Globalization and Productivity Growth: Theory and Evidence

Globalisation and Productivity Growth Theory and Evidence

Edited by

Holger Görg, David Greenaway and Richard Kneller

Globalisation and Productivity Growth

This page intentionally left blank

Globalisation and Productivity Growth Theory and Evidence Edited by

Holger Görg, David Greenaway and Richard Kneller Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham

Selection and editorial matter © Holger Görg, David Greenaway and Richard Kneller 2005 Individual chapters © contributors 2005 All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No paragraph of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, 90 Tottenham Court Road, London W1T 4LP. Any person who does any unauthorised act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The authors have asserted their rights to be identified as the authors of this work in accordance with the Copyright, Designs and Patents Act 1988. First published in 2005 by PALGRAVE MACMILLAN Houndmills, Basingstoke, Hampshire RG21 6XS and 175 Fifth Avenue, New York, N.Y. 10010 Companies and representatives throughout the world. PALGRAVE MACMILLAN is the global academic imprint of the Palgrave Macmillan division of St. Martin’s Press, LLC and of Palgrave Macmillan Ltd. Macmillan® is a registered trademark in the United States, United Kingdom and other countries. Palgrave is a registered trademark in the European Union and other countries. ISBN-13: 978–1–4039–3499–4 hardback ISBN-10: 1–4039–3499–1 hardback This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. A catalogue record for this book is available from the British Library. Library of Congress Cataloging-in-Publication Data Globalisation and productivity growth : theory and evidence / edited by Holger Görg, David Greenaway, and Richard Kneller. p. cm. Includes bibliographical references and index. ISBN 1–4039–3499–1 (cloth) 1. Industrial productivity. 2. Investments, Foreign. 3. International business enterprises. 4. Exports. 5. Globalization – Economic aspects. I. Title: Globalization and productivity growth. II. Görg, Holger. III. Greenaway, David. IV. Kneller, Richard. HC79.I52G57 2005 338
.06
09051—dc22 10 9 8 7 6 5 4 3 2 1 14 13 12 11 10 09 08 07 06 05 Printed and bound in Great Britain by Antony Rowe Ltd, Chippenham and Eastbourne

2005048114

Contents List of Tables

vii

List of Figures

ix

Acknowledgements

x

Notes on Contributors

xi

1 Introduction and Overview Holger Görg, David Greenaway and Richard Kneller

Part I

1

Aggregate Productivity Growth

2 Globalisation and Aggregate Productivity Growth Michael Henry and Chris Milner

9

3 Liberalisation and Aggregate Productivity Growth David Greenaway and M. Emranul Haque

28

Part II Cross Border Investment and Productivity 4 Foreign Direct Investment and Productivity Growth: Theory Ben Ferrett 5 Empirical Evidence on Foreign Ownership and Productivity Growth Sourafel Girma, Steve Thompson and Peter Wright 6 Multinational Enterprises and Spillovers Holger Görg and Alexander Hijzen

49

79

92

Part III Exporting and Productivity 7 Exporting and Productivity Growth: Theory Rod Falvey and Zhihong Yu 8 Empirical Evidence on Exporting and Productivity Growth in the UK David Greenaway and Richard Kneller v

117

136

vi

Contents

9 Multinationals, Exporting and Overseas Production Sourafel Girma, Richard Kneller and Mauro Pisu Index

156

177

List of Tables 2.1 Results of OLS regressions cross section estimations – dependent variable: TFP growth (average 1980–90) 2.2 Results of OLS regressions cross section estimations – dependent variable: TFP growth (average 1980–90) 2.3 Instrumental variables (2SLS) regressions cross section estimations – dependent variable: TFP growth (average 1980–90) 2.4 Instrumental variables (2SLS) regressions cross section estimations – dependent variable: TFP growth (average 1980–90) 2.5 Panel estimates (1960–90) 5-year averages – dependent variable: TFP growth 2.6 Panel estimates (1960–90) annual – dependent variables: TFP growth 3.1 Components of liberalisation episodes: the Papageorgiou et al. study 3.2 Elements of recent trade liberalisations 5.1 Average employment, wages and labour productivity of domestic- and foreign-owned firms in UK manufacturing industry 5.2 The impact of ownership on labour productivity 5.3 The impact of ownership on total factor productivity 5.4 The impact of ownership changes on wage rates by type of acquisition 6.1 Summary statistics 6.2 Descriptive statistics by export behaviour, 1988–96 6.3 Summary statistics by nationality of ownership 6.4 Basic regression results by export activity 6.5 Regression results by export and domestic market orientation MNEs 6.6 Regression results with absorptive capacity and intangible assets 6.7 Regression results by nationality 7.1 Summary of model differences 8.1 Percentage difference between exporters and non-exporters and their statistical significance vii

14 15

17

18 21 23 37 39

82 84 86 87 101 102 103 105 106 108 109 128 138

viii List of Tables

8.2 8.3 8.4

Probability model of exporting Transitions in and out of exporting Characteristics of new exporters versus established exporters and non-exporters 8.5 Probit model of export market entry 8.6 Probit model of export market entry: agglomeration effects 8.7 Sample characteristics of exporters and non-exporters in a matched sample of firms 8.8 Effect of export market entry on firm performance for a matched and unmatched sample of firms 9.1a World merchandise exports/GDP 9.1b Foreign assets/world GDP 9.2 Summary statistics 9.3 The determinants of MNEs’ exports (UK)

141 142 142 143 145 150 151 166 166 171 171

List of Figures 4.1 4.2 4.3 4.4 4.5 7.1 7.2 9.1

North–South FDI flows in general equilibrium à la Solow Game tree for simplified Fosfuri/Motta/Rønde model The MNE’s equilibrium period-1 choices Equilibrium industrial structures in the international duopoly game Equilibrium industrial structures Flow chart showing productivity uncertainty and firm entry/exit Productivity heterogeneity and with-in industry reallocation Probability density of productivity and firm choice

ix

52 59 61 64 68 125 126 163

Acknowledgements The work in this volume was completed at the Leverhulme Centre for Research on Globalisation and Economic Policy at the university of Nottingham and supported by funding from the Leverhulme Trust under Programme Grant F114/BF. The editors and authors gratefully acknowledge that financial support.

x

Notes on Contributors Rod Falvey is Professor of International Economics, University of Nottingham Ben Ferrett is Research Fellow, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham Sourafel Girma is Reader in Industrial Economics, Nottingham University Business School Holger Görg is Reader in International Economics, University of Nottingham David Greenaway is Professor of Economics and Director of the Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham M. Emranul Haque is Research Fellow, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham Michael Henry is Research Fellow, Aston University Business School Alexander Hijzen is Research Fellow, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham Richard Kneller is Lecturer in Economics, University of Nottingham Chris Milner is Professor of International Economics, University of Nottingham Mauro Pisu is Research Fellow, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham Steve Thompson is Professor of Strategic Management, Nottingham University Business School Peter Wright is Reader in Economics, University of Nottingham Zhihong Yu, is Research Fellow, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham

xi

This page intentionally left blank

1 Introduction and Overview Holger Görg, David Greenaway and Richard Kneller

It is widely acknowledged that over the last 30 years or so we have experienced a dramatic increase in the globalisation of economic activity. International trade, cross-border investment, portfolio capital flows and migration have all increased. As a consequence economies have become more open and more interdependent. Because economies have become more ‘joined-up’, national welfare has become more sensitive to events and developments beyond national frontiers, which is one of the reasons why globalisation has become a controversial issue. There is widespread acknowledgement that, in most circumstances, the longerrun net benefits of globalisation is positive; most discourse centres on the intra-national and international distribution of gains. Standard textbook analyses of gains from economic integration (i.e. globalisation) focus on static gains from specialisation and exchange. Empirically these tend to be of relatively small orders of magnitude, dwarfed by an order of magnitude by dynamic gains associated with stimulating economic growth. Given the centrality of economic growth to wealth creation and poverty reduction, we know less about the process than we should. One thing we know for sure, however, is that factor productivity is generally far more important to the process than factor accumulation. Globalisation and productivity growth are inextricably linked. On the one hand, globalising more could provide a boost to productivity, for example, by stimulating more competition; on the other hand, improved productivity could promote more globalisation by giving firms the necessary edge to enter global markets. The papers in this volume focus on a number of dimensions of these important interactions. 1

