing the deficits of the publicly managed social security system in Turkey. Facing even larger deficits to be funded than before and under intense pressure from the IMF which tied the renewal of the Stand-By Agreement under negotiation with Turkey to the condition of a major social security reform, the governments that came to power after the November 2002 elections prepared a new draft for a social security reform bill in July 2004. The bill focused on another parametric reform ruling out the possibility of major restructuring such as the introduction of a third pillar through compulsory individual retirement accounts. Despite concrete suggestions spelled out in a report commissioned by the Turkish Association of Industrialists and Businessmen (TUSIAD, 2004) and some lobbying, the government insisted on parametric reform on account of the lack of initial financing to meet the costs to be incurred during transition from a two-pillar system to a three-pillar one – made up of public pension schemes and one compulsory and one optional pillar for pension plans to be provided by privately managed but regulated insurance companies. So, the rest of the discussion in this section focuses on parametric reform alternatives alone, and as such, contributes to the literature on identification of parameters for rehabilitating a PAYG-based pension system as briefly surveyed by Chand and Jaeger (1996) and Sayan and Kiraci (2001b). Especially relevant studies focusing on the Turkish case include ILO (1996), Akmaz (1999) and Sayan and Kiraci (2001a,b). The investigation of alternative reform parameters here is carried out in reference to old-age insurance operations of SSK, the largest of three publicly managed social security agencies in Turkey. The results are obtained by solving an actuarial model through a genetic algorithm (GA) developed for this purpose
264
Serdar Sayan
(Sayan and Turhan-Sayan, 2001).11 The GA determines optimal values of pension program parameters by realistically allowing for transitional and long-term changes in entitlement ages over time and it significantly improves computational efficiency over other studies dealing with the same problem (see, for example, Sayan and Kiraci, 2001a). Despite partial equilibrium nature of the actuarial model used, the investigation here also links up well with the literature studying social security reform issues using overlapping generations general equilibrium models – see Joines et al. (1999) for a survey, and Miles (1999) or Huang et al. (1997) for examples to studies on other countries, and Sayan and Kenc (1999) and TUSIAD (2004) for examples to studies on the Turkish social security system. Formulation of the problem The search for new pension parameters that intertemporally balance expenditures and revenues of SSK begins by defining the difference between total inflationadjusted present discounted value (PDV) of future pension payments that need to be made to retirees and future contribution receipts from active workers who pay their dues. Letting D represent this difference for the period under consideration, we can write D
(1 1 )
tt0
t
le
mwa
aAt
aa0
RR rw a,tra,t CR rwa,t wa,t ,
(10.1)
where CR : Average contribution rate – including the supplementary contributions that the employers are legally required to make (0 CR 1), RR : Average replacement rate showing the percentage of pension income to replace the average wage/salary income earned prior to retirement (0 RR 1), At : Statutory entitlement age (minimum retirement age for female workers) at time t (At mwa, where mwa is maximum working age), : Discount rate, wa,t : Projected number of contributing workers at the age of a at time t, —a,t : Average real wage income projected to be earned by active workers at the rw age of a at time t – adjusted for the earnings cap or wage ceiling that determines the maximum amount out of which contributions are collected, ra,t : Projected number of pensioners who are aged a at time t, rwa,t : Average real work time income of pensioners aged a at time t, a : Age index running from the beginning of working life, a0, to le, life expectancy in years (le mwa), t : Time index running from the initial period, t0, to , the end of model horizon, and t t t0. The values of all variables other than At, CR and RR in (10.1) are determined by demographic or labor market conditions and, hence, cannot be controlled by
265
Political economy of pension reform
pension authorities. By this notation, total pension payments to be made by the pension fund at any year t are calculated by multiplying the number of retirees, distinguished by their ages, with the applicable pension for the corresponding age group. The applicable pensions at time t, in turn, are calculated through an indexation scheme requiring each retiree to be paid a certain proportion, RR, of the average of real wages earned prior to retirement.12 Each active worker aged a at time t, on the other hand, is paid rwa,t and a fraction, CR, of this income adjusted for the wage ceiling (i.e. the maximum level of wages/salaries by which contributions and pension payments are calculated) is collected as contributions. D in (10.1) therefore shows the inflation-adjusted PDV of the difference between pension payments (projected) to be made and total contributions (projected) to be collected by the end of time horizon considered. Equation (10.1) may be taken as the objective function in a constrained minimization problem requiring to find CR and RR values and minimum retirement ages to be introduced at different points in time, At, so as to keep D as low as possible under exogenously given, projected values of real wages, retiree, and active worker populations and other relevant constraints. Given that publicly managed pension systems do not typically aim to run surpluses (with D 0),13 a more realistic and relevant problem to consider would be to identify CR, RR, and At values without allowing D to be negative over the period under consideration. Since expenditures exceed receipts when D 0, the choice of a positive target value for D implies that the pension system is allowed to run a deficit. To consider this more realistic version of the pension reform problem for Turkey, the time horizon for the runs with the GA was chosen to be the period from 2000 to 2060, and the target level of pension deficit over this period was set equal to zero. Since the losses generated by SSK (and by the social security system at large) had already reached alarming proportions prior to the reform act of 1999, the target level of pension deficit was chosen to keep the existing debt from growing. In fact, getting |D| as close to zero as possible is equivalent to the maintenance of the real deficit of SSK at its initial level. As the Turkish economy would continue to grow, this must be expected to lower the real GDP share of SSK’s pension deficit to a negligible level by the year 2060. Furthermore, since Turkish population is projected to remain stable beyond 2060 (TUSIAD, 1999), the program parameter configurations suggested to minimize deficits over 2000–2060 would not need to be modified afterwards. Given this target, the problem for the relevant time horizon could formally be defined as14 Minimize D
At, CR, RR
subject to
2060
le
mwa
1 RR rwa,t ra,t CR rwa,t wa,t t aa0 t2000 (1 ) aAa
wa,t for a0 a At, t pa,t wa,t ra,t for At a a mwa, t ra,t for mwa a le, t
0 CR CR CR 1,
0 RR RR RR1,
(10.2)
266
Serdar Sayan
where CR (RR) and CR (RR), respectively, denote the lower and upper bounds defining the range within which policy makers would like to adjust the contribution rate (replacement rate), and pa,t represents the projected year t population of the people in age group a (a a0) and covered by SSK. Hence, the equality constraint in (10.2) states that anyone who is between the minimum retirement age At and maximum working age, mwa, at time t must either be a worker or a retiree, whereas any member of this group who, at time t, is younger (older) than At (mwa) will necessarily be a worker (retiree). Given the way minimum retirement age, At, defines the limits for relevant age brackets in the equality constraint, however, any increase in At introduced as part of a pension reform would increase the future size of working population at the expense of the retiree population. In other words, even though the policy makers/pension authorities cannot control pa,t itself, they may affect its composition by increasing minimum retirement age in such a way to reduce the number of retirees to collect benefits – while simultaneously increasing the number of workers to pay contributions. Since At marks the lower limit for the summation sign in the objective function, each alternative value of At considered within the context of a parametric reform implies a different value for |D|, even when CR and RR are held constant. This is, in fact, what raises the complexity of identifying the [CR, RR, At] configurations minimizing |D| (Sayan and Kiraci, 2001b). Yet, the GA used in this chapter overcomes the associated computational difficulties quite efficiently (Sayan and Turhan-Sayan, 2001). Description of the data and implementation Since the time horizon for the computational exercises was chosen to be the 61-year period between 2000 and 2060, projections on active worker/retiree populations by ages (aggregated over genders) were used to construct a (mwaa01) 61 matrix for the working population, and a (leA01) 61 matrix for the retiree population over 2000–2060, where A0 is the initial value of minimum retirement age. Based on data on SSK membership in ILO (1996), maximum working age, mwa, was taken to be 75 and the earliest working age, a0, to be 15. Likewise, following ILO (1996), the life expectancy, le, was set equal to 80 allowing retirees to continue to collect benefits by the age of 80 or more. Since ILO (1996) projections on working and retiree populations assumed the maintenance of the actual (pre-1999) minimum retirement age, A0, was taken to be 38. These values yielded two demographic projection matrices for 2000–2060: WAT [wa,t]6161 for the active (contributing) worker population by age, and RAT [ra,t]4361 for the retiree population by age. Other data available in ILO (1996) included real wage and pension incomes of actively working and retired members of SSK by ages, all in terms of Turkish Liras (TL) at constant 1995 prices. Whenever necessary, the available series were extended into the year 2060 by taking into account the likely growth in economy wide labor productivities as well as productivity differentials due to seniority, and by using the average replacement rates given in the same source. The resulting work and pension income matrices, WPAY [rwa,t]6161 and RPAY [rwa,t]4361, were then constructed and
Political economy of pension reform
267
Deficits (in 100 tril. TLs at 1995 prices)
used together with the corresponding demographic projection matrices, WAT [wa,t]6161 and RAT [ra,t]4361, to calculate the initial (pre-reform) value of |D| by taking CR 0.20, RR 0.65 and A0 38 (i.e. average contribution and replacement rates, and minimum retirement age prior to the reform of 1999) and 0.05. Figure 10.1 generates the paths SSK’s pension balances would follow over time under three different policy scenarios by employing the projections used in solving the problem described earlier: Continuation of pre-1999 policy without any changes to pension program parameters; the actual reform of 1999 with no changes to CR and RR but transitional and permanent increases introduced to entitlement ages, and under revisions to the entitlement age increase scheme as required by the relevant ruling of the Turkish Constitutional Court (see also Table 10.4). The area between each plot and zero line at the top of the figure corresponds to total deficit over 2000–2060 as captured by objective function variable |D| in the problem described earlier. Since plots in Figure 10.1 assume away compliance problems, however, the path plotted for each policy scenario should actually be expected to lie below the corresponding plot in Figure 10.1 due to the leakages in the form of underreported payments to workers or contributions not transferred to SSK by employers. Table 10.6 reports total deficit values implied by plots in Figure 10.1, as well as by considering a 10 percent leakage rate assumed to be maintained throughout the time horizon considered so as to highlight direct costs of low compliance. The numbers in Table 10.6 indicate that the pension reform act of 1999 could have achieved considerable success in reducing the real PDV of the pension deficit likely to be incurred by SSK over the 2000–2060 period, if it had been implemented without modifications. Deficit over the whole period would have
0 –1 –2 –3 –4 –5 –6 –7 –8 –9 Years No intervention Law no. 4447
Post–2002 revision to law no. 4447
Figure 10.1 SSK’s pension balances under different policy scenarios.
268
Serdar Sayan
Table 10.6 Pension deficits of SSK under different policy scenarios: 2000–2060 (in trillions of 1995 TLs) Compliance scenarios
Policy scenarios No. reform
Full compliance Low compliance (10% leakage) Cost of low compliance
Law no. 4447 (original formulation) I
Revision of law no. 4447 (present situation) II
Cost of the revision III
38,013
7,561
11,060
3,499
39,050
9,180
12,610
3,430
1,037
1,619
1,550
69
been declined from 38,013 to 7,561 billion TLs at 1995 prices, representing an 80 percent reduction in real terms. The actual savings, however, remained more modest because of the modifications to the way entitlement age was raised due to the Constitutional Court’s decision on the appeal by the main opposition party. Furthermore, low compliance assumed to be represented by a conservative 10 percent leakage rate accounts for more than 10 percent of deficit implying considerable direct costs to SSK and hence to public deficits. Simulation experiments and results The search for optimal pension program parameters was carried out through the GA-based optimization program developed in MATLAB which updated WAT and RAT matrices continuously for each increase introduced to the entitlement ages, generating a new projection for supply of labor to be covered by SSK until 2060 as described in Sayan (2002). Since GAs are designed to search for the global optimum without requiring a good initial guess, the GA approach was particularly useful for the problem at hand. This section reports results from simulation experiments carried out by solving the problem in the previous section under alternative ranges imposed on the values that parameters of interest are allowed to take. In the case of CR and RR, a single range is defined for the entire model horizon whereas the minimum retirement age, A, was assumed to be increased at different points in time between 2000 and 2060 for each simulation experiment. This treatment is generally compatible with pension reform practices in different countries (Kohl and O’Brien, 1998) including the parametric pension reform of 1999 in Turkey. The simulation scenarios described later seek an answer to what would have happened to SSK’s pension balances if alternative parameter configurations had been introduced instead: Scenario 1 The transition period in the original formulation of law no. 4447 is maintained under this scenario but the entitlement ages for each year during the
Political economy of pension reform
269
transition period are allowed to take values within given age ranges. The upper limits of ranges are set in equal to the ages fixed by law no. 4447 so as not to allow for age increases higher than in the actual reform during the transition period. To take the upward trend of life expectancies into account, the period from the end of the original transition period (2009) to the end of model horizon (2060) is divided into two sub-periods, 2010–2023 and 2024–2060, with higher entitlement ages being allowed in the last sub-period. The year 2024 is taken as the beginning of this last period so as not to require anyone who is in the workforce and covered by SSK at the time of legislation to postpone retirement any longer than required by the actual reform act. Given that the reform act forced the youngest workers who began to work at the age of 15 in 2000 to postpone collection of pension benefits until the age of 58 (instead of 38) or by the year 2043 (instead of 2023), entitlement ages exceeding 58 are allowed to be applicable only to those who are to begin working after 2024, that is, those who are 14 or younger in the year 2000. Furthermore, this year also marks the closing of the demographic window of opportunity for Turkey (TUSIAD, 1999): Since the share of working age population is expected to start going down afterwards, the need for higher entitlement ages would be stronger in the second sub-period. Unlike the actual reform act that maintained average contribution and replacement rates, a 25 percent band around initial values of CR and RR is considered. In terms of the notation of the problem defined above, values assigned to CR and RR, and CR and RR define the ranges within which CR and RR may be adjusted.15 The relevant ranges and the resulting [CR, RR, At] configurations obtained with the GA are given in Table 10.7. The results in Table 10.7 reveal that even more impressive savings could have been achieved by using different configurations of pension program parameters. The optimal configurations in the table indicate that it would have been possible to accomplish an even higher reduction in |D| by (i) slightly increasing CR by one percentage point and decreasing RR by about 1.1 points (Column 4); or (ii) decreasing both CR and RR by about 2.3 and 3.3 percentage points, respectively (Column 5), with minor adjustments in the gradual increases in the entitlement age during the transition period, and a higher jump after 2024. Both sets of optimal configurations point to huge cuts in |D| reducing it to negligible levels over the period from 2000 to 2060. Yet, it would be impossible to define some objective criteria to choose one of the optimal configurations for implementation, since neither is strictly better than (or Pareto superior to) the other. Entitlement age increases under configuration (ii) are smaller than or equal to those under configuration (i) during all years of the transition period except 2008, but this configuration requires a higher age increase after 2024. Still, at least as far as entitlement age increases alone are concerned, one may argue that it would be easier to implement (ii) politically, as people who would be directly affected
270
Serdar Sayan
Table 10.7 Results under scenario 1 with the Genetic Algorithm Pension parameters
Average RR (%) Average CR (%) A for 2000 A for 2001 A for 2002 A for 2003 A for 2004 A for 2005 A for 2006 A for 2007 A for 2008 A for 2009–2023 A for 2024–2060 |D|a RR/CR
No. reform
65 20 38 38 38 38 38 38 38 38 38 38 38 38,013.0 3.25
Law no. 4447
65 20 38 41 43 45 47 48 49 50 51 58 58 7,561.7 3.25
Possible parameter configurations (i)
(ii)
63.92 21.00 38 41 42 45 47 48 49 50 50 58 62 nil 3.04
61.75 17.67 38 40 42 44 46 48 48 50 51 57 64 nil 3.49
Ranges allowed for variables under scenario1 48.8–81.3 15.0–25.0 38 38–41 42–43 44–45 46–47 47–48 48–49 49–50 50–51 51–58 59–64
Note a In billions of TLs at constant 1995 prices.