2

Holger Görg, David Greenaway and Richard Kneller

1.1

Aggregate productivity growth

We begin, by way of context with two chapters on aggregate productivity growth. In the first chapter, Michael Henry and Chris Milner note the longstanding debate about the nature of the correlation between aggregate openness and (aggregate) total factor productivity growth. One dimension of this debate centres on the measurement of openness – it is not straightforward to do this and research has shown that the measure used can make a difference to results generated. That finding has been helpful in moving the literature forward. Henry and Milner point out that the measurement of total factor productivity is also not straightforward, but the impact of different measures on outcomes has not been subject to the same scrutiny. They estimate a range of growth equations for a large number of countries, both in cross section and in a panel framework. In the process they deploy three different measures of productivity. What they show is that the performance of different openness measures is sensitive to how productivity is measured, with greater fragility in a cross-sectional setting (as one would expect). They conclude, therefore, that one must pay as much attention to the way productivity is measured as one does to the way openness is measured. The second chapter by David Greenaway and M. Emranul Haque shifts the focus to changes in openness, that is, liberalisation and the linkage to growth. The chapter begins with a review of the theoretical literature, which is actually quite limited in scope. Moreover, it is also ambivalent about the links between liberalisation and growth: in some circumstances it can be growth promoting; in others it could equally well be growth retarding. Empirically, this is also a complicated area. Liberalisations are not always that straightforward to identify. This is partly because different analysts can have different concepts in mind (straightforward tariff liberalisation, reducing anti-export bias, replacing less efficient with more efficient instruments of protection), and partly because there is no straightforward and widely accepted metric of liberalisation. Moreover because response lags vary from economy to economy, knowing exactly when to take a snap-shot is a further complication. Notwithstanding this, there is now a large empirical literature which the authors review. In a sense outputs from the empirical literature mirror the ambiguity in theoretical models – in some cases liberalisation seems to have been associated with faster growth, in others it has not. But actually the literature tells us more than that: it also tells us something about the factors likely to be associated with success, in particular,

Introduction and Overview 3

initial conditions and sequencing of reforms. Recent work also points to the importance of distinguishing between short-run and long-run effects.

1.2 Cross-border investment and productivity Cross-border investment (or Foreign Direct Investment (FDI)) is a key driver of globalisation and its growth has accelerated over the last two decades. By definition FDI is footloose and Governments the world over invest public funds in an attempt to attract FDI motivated by the hope that, as well as bringing jobs and income, their activities promote productivity growth through spillovers of one form or another. The three chapters in this part of the volume focus on aspects of this relationship. Ben Ferrett centres on theoretical perspectives. His starting point is that one can envisage two mechanisms via which firm productivity can increase, either by acquiring new technologies or ways of doing things from foreign firms, or by developing these in competition with foreign firms. He reviews general equilibrium perspective on the FDI/productivity relationship. He shows how in an imperfectly competitive setting one can account for the influence of endogenous Research and Development (R&D) decisions and specialisation. However, such models miss many of the microeconomic features of FDI and multinationals which we know to be important in practice. So he then goes on to illustrate how a range of partial equilibrium perspectives can yield insight into the strategic effects of spillovers on choices between FDI and exporting; and interactions between FDI flows and R&D investments. In the final part of his chapter, Ferrett disaggregates FDI into greenfield and acquisition to examine their relationships to productivity growth. He examines both the links between these alternative forms of FDI flows and industry performance; and how FDI flows help shape national productivity distributions across plants. Both can inform empirical investigation. In Chapter 5 Sourafel Girma, Steve Thompson and Peter Wright review empirical evidence on foreign takeovers and productivity growth. Their starting point is the oft-cited empirical observation that multinational firms are generally more productive than domestic firms, which is generally rationalised by reference to the exploitation of firm specific proprietary assets. The latter might be access to superior technology or better organisational and managerial inputs. These confer advantages over domestic firms which manifest in higher productivity. Girma, Thompson and Wright’s objective is to establish whether the

4

Holger Görg, David Greenaway and Richard Kneller

evidence really does confirm that multinationals have a real productivity advantage over domestic firms. They begin by asking whether methodology and data matter and their review of the literature concludes that they do. In other words, compositional effects and omitted variables can explain a good part of the apparent productivity advantage of Multinational Enterprises (MNEs). However, they also conclude that even when one controls appropriately for all of the factors which can influence productivity differentials, on average a productivity advantage is still evident. In Chapter 6 Holger Görg and Alexander Hijzen shift the focus away from relative productivity to the issue of whether the productivity advantages of foreign firms spill over to domestic firms. It seems reasonable to suppose that policy makers believe that they do – otherwise how would one explain the significant amounts of taxpayers’ monies being deployed by most governments in trying to attract MNEs. One can think of reasons why spillovers occur: through learning, competition effects or labour mobility; but what does the evidence tell us? Görg and Hijzen review the broader cross-country evidence, then focus specifically on the United Kingdom. With respect to the former, a huge literature has emerged, but is inconclusive, in part because it is only in recent years that robust tools have been applied to investigating microlevel data. The work they report on in the United Kingdom is based on micro level data and the results here suggest they are more likely to deliver positive spillovers.

1.3 Exporting and productivity The literature on links between exporting and productivity has exploded in recent years, with a growing number of empirical studies on a growing number of countries, and the development of theoretical models which embed firm level exporting activity. In Chapter 7 Rod Falvey and Zhihong Yu focus on theoretical models of links between exporting and productivity. They start from the observation that, at the aggregate level this has been a focus for academic and policy-focused research for a long time, whilst it is only in recent years that firm level links have been subject to similar scrutiny. Several important models have emerged identifying channels whereby entry to export markets stimulates higher productivity growth at the firm level. One such channel is self-selection: because export market entry incurs fixed costs (associated with market research, setting up distribution networks etc.), firms have to become more productive before they enter export

Introduction and Overview 5

markets. Another possible channel is learning by exporting: exposure to foreign competition leads to the adoption of new technologies, new techniques, and so on, leading to a productivity boost after entry. Falvey and Yu also review and evaluate new models which link exporting and industry productivity growth. Here the key channels are competitive pressures which force the least productive firms to exit; allow the most productive firms to expand; and reallocates resources towards more productive firms. These reallocation effects are not only embedded in general equilibrium models with heterogeneous firms, but also in models where comparative advantage drivers are also present. In Chapter 8 the emphasis switches to empirical analysis, with a contribution from David Greenaway and Richard Kneller on exporting and productivity in the United Kingdom. They ask whether exporting firms are different and do have higher productivity than non-exporters and whether there is evidence of learning by exporting in the United Kingdom. Greenaway and Kneller report evidence which supports self-selection – in other words exporting firms do appear to be more productive prior to entry. They also find that clusters are important – being co-located in the same industry and/or region raises the likelihood of export market entry. They also report some evidence pointing to post-entry effects, suggesting that learning by exporting may indeed take place. In the final chapter, Sourafel Girma, Richard Kneller and Mauro Pisu begin by noting that exports and FDI have grown contemporaneously, which raises the question: are they growing contemporaneously because they are alternative routes for firms which are globalising or are they complementary? In the case of the former, firms engage in one or the other; in the case of the latter they engage in both. Girma, Kneller and Pisu explain how many theories, building on the proximity-concentration trade-off point to a substitute relationship. In recent years, however, new theories either emphasise strategic interaction or productivity heterogeneity to underpin models which predict a complementary relationship. The authors then go on to review the empirical literature and conclude that firm level evidence favours complementary – firms that both export and invest overseas, with the latter often preceding the former.