sooner would presumably react more strongly to age increases than people who would be affected later – including those who are yet to begin working and yet to become voters. However, the differences between configurations (i) and (ii) are not limited to age increases. Of the two configurations, (i) would be more (less) favorable to current retirees (workers) since it requires a smaller cut (rise) in RR (CR). Thus, even if the entitlement age increase schemes are assumed to be similar enough to leave them out of consideration in making a choice between the two configurations, law makers would still have to weigh potential reaction from retired voters against that from voters who are in the workforce. Certainly, the age composition of voters would be a major factor to consider in this comparison, but how well-organized the lobbying groups are would also affect the outcome. Alternatively, the ratio of RR to CR under each configuration can serve as a rule of thumb since an incumbent government considering a number of parametric reform alternatives will be likely to prefer the configuration with the highest RR/CR ratio for any given entitlement age vector over time and a comparable level of pension deficits. In the light of this discussion, configuration (ii) would probably have been a better choice for a parametric reform in Turkey politically. Given that both solutions yield 2- or 3-year increases in the entitlement ages at first and one year increases later during the transition period, however, these schemes too would have been subject to the same criticism raised by the Turkish Constitutional Court that the law’s scheme for gradual entitlement age increases
Political economy of pension reform
271
during the transition period violated the equality principle. To meet this criticism, the next scenario considers an alternative reform configuration. Scenario 2 In this scenario, a transition period allowing for equally-spaced but faster increases in the entitlement ages is considered. By the scenario, entitlement age increases were to be speeded up through 3-year age increases after every 2 years during the transition period, and would be increased by 1 year for every 10 years thereafter to account for increasing life expectancies. The ranges within which program parameters are allowed to vary are given in Table 10.8 along with the results. Once again, the results indicate that SSK’s 2000–2060 deficit could have been practically eliminated, if accelerated increases in the entitlement age could be matched with reductions in CR (1 percentage point) and RR (5 percentage points). Figure 10.2 shows how overall deficit would have been eliminated over time as improvements to SSK balances at later periods following reform compensate for losses incurred during earlier periods. Admittedly, this configuration would have been more difficult (but not impossible) to sell to the retired voters than the actual reform. Table 10.8 Results under scenario 2 with the Genetic Algorithm Pension parameters
No. reform
Law no. 4447
Revision of law no. 4447a
Alternative parameter configurations
Ranges allowed for variables under scenario 2
Average RR (%) Average CR (%) A for 2000 A for 2001–2002 A for 2003–2004 A for 2005–2006 A for 2007–2008 A for 2009–2010 A for 2011–2012 A for 2013–2014 A for 2015–2016 A for 2017–2018 A for 2019–2020 A for 2021–2024 A for 2025–2034 A for 2035–2044 A for 2045–2054 A for 2055–2060 |D|b RR/CR
65 20 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38 38,013.0 3.25
65 20 38 41–43 43–47 48–49 50–51 58 58 58 58 58 58 58 58 58 58 58 7,561.7 3.25
65 20 38 38 39–40 41–42 43–44 45–46 47–48 49–50 51–52 53–54 55–56 58 58 58 58 58 11,060.0 3.25
60 19 38 41 44 47 50 53 56 59 60 60 60 60 61 62 63 64 0.0 3.16
48.8–81.3 15.0–25.0 38 41 44 47 50 53 56 59 60 60 60 60 61 62 63 64
Notes a As required by the Constitutional Court. b In billions of TLs at constant 1995 prices.
Serdar Sayan
SSK pension balance (in 100 tril. TL at 1995 prices)
272
2 1 0 –1 –2 –3 –4 Years Present situation
Alternative reform scenario 2
Figure 10.2 SSK’s pension balances under reform scenario 2.
Conclusions This chapter describes the recent pension reform process in Turkey and discussed the nature and magnitude of financial difficulties facing the publicly managed pension system in the country by highlighting the role that politically motivated interventions to the system have played in generating pension losses underlying the reform efforts. Pre-election entitlement age cuts introduced by different governments on the pretext of creating vacant positions for the younger generation seeking employment were identified as the primary culprit behind the pension crisis that Turkey experienced prematurely. Another factor cited as contributing to the crisis was arbitrary pre-election bonuses paid to retirees by almost every government despite the absence of matching contributions, adding to the generosity of retirement benefits that are linked only weakly to contributions made by civil servants covered by ES, the pension fund administration of white-collar workers employed by local and central governments. Coupled with the misuses of surplus funds once generated by the pension system for political rent seeking, these practices were argued to facilitate the growth in pension deficits which eventually reached alarming proportions during the 1990s. Finally, inability and political unwillingness of governments to deal with the problem of low compliance with pension laws and regulations, particularly the lack of action to reduce the substantial population of unregistered workers were noted as a significant contributor not only to the growth in pension deficits but also to serious economic and social problems in other areas. The chapter also investigated parametric alternatives that can be employed to control deficits that are likely to be incurred in the decades ahead by SSK, the
Political economy of pension reform
273
agency that run the largest pension fund in Turkey. The issue of identification of parametric policy options before SSK was tackled by solving an actuarial model with the help of a GA developed for this purpose. The resulting parameter configurations were shown to have the potential to eliminate pension deficits of SSK and political feasibility of implementing them was discussed. The contribution rate-replacement rate pair found by the GA under second reform scenario involving no change in the entitlement age increase scheme currently in effect pointed to a five points reduction in the replacement rate and a one-point reduction in the contribution rate. This is largely compatible with the draft reform bill that is currently under discussion for legislation to go into effect in 2006. The replacement rate of 60 percent found through the genetic algorithm is exactly the same as the rate the draft bill proposes to implement starting from 2016, following an immediate 2.5 point reduction from 65 to 62.5 percent. Different from the optimal solution found here, however, the bill foresees no reduction in the contribution rate, suggesting instead to raise entitlement ages after 2036, and raising doubts about the effectiveness of upcoming pension legislation to overcome compliance problems of the past. Still, the new bill reflects (an IMF-supported) determination to put an end to privileged status of civil servants by harmonizing conditions for coverage across social security agencies. Whether the current government will be able to achieve this politically risky goal remains to be seen.
Acknowledgment The author appreciates useful discussions with colleagues at the Research Department of the IMF where he was a visiting scholar during the time first draft of this chapter has been written, and expresses his gratitude to Prof. Gonul Turhan-Sayan of Middle East Technical University who wrote the genetic algorithm code used here.
Notes 1 A PAYG-based pension fund starts out with many contributors and very few, if any, retirees collecting benefits. That’s why benefits are typically very generous relative to contributions at the initial stages following establishment of the fund. As time passes, the ratio of contributing workers to retirees begins to fall gradually as covered workers start retiring. The decline in this critical ratio continues at an accelerating pace as populations age due to declining fertility and increasing life expectancies in the course of demographic transition of nations. 2 The CR determines the amounts to be collected from earnings of workers as old-age insurance premiums, whereas the RR determines the share of average work time earnings to be replaced by pension income after retirement. The entitlement age (EA) sets the minimum age when individuals may begin collecting pensions and other retirement benefits. 3 In addition to immediate effects on the living standards of working and retired members of the society, the parameter configuration to be introduced will also have implications for intergenerational distribution of the reform’s burden on presently living and future generations (see, for example, Boll et al., 1994). 4 For more detailed information about the structure of Turkish social security system, see Topal (1999) or Akmaz (1999).
274
Serdar Sayan
5 See table 44 at http://www.ssk.gov.tr/wps/sskroot/istatistik/2000/t43–60.htm, accessed on October 25, 2004. 6 Low entitlement ages shortened average contribution period and hence lowered contributions per worker, while at the same time extending the average retirement period and hence raising total benefits paid per retiree. 7 In many cases, young employees would retire by a certain weekend and come back to their workplace and resume their responsibilities the next Monday without violating any laws or regulations. 8 As of 1990, the average age of Turkish population was 26.4 whereas the median age was 22.2 (http://www.frekans.com.tr/html/4tr_istatistikler.asp, accessed on October 25, 2004). Despite its tendency to age gradually, the UN data show that Turkish population is still young with 29 percent of the population between 0 and 14 years of age, 64 percent between 15 and 64 and only 7 percent older than 64 as of 2000. 9 So, the government opted to cut the number of pensioners rather than average pensions which might have led to increased poverty among retirees (Barr, 2002). 10 For more detailed information on the structure of the market for private pension plans and the underlying legal and regulatory framework, see Elveren (2003) that is available at http://www.bireyselemeklilik.gov.tr/seminer_konferans/SK_PerYon.ppt, accessed on April 2, 2005. 11 GAs are adaptive parallel search techniques that are of probabilistic nature. The reader may consult to a wide selection of books and papers for more information on GAs. A formal description of genetic methods is given in Judd (1998). For implementation details, Goldberg (1989) and Banzhaf et al. (1998) are particularly good references. n rwai,t(ara)i where ra is the actual retirement age 12 To be more precise, rwa,t (1/n)i1 (ra A) and n is the number of years in the applicable averaging period. 13 Because many governments view pension systems as a channel to make income transfers to working and/or retiree populations, they would often be willing, in fact, to allow public pension systems to run “reasonable” deficits – with the definition of reasonable varying across governments and macroeconomic conditions. 14 With the absolute value operator, the global minimum for |D| would be zero. Hence, the formulation of the objective function rules out the possibility that SSK would run a significant surplus while providing pension coverage over the 2000–2060 period. 15 Admittedly, the selected ranges for CR and RR in the simulation experiment here are somewhat arbitrary and may actually be outside the ranges that policy makers deem politically feasible or desirable. But our GA-based approach is flexible enough to allow for a rather quick computation of parametric reform alternatives under alternative ranges.
References Akmaz, M. (1999). Systemic Analysis of and Policy Generation for the SSK Pension Scheme in Turkey, Unpublished Thesis, Ankara: Middle East Technical University. Ayas, E. (1998). “Türkiye’de sosyal güvenlik sisteminin sorunları ve çözüm önerileri,” Isletme ve Finans 13(150): 41–50. Banzhaf, W., P. Nordin, R.E. Keller, and F.D. Francone (1998). Genetic Programming: An Introduction, San Fransisco, CA: Morgan Kaufman Publishers. Barr, N. (2002). The Pension Puzzle: Prerequisites and Policy Choices in Pension Design, Washington, DC: IMF. Boll, S., B. Raffelhüschen, and J. Walliser (1994). “Social security and intergenerational redistribution: a generational accounting perspective,” Public Choice 81: 79–100. Chand, S.K. and A. Jaeger (1996). “Aging Populations and Public Pension Schemes” (Occasional Paper 147), Washington, DC: IMF.