This page intentionally left blank

Part I Aggregate Productivity Growth

This page intentionally left blank

2 Globalisation and Aggregate Productivity Growth Michael Henry and Chris Milner

2.1

Introduction

The relationship between trade openness and growth is a contentious issue. Advocates of free trade and outward oriented trade policies have advanced both theoretical and empirical evidence demonstrating that greater openness results in better long-run economic performance, measured either in terms of higher per capita gross domestic product (GDP) or total factor productivity (TFP) growth. However, there are some researchers who remain sceptical in the face of this evidence. There are a number of issues which lie at the heart of the controversy. The first relates to the suitability of the indices commonly used in empirical trade and growth studies to proxy a country’s trade regime (Edwards, 1993; Rodrik, 1995; Rodriguez and Rodrik, 2000). For example, Rodrik (1995) argues that in most studies of openness and growth, ‘the trade regime indicator is typically measured very badly’ and ‘openness in the sense of lack of trade restrictions is often confused with macroeconomic aspects of the policy regime’ (p. 2941). Additionally, Pritchett (1996) finds that the commonly used trade policy measures are uncorrelated among themselves. The second, is the lack of good quality trade policy information with broad country and time coverage, particularly for developing countries, to construct satisfactory measures of trade policy (Edwards, 1998; Harrison and Hanson, 1999; Baldwin, 2003). Other concerns relate to the econometric models employed and the fragility of the results to alternative specifications and sensitivity checks (Levine and Renelt, 1992; Edwards, 1998; Rodriguez and Rodrik, 2000; Wacziarg, 2001); the time period and country coverage (Wacziarg, 2001); the direction of causality between trade and growth (Harrison, 1996) and the absence of a convincing theoretical framework linking commercial policy and trade orientation to growth (Edwards, 1993, 1998). 9

10 Michael Henry and Chris Milner

Missing so far from the debate is a concern with the way in which productivity growth is measured. Yet, until recently, there has been an intense debate within the narrower confines of the productivity literature over the magnitude of economy-wide productivity growth rates estimated for some of the South East Asian countries (see Young, 1995; Nelson and Pack, 1999). The wide variations in estimates produced by different researchers for the same countries, using the same data and for the same time period, leads Felipe (1999) to conclude that the resulting TFP growth estimates are largely conditioned on the assumptions made, estimation techniques employed and functional form specified. In light of this, the natural question to ask is, to what extent are the results from cross-country regressions on the trade openness–productivity relationship affected by how TFP growth is measured?

2.2 Measuring openness and TFP growth In this section, we use three alternative residual estimates of TFP growth to test the sensitivity of various measures of trade openness to these alternative estimates. To do so, we employ both cross section and panel data estimation. This allows comparisons with existing studies of both types. The often-cited studies of Dollar (1992), Edwards (1992, 1998) and Sachs and Warner (1995) are of the cross-sectional type, while in terms of panel estimations, the studies by Miller and Upadhyay (2000) and Dollar and Kray (2001) are examples. Harrison (1996) employs both methodologies. For our cross section analysis, we use data from Rodriguez and Rodrik (2000) to specifically test the robustness of their critique of Edwards (1998). Doubts however, have been expressed over results which find a positive association between greater openness and economic growth on the basis of studies which use cross-sectional averages or starting values for time series data (see Harrison, 1996). Harrison argues that the use of cross section data makes it impossible to control for unobserved countryspecific effects, possibly biasing the results. She further argues that longrun averages or initial values for trade policy variables – particularly in developing countries – ignore the important changes which have occurred over time for the same country. Additionally, Söderbom and Teal (2001) argue that cross section estimations while informative of the correlations that can be established from the data, are uninformative as to the determinants of growth unless convincing instruments can be found. On the issue of instrument validity, they point to Rodriguez and

Globalisation and Aggregate Productivity Growth 11

Rodrik’s finding that the instruments used by Edwards (1998) were generally not valid. In light of these arguments, we utilise pooled cross section, time-series data to broaden our investigation of the robustness of measures of openness to alternative measures of TFP growth. 2.2.1

Measurement of TFP growth

We use three alternative residual measures of TFP growth constructed using the estimated factor elasticities from an aggregate Cobb–Douglas production function. These three measures are representative of the existing residual TFP measures at the country level, generated on the basis of econometric estimation of an aggregate production function.1 Two of the measures are based on the polar assumptions of TFPGheterogeneity (TFPG-HET) and homogeneity, TFPG – Generalised Method of Moments (TFPG-GMM) respectively, in the production parameters across countries. Additionally, the latter measure also controls for the possibility that the factor inputs are endogenous. The third, is an intermediate measure (TFPG-AMG) where we allow for heterogeneity in the production parameters across countries but homogeneity in the factor elasticities (used to estimate TFP growth) within regions. This measure is based on the principle of the mean group estimator of Pesaran and Smith (1995) and is our preferred measure of TFP growth.2 Further, while the two heterogeneous type TFP measures, TFPG-HET and TFPG-AMG, are derived using the estimated factor elasticities from an aggregate production function that assumes constant returns to scale, the homogenous measure (TFPG-GMM) is derived using the estimated factor elasticities from a production function that assumes non-constant returns to scale. Moreover, the assumptions underlying the estimation of the aggregate production function reflect some of the competing arguments regarding econometric estimation of crosscountry regressions in general, and the production function in particular. One such debate pertains to the performance of heterogeneous type estimators vis-à-vis homogenous type estimators when the sample is manifestly heterogeneous in character. 2.2.2

Measures of trade openness

A large literature has evolved proposing and evaluating alternative measures to capture the concept of openness. Since the focus of this chapter, however, is to examine the robustness of the trade openness–TFP growth relationship to alternative measures of the latter, we employ several of the openness measures frequently used in the literature. For our

12 Michael Henry and Chris Milner

cross section analysis, we use the 9 indices of trade policy openness (distortions) and the composite index of openness employed first in Edwards (1998) and in the later critique of that study, among others, by Rodriguez and Rodrik (2000).3 The 9 indices are: (i) the Sachs–Warner Openness Index (SWOPEN); (ii) the World Development Report Outward Orientation Index (WDR); (iii) Leamer’s Openness Index (LEAMER); (iv) the Average Black Market Premium (BLACK); (v) the Average Import Tariff on Manufacturing (TARIFF); (vi) the Average Coverage for NonTariff Barriers (QR); (vii) the Heritage Foundation Index of Distortions in International Trade (HERITAGE); (viii) Collected Trade Taxes Ratio (CTR); (ix) Wolf’s Index of Import Distortions (WOLF). The composite index (PCF) is computed as the first principal component of (i), (iv), (v), (vi) and (ix). For our panel estimations, we use the ratio of exports plus imports to GDP (OPEN1); the Sachs–Warner openness index (SWOPEN); and the local price deviation from PPP (PRIDEV) as the measures of openness (trade orientation). The price deviation measure is a measure of the price level of GDP (%) in PPP prices, relative to the US dollar exchange rate.4

2.3 Cross section estimations and results Our starting point is the open economy TFP growth model postulated by Edwards (1998). The determinants of TFP growth are hypothesised to be initial conditions (initial levels of GDP per capita and human capital) and trade policy openness, such that: TFPGi 
0 
1GDP65i 
2HUMAN65i 
3OPENNESSi  i

(1)

where TFPG is proxied by three alternative measures of the average growth rate of total factor productivity for the decade 1980–90, GDP65 is the log of initial per capita GDP in 1965 which is intended to capture the existence of TFP (conditional) convergence and so its coefficient is expected to be negative, HUMAN65 is the log of the initial level of human capital (average years of schooling) in 1965 and is intended to reflect the fact that countries with a more developed educational system have a greater ability to innovate and absorb new ideas. Our a priori expectation is for the coefficient of this variable to be positive. OPENNESS is a variable proxied by alternative indices of trade policy openness (distortions) as well as a composite index (see Section 2.2), all of which enter Equation (1) sequentially. It is on the coefficients of these indices that we mainly focus. Our expectations for these coefficients are that they are