Political economy of pension reform
275
Elveren, A.H. (2003). “Bireysel emeklilik sistemi: Ivleyivi ve gelivmeler,” paper presented at the 11th Congress of the Turkish Association of Human Resource Management, October 3–4, Istanbul. Gillion, C. and M. Cichon (1996). “Pension reform in Turkey,” Hazine Dergisi (Special Issue on Social Security and Health Insurance Reform): 40–56. Goldberg, D.E. (1989). Genetic Algorithms in Search, Optimization and Machine Learning, Reading: Addison Wesley. Huang, H., S. Imrohoroglu, and T. Sargent (1997). “Two computations to fund social security,” Macroeconomic Dynamics 1: 7–44. ILO (1996). Social Security and Health Insurance Reform Project: Final Report, Geneva: ILO. Ivıklı, A. (1998). “Türkiye’de ivgücü piyasası ve sosyal güvenlik,” in A. Vahinöz (ed.), Türkiye Ekonomisi Sektörel Analiz, Ankara: Turkiye Ekonomi Kurumu. Joines, D.H., A. Imrohoroxlu, and S. Imrohoroxlu (1999). “Computing models of social security,” in R. Marimon and A. Scott (eds), Computational Methods for the Study of Dynamic Economies, London: Oxford University Press. Judd, K.L. (1998). Numerical Methods in Economics, Cambridge: MIT Press. Kenar, N., T. Teksöz, and A.H. Çovkun (1996). “Sosyal guvenlik sistemimiz ve reform,” Hazine Dergisi (Special Issue on Social Security and Health Insurance Reform): 5–25. Kenc, T. and S. Sayan (2001). “Demographic shock transmission from large to small countries: an overlapping generations CGE analysis,” Journal of Policy Modeling 23(6): 677–702. Kılıçdaroxlu, K. (1998). “Türkiye’de Sosyal Güvenlik Sisteminin Sorunlari ve Çözüm Önerileri,” Ivletme ve Finans 13(150): 35–40. Kohl, R. and P. O’Brien (1998). “The Macroeconomics of Ageing, Pensions and Savings: A Survey” (Economics Department Working Paper 200), Paris: OECD. Miles, D. (1999). “Modelling the impact of demographic change upon the economy,” Economic Journal 109: 1–36. Sayan, S. (1999). “Türkiye’deki yavlılık (emeklilik) sigorta prim kaçaklarının toplumsal ve bireysel maliyetleri,” ASO Medya (Eylul): 40–54. Sayan, S. (2002). “Turning Potential Retirees into Workers: Pension Reform Act of 1999 and Beyond in Turkey,” Middle East Business and Economic Review, 14(2): 54–70. Sayan, S. and T. Kenc (1999). “Long-term consequences of rehabilitating a financially troubled pension system: an overlapping generations, general equilibrium analysis for Turkey” (Economic Research Forum Working Paper No. 9914), Cairo: ERF. Sayan, S. and A. Kiraci (2001a). “Parametric pension reform with higher retirement ages: A computational investigation of alternatives for a Pay-As-You-Go based pension system,” Journal of Economic Dynamics and Control 25(6–7): 951–966. Sayan, S. and A. Kiraci (2001b). “Identification of parametric policy options for rehabilitating a Pay-As-You-Go based pension system: an optimisation analysis for Turkey,” Applied Economics Letters 8(2): 89–93. Sayan, S. and G. Turhan-Sayan (2001). “In search of parametric reform alternatives for a pay-as-you-go pension system: a genetic algorithm approach,” Unpublished Manuscript. Teksöz, T. and S. Sayan (2002). “Simulation of benefits and risks after the planned privatisation of pension system in Turkey: is the expected boost to financial markets feasible?,” Emerging Markets Finance and Trade 38(5): 23–45. Topal, N. (1999). “Is the Turkish social security system in need of reform?” (ESDS Employment Studies Working Paper No. 22), Hertfordshire: University of Hertfordshire. TUSIAD (1999). Türkiye’nin Fırsat Penceresi: Demografik Dönüvüm ve Izdüvümleri, Istanbul: TUSIAD. TUSIAD (2004). Türk Emeklilik Sisteminde Reform: Mevcut Durum ve Alternatif Stratejiler, Istanbul: TUSIAD.
11 The Turkish banking sector A rough ride from crisis to maturation Ali Tükel, Murat Üçer, and Caroline Van Rijckeghem Introduction The Turkish Banking Sector (TBS) has gone through a remarkable process of crisis, resolution, and recovery in the past few years. Having left much of its excesses behind, the sector now stands at a delicate juncture, and is faced with many challenges. The purpose of this chapter is to document the salient features of this process, and provide a flavor of the challenges ahead. The chapter starts with a brief description of the poor state of the banking sector in the lead up to the November 2000–February 2001 financial crises and explains how anxiety over the state of the banking sector helped trigger the crisis. The section on The banking sector rehabilitation program describes the effect of the crisis on the banking sector, the major consolidation that happened in the wake of the crisis, the various banking sector operations undertaken to re-capitalize the banking system, and the regulatory changes introduced since the disinflation program of late 1999. The past few years saw a significant recovery in profits, albeit largely attributable to trading profits, as the section on Recovery from crisis documents. The chapter then proceeds in the section on Looking forward to the challenges, to explore the growth prospects for the TBS. This section also discusses the recent entry of a few major international banks into the TBS, in the cleaner banking sector environment. The discussion is based, in part, on the results of a questionnaire conducted by Steinherr et al. (2004) with a number of Turkish bankers. Longstanding impediments to growth of banking in Turkey, notably under-capitalization of the sector and chronic macroeconomic instability, have lost in importance. Another major impediment, the high taxation of the sector, is beginning to be addressed through plans to phase out financial transaction taxes and the tax on consumer credits beginning from 2006. The absence of a clear strategy on state banks, which still make up about one-third of the sector’s assets, appears to be a key problem, not only for efficiency reasons, but also because such strong presence unavoidably hinders a “level playing field” for the sector’s private banks. But the greatest challenge ahead for the TBS is arguably achieving further growth, in a way that does not undermine financial stability. In this respect, given the low degree of financial intermediation in Turkey, there appears to be growth
The Turkish banking sector 277 potential, notably in the small and medium sized enterprises (SME) and consumer segments of the market, though the large degree of informal sector activity impedes analysis of credit worthiness and exacerbates asymmetric information problems, while legal impediments to liquidating collateral inhibit lending to SMEs as well. There is not much scope for profitable lending to the large enterprise sector since, as is typical for emerging markets, this sector already obtains ample finance from foreign banks and from capital markets abroad. Sustaining a single digit inflation-cum-high growth environment, through structural reforms, appears to be the only way out of this conundrum.
Turkish banks in the lead-up to the 2000–2001 financial crisis The December 1999 IMF-supported disinflation program sought to address Turkey’s chronic macroeconomic instability, including a growing public debt problem, at a time when the situation had become largely unsustainable. The program aimed at reducing inflation using an exchange-rate anchor. But it sought to balance the risks associated with exchange-rate based inflation stabilization programs by incorporating an “exit strategy” from the peg (which entailed a widening of the band 18 months into the program) and supporting the program by an ambitious package of structural reforms and a sizeable fiscal adjustment. As for the banking sector, the program strategy was to reform the sector gradually, by taking over some “lemon” banks at the outset, to be followed by a gradual improvement in banking supervision and regulation. At the heart of this transformation was a new Banking Act, which called for the establishment of an independent Banking Regulation and Supervision Agency (BRSA) to take over the supervisory and regulatory functions from the Treasury. In the late 1990s, any observer of the TBS would have agreed that the sector was marred by several deep-seated problems: a politicized regulatory structure and weak enforcement, which had culminated in pervasive connected-lending practices and risky balance sheets, and a low degree of financial intermediation of private activities, of only 20 percent as measured by private bank loans-toGNP ratio (Figure 11.1).1 The public sector’s financing need, or the savinginvestment imbalance, was at the heart of many of these problems, as it was associated with crowding out of private sector borrowing, macroeconomic instability and regulatory forbearance (Figure 11.2).2 The TBS had been unprofitable for many years, chiefly on account of connected lending to unprofitable projects (including pure embezzlement) and fraud in a number of banks, reflecting a lax and politicized supervision process, but also because of high taxation of domestic deposits. Troubled banks could continue to attract depositors, because of deposit insurance, imposing unfair competition on healthier banks. Banking licenses were given very easily and decision-making was politicized. Balance sheets were also very risky because of maturity mismatches, the very short-term nature of borrowing (notably through the so-called “repos”
100
31 Assets (left scale) Loans (right scale) 29
90
27 80 25 70 23 60 21 50 19 40
17
30
15 1990
1992
1994
1996
1998
2000
2002
2004
Figure 11.1 Banking sector total assets and credit (% of GNP).
20 Public 15
Private S-I
10 5 0 –5 –10 –15 –20
1990
1992
1994
1996
1998
2000
Figure 11.2 Saving-investment (non-financial) balances (% of GNP).
2002
2004
The Turkish banking sector 279 (repurchase operations) with customers and banks), and large open foreign exchange positions. The latter were the result of the lucrative carry trade, whereby banks funded government debt by borrowing and taking deposits denominated in foreign exchange. In the process, banks circumvented foreign exchange regulations, which postulated a limit for open positions of 20 percent of equity. Supervisors closed their eyes to such transgressions in a live-and-let live game intended to lower interest rates on government debt, since borrowing from abroad or in foreign exchange was available at relatively low interest rates. These weaknesses – lack of profitability and risky balance sheets – were accentuated during the IMF program, partly because of now well-understood adverse incentives provided by an exchange rate anchor.3 On the one hand, the stabilization program was expected to reduce profitability of banking, at least in the shortrun. The projected drop in inflation would lead to a reduction of inflation tax revenues on the “float,” that is on demand deposits not earning (or earning very little) interest. Real interest rates would be falling and the shift to more external borrowing planned under the IMF program would reduce the profitability of lending to the government.4 Banking supervision was being strengthened, notably through the introduction of the earlier-mentioned supervisory agency, BRSA, with operational independence; and through strict regulations on consolidated reporting (which would make it more difficult to hide open positions abroad), market risks, related party lending, as well as open positions in excess of legal limits. While most banks had large windfall gains on their security portfolios when interest rates dropped at the beginning of the IMF program, the IMF had captured part of these gains by taxing windfalls gains, partly to make stick the program’s ambitious primary surplus target. Given all this, survival, in the view of many, was predicated on the ability to move to real banking (consumer and business lending) from financing the government and Turkey’s “over-banked” environment spelt consolidation. The following quotes sum up the prevailing sentiment:5 In the next 18 months, the Turkish banking sector will be undergoing painful changes in its structure, in the strategies of banks and their performances. [. . .] Consolidation within the Turkish banking sector is inevitable in the long-run, with smaller banks finding it more difficult to adapt to the new environment.6 The expected transition to an era of “real banking” – including tougher regulation, lower inflation and shrinking interest margins – has already pushed 10 out of 40 Turkish commercial banks into state administration.7 This is the Darwinian age of Turkish Banking. [. . .] The big four are fine. They have enough capital, people, technology. [. . .] But for the rest: who knows?8 On the other hand, given the initial success of the IMF program, banks gained increased access to syndicated loans and financial structures with foreign investment banks, leading to an increase in their open foreign exchange positions. This was paired with both a reduction in domestic interest rates and a lending boom,
280
Ali Tükel et al.
as consumer and business loans rose by over 50 percent from about $18 billion at end-1999 to almost $30 billion at end-2000. Consumer loans themselves more than doubled, from some $4 billion to almost $10 billion.9 The situation came to a head in the fall of 2000. The Banks’ Association released figures for on-balance sheet open positions in the second quarter of 2000, which were shockingly high, at $19 billion, well in excess of the 20 percent of capital legally permitted.10 This was a “gross” figure relating only to on-balance sheet items, and thus excluded any forward contracts which served as a hedge of the open positions as well as structures with London investment banks which masked the true size of open positions. On balance, given that many of the forward contracts were fake contracts (so-called “drawer contracts”) with the sole purpose of window-dressing,11 open positions were probably even larger. There were concerns at the time that these open positions would be closed in a disorderly way (possibly following intensified scrutiny by supervisors), pressuring interest rates and foreign exchange reserves/the exchange rate. Some investment banks started to issue warnings to clients suggesting “they exit Turkey.” At about the same time, concerns developed over the large number of weak banks that remained in the system and over the cost to the Treasury of handling these remaining lemons. One of the first actions of Zekeriya Temizel, the head of the new BRSA, whose surname literally meant “Clean Hands,” was to jail a number of allegedly corrupt bankers, as part of “Operation Hurricane.” Triggering a “Will I be next?” syndrome among the sectors’ key players, this raised question marks regarding the extent of problems in the banking sector and the budgetary cost of potential bank takeovers. The magazine Euromoney reported that between 6 and 10 other banks could soon be taken over. Late Sakip Sabanci, head of one of Turkey’s most-respected and largest conglomerates would go as far as to openly declare, in his characteristically candid style, “If you take your binoculars out, you will see another 20 . . . There is no country in the world as corrupt as Turkey; in this, we must be champions.”12 Subsequent to their take-over by the Savings Deposit and Insurance Fund (SDIF), the true extent of non-performing loans on the balance sheets of banks taken over by the SDIF the previous year had also been revealed.13 In the most glaring example, the ratio of non-performing to performing loans of Yurtbank (which was taken over at end-November 1999 as a prior action under the IMF program) increased from 6 percent of the loan-portfolio in 1998 to almost 20 times that ratio at end-1999, and then to 75 times that ratio at end-2000 (Özatay and Sak, 2002). This made observers realize that official statistics on non-performing loans only reflected the tip of the iceberg, and added to concerns about the ultimate cost of bank-takeovers to the budget.14 Interest rates were already on a rising course by this time, reflecting lack of confidence in the program – the third review of the IMF program had not been concluded for instance.15 This led to yet more stress in the banking sector, given the maturity and liquidity risk in the system. Notably, there were worries that a medium-sized bank, Demirbank, which was financing its large portfolio of government securities largely in the money markets would eventually run in to
The Turkish banking sector 281 liquidity and/or solvency problems, which could spread through the banking sector in the form of bank runs, or sales of T-bills by Demirbank which would pressure interest rates. Though there is as yet no consensus, in the view of many the loss of credibility of the banking sector was the key factor causing the November crisis. A spokesman for those who were among the first to exit Turkey referred to “lack of privatization, cost of bank bailouts to the banking sector, and the banking rumors.” Following an initial foreign exchange outflow of $1.5 billion on the so-called “Black Wednesday” (November 22), further outflows totaling $5.5 billion occurred as financial structures were unwound and investors fled on fears of devaluation of the exchange rate. The November crisis weakened the large segment of the banking sector that funded itself in money markets given the rise to astronomic levels of money market rates (simple overnight rates reached some 1,250 percent, or 23 106 compounded annual rate, as the Central Bank defended the exchange rate). Together with the erosion of the credibility of the currency peg, this led to a preference for liquidity on the part of these banks and a reduced appetite for Turkish T-bills in the months following the November crisis, which in turn was a factor in the February devaluation.16 The February floatation of the exchange rate (a depreciation from $1 670,000 to 1,300,000) then dealt the “coup de grace” to many banks, given their large open foreign exchange positions. Astronomical interest rates in the days before the floatation (4,500 percent simple annual rate) also dealt a serious blow to those banks funding themselves in the overnight market. The crisis led to a forced consolidation and a massive injection of public funds into the TBS. The cost of restructuring amounted to over $50 billion or some 36 percent of 2001 GNP, most of which was borne by the Treasury and some of which was borne by the private sector (Table 11.1). As the table shows, this figure includes the restructuring cost to the Treasury of state banks of 15 percent of GNP. The restructuring cost on account of state banks originated from two Table 11.1 The cost of the banking sector crisis
Cost to the Treasury Restructuring of state banks “Duty losses” Recapitalisation For private banks transferred to the SDIF Cost to the private sector Cost borne by the SDIF Capital injection by shareholders Total
US$ (billion)
% of GDP
43.7 21.9 19 2.9 21.8
29.5 14.8 12.8 2 14.7
9.5 6.7 2.8 53.2
6.4 4.5 1.9 35.9
Source: BRSA, Progress Report (various issues); modified to include $6 billion of Imar Bank-related losses.
282
Ali Tükel et al.
sides: (1) compensation for so-called “duty-losses” or the cost of providing subsidized credit to agriculture by state banks on behalf of the Treasury, for which the Treasury had not compensated state banks for several years. These duty losses ballooned during the crisis because state banks financed themselves at exorbitantly high overnight interest rates; and (2) recapitalization, intended to fully provision for non-performing loans of state banks, which originated in politically motivated lending to companies or groups with low creditworthiness. The restructuring cost borne by the Treasury on account of private banks taken over by the SDIF was also 15 percent of GNP. On the other hand, the restructuring costs borne by the private sector, including a sum met from the SDIF’s own resources, was a relatively small chunk of the total restructuring costs (about 6.4 percent of GNP).17 The restructuring costs stemming from non-performing loans involving connected lending in the banks taken over by the SDIF (i.e. lending by banks to companies controlled by the majority shareholders of the banks) were $11 billion.18 Asset collections from former owners, or more generally, asset resolutions have gone fairly slowly. The SDIF has signed long-term repayment agreements with controlling shareholders of a number of banks for a total of about $3.5 billion19 and to improve asset collections the SDIF was separated from BRSA. However, no serious attempt has been made to auction off bad loans. These figures make the Turkish banking crisis one of the most costly in recent history. The good news is that this painful process of collapse and recovery has led to two major improvements: the sector has now become much leaner, and the regulatory structure has become more aligned with international standards.20
The banking sector rehabilitation program The May 2001 rehabilitation program, formulated and executed by the BRSA in consultation with the IMF/World Bank, took a four-pronged approach to banking rehabilitation: it sought to restructure state banks, resolve banks taken over by the SDIF, strengthen the financial structure of private banks and further improve the regulatory and supervisory framework.21 State bank restructuring was completed in the early phases of the program and the modification of the regulatory structure advanced broadly on schedule, but progress as regards the strengthening of private banks and resolution of SDIF banks was relatively slow. The BRSA opted for restructuring and the sale of banks taken over by the SDIF rather than outright liquidation and paying off depositors (other than Imar Bank) because of multiple concerns (e.g. to avoid bank runs, to minimize interest and exchange-rate volatility, and to prevent difficulties in debt management) (BRSA, 2003a). The downside was that this, in turn, contributed to a substantial slowing down of the process.22 As for private banks not taken over by the SDIF, and which suffered from high non-performing loans (NPLs) because of the recession after the crisis, the BRSA’s approach was completely passive in the first year of the program, leaving the
The Turkish banking sector 283 recovery in bank assets largely to the resumption of economic growth and capital strengthening to injections by shareholders or voluntary mergers. In late 2001, a prolonged recession and continued decline in credit forced the BRSA to become more proactive by officially undertaking a recap operation in June 2002. Around the same time, the BRSA formulated and launched the legal framework to enable the smooth functioning of two NPL workout mechanisms – the so-called “Istanbul Approach” and the establishment of the Asset Management Corporation (AMC).23 Even then, however, there has been limited success: the recap operation provided an opportunity for an in-depth penetration into private bank balance sheets and greatly enhanced transparency, but the actual capital injections needed, somewhat surprisingly, turned out to be much smaller than envisaged. As for the NPL rescue mechanisms, the Istanbul Approach continued to operate very slowly, while the AMC never took off.24 Yet, in the aftermath of a crisis-ridden period, it is perhaps fair to argue that Turkey has largely paid its dues and that the restructuring program was broadly successful in cleaning up the banking environment from its previous excesses and weaknesses. Some 21 private banks have been taken over (19 of these between the onset of the disinflation program in late 1999 and mid-2003) and the number of domestic private commercial banks in the system has declined to 18. Moreover, the capital base of the remaining private banks had largely been replenished, and the state banks had been operationally restructured under new management and ceased to be a major source of “liquidity risk” for the markets. Indications are that the remaining “contingent liabilities” in the sector are unlikely to exceed 5 percent of GNP – a manageable level compared to the overall cost of the crisis. The share of non-performing loans in the loan portfolio has declined substantially, and provisioning against non-performing loans is some 90 percent for the TBS as a whole and 80 percent for the private segment (Table 11.2). As for the capital adequacy of the sector, there is clearly a significant improvement since the May 2001 rehabilitation program and the sector appears to be very well capitalized (Table 11.3). Taking private banks, for instance, the capital adequacy Table 11.2 NPL ratio and provisions in the Turkish banking sector NPL Ratioa (%)
State banks Private banks Foreign banks Investment banks SDIF banks Total
Provisions/NPL (%)
2001
2002
2003
2004
2001
2002
2003
2004
37.3 27.6 5.5 10.7 67.3 29.3
37.4 8.9 4.9 4.0 69.4 17.6
26.2 6.5 4.3 3.5 53.8 11.5
10.3 4.8 3.2 3.3 91.6 6.0
62.7 31.0 76.0 82.2 89.1 47.1
73.9 53.0 77.6 64.0 60.5 64.2
97.6 80.0 78.5 90.3 75.4 88.5
95.6 84.5 74.8 87.2 87.3 88.7
Source: BRSA and author’s calculations. Note a Pass-due loans (gross) as percentage of total loans including pass-due loan.
284
Ali Tükel et al.
Table 11.3 Selected items from the balance sheet of the banking sector
Total assets (in quad TL) (12-month real growth; %) (% of total) Public banks Private banks Foreign banks Inv. and dev. banks Loans (% of assets) (% of total) Public banks Private banks Nonperforming loans (as % of gross loans) Public banks Private banks Securities bortfolio (% of total assets) (% of total) SDIF Public banks Private banks Deposits (% of assets) (% of total) SDIF Public banks Private banks Net earnings (as % of total assets) (% of total) SDIF Public banks Private banks Open F/X positions (gross; $ billion) o/w: private banks Capital adequacy ratio (%) Total Public Private Foreign
2004
2003
2002
306.5 12.3
249.7 0.8
212.7 2.9
34.9 57.4 3.4 3.7 32.4
33.3 57.0 2.8 4.1 26.5
31.9 56.2 3.1 4.4 23.0
21.1 68.9 0.7 0.5 0.8 40.4
18.3 69.1 1.5 0.8 1.4 42.8
15.5 69.4 7.6 15.6 4.6 40.5
1.2 50.0 45.6 62.3
4.6 44.7 48.2 62.2
5.4 45.6 46.2 64.9
0.0 42.4 55.1 2.1
2.7 38.5 56.8 2.3
4.2 35.1 58.4 1.4
3.9 42.6 44.7 0.3 0.1
4.8 31.5 51.4 0.1 0.5
38.9 36.4 83.7 0.6 0.5
28.8 41.4 22.3 26.9
30.9 56.3 23.5 36.2
25.1 48.3 19.6 32.8
Source: BRSA and authors’ calculations.
ratio (CAR) has improved further from 19.6 percent in 2002 to some 22–23 percent more recently, way above the legal requirement of 8 percent. The regulatory framework also has gone through several amendments, six to be precise since mid-1999, broadly coming to par with EU standards.25 The “blanket deposit guarantee” is now terminated, with the coverage having been reduced to EU-comparable levels as of July 2004.26 A new law is in the works aiming to strengthen banking supervision and tightening ownership conditions.27 However, enforcement of regulations remains inherently difficult. As already mentioned,
The Turkish banking sector 285 banks successfully hid open positions from regulators in the run-up to the November crisis by engaging in financial structures with London counterparts and through “drawer future contracts,” whose true nature can only be revealed upon expiration of the contract.28 The recent “Imar Bank episode” dealt a serious blow to the credibility of the regulatory framework in general, and to that of the BRSA in particular. In some sense, this was a case of large-scale “fraud” with double-bookkeeping, examples of which are also to be found in several advanced countries. Yet a report of inquiry conducted by independent experts argued that the episode revealed several weaknesses of the existing supervisory structure.29
Recovery from crisis Return on assets (ROA) and equity (ROE) have been reasonably strong since 2002, helping the sector to recoup some of the losses from the crisis, combined with steady improvement in the loan-to-deposit ratio (Table 11.4).30 The main driver of profitability in 2002–2003, however, was “trading activities” that mainly comprise capital gains associated with the rally in real interest rates since the end of the Iraqi war (Figure 11.3). For instance, in 2003, out of some TRY 8 billion of net income before tax, some TRY 6.7 billion, or some 80–90 percent of total net income (Figure 11.4), comprised net trading gains. In 2004, with lower interest rates, trading gains (net) declined sharply, to TRY 2.2 billion out of TRY 9 billion net income, or to some 30 percent. Even though banks managed to maintain a relatively stable net income profile through 2004, profitability was arguably driven by one-off factors in 2004 as well, such as lower provisioning and reduction in interest expenses.
Table 11.4 Performance of ISE-listed banks, 2002–2003 ROA (%)
ROE (%)
NPL ratio (net, %)
Free capital (quad. TL)
Loans/deposits (%)
2004 2003 2004 2003 2004 2003 2004 2003 2004 Akbank 2.9 4.5 Wv Bankası 1.6 1.4 Garanti Bankası 1.7 1.3 YKB 0.2 0.8 Dıv Bank 1.5 3.3 Finansbank 6.6 7.8 TEB 0.4 0.7 Vekerbank 2.6 2.3 Tekstilbank 0.3 0.5 TSKB 4.0 2.9 Denizbank 1.8 2.0 Alternatifbank 0.4 1.1
16.4 8.3 14.2 1.3 10.8 54.1 3.2 27.8 3.1 23.9 14.4 3.9
26.3 7.5 12.3 4.6 19.6 59.5 6.5 33.4 5.4 16.5 18.4 11.3
0.0 0.0 1.8 1.9 1.5 0.0 0.6 0.0 0.3 0.0 0.5 2.8
Source: Istanbul Stock Exchange, author’s calculations.
0.0 0.0 2.4 2.5 1.2 0.0 0.7 6.8 0.2 0.0 1.1 5.2
5.1 4.6 65.0 0.9 0.3 51.2 0.1 0.7 59.6 0.2 0.6 70.4 0.5 0.6 88.5 0.5 0.3 101.9 0.2 0.2 70.1 0.2 0.0 56.7 0.0 0.0 95.5 0.3 0.2 — 0.5 0.2 63.2 0.1 0.1 93.4
2003 45.6 43.9 47.7 59.5 74.1 72.4 59.1 36.0 68.6 — 49.1 54.3
35
30
25
20
15
10
5 *(1 + nominal secondary market rate)/(1+12-month ahead inflationary expectation)
0 Jan-02 May-02 Sep-02
Jan-03 May-03 Sep-03
Jan-04 May-04 Sep-04
Jan-05
Figure 11.3 Real interest rate (%).