Globalisation and Aggregate Productivity Growth 13

consistent with the hypothesis that more openness (less trade distortions) is associated with higher rates of TFP growth.  is a disturbance term and i indexes 93 developed and developing countries. Edwards uses weighted least squares (WLS) and instrumental variable weighted least squares (IVWLS) to estimate Equation (1), and generally finds that countries with more open (less distorted) trade policies experience faster TFP growth. This finding, he argues, is robust to the use of openness indicator, estimation technique and functional form. He also finds a positive and significant coefficient for the initial level of human capital in all 19 regressions and support for conditional convergence in 18 of them. One of the criticisms levelled by Rodriguez and Rodrik (2000) against the findings of Edwards (1998) is that the latter’s claim of robustness of his regression results to alternative indicators of openness (distortions) is ‘largely an artifact of weighting and identification assumptions that seem … inappropriate’ (Rodriguez and Rodrik, 2000, p. 37). When they repeat Edwards’s regressions using the natural log of per capita GDP (1985) as the weighting variable (instead of per capita GDP (1985)), a number of the coefficients on the openness indicators have the ‘wrong’ sign, while some lose their statistical significance. They reason that by using WLS estimation, Edwards is presumably correcting for possible heteroscedasticity in the residuals. Further, by using per capita GDP as his weighting variable, he is implicitly assuming that the form of the scedastic function is known. To put aside doubts about the appropriateness of alternative assumptions regarding the nature of the scedastic function, Rodriguez and Rodrik (2000) re-estimate Edwards’s regressions with White’s heteroscedasticity consistent standard errors which are robust to the form of heteroscedasticity. From among the 19 different specifications, they find evidence that more openness is significantly associated with higher productivity growth for only three of the indices. These are: the Collected Taxes Ratio (CTR), the World Development Outward Orientation Index (WDR) and the Heritage Foundation Index of Trade Distortions (HERITAGE). Of these three, Rodriguez and Rodrik described the last two as subjective indices and thus are likely to suffer from the judgement biases of their authors. Rodriguez and Rodrik conclude that Edwards’s results are sensitive to the choice of the weighting variable. We therefore take their results as our starting point and first estimate Equation (1) by OLS with heteroscedasticity consistent standard errors. However, unlike Rodriguez and Rodrik who took Edwards’s measure of TFP growth as given and thus focused

14 Michael Henry and Chris Milner Table 2.1 Results of OLS regressions cross section estimations – dependent variable: TFP growth (average 1980–90) (TFPGAMG and TFPG (R & R)) Openness indicators (expected sign)

TFPG-AMG (1)

TFPG (R & R) (2)

1. SWOPEN () 2. WDR () 3. LEAMER () 4. BLACK () 5. TARIFF () 6. QR () 7. HERITAGE () 8. CTR () 9. WOLF () 10. PCF ()

0.0222** (2.57) 0.0112*** (4.86) 0.0035 (0.55) 0.0163** (2.16) 0.0121 (0.86) 0.0046 (0.50) 0.0076*** (3.15) 0.1654 (1.13) 2.1E-5 (0.16) 0.0070* (2.31)

0.0102 (1.54) 0.0068*** (3.67) 0.0041 (0.82) 0.0098* (1.79) 0.0114 (0.88) 0.0036 (0.43) 0.0064*** (2.87) 0.2676** (2.25) 4.1E-5 (0.36) 0.0043 (1.37)

N 52 32 44 76 68 67 58 45 53 35

Notes: Each row corresponds to two separate TFP growth regressions for the respective openness (distortion) indicator. Each equation also includes a constant and the log of GDP per capita in 1965, and schooling in 1965 as additional regressors. Their values, however, are not reported due to space limitations. The numbers in parentheses are t-statistics based on heteroscedasticity consistent standard errors. *** means significant at 1%; ** means significant at 5%; * means significant at 10%. Source: Authors’ estimates; Rodriguez and Rodrik (2000), Table V.I.

their critique mainly on the proxies of trade policy openness, our emphasis is primarily on the left hand side variable: TFP growth.5 Table 2.1 presents results from cross section estimations using our preferred measure of TFP growth, TFPG-AMG, and the nine indicators of openness (distortions) as well as the composite index (PCF). Generally, Rodriguez and Rodrik’s results (column 2) are robust to our measure of TFP growth based on OLS estimations. Of the four indicators of trade policy openness they find to be significantly correlated to TFP growth, three are also shown to be significantly correlated to the augmented mean group measure of TFP growth (TFPG-AMG). However,

Globalisation and Aggregate Productivity Growth 15

some of the proxies of trade policy openness appear to be sensitive to the way TFP is measured. For example, we find both the Sachs–Warner openness index and the composite openness measure constructed by Edwards (1998) to be significantly related to TFPG-AMG. In contrast, using the TFP growth estimates derived in Edwards (1998), Rodriguez and Rodrik do not. This pattern of results is reversed for the collected taxes ratio (CTR) of openness (distortion). To further explore these findings we present results using the two alternative TFP measures. These are shown in Table 2.2 along with those of Rodriguez and Rodrik (2000) for comparative purposes. The sensitivity of the openness proxies identified above to alternative measures of TFP

Table 2.2 Results of OLS regressions cross section estimations – dependent variable: TFP growth (average 1980–90) (TFPG (R & R), TFPG-HET and TFPG-GMM) Openness indicators (expected sign) 1. SWOPEN () 2. WDR () 3. LEAMER () 4. BLACK () 5. TARIFF () 6. QR () 7. HERITAGE () 8. CTR () 9. WOLF () 10. PCF ()

TFPG (R & R) (1)

TFPG-HET (2)

TFPG-GMM (3)

0.0102 (1.54) 0.0068*** (3.67) 0.0041 (0.82) 0.0098* (1.79) 0.0114 (0.88) 0.0036 (0.43) 0.0064*** (2.87) 0.2676** (2.25) 4.1E-5 (0.36) 0.0043 (1.37)

0.0096 (1.46) 0.0172*** (7.36) 0.0051 (0.66) 0.0222*** (3.10) 0.0140 (1.21) 0.0042 (0.38) 0.0074* (1.85) 0.0072 (0.04) 0.2E-5 (0.72) 0.0071** (2.07)

0.019** (2.24) 0.014*** (6.57) 0.005 (0.91) 0.018** (2.34) 0.017 (1.45) 0.008 (0.90) 0.008*** (2.81) 0.124 (0.91) 0.5E-4 (0.44) 0.007** (2.24)

N 52 32 44 76 68 67 58 45 53 35

Notes: Each row corresponds to three separate TFP growth regressions for the respective openness (distortion) indicator. Each equation also includes a constant and the log of GDP per capita in 1965, and schooling in 1965 as additional regressors. Their values, however, are not reported due to space limitations. The numbers in parentheses are t-statistics based on heteroscedasticity consistent standard errors. *** means significant at 1%; ** means significant at 5%; * means significant at 10%. Source: Authors’ estimates; Rodriguez and Rodrik (2000), Table V.I.

16 Michael Henry and Chris Milner

growth appears to be confirmed by the results shown in Table 2.2. Now the Sachs–Warner measure of openness is statistically insignificant in the regression with the completely heterogeneous TFP measure (TFPGHET) but significant in the regression using the GMM-based measure of TFP (TFPG-GMM) as the dependent variable. In contrast to Rodriguez and Rodrik, the composite openness indicator (PCF) is significantly correlated to the two alternative TFP measures while for the collected taxes ratio (CTR) it is insignificant. To control for the possibility of causality running from TFP growth to trade openness rather than vice versa, Edwards also performed IVWLS estimations. These were also replicated by Rodriguez and Rodrik, but again because of concerns over the choice of weighting variable used in the original study, the latter authors chose the results obtained from IV (2SLS) regressions with robust standard errors as their preferred results. Once again we take Rodriguez and Rodrik’s results as our starting point and therefore undertake IV (2SLS) estimations with robust standard errors for our three measures of TFP growth. The results from these estimations are presented in Tables 2.3 and 2.4, respectively. The results in column (1) of Table 2.3, which uses TFPG-AMG as the measure of TFP growth, closely mirror the results obtained on the basis of OLS estimations and by extension the assumption that the trade indicators are exogenous. Consequently, Rodriguez and Rodrik’s results are generally robust to the augmented mean group TFP measure even when the openness indicators are assumed endogenous. Interestingly, the signs on the openness coefficients correspond exactly for the two TFP measures. However, as occurred when we assumed the measures of openness to be exogenous, there is evidence of sensitivity in the relationship between some openness indicators and TFP growth. For example, the average black market premium loses statistical significance in Rodriguez and Rodrik’s IV estimation compared to their OLS estimation. For us, this variable is still shown to significantly explain variations in cross-country TFP growth. In the case of the Sachs–Warner index, the sensitivity appears to be caused by a combination of the measure of TFP, and whether one assumes this particular openness index to be exogenous or endogenous. Recall, however, that Rodriguez and Rodrik find it to be insignificant regardless of the assumption made. For us, however, the assumption about its exogeneity or endogeneity seems important for our preferred measure of TFP growth. Table 2.4 broadens the comparison by showing the results obtained when using the alternative TFP measures: TFPG-HET and TFPG-GMM. The results further highlight the sensitivity of some of the trade openness measure to alternative measures of TFP. For example, when we use the