120 *4-quarter rolling 100
80
60
40
20
0
–20 2000Q4 2001Q2 2001Q4 2002Q2 2002Q4 2003Q2 2003Q4 2004Q2 2004Q4
Figure 11.4 Net trading gains (% of total income).
The Turkish banking sector 287 Perhaps another not so favourable piece of news concerns “free capital,” which would appear to be very important in Turkey’s volatile financial environment.31 As shown in Table 11.4, ISE banks keep their free capital at levels that are far from providing flexibility against risk. In the bigger scheme of things, these financials carry some good and bad news, which arguably also makes up the key challenge for the TBS, going forward. On the one hand, the sector has undergone an impressive recovery process. But on the other hand, its profitability is already under pressure – a problem paradoxically likely to get exacerbated as macro stability gets rooted, reducing opportunities for trading profits.
Looking forward to the challenges32 Despite a much-improved regulatory framework and a leaner environment, the TBS is faced with many challenges, and arguably chief among these is achieving high growth and profitability in the sector in a way that does not undermine financial stability. But before we explore growth issues, we begin with an overview of the most typical problems that the TBS has been faced with until recently, namely macroeconomic instability and capital inadequacy, heavy taxation of the sector, and the absence of a level playing field vis-à-vis state banks.33 We then explore the potential market for Turkish banks and discuss the latest state of play as regards foreign bank entry. For the sake of completeness, we explore efficiency issues as well, in an appendix (see Appendix I).34 Macroeconomic instability and capital inadequacy A track record for macroeconomic stability is slowly being established with inflation forecast at 8 percent in 2005, and real interest rates around the 10 percent mark, compared with historical averages of around 70 and 25 percent respectively (Figure 11.5). As already noted, non-performing loans have been much reduced and provisioned for, while capital ratios are very high. High capital ratios of course reflect the dominance of government securities in banks’ balance sheets (which receive a risk weight of 0 percent), which would be lowered by a change in the composition of bank assets in favour of loans to the private sector (with a 100 percent risk weight).35 For example, if private banks increased their loan share from 33 to 50 percent of total assets (offset by a reduction in government securities) the CAR would decline from 23 to 15 percent.36 This is a substantial drop, but nevertheless still in line with the CARs of leading international banks. If, in addition (as is to be hoped), Turkey embarks on a path of strong growth, then bank lending will not only produce a reallocation of a constant balance sheet, but generate growing bank assets to be backed by more equity. The higher the growth, the more bank capital will be needed. Then again, in such an environment of improved systemic confidence, greater macro-stability and more robust growth, it should not be difficult to find the required equity.
288
Ali Tükel et al.
130 120 110 100 90 80 70 60 50 40 30 20 10 0 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
Figure 11.5 Inflation: the long view (end-period; %).
A key issue in this regard is how the TBS will be affected by the Basle II standards. The concern is that Basle II, which attributes higher capital coverage for risky assets, including government debt, will present a difficult challenge for Turkish banks. Under Basle II, Turkish debt which is rated below investment grade will have a risk-weight of 100 percent and hence a capital requirement which applies in full force. Though, in principle, the BRSA committed to adopt Basle II as of 2007, this rule does not automatically apply to the banks incorporated in Turkey since under Basle II it is at national discretion to set the riskweights for national debt in national currency held by banks incorporated in the country.37 This implies that the Turkish authorities can maintain, if they choose so, these risk weights at 0 percent implying no change in capital requirements. The only additional requirement would be capital backing for operational risk.38 Taxation of the sector The heavy taxation of the financial sector has received much attention in the past few years. Reserve requirements (which are unremunerated) and other taxes on financial intermediation are much above those generally observed in the EU (ECB, 2003). In a comprehensive and carefully prepared report, the BRSA shows that for 2002, various taxes and fees on financial intermediation (including reserve requirements) raised the cost of TL-credit by over 50 percent and the cost of
The Turkish banking sector 289 foreign exchange denominated loans by about as much. The study finds that some 85 percent of these costs are passed on to the consumer (BRSA, 2003b). Heavy taxation, like all excessive taxation, causes two problems. It creates a form of “financial repression” by causing high intermediation margins, impeding the growth of the sector.39 Also, it causes the migration of financial intermediation abroad especially given Turkey’s liberalized capital account. Evidence of the latter is two-fold – the high level of non-financial short and medium term private sector external debt ($45 billion at the end of 2004); and the share of credits extended through foreign branches and subsidiaries of local banks in total credit (some 40 percent in late 2004). In other words, the TBS has reacted to heavy taxation with the migration of both deposits and loans to subsidiaries or branches abroad. Almost every Turkish private bank worth its name has established a subsidiary bank in Europe, with the Netherlands as the favoured location. As of the end of 2002, 40 percent of all loans by Turkish banks had been booked offshore, mainly to companies domiciled in Turkey. International branches and affiliates accounted for 22 percent of total assets and 16 percent of deposits. The asset and liability migration initially created a supervision problem as the international subsidiaries were outside the jurisdiction of Turkish regulatory authorities. This problem has now been solved by the BRSA auditing the international subsidiaries and branches along with the production of consolidated financial statements, which cover these subsidiaries. Nevertheless, the burden continues on the TBS in the sense of allocating capital and resources to these subsidiaries.40 The good news is that the government has now formulated a strategy to phase out and harmonize some financial taxes. The May 2005 IMF Letter of Intent envisages the elimination of two key financial intermediation taxes, a financial transactions tax by 2006, and a tax on consumer credits by 2008, conditions permitting. Absence of a level playing field between private and state banks State-owned banks make up a large segment of the Turkish banking sector (about 42 percent of total deposits and some 20 percent of total loans). Subsequent to the crisis, the state banks (Ziraat, Halk, and Emlak) have gone through successful operational restructuring as part of the banking sector reform program (e.g. the merger of Emlak with Ziraat and the appointment of a joint management board; the downsizing of branches and employment; and the passing of legislation preventing “duty losses”). But now the next phase of reform is marred by uncertainties, and the government (at least until now) has provided mixed signals regarding its intentions on state bank privatization.41 One reason could be that the Turkish experience with bank privatization has not been encouraging. Turkey privatized a number of state banks in the mid-1990s, but most of these banks were among the first to be taken over by the SDIF during the crises years (e.g. Sumerbank). Regardless of the government’s intentions and fair concerns regarding the privatizations of state banks, the presence of these banks adversely affects the
290
Ali Tükel et al.
smooth functioning of the banking sector in a number of ways. With less concern for the bottom line, state banks can offer higher interest rates on deposits than private banks, putting private banks at a competitive disadvantage. According to industry sources this continues to be a problem, as it was before the 2001 crisis. Further, following the imposition of a ceiling on the government guarantee on deposits, state banks may now be perceived as having an “implicit” blanket guarantee while private banks no longer do. This may also lead to a migration of deposits.42 One way to address this problem, according to banking sources (mainly representing private banks) is to limit state bank functions to their original public functions, notably agricultural and SME-lending. In order to continue with these functions, state banks need not continue as universal banks with a deposittaking license. The counter argument, expressed by the current management of Ziraat and Halk, asserts that state banks must, to the contrary, expand certain activities and improve their results in order to become more attractive to potential buyers so that privatization can eventually go ahead. No potential buyer would be interested in the state banks as they are. The recent forays of these banks into consumer products, especially into the credit card business should be interpreted in this context. Second, state banks hold a monopoly on banking with state enterprises and foundations in Turkey, which are barred from opening an account with private banks and can only bank with state banks. This practice gives state banks a complete monopoly on the cash management of state enterprises and foundations. Arguably, the practice also deprives this captive customer base from using the more advanced financial products offered by private and foreign banks. Growth prospects, potential size of the market, foreign entry Credit demand in Turkey is derived from three segments: large corporates, SMEs, and consumers. The TBS currently faces different problems in profitably expanding in all three segments. Large corporates have never experienced any problems in terms of access to credit. In the completely liberalized financial environment of the Turkish economy, not only domestic banks, but also foreign banks (with or without a local presence) have been willing to lend to Turkish corporates with good credit standing. In other words, lending to corporates has been a “borrower’s market.” Consequently, spreads in this segment have reportedly declined to breakeven levels, with the implication that, short of a pick up in growth, there seems to be little room for further loan growth in this fiercely competitive segment. The SMEs on the other hand have been notoriously underserved in terms of investment credit, which seems to be explained by a combination of demand and supply-side factors. Even for Halkbank, which is the state bank with the designated mission of lending to SMEs, the loans/total assets ratio is less than 10 percent (2002 figures). One reason is that SMEs have not enjoyed a comfortable relationship with private banks, which in the past have tried to expand SME lending. During
The Turkish banking sector 291 the extremely volatile macroeconomic environment in the last decade, however, at the very first sign of a liquidity problem, banks either called in SME loans before maturity or increased interest rates on the existing loans drastically. This practice was repeated in the crisis years of 1994, 1999, and 2001, and created a problem of trust between SMEs and private banks, which subsequently started counting on equity finance for expansion. Under the current more stable macroenvironment, the trust between the SMEs and the TBS is likely to be restored, but this will take time. On the supply side, banks face a problem of evaluating SME creditworthiness and overcoming the asymmetric information problem faced by all banks in dealing with SMEs. In many sectors where SMEs dominate, non-registered transactions are the norm and the actual activity level of the prospective borrower is rarely reflected in its financial statements. This makes it difficult to make valid credit decisions about SMEs. This problem is especially great for foreign banks, which have to comply with their global credit practices. In more developed markets most of the lending to SMEs is secured by mortgages and pledges on property (merchandise, machinery, trucks, cars, ships, etc.) or revenue streams. In Turkey, these legal instruments also exist and are widely used, but the problem lies with the legal procedure for converting these collaterals into cash, which takes a long time and can be obstructed easily by the defaulting borrower. The banking sector is lobbying to change to articles of the law to make liquidation easier. Given these problems with lending to enterprises, be they blue chip or SME, the third segment, retail lending, becomes the only possible loan growth area. One problem though is that this is a relatively new area for the TBS, the first one of which ended in tears. The first consumer-lending boom in recent history took place in 2000, and ended in the 2001 crisis. Banks started to lend again to consumers in late 2003 and total consumer loans (including car and house loans) tripled, restoring the share of retail loans to the pre-crisis levels of 15 percent (Figure 11.6). At the end of 2004 there were some 3 million consumer loan contracts outstanding. In a country with an estimated 15 million households, this figure may indicate that there is a lot of room for growth. Yet industry sources assert that this represents the bankable population of Turkey at the moment. The remaining majority of the population does not demand banking products, including loans. The reasons frequently cited are low levels of per capita income, unequal distribution of income, the relatively recent history of urbanization, and low levels of education. The increasing foreign bank presence in Turkey Foreign banks have operated in Turkey since the financial liberalization of the 1980s. All legal barriers to entry were gradually removed and the Treasury has traditionally been accommodating in granting licenses, be it to open a single branch or obtain a license as a bank incorporated in Turkey. Despite this open-door policy, the presence of foreign banks until recently was limited, in most cases to a single
292
Ali Tükel et al.
270
30
240
12-month real growth (%; left scale))
210
as % of total credits by private banks (right scale)
27
180
24
150 21
120 90
18 60 30
15
0 12
–30 –60
9
–90 –120 Dec–00
Oct–01
Aug–02
Jun–03
Apr–04
6 Feb–05
Figure 11.6 Consumer credit growth (%; real).
branch in Istanbul doing corporate banking. The only exception with a retail branch network was Citibank. Moreover, its market share did not exceed 1 percent. Citibank first tried organic growth by opening branches, then negotiated with some local banks to acquire market share, but none of the deals were concluded. The entry of foreign banks to the retail market started after the 2000–2001 crisis. HSBC led the way by purchasing the branch network of the failed Demirbank (minus its securities portfolio) from the SDIF. Then Uni-Credito Italiano bought 50 percent of Koçbank, which was a wholly owned subsidiary of one of Turkey’s leading conglomerates. Garanti, Turkey’s third largest private bank, had foreign suitors, and enter into talks twice, in 2001 and then in 2004, with Intesa BCI of Italy. Even though the acquisition did not go through, interest from various parties continues. In early 2005, BNP Paribas purchased 50 percent of TEB, a mid-size bank, while the Dutch-Belgian financial group Fortis bought Dıvbank, Turkey’s seventh largest private bank. The biggest deal came when the Koç-Credito Italiano partnership bought a controlling share in Yapı Kredi, one of Turkey’s largest banks, in May 2005. This brought the market shares of banks with foreign partners to about 20 percent, up from around 1 percent before the crisis. Why has foreign bank participation been so limited for so long and why has it started to change recently? Some bank executives we have interviewed argued that foreign banks have been interested in the Turkish market for two main reasons. One was to benefit from the growing business of Turkish corporates, the
The Turkish banking sector 293 other was to exploit the retail banking business, which offered significant upside potential. Concerning the first motive, our interviewees further asserted that, under the fully liberalized financial regulations of Turkey, a foreign bank that wants to do business with local corporates can do so with one branch or even one representative office. Although it is not possible to support it with figures, it is a wellknown fact that foreign banks lend a lot more to Turkish corporates than it appears on the balance sheet of their Turkish operations. There really is no incentive to expand the balance sheet in Turkey and add “currency risk” in addition to the country risk they have already assumed. As to the second motive, there was no successful entry to the retail sector, that is, lending to SMEs and individuals. The main reason for that was that inflated valuations of Turkish banks acted as a natural barrier to entry. Such inflated valuations in turn reflected the lucrative but risky reliance on treasury holdings in an environment of macro instability, and corrupt banking practices. Potential foreign acquirers felt unsure about their ability to sustain high levels of profitability in this environment through “operational excellence” alone (McKinsey, 2003). Only Citibank attempted to grow organically in this sector, by expanding its own branch network and its relative lack of success seems to have acted as a deterrent for other prospective entrants. The recent interest in retail banking is mainly a result of changes in the operating environment and of Turkey becoming relatively stable. Factors that have made acquisition of Turkish banks less attractive are gradually being eliminated: legal uncertainties, the tradition of having portfolios of non-financial holdings and the exposures of banks to their parent groups. The TBS looks and feels more like a normal banking industry today than it did in 2000. Once a track record of stability is established and certainly when EU accession becomes a concrete vision of the future, further foreign bank entry naturally will follow. European banks in particular, have been watching the market closely. A final factor for those banks, which hesitate to enter the Turkish market, seems to be the fear that in Turkey foreign banks may not have an “innovation advantage.” European banks’ last series of acquisitions were made in Central and Eastern Europe, which were essentially virgin markets in terms of retail and SME banking products. The products they offered routinely in their home markets were innovations in the new markets they entered. This is not the case in Turkey. The retail banking product range in the domestic market does not differ much from the products offered in the EU-15. The innovation advantage Central and Eastern Europe provided to foreign banks does not exist.
Conclusion The TBS has made major strides in terms of its transparency, capital adequacy, and the quality of the regulatory framework and supervision, and is facing a more stable macroeconomic environment. The high degree of taxation of the sector is a problem. Reserve requirements (which are unremunerated) and other taxes on financial intermediation are much above those generally observed in the EU (ECB, 2003). As is well known, as banks manage to shift this tax onto borrowers,
294
Ali Tükel et al.
it acts as a drag on borrowing, hence on financial intermediation and investment. Reduction of taxes on financial intermediation, some of which are planned for 2006 and beyond, could help repatriate operations which have driven large borrowers and depositors abroad and to level the playing field for entry into the European banking market. An uneven playing field vis-à-vis the state banks is also an important impediment. State banks enjoy certain privileges that distort competition, and allow them to account for close to 40 percent of deposits. Chief amongst these are the perception that there is an implicit guarantee by the state on state bank deposits, high interest rates on deposits offered by state banks which are less constrained about the bottom-line than private banks, and the restriction that state and foundation funds be exclusively deposited with state banks. The key problem however appears to be the limited prospects for loan growth. Large enterprises are unlikely to be a source of credit demand given their access to low-cost international financing, while lending to the SME sector is hampered by the difficulties in evaluating SME creditworthiness and overcoming asymmetric information problems, in particular because non-registered transactions are the norm and the actual activity level of the prospective borrower is rarely reflected on the financial statements, while cumbersome procedures for liquidation of collateral also limit SME lending. This leaves consumer lending as the main source of growth looking forward, and here the low level of income in Turkey may act as a damper on growth. After many years of very little presence despite no explicit legal barriers, the interest from foreign banks appears to be significantly increasing. The reasons for this are Turkey’s improved macroeconomic stability, a cleaner banking sector, and EU perspective. Therefore, sustaining this, and hence locking in Turkey’s current single digit inflation-high growth environment to an irreversible path through structural reforms throughout the economy and including in the banking sector (reduction of taxation of the sector, addressing legal impediments to liquidation of collateral), would seem to be the only realistic growth strategy for the sector, no matter how generic this may sound. In this connection, able management of this process by the BRSA would appear to be a critical policy challenge.
Appendix 1: efficiency, productivity, and competitiveness of the TBS43 In this appendix, we explore issues of efficiency. Admittedly, this is a very broad and data-demanding area, and a thorough examination goes beyond the scope of this chapter.44 Instead, we do three things: we look at a few broad indicators of efficiency, summarize the findings of a comprehensive productivity study by the consulting group McKinsey on retail banking and share some anecdotal evidence on the international competitiveness of Turkish banks in the current environment. If efficiency is an issue, one might expect Turkish corporates to turn to foreign banks rather than branches of Turkish banks abroad. Foreign bank entry in the Turkish market should also be relatively high, especially in the context of
The Turkish banking sector 295 more realistic bank valuations since the crisis. There is indeed evidence to that effect in certain segments of the market, as we described in a separate section above. We conclude that the record is somewhat mixed. Despite weak “financial intermediation” and excessive reliance on government securities, Turkish private banks come across as relatively advanced retail franchises.45 They have made large investments in IT infrastructure and exploited the efficiency gains offered by alternative distribution channels. Internet-based and telephone banking have become standard options, with an increasing proportion of transactions moving to these channels from the traditional branch network. Most of the banking executives we interviewed were of the opinion that private Turkish banks are as advanced as any bank in the EU with regard to retail banking, and that given the skilled resources they employ, they were capable of developing and maintaining this level of service. This would appear to be an important asset as Turkey prepares to enter a common market for financial services, as called for by the EU’s Second Banking Directive. But all is not well. As a McKinsey study shows, the sector seems to suffer from low levels of productivity compared with both the benchmark country (the United States) and its own potential, while our simple, illustrative indicators suggest that efficiency gains in the sector are a recent phenomenon. Moreover, anecdotal evidence suggests that Turkish banks are already suffering from international competition. Measuring efficiency through illustrative indicators There is little doubt that compared with the 1990s the TBS is now much leaner in terms of standard efficiency measures such as deposits and assets per employee and per branch. Perhaps the most striking statistic is that, despite the tripling in both deposits and assets in the 1990–2003 period, the total numbers of employees and branches are now lower than they were at the beginning of the last decade (see Table 10A.5). Specifically, the sector employed 154,000 staff in 6,500 branches in 1990; it now employs 123,000 staff in 6,000 branches. The numbers peaked in 2000 with 170,000 staff and 7,800 branches, and then declined rapidly. The productivity gains were achieved in the post-crisis 2001–2003 period during which personnel numbers declined by 28 percent and branch numbers by 24 percent. In terms of branch density, Turkey has nine branches for every 100,000 persons (a common measure of branch density). Among the EU members, Germany has 60 branches and the UK, Finland and Sweden have less than 25. Among the new members Poland has 6 and Hungary has 11. Employees per branch now stand at 21 and assets per branch at about $30 million. Assets per employee, which were less than $400,000 in 1990, are now nearly $1.5 million. Contrary to what may be expected, the productivity figures of state banks are in line with those of private banks, possibly reflecting the favourable impact of restructuring during 2001–2002. The improvement in the productivity figures of the sector was mainly the result of the restructuring of state banks and the liquidation of the SDIF banks. The
296
Ali Tükel et al.
Table 11A.1 Efficiency indicators
Deposits ($ billion) Assets ($ billion) Deposits/assets (%) Deposits/branch ($ million) Deposits/employee ($ million) Employee/branch Assets/employee ($ thousand) Assets/branch ($ thousand)
1990
1995
1999
2000
2001
2002
2003
2004
32.5 58.1 55.9 5.0
43.5 67.0 64.9 7.0
89.1 133.2 66.9 11.6
87.7 155.2 56.5 11.2
76.6 117.3 65.3 11.1
84.4 130.1 64.9 13.8
111.3 178.9 62.2 18.7
142.4 228.3 62.3 23.3
0.2
0.3
0.5
0.5
0.6
0.7
0.9
1.1
23.5 377
23.2 463
22.6 766
21.7 911
19.9 853
20.2 1056
20.7 1451
20.8 1796
8,857 10,735 17,319 19,803 16,984 21,310 29,984 37,397
Sources: TBA, BRSA and authors’ calculations.
survivors also improved their productivity through an emphasis on alternative delivery channels. No data are available, but industry sources assert that nearly half of all the transactions of large private banks take place through telephone and online banking. In terms of non-interest costs, Turkey’s four largest banks compare favourably with the transition economies. Their “non-interest cost/total asset” ratio ranged between 3.4 and 4.6 percent in 2003. In terms of the most common measure of bank efficiency, the operating costincome ratio, 3 of the 4 largest Turkish banks had a good ratio, in the 30–70 percent range. The EU average is 61 percent. Only Yapı Kredi, with a 95 percent ratio, appears to be “inefficient” by this criterion. Given that 2003 was a year when one-off trading results had a big impact on bank earnings, one alternative is to normalize by excluding trading gains. Then, only Akbank and Wvbank appear to fall in the acceptable range of efficiency. Under conditions of high interest-rate volatility, the stability of a bank’s earnings depends to a large extent on the income from commissions and fees for the banking services they provide. Historically, with the high interest margin earned from investments in government securities, Turkish banks have subsidized other banking business by offering zero-fee transactions to gain market share. This practice has been discontinued and fees and commissions have started to contribute to the “bottom line.” The weighted average (i.e., weighted with assets) commission was 1.4 percent in 2003. The bulk of commission income is from the consumer business, mainly from credit cards and asset management fees. A productivity assessment: the McKinsey study Measuring productivity, defined essentially as output per unit of input(s), is clearly a difficult exercise in the absence of homogenous inputs and outputs, requiring a clear methodology and much data gathering. Luckily, as part of a recent
The Turkish banking sector 297 comprehensive study on Turkish productivity, the consulting group McKinsey examines the productivity levels in the retail-banking sector as one of the 13 sectors (McKinsey, 2003).46 The study concludes that Turkey has a long way to go in terms of matching the productivity level of the benchmark country (the United States) as well as its own potential. Specifically, it finds that labor productivity in the TBS is only 42 percent of the benchmark US case, and some 46 percent below its own potential (which the study calculates is 88 percent of the United States). The study takes a production function approach to the calculation of labor productivity, whereby three main categories – payment transactions, deposit accounts, and loans – are considered as physical outputs and their relevant employment figures as inputs.47 The analysis finds the “organization of functions and tasks” (OFT) in the “payment transactions” category, as the biggest operational contributor to the productivity gap (see Exhibit 4 on p. 199 in McKinsey, 2003). (Other contributors to the productivity gap include capital intensity/technology, scale, capacity utilization, and customer demographics and behavior, with relatively smaller contributions from these categories except perhaps for customer demographics and behavior.) This, in turn, is explained by a payment mix dominated by labor intensive, branch-level transactions, with the latter accounting for 46 percent of all payment transactions (compared to 3 percent in the US and 7 percent in the Netherlands). Although OFT-related weaknesses are common to all banks, the study finds that state banks have two additional challenges: the infrastructure for “alternative delivery channels,” and branch design and practices. In fact, low productivity levels in state banks explain 21 percentage points of the 46 percentage point gap between the TBS’s current and potential productivity levels. Just aligning their operational efficiency with that of private banks, the study finds, would improve overall productivity by 8 percentage points. The analysis comes up with three recommendations for policy-makers. Two are related to state banks alone: the introduction of productivity-related performance measures into state banks; and removing state banks’ unfair advantages by levelling the playing field. The third involves creating the enabling legislation that encompasses the promotion of alternative delivery channels and improves efficiency in loan application processes. All this does not mean however that the TBS is notably less efficient than some of its European competitors. A study of the efficiency of European banks by Wagenvoort and Schure (1999) shows that the average of banks in some EU countries deviates as much as 60 percent from the “optimal-cost curve” (i.e., they suffer from a so-called “x-inefficiency”). By that standard, Turkey is not an outlier in the European sample. Sectoral competitiveness The TBS is clearly at a competitive disadvantage vis-à-vis international banks in meeting the demands of large Turkish corporates, especially in terms of capital
298
Ali Tükel et al.
market products. The frequent crises and the short-term outlook of the TBS developed under the volatile macro-conditions of the past decade left only a few tools in their arsenal. Under an open capital account, large Turkish corporates with increasingly international businesses have turned to foreign banks and started using financial products that were not available to them locally. Anecdotal evidence abounds that Turkish corporates give only their most “mundane businesses” to the TBS. On the other hand, a wide spectrum of more sophisticated tasks with higher profit margins, ranging from Vestel’s asset-backed bond issue in international markets to Erexli Demir Çelik’s long-term investment finance syndicated loan, were all arranged by foreign banks. Local banks have so far managed to serve the other two segments of the customer base, the SMEs and consumers. The decade-old experience of the only foreign bank to venture into the local consumer business suggests that local knowledge still counts in the Turkish market. Their credit card, otherwise a valuable global brand, has under-performed in the market.
Acknowledgment The authors thank Dr Hasan Ersel for his very useful comments and Ozlem Derici for help with the data and charts.