Globalisation and Aggregate Productivity Growth 17 Table 2.3 Instrumental variables (2SLS) regressions cross section estimations – dependent variable: TFP growth (average 1980–90) (TFPG-AMG and TFPG (R & R)) Openness indicators (expected sign)

TFPG-AMG (1)

TFPG (R & R) (2)

1. SWOPEN () 2. WDR () 3. LEAMER () 4. BLACK () 5. TARIFF () 6. QR () 7. HERITAGE () 8. CTR () 9. WOLF ()

0.0131 (1.62) 0.0167*** (3.42) 0.0039 (0.21) 0.0173** (2.39) 0.0095 (0.27) 0.0323 (0.37) 0.0240*** (3.85) 1.3072 (1.16) 3.8E-4 (1.23)

0.0078 (1.06) 0.0126** (2.13) 0.0033 (0.32) 0.0027 (0.54) 0.0079 (0.28) 0.0401 (0.79) 0.0202*** (3.24) 1.8368 (1.06) 3.3E-4 (1.21)

N 48 30 43 70 63 62 56 42 51

Notes: Each row corresponds to two separate TFP growth regressions for the respective openness (distortion) indicator. Each equation also includes a constant, the log of GDP per capita in 1965 and schooling in 1965 as additional regressors. Their values, however, are not reported due to space limitations. The numbers in parentheses are t(z)-statistics based on heteroscedasticity consistent standard errors. *** means significant at 1%; ** means significant at 5%; * means significant at 10%. The instruments used (not all in the same equation) are: TFP growth in the 1970s (for each TFP measure), SWOPEN in the 1970s, the ratio of imports/GDP in the 1970s, the ratio of exports/GDP in the 1970s, the average black market premium in the 1970s, the Heritage Foundation index of property rights protection and the change in the terms of trade. Source: Authors’ estimates; Rodriguez and Rodrik (2000), Table V.I.

completely heterogeneous measure (TFPG-HET), Leamer’s openness index (LEAMER) is highly significant (at less than 1 per cent). In contrast, it is not significantly different from zero when the GMM-based measure is used. Additionally, the latter TFP measure yields the unexpected result that higher average tariffs on manufacturing (TARIFF) leads to higher TFP growth. Moreover, the average black market premium is again found to

18 Michael Henry and Chris Milner Table 2.4 Instrumental variables (2SLS) regressions cross section estimations – dependent variable: TFP growth (average 1980–90) (TFPG (R & R), TFPG-HET and TFPG-GMM) Openness indicators (expected sign) 1. OPEN () 2. WDR () 3. LEAMER () 4. BLACK () 5. TARIFF () 6. QR () 7. HERITAGE () 8. CTR () 9. WOLF ()

TFPG (R & R) (1)

TFPG-HET (2)

TFPG-GMM (3)

0.0078 (1.06) 0.0126** (2.13) 0.0033 (0.32) 0.0027 (0.54) 0.0079 (0.28) 0.0401 (0.79) 0.0202*** (3.24) 1.8368 (1.06) 3.3E-4 (1.21)

0.0097 (1.50) 0.0259*** (3.34) 0.0346*** (3.30) 0.0274*** (2.99) 0.0554 (1.34) 0.0613 (1.19) 0.0240*** (2.93) 1.1542 (0.92) 0.0002 (0.75)

0.0117* (1.66) 0.0264*** (3.62) 0.0047 (0.26) 0.0203*** (2.72) 0.0790** (2.09) 0.0631 (1.45) 0.0141*** (2.79) 0.1857 (0.29) 2.5E-5 (0.13)

N 48 30 43 70 63 62 56 42 51

Notes: Each row corresponds to three separate TFP growth regressions for the respective openness (distortion) indicator. Each equation also includes a constant, the log of GDP per capita in 1965 and schooling in 1965 as additional regressors. Their values, however, are not reported due to space limitations. The numbers in parentheses are t(z)-statistics based on heteroscedasticity consistent standard errors. *** means significant at 1%; ** means significant at 5%; * means significant at 10%. The instruments used (not all in the same equation) are: TFP growth in the 1970s (for each TFP measure), SWOPEN in the 1970s, the ratio of imports/GDP in the 1970s, the ratio of exports/GDP in the 1970s, the average black market premium in the 1970s, the Heritage Foundation index of property rights protection, and the change in the terms of trade. Source: Authors’ estimates; Rodriguez and Rodrik (2000), Table V.I.

be a significant variable explaining differences in productivity growth when using the two alternative TFP measures. With respect to their IV regressions, Rodriguez and Rodrik allude to the sensitivity of the only two significant coefficients – WDR and HERITAGE – to the specification of the instrument lists. They further indicate that these two regressions are among the only three in which the Heritage Foundation Index of Property Rights Protection was used as an instrument by Edwards (1998), and for which the Sargan test of overidentification shows the instruments to be valid. Similarly for us, the

Globalisation and Aggregate Productivity Growth 19

coefficients on the trade policy indicators in our IV estimations are sensitive to the instrument lists specified. Consistent with the finding of Rodriguez and Rodrik, only the regressions with the WDR and HERITAGE indices did the Hansen J (chi-square) statistic fail to reject the null hypothesis that the instruments are valid. Our overall conclusion from the cross section estimations is that the results obtained by Rodriguez and Rodrik (2000) based on their critique of Edwards (1998) are generally robust to our preferred measure of TFP growth and to a lesser extent the two alternative measures of TFP constructed by us. However, some trade policy indicators appear to be sensitive to the way TFP is measured. In terms of the more commonly used openness measures in the trade and growth literature, these include the SWOPEN, BLACK and to a lesser extent the CTR and the TARIFF. In light of the latter findings, some of Rodriguez and Rodrik’s analysis and conclusions are therefore sensitive to how TFP growth is measured. For example, while Rodriguez and Rodrik (2000) find BLACK to be insignificant in their TFP growth regressions, we find this openness measure to be statistically significant in the different regressions using all three measures of TFP growth. Thus both for our preferred measure of TFP growth and the two alternative measures there is support for a positive trade liberalisation effect on TFP growth for more measures or proxies of liberalisation than those found in Rodriguez and Rodrik (2000). Finally, for our three measures of TFP growth there is no consistent support for conditional convergence and the initial level of human capital being a significant determinant of TFP growth.6 To determine whether this pattern of results holds across estimation methodology, we now extend our robustness checks to estimations based on pooled cross section, time-series data.

2.4 Panel estimations and results We specify the following estimating equation, that mirrors Equation (1):7 TFPGit  0  1GDP0it  2HUMAN0it  3OPENNESSit 

6

6

兺  兺

j1

j

t1

t

 it

(2)

where TFPG once again proxies three alternative measures of TFP growth, GDP0 is the log of per capita GDP at the beginning of each 5-year period from 1960–90, HUMAN0 is the log of human capital at the start of each 5-year period, OPENNESS is as defined in Equation (1) but

20

Michael Henry and Chris Milner

in this instance proxied by three aspects of trade openness (see Section 2.2), i indexes countries, t indexes time periods, j is a region-specific fixed effect, t is a period-specific fixed effect and it a common independent and identically distributed term. The sample comprises a maximum of 79 developed and developing countries and covers the period 1960–90. Unlike as for the cross section estimations where the openness indicators were entered into the regressions sequentially, for our panel data analysis we consider the nexus between measures of trade openness (orientation) and TFP growth simultaneously. The advantage of this is that since all the trade indicators used as proxies for openness do not capture the same aspect of trade policy, then considering them simultaneously reduces the possibility of information being lost.8 This is the approach adopted by Miller and Upadhyay (2000). The disadvantage however, is that because most of the existing trade policy measures do not have a lengthy time dimension and broad country coverage, the number of openness measures available for our analysis is substantially reduced. We estimate Equation (2) first using the fixed effects (within) estimator and then the feasible efficient two-step GMM-IV estimator. Additionally, together with using this estimator, we also allow for unspecified correlation of error terms within groups (i.e. countries) but not across groups. Results for both sets of panel estimations are shown in Table 2.5. Taking the fixed effects regressions first (i.e. regressions 1–3), both the trade intensity and trade policy measures of openness have the expected signs in all three regressions and are significant at the 10 per cent level or lower in two of them. Only in regression 1 is there evidence of sensitivity between one of the trade openness and the way TFP is measured. This relates to the price distortion measure which is not significantly different from zero when the augmented mean group measure of TFP (TFPGAMG) is used. Therefore, when the trade measures are assumed exogenous then they are largely robust to alternative measures of TFP growth. The results for OPEN1 and PRIDEV in regression 3 are consistent with the findings of Miller and Upadhyay (2000) who, on the basis of fixed effects estimations, find both the share of exports in GDP and the local price deviation from PPP to be significantly correlated with the level of TFP for a sample of 83 countries over a similar time period. Finally, with regard to the variables proxying for the effects of initial conditions on TFP growth, the results are mixed. There is support for conditional convergence in regressions 1 and 3, and for the initial stock of human capital at the start of each period explaining differences in TFP growth in regressions 1 and 2. However, in regressions 1 there is an unexpected negative sign on the variable that proxies for human capital.9

Globalisation and Aggregate Productivity Growth 21 Table 2.5 Panel estimates (1960–90) 5-year averages – dependent variable: TFP growth Fixed effects TFPGAMG (1) GDP0 HUMAN0 OPEN1 (X  M/ GDP) SWOPEN

TFPGHET (2)

GMM-IV TFPGGMM (3)

0.0293*** 0.0039 0.0289*** (2.91) (0.92) (3.04) 0.0138* 0.0133*** 0.0059 (1.89) (3.61) (0.82) 0.0254*** 0.0085*** 0.0207*** (3.40) (2.58) (2.94)

PRIDEV Trade measures endogenous? Hansen J 2 (p-value) R2 # of observations # countries

TFPGAMG2 (4) 0.0036 (1.40) 0.0019 (0.95) 0.0002 (0.08)

TFPGHET (5)

TFPGGMM (6)

0.0017 0.0034 (0.45) (1.17) 0.0006 0.0025 (0.20) (1.19) 0.0027 0.0001 (1.03) (0.04)

0.0098** 0.0064* 0.0122** 0.0133*** 0.0092* 0.0129** (2.02) (1.86) (2.55) (2.57) (1.93) (2.47) 0.0168 0.0083** 0.0187* 0.0070 0.0209*** 0.0076 (1.42) (2.48) (1.69) (1.24) (3.41) (1.27) NO NO NO YES YES YES

0.40 463

0.74 463

0.39 463

0.048 (0.827) 0.22 384

79

79

79

79

30.637 (0.000) 0.36 384

0.244 (0.621) 0.22 384

79

79

Notes: The numbers in parentheses are robust t (fixed effects regressions) and z (GMM-IV regressions) statistics. *** means significant at 1%; ** means significant at 5%; * means significant at 10%. All variables are in logs except the SWOPEN. All regressions include a constant as well as country and period dummies. These are not reported due to space constraints. The instruments for GMM-IV regressions are: TFP growth, the log of the share of trade in GDP (OPEN 1), the SWOPEN, and the log of PPP, all lagged one period (5 years), as well as region and period dummies.

However, before firm conclusions about the significance (or insignificance) of the trade variables in explaining differences in TFP growth, we first control for the possibility of the trade measures being endogenous. This possibility is particularly strong for the trade share measure of openness (OPEN1) which can be heavily influenced by geographical factors. In other words, causality can be going from productivity to trade, not vice-versa. The problem of reverse causation however, is not limited to outcome measures such as trade volumes. Trade policies can

22 Michael Henry and Chris Milner

themselves be endogenous to growth with countries perhaps more likely to liberalise during a period of higher economic growth. In light of the above arguments, we also estimate Equation (2) for the three measures of TFP growth using the GMM-IV estimator. The results from these are shown as regressions 4–6 in Table 2.3. As the table shows, there is a marked difference in these results compared to those obtained when we assumed the trade measures to be exogenous. OPEN1 is now insignificantly correlated with TFP growth for all three measures of TFP and has the ‘wrong’ sign for two of the three measures. These results differ from those obtained by Dollar and Kray (2001) for this variable, based on per capita growth regressions with panel data. It should be noted however, that the latter authors used the conventional IV estimator and also estimated a different model to ours. Further, despite having the expected sign, the price distortion measure (PRIDEV) is only significant when we employ the completely heterogeneous measure of TFP growth. For the other two TFP measures, this variable loses significance and the magnitude of its coefficient falls. In fact, from among the three trade indicators, the SWOPEN appears to be the most robust. This variable is statistically significant at the 1 per cent and 5 per cent levels for the regressions with the intermediate and homogenous TFP measures respectively, and is almost significant at the 10 per cent level for the completely heterogeneous measure. Further, the value of its coefficient is less variable than the other two trade indicators. Based on the point estimates obtained in regressions 5 and 6, it can be argued that countries that are open have on average a TFP growth rate of between 1.3 and 1.5 per cent higher than countries that are closed based on the Sachs and Warner (1995) criteria. This compares with the average per capita GDP growth effect of 2.7 per cent found by Greenaway et al. (2002) for this openness index based on their Levine and Renelt type ‘core variables’ specification. Additionally, the finding that SWOPEN is the most robust of the trade measures is consistent with Sala-i-Martin (1997). In proposing an alternative robustness test to Levine and Renelt (1992), he finds this measure to be the only significant openness measure when confidence intervals were constructed for the entire distribution of coefficients for different growth determinants. Finally, both initial level of GDP and initial stock of human capital are statistically insignificant. Further, based on Hansen’s test for the validity of overidentifying restrictions, only the regressions using the intermediate (mean group) and homogeneous measures of TFP growth are validly overidentified and hence have valid sets of instruments.

Globalisation and Aggregate Productivity Growth 23

To complete our empirical analysis, we use annual data. This requires us to modify Equation (2) by dropping the two variables proxying for initial conditions. We however include a contemporaneous measure of human capital that allows us to test whether this variable has a significant impact on TFP growth (as suggested by Benhabib and Spiegel, 1994; Islam, 1995) rather than on output growth. This model is similar to the core model of Miller and Upadhyay (2000) but with fewer trade indicators. Once again we undertake both within (fixed effects) and GMM-IV estimations. The results from these estimations are shown in Table 2.6. Taking the fixed effects estimations first, all the openness measures have the expected signs and are all statistically significant at the 10 per cent level at least, except for the trade share measure when the intermediate TFP measure is employed. Therefore, for these regressions the openness variables are generally robust to the choice of TFP measure. However, only for the heterogeneous measure of TFP growth (TFPG-HET) is human capital coefficient ‘rightly’ signed and significantly correlated with TFP growth. Turning now to the IV estimations where we assume all the variables (both trade and human capital) to be endogenous, all the openness measures once again have their expected signs. However, similar to the

Table 2.6 Panel estimates (1960–90) annual – dependent variable: TFP growth Fixed effects

GMM-IV

(Expected sign)

TFPG-HET (1)

TFPG-MG (2)

TFPG-GMM (3)

TFPG-HET (4)

TFPG-MG (5)

TFPG-GMM (6)

HUMAN () OPEN1 (X  M/GDP) () SWOPEN () PRIDEV ()

0.0103*** (3.94) 0.0106*** (5.74) 0.0060*** (3.44) 0.0055** (2.00)

0.0010 (0.10) 0.0107 (1.22) 0.0166** (2.53) 0.0309*** (2.17)

0.0004 (0.05) 0.0128* (1.68) 0.0168*** (3.38) 0.0293** (2.21)

0.0001 (0.02) 0.0019 (0.62) 0.0106** (2.53) 0.0124** (2.24)

0.0008 (0.33) 0.0020 (0.75) 0.0108*** (2.98) 0.0106** (2.04)

0.0003 (0.12) 0.0011 (0.69) 0.0100*** (2.73) 0.0122** (2.37)

NO

NO

NO

YES

YES

YES

2,057 79 0.55

2,057 79 0.25

2,057 79 0.14

1,897 79 0.26 35.252 (0.000)

1,897 79 0.24 1.303 (0.253)

1,897 79 0.10 0.512 (0.474)

Regressors endogenous? # of observations # countries R2 Hansen J 2 (p-value)

Notes: The numbers in parentheses are robust t (fixed effects regressions) and z (GMM-IV regressions) statistics. *** means significant at 1%; ** means significant at 5%; * means significant at 10%. All variables are in logs except the SWOPEN. All regressions include a constant as well as country and time dummies. These are not reported due to space constraints. The instruments for the GMM-IV regressions are: TFP growth, the log of human capital, the log of the share of trade in GDP (OPEN1), the SWOPEN, and the log of the price deviation from PPP, all lagged two periods, as well as region and time dummies.