Notes 1 This, however, partly owes to the pervasiveness of non-bank credit channels in the Turkish economy, a point recently made by Türkan (2004). Türkan (2004) shows that combining the bank and non-bank channels, financial intermediation in Turkey amounts to some 50 percent of GNP. 2 It may be worth adding that in the absence of structural reforms and a strong fiscal adjustment, monetary policy remained largely accommodative in the 1990s, providing the liquidity needs of the market through foreign exchange purchases in periods of strong inflows, and open market operations at times these flows weakened or reversed. 3 This is not to say that program designers were unaware of the trade-off between vulnerabilities caused by the pre-announced crawl, and its effectiveness in reducing inflation expectations and hence real interest rates. 4 See Van Rijckeghem (1999), IMF (2000), and Alper et al. (2001). 5 These quotes are drawn from Üçer and Van Rijckeghem (2005). 6 Moody’s Investors Service, “Turkey: Banking System Outlook,” August 2000. 7 Financial Times, November 20, 2000. 8 Interview with David Edgerly, board member Garanti Securities, as reported in Euromoney, “The Old Economy is Dead, Let’s Bury It,” April 2000. 9 The Bankers’ Association of Turkey. 10 The regulations also stipulate that the ratio of foreign exchange assets to foreign exchange liabilities be between 75 and 115 percent. 11 This issue was conclusively resolved after the February devaluation when the doubtful forwards were recontracted at higher exchange rates and banks announced publicly that they would not insist on settlement of the forward contracts, announcing that they had “benefited from the presence of these open F/X positions and thus should bear the resulting damage as well.” (BSI Turkish Research, “Turkish Banks,” April 12, 2001). Interestingly the issue has not received much attention, and the shareholders of the banks that did not insist on settlement of the forward contracts did not object, to
The Turkish banking sector 299
12 13
14 15
16 17
18
19
20 21
22
23
our knowledge. This probably reflected the prevailing atmosphere of the period, where many banks, companies and families were fighting for economic survival. Euromoney, “Treasury Bill Famine Drives Consolidation,” November 2000. The SDIF, originally annexed to the Central Bank, was given a somewhat awkward but more central role during and before the crisis, in the aftermath of its switch to the BRSA. When the BRSA deemed a bank nonviable, it would be taken over by the SDIF, which would then function as the asset recovery arm of the BRSA. Until a recent law, the SDIF did not have a juristic personality and was represented by the BRSA. In fact, the Turkish crisis has been explained with respect to the academic literature on monetization of prospective costs of bank bailouts (Burnside et al., 2001). The economy was overheating, but fiscal measures to address this envisaged in the program were not forthcoming. There were a number of difficulties on the privatization front as well, with Turkish Telecom turning into a disastrous saga. In the banking area, there were delays in introducing consolidated reporting, in introducing regulations on internal risk management systems and an amendment to capital requirements to properly reflect market risk, and in introducing legislation on the commercialization of state banks. See Üçer and Van Rijckeghem (2005). To clarify, the SDIF met a small chunk of the costs from its own resources, that is through premiums paid by the banking sector that had accumulated at the SDIF over time, but the remainder was met through securities on loan to the SDIF in return for “IOUs” and future collections. The SDIF has recently been separated from the BRSA, with a view to stepping up collection efforts. Thus far, the SDIF has managed to collect about $2 billion and there is occasionally talk of the SDIF’s debt to the Treasury being written off. Connected lending has traditionally been a major problem of the TBS. Prior to the 2001 crisis, the problem had been aggravated by both the weak enforcement of the regulations and by legal loopholes, which allowed the controlling shareholders to lend to their non-financial businesses without exceeding the legal lending limits. Treasury auditors had been aware of the over-lending to subsidiaries or affiliated companies. Some bank shareholders infamously invested in new ventures where the only lender was the owner’s bank. In addition, shareholders used cross-holdings, relatives, and friends, etc. to further their interests. By the late 1990s, the problem had reached proportions that constituted a systemic risk to the sector as a whole. Excluding the Cukurova deal, signed in early August 2004. The Banking Law was changed to allow legal action to seize personal assets of SDIF banks’ former shareholders. The first implementation of this was the seizure of the assets of the Uzan family, owners of the infamous Imar Bank, in February 2004. Yet, as we discuss later, the failure of Imar Bank in 2003 led to a questioning of the adequacy as well as the implementation of this framework. The BRSA kept the public informed about the banking sector restructuring process through regular reports, hence we provide only a brief overview here. On the approach taken, see various IMF letters of intent and BRSA presentations, as well as Akçay (2003) and World Bank (2003). The BRSA’s approach followed the insights of the voluminous literature on banking crisis resolution – see, for instance, Dziobek and Pazarbavıoxlu (1997) and Tang et al. (2000). It is notable that both the political authority and the current BRSA administration seemed to favor outright liquidation rather than takeover – an issue that has, among others, delayed the adoption of the new Banking Act. In Turkey’s circumstances, the right approach would appear to keep both options open, as the current draft law stipulates. For resolution of the NPLs of private banks, the Bank Association of Turkey, with the approval and support of the BRSA, initiated the Financial Restructuring Framework Agreements to work out the corporate debt problem. The platform was named the
300
24
25
26
27
Ali Tükel et al.
“Istanbul Approach” (after the London Approach of the 1970s, where banks and corporates gathered under the auspices of the Bank of England). Though some 25 banks and 17 non-bank financial institutions signed the agreement, the Istanbul Approach essentially evolved into a platform of postponing rather than restructuring debt. That the private banks have not received any financial support also contributed to this outcome. In another attempt to tackle the NPL problem, asset management companies, which would take over the bad loans of the TBS, could develop as another method. Despite legislation and regulations being finalized in 2002, no asset management companies have been established yet. The recap plan had three phases that extended through the end of June 2002, from the completion of three separate audits on the basis of December 2001 balance sheets to the injection of public funds. The two audits were conducted by the banks’ own auditors and an externally designated auditor for each bank, respectively, and the final audit was essentially a judgment call by the BRSA. The reform of the regulatory structure picked up in the run-up to the 1999 disinflation program. In June 1999, a new Banking Act was passed (No. 4389), which, among others, called for the establishment of a new autonomous banking agency (BRSA) and the beginning of its operations in September 2000. This was a welcome step given that, while the regulatory framework needed much improvement (especially with regard to connected-lending practices, transparency and consolidated accounts, etc.), the key problem had been the lack of enforcement rather than the regulatory framework as such. After some haggling over the selection of board members, the coalition government managed to put together a board in late July 2000 and appointed the head of the BRSA in August. Inevitably though, the financial crisis of 2001 shifted the focus of the newly born BRSA from supervision to restructuring and rehabilitation. The guarantee has been around for ten years and caused much “moral hazard” on the part of depositors and distortion in the sector, but at the same time it may have helped to prevent wide-scale “bank runs” – a phenomenon never observed in Turkey despite the many financial crises the country underwent. The blanket state guarantee was first introduced in the wake of the 1994 currency crisis. Though the initial plan was to phase out the unlimited guarantee once the situation normalized, the temporary policy turned into a permanent one. The moral hazard this created is generally considered to be the main reason why many private banks, which ended up at the SDIF, continued to be able to take deposits. In July 2000, as part of the disinflation program, the coalition government started to phase out the blanket guarantee, by lowering the ceiling to TL100 billion (more than US$150,000 at the prevailing exchange rate). The blanket guarantee was re-introduced six months later in December 2000 during the liquidity crisis. During the collapse of Imar Bank in July 2003, the BRSA said it would adhere to the blanket guarantee, but that it would reduce the coverage to deposits that are below TL 50,000 billion (around €30,000) by July 2004, which was executed as planned. This ceiling is broadly in line with the deposit guarantee in EU countries, where the level of deposit protection varies between €20,000 and €60,000, but is a bit high for Turkey, given that its average income per capita is about one-fifth of that in the EU. The ceiling covers over 90 percent of the accounts by the number of accounts, but only about 60 percent by the size of accounts. As explained in the latest letter of intent (IMF, May 2005), the law focuses on strengthening the legal framework in the following areas: (i) “fit and proper” criteria for bank owners; (ii) on-site inspections; (iii) lending to related parties; (iv) legal protection for BRSA and SDIF board members and staff for actions taken in good faith during the course of their duties; (v) delineating the responsibilities of BRSA and SDIF respectively and providing for their effective coordination; and (vi) granting SDIF the power to determine deposit insurance premiums in coordination with the BRSA, the CBT, and the Treasury.
The Turkish banking sector 301 28 As an IMF banking report on Turkey noted, “ ‘fake’ would be an appropriate term to use only in cases where, at the outset, it was manifestly impossible for the counterparty to fulfill its side of the bargain, or there was no intention of doing so. The first point would be picked up by the SBAs, who review counterparty credit analysis as mentioned above. The second is more difficult to assess, particularly at the outset, but should be revealed on maturity of the deal. The SBAs should pick up unexplained failures to pay on the part of counterparties that are not appropriately dealt with by the bank concerned.” (IMF, 2001, p. 9). 29 See report by Fort and Hayward (2004). 30 We focus here on ISE-listed banks only, which make up some 80–90 percent of the sector. 31 As there is no standard definition of free capital, we calculated it, for illustrative purposes, by deducting NPLs net of provisions, subsidiaries, participations and fixed assets (land, buildings, etc.) from total equity. 32 This section is a heavily revised and streamlined version of Steinherr et al. (2004). 33 These factors were partly identified through interviews with a number of bank executives. The questionnaire used in these interviews is available from Steinherr et al. (2004). 34 It may be worthwhile noting that Turkey has yet to undergo an IMF/World Bank joint Financial Sector Assessment Program, which is scheduled for 2006. The FSAP should assist in identifying in more detail potential risks and development challenges in the sector. 35 Moreover, as noted above, the so-called free capital is much lower as well. 36 See Steinherr, et al. (2004). 37 Foreign banks, however, may not follow the same route due to their own regulations. 38 Albeit few, some advantages derive from Basle II as well. One such advantage is that Basel II eliminated the “sovereign floor,” which implies that banks can now be rated higher than their government, with a minimum risk-weight of 20 percent remaining. 39 There are no readily available data on intermediation margins in Turkey that could be used for comparison with EU countries and a bare look at the margins could be somewhat misleading owing to several factors (e.g. interest-rate volatility or heavy taxation of the financial sector). The margins look high nevertheless: for instance, for consumer loans, the margin was over 6 percentage points until recently – well above EU averages (Consumer loans were made at interest rates close to 40 percent while savings deposits fetched less than 23 percent interest). 40 In 2002, fiscal revenue from this taxation amounted to about 3 percent of GDP; meanwhile, with declining interest rates, total fiscal revenue from the financial sector appeared to have been halved in 2003, and simulations suggest that it would drop to about 0.5 percent of GDP should the interest rate decline to European levels of around 10 percent. Tax “rate” would decline as well, although still remain significant at almost 40 percent even at 10 percent interest rate. 41 Numerous examples of these mixed signals can be found in the local press. Another interesting issue is the recent attempts by the Turkish post office to obtain approval from the BRSA to offer banking services, which appear to contradict the broader objective of reducing the role of the public sector in banking. 42 In early 2005, state banks had some 40 percent of all deposits. 43 This section draws heavily on Steinherr et al. (2004). 44 The terms “efficiency” and “productivity” for our purposes are very close to each other – essentially, productivity is a measure of economic efficiency, which shows how effectively economic inputs are converted into output. On the other hand, competitiveness is, by definition, a relative position, but arguably it also boils down to productivity/efficiency.
302
Ali Tükel et al.
45 State banks suffer from low-skilled human resources, an underutilized IT infrastructure and a lethargic service level, typical of many state-owned enterprises. 46 This is by far the most comprehensive productivity study on Turkey that we are aware of, whereby the standard McKinsey Global Institute methodology is employed. 47 See the appendix to the McKinsey (2003) chapter on p. 219 for an exact description of the methodology used in measuring labor productivity.
References Akçay, C. (2003), “The Turkish Banking Sector Two Years after the Crisis: A Snapshot of the Sector and Current Risks,” in Öniv, Z. and B. Rubin (eds), The Turkish Economy in Crisis, London: FrankCass. Alper, E., H. Berument, and K. Malatyali (2001), “The Impact of the Disinflation Program on the Structure of the Turkish Banking Sector,” Russian and East European Finance and Trade, V. 37, No. 6, pp. 81–95. Bank for International Settlements (2004), Foreign Direct Investment in the Financial Sector of Emerging Market Economies, Basle, March. Banking Regulation and Supervision Agency (BRSA) (2003a), “Efforts Strengthen the TBS [after] the IMAR Bank Episode,” presentation by Ercan Türkan, Vice-President, BRSA, Ankara, October (in Turkish). Banking Regulation and Supervision Agency (BRSA) (2003b), On the Reduction of Intermediation Costs in Turkish Banking, BRSA, Ankara, November (in Turkish). Banking Regulation and Supervision Agency, Banking Sector Restructuring Program: Progress Report, various issues, BRSA, Ankara. Burnside, C., M. Eichenbaum, and S. Rebelo (2001), “Prospective Deficits and the Asian Currency Crisis,” Journal of Political Economy, V. 109, pp. 1155–1197. Dziobek, C. and C. Pazarbavioxlu (1997), “Lessons and Elements of Best Practice,” in W.E. Alexander, J.M. Davis, L.P. Ebrill, and C.-J. Lindgren (eds), Systemic Bank Restructuring and Macroeconomic Policy, International Monetary Fund, Washington, DC, pp. 75–143. European Central Bank (ECB) (2003), The Transformation of the European Financial System, ECB, Frankfurt. Fort, J.-L. and P. Hayward, The Supervisory Implications of the Failure of Imar Bank, Commission Inquiry Report, August 2004. IMF (2000), “The Role of Foreign Banks in Emerging Markets,” in International Capital Markets: Developments Prospects and Key Issues. IMF (2000), “Implications of Disinflation for Banks Profitability in Turkey,” Turkey Selected Issues and Statistical Appendix, IMF Country Report No. 00/14, pp. 37–51. IMF(1999–2005), Turkey: Letters of Intent (various documents from 1999 to 2005), IMF, Washington, DC. McKinsey Global Institute (2003), “Overview” (Section 1) and “Retail Banking” (Chapter 3), Turkey: Making the Productivity and Growth Breakthrough, MGI Report, February (retrievable from http://www.mckinsey.com/knowledge/mgi/turkey/perspective.asp). Özatay, F. and G. Sak (2002). “The 2000–2001 Financial Crisis in Turkey,” Paper presented at the Brookings Trade Forum 2002: Currency Crises, Washington, DC. Steinherr, A., A. Tükel, and M. Üçer (2004), “The Turkish Banking Sector: Challenges and Outlook in Transition to EU Membership,” CEPS EU-Turkey Working Papers No. 4 (August).
The Turkish banking sector 303 Tang, H., E. Zoli, and I. Klytchnikova (2000), Banking Crises in Transition Countries: Fiscal Costs and Related Issues, World Bank Policy Research Paper No. 2484, World Bank, Washington, DC. Türkan, E. (2004), “Türk Ekonomisinde Makro Kredi Kanali: Ölçek ve Kalite Açisindan Bir Dexerlendirme,” Central Bank of Turkey (November). Turkish Banks Association, Banks in Turkey, various issues. Üçer, M. and C. Van Rijckeghem (2005), Chronicle of the Turkish Financial Crises of 2000–01 (Bogazici University Press). Van Rijckeghem, C. (1999), “The Political Economy of Inflation: Are Turkish Banks Potential Losers from Stabilization?” Istanbul Stock Exchange Review, V. 3, No. 10, pp. 1–16. Wagenvoort, R. and P. Schure (1999), “Who are Europe’s efficient bankers?,” EIB Papers, V. 4, No. 1, pp. 105–26. World Bank (2003), “Banking Sector,” Chapter 3, Country Economic Memorandum, World Bank, Washington, DC, July.
Index
adjustment costs 150, 158 agency costs 149, 178, 193, 207 agency problems 199, 202, 206 agency view 209, 228 agriculture 10, 17–18, 45–46, 51 Akaike Information Criterion (AIC) 119 Akbank 286 Asset Management Corporation (AMC) 283 asymmetric information 172, 177, 189, 193, 277, 291, 294 auctions 232 Bax-Kur 256; see also BK bakkal 131 Banking Regulation and Supervision Agency (BRSA) 277, 282–283, 285, 288, 294; and consolidated reporting 278, 299 banking supervision 277–279 Banks’ Association 280 Basle II 288 BK 253, 262; see also Bax-Kur blockholders 202; see also large shareholders block sales 232, 233, 237 BNP Paribas 291–292 borrowing constraints 78, 79 business cycles 15, 46, 48 business group 173, 179, 183, 193, 199–201, 207, 211–212, 217, 221 capital accumulation 38, 40 capital adequacy 283, 287, 293; capital adequacy ratio (CAR) 283–284, 287; free capital 287, 301 capital deepening 38 cash flow rights 173, 181, 184, 200, 203, 209, 211
cement industry 230, 233, 242 chaebol 179, 182, 207 Citibank 291–293 collateral 277, 291, 294 competition 229, 231 competitiveness 40, 287, 294 concentrated ownership 199, 201–203, 212–217, 221; see also ownership concentration connected-lending 277, 299 consolidation 279, 281 Constitutional Court 260 consumer loans 280, 291 Consumer Price Index (CPI) 69, 87 consumption expenditures 68, 69, 86, 87; nondurable consumption expenditures 69, 73; standard deviation of logarithm of 69 consumption inequality 64, 66, 68, 69, 70, 88 contribution rate 253 controlling shareholders 199, 201, 207, 209, 211, 221 control or voting rights 200, 209 convergence: conditional convergence 23, 46; convergence clubs 16; convergence coefficient 23, 46; regional convergence 44–46 corporate governance 172, 179, 193, 194, 201–203 corruption 280 cost of capital 185 cross-holdings 200, 209 Customs Union Agreement 22, 28 debt 185 Demirbank 280–281, 292 Democratic Left Party 112 Democrat Party 19
306
Index
demographic trends 97 deposit insurance, moral hazard 277, 290, 294 differential fertility 79 Dıvbank 292 dispersed ownership distribution of prices 142 dividends 178, 209, 211 drawer contracts 280, 285; see also fake forwards dual class shares 180, 184 duration of prices 130, 133 IV estimation 107, 159 East Asia 38–40, 203 Education 63, 64, 69, 70, 71–72, 74–88 educational attainment of husbands and wives 72, 75 efficiency 25–26, 51; allocative efficiency 229, 231; productive efficiency 229 Emekli Sandixi 256; see also ES employers 103 employment 10, 43–44, 48–49; composition of 94; feminization of 94, 96; public and private 96 endogeneity 150,159, 200, 204–205; allocative efficiency 229, 231; efficiency 227, 230, 240, 287, 294; productive efficiency 229 entitlement age 253, 255, 259, 267, 271; see also retirement age equivalence scales 66, 70 error components model 43, 143 ES 253, 258, 262; see also Emekli Sandixi etatism 17 European Union 3, 15, 22, 48, 67, 284, 293, 295 export-led growth 21, 37; see also exports, export orientation exports, export orientation 37, 94; see also export-led growth expropriation 199, 203, 209, 211 fake forwards 280, 285; see also drawer contracts family-controlled firms 202, 207 family ownership 173, 181, 190 female labor force participation 72, 73; see also labor force participation rate female literacy rates 82, 89 fertility 72, 75–76, 79, 82, 88, 92, 98 financial crisis 21, 66–67, 87, 93, 96 financial liberalization 20 financing 179, 193, 195
financing constraints 172, 179 firm performance 200, 203–204, 206, 217, 221, 229 firm size 41–44,149, 155, 157 foreign bank(s) 277, 291–292, 293; foreign bank entry 287, 291, 294 Fortis 291 Funds 189, 191, 192, 199, 200, 205, 206, 208, 209 Garanti Bank 291–292 generalised method of moments (GMM) 150, 159,161 gender gap 98 genetic algorithm 263 Gini coefficient 64, 66, 69 government transfers 21, 76 growth: endogenous growth 23, 36; exogenous growth 22; growth accounting 25, 27 growth opportunities 156, 157 Heckman-Lee two-step estimation 106 heterogeneity of prices, 133, 143 holding companies 180, 183, 202, 205, 207 home production 99, 103 hours equation 106, 117 hours of work 102, 117; actual and usual 108 household formation 64, 75–78 household income 64–66, 73, 75, 77, 79, 87; household income per adult equivalent person 66, 88 Household Income and Consumption Expenditure Surveys (HICES) 64, 69, 74, 79, 86–88 Household Labor Force Survey (HLFS) 93, 96, 99, 103, 108, 112 household size 71–72, 88 HSBC 291–292 human capital 36–37, 63, 70, 74, 79–81, 83, 86, 98, 116 ICT investment 40 identification 107 Imar Bank 285 IMF-supported disinflation (or stabilization) program 19, 22, 277; and November crisis 281 imperfect competition 25, 34 import-substituting industrialization (ISI) 17, 19–20 income distribution 64, 66, 69, 70–76, 85 income inequality 63–70, 72, 74–76, 82, 84
Index inflation 21, 128 innovation 293 input choice 230 intergenerational transmission of inequality 64, 84 internal capital markets 190 international comparisons 68 International Labour Organization (ILO) 103 Intesa BCI 292 investment 21, 41, 54, 172, 173, 174, 184, 206; equation 177, 178, 190; infrastructure investment 21, 41; opportunities 177, 185, 190; share in GDP 174 Wvbank 290 Istanbul approach 283 Istanbul Stock Exchange (ISE) 151, 156, 185, 199, 204 “jobless growth” 10, 93 Justice and Welfare Party 11, 22 keiretsu 179, 182, 183 kernel density estimates 108 Knowledge Economy 40 Koçbank 292 labor force participation model 104, 112, 119 labor force participation rate 99; see also female labor force participation rate labor supply: elasticities 117; extensive and intensive margins of 100 large shareholders 153, 200, 204–205, 208–209, 221; see also blockholders Latin America 63, 67, 72, 80, 88 Law no. 4447, pages 259, 268 legal protection 154, 202–203, 209 legal system 184 lemon banks 277 leverage 150, 156, 178, 206 leverage buyout 203 life cycle 112 liquidity 172, 184, 207 liquidity crisis 280–281 long-term multipliers 165 McKinsey 293–295 macroeconomic instability 22, 31, 41, 43, 51, 277 Malquist productivity index 26, 51 managerial discretion 172, 178
307
marginal q 177, 193 margins, interest margins 279, 289, 298, 301 marital sorting 78–79, 88 market: failures 199; imperfections 199; internal market 201; see also funds or financing; structure 231 markups 25, 34 marriage 98 maximum likelihood 106, 112 May 2001 rehabilitation program 282–284 menu costs 128 migration 94, 99,112 military expenditures 19, 41 Mincer coefficient 74, 79, 87, 88 minority shareholders 154, 174, 182, 191, 203, 213 Motherland Party 20; see also Özal, Turgut mutual funds 199 natural monopoly 228, 249 non-performing loans 280, 283, 287 normal distribution 105 open foreign exchange positions, open positions 279–280 openness 36–38 optimal money supply rule 141 ordinary least squares (OLS) 158, 161, 201, 217 ownership concentration 150,153, 199, 203, 213, 217, 221; see also concentrated ownership ownership structure 179, 184, 193, 200 Özal, Turgut 20–21,41 parametric reform 263 parental resources 76, 80, 81–82, 86 Pay-As-You-Go 252; see also PAYG PAYG 252; see also Pay-As-You-Go pazar 131 pecking order theory 155, 163, 206 pension balances 254 pension deficits 254, 267 per capita household income 65, 88 per child expenditure on education 81 persistence of income inequality 75, 76, 80, 83, 88 population 97 price dispersion 129 price rigidity 133, 139 price stickiness 130 price volatility 129
308
Index
privatization 96, 227, 229, 289; prices 232, 248 production frontier 51 productivity 15, 287, 295–298; capital productivity 38; intra-sectoral productivity growth 49, 51; labor productivity 15, 17, 43; measurement 25; relative sectoral productivity 17, 21 profitability 155 156 public education 63, 82, 84 public ownership 227 public sector 232; public sector firms 27, 34 public share offerings 231 pyramiding 173, 180, 182, 184, 200, 202 real wages 21, 28 relative price variability 129 replacement rate 259, 266 Republican People’s Party 112 reserve requirement 288, 293 retail sector 291 retirement age 253, 255, 259, 267, 271; see also entitlement age returns to education 74, 81, 117 returns to scale 25 Savings Deposit and Insurance Fund (SDIF) 280, 282 scale effects 34 search costs 128 sectoral re-allocation 31, 49 selectivity 107 self-employment 103,104 share issue privatizations 248 signal extraction models 128 skill premium 78, 79 Small- and Medium-sized Establishments (SME’s) 43, 277, 290–291, 298; SME lending 44, 290, 294 social security 101,122 social view 228 Solow residual 25; see also dual residual 39; primal residual 25; TFP Sosyal Sigortalar Kurumu 253, 262; see also SSK spending on education: private spending 64, 86; public spending 64, 83 SSK 253, 262; see also Sosyal Sigortalar Kurumu STATA 8.0 112
state banks 289; and duty losses 282; and level playing field 287, 289; and recapitalization cost 282 State Economic Entreprises (SEE’s) 19, 28; see also State Owned Enterprises (SOE’s) State Institute of Statistics (SIS) 28, 86, 88, 89 State Owned Enterprises 228, 232 (SOE’s); see also State Economic Enterprises (SEE’s) steady state 23, 46 stochastic production frontier 43 store types 131 structural transformation 51, 53 supermarket 131 synchronization of price adjustments 133, 140 tangible assets 155, 157; taxation of the financial sector 288; and financial repression 289 technical change 25, 51 telecommunications industry 230, 234 Temizel, Zekeriya 280 test: of exogeneity 124, 159; of overidentifying restrictions 116, 159 Tobin’s q 177, 179, 187, 193, 204, 206, 217 total factor productivity (TFP) 15, 27–36, 40; see also Solow residual trade liberalization 1, 20, 28, 31 Turkish Demographic and Health Survey (TDHS) 98 Turk Telekom 234 two stage least squares (2SLS) 203–204, 217 unemployment 21, 103,105 Uni-Credito Italiano 291–292 unobservable heterogeneity 103, 114, 117, 159 value-added 28, 44 venture capitalists 199 wage and salary work 96, 103, 104 wage equation 106, 116 Welfare Party 112 women’s education 64, 72, 81 workaholics 117 worker-to-retiree ratios 252, 254 YKB bank 291
eBooks – at www.eBookstore.tandf.co.uk
A library at your fingertips!
eBooks are electronic versions of printed books. You can store them on your PC/laptop or browse them online. They have advantages for anyone needing rapid access to a wide variety of published, copyright information. eBooks can help your research by enabling you to bookmark chapters, annotate text and use instant searches to find specific words or phrases. Several eBook files would fit on even a small laptop or PDA. NEW: Save money by eSubscribing: cheap, online access to any eBook for as long as you need it.
Annual subscription packages We now offer special low-cost bulk subscriptions to packages of eBooks in certain subject areas. These are available to libraries or to individuals. For more information please contact
[email protected] We’re continually developing the eBook concept, so keep up to date by visiting the website.
www.eBookstore.tandf.co.uk