24 Michael Henry and Chris Milner

GMM-IV estimations based on 5-year averages of the data, OPEN1 is insignificantly correlated with TFP growth for all three measures of this variable. In contrast, both SWOPEN and PRIDEV are statistically significant at the 5 per cent level at least, for all three TFP measures. Therefore, it appears that for annual data, the latter two trade policy measures are robust to the measurement of TFP growth and also to the assumption of exogeneity or endogeneity. For the SWOPEN this is consistent with the finding on the basis of five-yearly data. All three TFP growth regressions pass the instrument validity test. Based on the point estimates for SWOPEN in these estimations, economies that are open have on average a TFP growth rate of between 0.9 and 1.1 per cent higher than economies that are closed. Finally, human capital is not significantly associated with TFP growth in all three regressions. The mixed results for human capital, including the negative coefficient, are consistent with the findings of many researchers in this literature (see Pritchett, 2001).

2.5 Conclusions The manner in which trade policies affect economic performance has generated a prolonged debate that shows little signs of abating, as well as a large empirical literature. Most of this literature has concluded that trade reforms which lead to an outward oriented trade regime are associated with superior economic performance either in terms of higher output or productivity (levels and growth). However, questions have been raised over the robustness of these findings to better measures of trade policy, the inclusion of other growth determinants in the specifying equations, appropriate econometric techniques and the like. In this chapter we explored another dimension of the robustness debate that has been neglected in the aggregate productivity growth literature, namely how productivity growth is measured. For our cross section estimations, using the results obtained by Rodriguez and Rodrik (2000) as a comparator, we find that while most of the measures of trade openness used in Edwards (1998) are generally robust to the three alternative measures of TFP growth derived on the basis of differing assumptions, some of the openness indicators are sensitive to alternative measures of TFP growth. These include SWOPEN and BLACK. On the basis of the relationship between the latter openness variables and TFP growth, we find support for a positive and significant effect from trade liberalisation on TFP growth for more of the proxies of trade policy openness compared to Rodriguez and Rodrik (2000). This is

Globalisation and Aggregate Productivity Growth 25

true both for our preferred measure of TFP growth and the two additional measures. In terms of our panel results, when the data are in 5-year periods only SWOPEN is robust to alternative measures of TFP and the alternative assumptions about the direction of causality between this variable and TFP growth. This finding is in direct contrast to that found for this variable based on cross section estimations. On the other hand, the price deviation measure is sensitive in both dimensions outlined above. Finally, the outcome measure (OPEN1) is robust in one dimension only, that is, to the way TFP is measured. For the three TFP measures, this measure is shown to be positive and significantly correlated with TFP growth. However, when we assume it to be endogenous it is statistically not different from zero in the three TFP growth regressions. For the panel estimations using annual data, the findings on both the SWOPEN and the local price deviation from PPP are robust to the three alternative measures of TFP growth and to the estimator used. For the outcome measure of openness (OPEN1), robustness is again established in one dimension as in the panel estimations using five-yearly data. Overall, one can conclude that the precise openness–productivity growth relationship identified in cross-country analysis is sensitive to how both openness and TFP growth are measured, to whether the model is estimated by cross section or panel methods, whether we use data in 5-year periods or annually, and whether openness is treated as an exogenous or endogenous variable. Our own preference is to use panel estimation methods and to include several alternative dimensions of trade openness and distortion simultaneously (some of which are more likely to be endogenous than others). It is difficult to conclude other than that our panel estimates provide support for the view that reduced trade policy distortion increases aggregate productivity growth, but that this effect may well not be appropriately captured by the standard trade to GDP measure of openness. Finally, over the different types of estimations (cross section and panel) there is no consistent support for conditional convergence and the initial level of the stock of human capital is a significant determinant of end of period TFP growth.

Notes 1. We employ the stochastic growth accounting methodology for measuring TFP growth rather than the non-stochastic approach (see Fischer, 1993; Collins and Bosworth, 1996).

26 Michael Henry and Chris Milner 2. The data for constructing these TFP growth estimates were obtained from the World Bank STARS database. 3. We thank Professor Dani Rodrik for making the data used in Edwards (1998) and Rodriguez and Rodrik (2000) available to us. 4. Data for these measures were obtained from the World Bank World Development Indicators CD ROM 2000 and the Penn World Tables (Mark 5.6) of Summers and Heston (1995). 5. Edwards (1998) derived estimates of TFP growth as residuals following a random effects estimation of a Cobb–Douglas aggregate production function with the factor elasticities constrained to sum to one (i.e. constant returns to scale is assumed). 6. Rodriguez and Rodrik (2000) did not report the values of the coefficients on the initial levels of GDP and human capital in the presentation of their results. 7. Our decision to include the initial levels of GDP and human capital (for each sub-period) in our estimating equation means that for this specification we are constrained to perform estimations using data based on period averages (5 years in our case) rather than annually. We do however modify the estimating equation in order to undertake panel estimations based on annual data. 8. Edwards (1998) sought to offset this potential loss of information by using a principal components approach to construct a ‘grand’ composite index. 9. Since the two variables proxying initial conditions are highly collinear (r  0.74), we checked whether the coefficient on both variables may be affected by the problem of multicollinearity. However, when we alternated in omitting one of the variables from our regressions while keeping the other, the results remained largely unchanged. Harrison (1996) cites the unsatisfactory results obtained when using initial GDP as the reason for excluding this variable from her fixed effects regressions based on five-yearly data.

References Baldwin, Robert (2003), ‘Openness and Growth: What’s the Empirical Relationship?’, NBER Working Paper 9578. Benhabib, Jess and M. Spiegel (1994), ‘The Role of Human Capital in Economic Development: Evidence from Aggregate Cross-Country Data’, Journal of Monetary Economics, 34(2), 143–73. Collins, Susan and B. Bosworth (1996), ‘Economic Growth in East Asia: Accumulation Versus Assimilation’, Brookings Papers on Economic Activity, 2, 135–91. Dollar, David (1992), ‘Outward-Oriented Developing Economies Really Do Grow More Rapidly: Evidence from 95 LDCs, 1976–1985’, Economic Development and Cultural Change, 40(3), 523–44. Dollar, David and Art Kray (2001), ‘Trade, Growth and Poverty’, Downloadable from www.worldbank.org/research/growth/Trade5.htm. Edwards, Sebastian (1992), ‘Trade Orientation, Distortions and Growth in Developing Countries’, Journal of Development Economics, 39, 31–57. Edwards, Sebastian (1993), ‘Openness, Trade Liberalization, and Growth in Developing Countries’, Journal of Economic Literature, XXXI(3), 1358–93. Edwards, Sebastian (1998), ‘Openness, Productivity and Growth: What Do We Really Know?’ The Economic Journal, 108, 383–98.

Globalisation and Aggregate Productivity Growth 27 Felipe, Jesus (1999), ‘Total Factor Productivity Growth in East Asia: A Critical Survey’, Journal of Development Studies, 35(4), 1–41. Fischer, Stanley (1993), ‘The Role of Macroeconomic Factors in Growth’, Journal of Monetary Economics, 32, 485–512. Greenaway, David, Wyn Morgan and Peter Wright (2002), ‘Trade Liberalisation and Growth in Developing Countries’, Journal of Development Economics, 67, 229–44. Harrison, Ann (1996), ‘Openness and Growth: A Time-Series, Cross-Country Analysis for Developing Countries’, Journal of Developing Economies, 48, 419–47. Harrison, Ann and G. Hanson (1999), ‘Who Gains from Trade Reform? Some Remaining Puzzles’, Journal of Development Economics, 59, 125–54. Levine, Ross and David Renelt (1992), ‘A Sensitivity Analysis of Cross-country Growth Regressions’, American Economic Review, 82(4), 942–63. Miller, Stephen M. and Mutki P. Upadhyay (2000), ‘The Effects of Openness, Trade Orientation, and Human Capital on Total Factor Productivity’, Journal of Development Economics, 69, 399–423. Nelson, Richard and Howard Pack (1999), ‘The Asian Growth Miracle and Modern Growth Theory’, Economic Journal, 109, 416–36. Pesaran, M.H. and R. Smith (1995), ‘Estimating Long-run Relationship from Dynamic Heterogeneous Panels’, Journal of Econometrics, 68, 79–112. Pritchett, Lant (2001), ‘Measuring Outward Orientation in Developing Countries: Can it be Done?’, Journal of Development Economics, 49(2), 307–35. Rodriguez, Francisco and Dani Rodrik (2000), ‘Trade Policy and Economic Growth: A Skeptic’s Guide to the Cross-National Evidence’. Downloadable from http://www//ksghome.harvard.edu/~.drodrik.academic.ksg/papers.html. Rodrik, Dani (1995), ‘Trade Policy and Industrial Policy Reform’, in Jere Behrman and T.N. Srinivasan, eds Handbook of Development Economics, 3B, Amsterdam: North Holland. Sachs, Jeffrey and Andrew Warner (1995), ‘Economic Reform and the Process of Global Integration’, Brookings Papers on Economic Activity, 1, 1–118. Sala-i-Martin, Xavier (1997), ‘I Have Just Ran Two Million Regressions’, American Economic Review, 87(2), 178–83. Söderbom, M. and F. Teal (2001), ‘Trade and Human Capital as Determinants of Growth’, WPS 2001–10, CSAE, University of Oxford. Summers and Heston (1995), ‘The Penn World Tables (Mark 5.6)’, World Development Indicators CD ROM 2000, World Bank. Wacziarg, Romain (2001), ‘Measuring the Dynamic Gains from Trade’, The World Bank Economic Review, 15(3), 393–429. Young, Alwyn (1995), ‘The Tyranny of Numbers: Confronting the Statistical Realities of the East Asian Growth Experience’, Quarterly Journal of Economics, 110, 641–80.

3 Liberalisation and Aggregate Productivity Growth David Greenaway and M. Emranul Haque

3.1 Introduction The focus of the previous chapter was potential linkages between openness and aggregate productivity growth. As Henry and Milner show, there is a large literature on this, stimulated by a longstanding (academic and policy) interest in the appropriate trade orientation strategy for developing countries. The empirical evidence on links between openness and productivity growth is controversial, with some influential commentators arguing that what one finds depends in large measure upon the way in which openness is measured. Henry and Milner actually ask a different question; namely, does the way in which productivity growth is measured make a difference and they show that it can. In this chapter we shift the focus back to openness, or to be more precise, changes in openness. More specifically, can we expect changes in openness, that is, liberalisation, to impact on productivity growth and economic growth? This is a non-trivial question and certainly not a question of purely academic interest. Over the last twenty years or so, well over one hundred developing countries have initiated liberalisation programmes; most have done so in the expectation of stimulating growth. In this chapter we first of all ask whether the theory gives us any firm grounds for expecting growth to follow trade liberalisation, and second, what we can deduce from available evidence.

3.2 International trade, trade liberalisation and growth: theory The conventional wisdom in favour of trade liberalisation is that it can lead to significant gains in productivity by promoting learning-by-doing, 28

Liberalisation and Productivity Growth 29

innovation, cost cutting and acquisition of new technology. Despite the intuitive appeal of these arguments, their analytical underpinnings were not always firm.1 For example, trade liberalisation advocates assert that tariffs raise entrepreneurial slack in import-competing industries because they raise the relative prices of these industries. But by the same logic, they could reduce slack in export-oriented industries. The net effect of trade protection on efficiency may, therefore not be so straightforward (see Kim, 2000). The advances in growth theory in the late 1980s enable us to address rigorously the possibility of a permanent change in productivity growth that may occur due to trade reforms. However, the blessings of the new growth theory have proved to be mixed (see the surveys by Rodrik, 1988, 1992 and Tybout, 1992). Even though we gain analytical rigour from these theories, they do not yield unambiguous predictions about possible links between openness, trade liberalisation and growth. By slightly changing key assumptions, such as the scope and extent of technology or knowledge spillovers, new growth theory could be used to support any set of possible outcomes between trade and economic growth. For example, if technological spillovers are national in scope, a country with a relative abundance of natural resources and unskilled labour could actually experience lower long-run growth if international specialisation encourages traditional activities and pulls resources away from more dynamic technology-intensive sectors. By contrast, if technological spillovers are international in scope, trade would have the opposite effect. 3.2.1 3.2.1.1

Theories of growth in an open economy context Learning-by-doing, trade and growth

Learning-by-doing is widely regarded as a key driver of growth. The simplest dynamic learning-by-doing model in an international context is Lucas (1988). He postulates a two-sector model with two consumption goods characterised by perfect competition. Learning is accidental in the sense that the workers learn and accumulate experience, or human capital, while working in a firm. That means the learning process is not an outcome of a deliberate choice on the part of workers. Human capital is sector specific, growth of which increases with the effort devoted to producing in that sector. One good is more ‘high-technology’ than the other and the elasticity of substitution is constant between them. In autarky the system converges to specialisation in one of the two goods if they are good substitutes. Alternatively, if the goods are poor substitutes,

30 David Greenaway and M. Emranul Haque

we get a stable stationary point with diversification in production because consumers prefer to consume positive quantities of both goods. Under free trade in the two final goods with a continuum of small countries (without government intervention), prices will equal world prices. Countries will tend to specialise, therefore each country will accumulate only the type of human capital specific to the good it produces. Countries do not converge in growth rates. This model implies that protection (at least for some initial phase of development) may be better than free trade. Here again equilibrium is not stable and depends on the elasticity of substitution. If the goods are good substitutes, producing (having a comparative advantage in) high-learning goods will lead to higher-than-average real growth. But in the case when goods are poor substitutes, the terms-of-trade effects of technological change dominate the direct effects on productivity, and the countries with rapid technological change experience the slowest real income growth. This model predicts that under certain conditions, it is possible to get support for the ‘infant industry’ argument for some initial stage of development. Eventually the country’s expertise grows to the point where it will have a comparative advantage in the ‘high-tech’ good under free trade and experiences higher growth. Stokey (1988, 1991) and Young (1991) develop models of learning technologies with north–south trade where ‘north’ means relatively well endowed with human capital. In such settings, labour is allocated to the sector with the highest current return, independent of learning rates. Under certain assumptions on consumer preferences, Stokey (1991) obtains a unique world equilibrium in which the south produces an interval of low quality goods while the north produces high quality goods under autarky. By contrast, under free trade poor countries import high quality goods, which reduce learning-by-doing initially. But in the long run, both countries will experience growth where poor countries will grow slower than rich countries. In a similar analysis, Young (1991) allows for spillovers across different industries. For example, advances in the pharmaceutical sector may benefit the biotech industry. There are many possible equilibria in Young’s framework, which depend on populations and the relative human capital holding of two sets of countries. A poor country upon entering free trade may experience a fall in the accumulation of human capital, because free trade reduces the rewards to the skilled labour of less developed countries (LDCs). There are possibilities of other equilibria in which the poor may catch up to the rich, but only when their larger population lets them enjoy greater scale economies.

Liberalisation and Productivity Growth 31

These models by Lucas, Stokey and Young have some interesting policy implications. Suppose a (poor) country has a long-run comparative advantage in the ‘high-technology’ good, but currently under autarky it has not reached its steady state and shows a short-run comparative advantage in the ‘low-technology’ good. If this country adopts a free trade policy, it will export the ‘low-technology’ good, become completely specialised in it and never produce the ‘high-technology’ good. The right policy for the government to achieve higher growth in the long run should be to restrict (or even prohibit) trade at first and let the economy adjust closer to its steady state. Trade can be liberalised only when the economy has reached its steady state and gained comparative advantage in a ‘high-technology’ good.2 3.2.1.2

Technological spillovers, trade and growth

In an influential study, Rivera-Batiz and Romer (1991a) have looked at a one-shot gain (a level effect) and a permanent growth effect of integration that takes place only between countries that are identical, that is, they do not consider the effects of economic integration between a poor LDC and a developed economy in order to focus on pure scale effects of integration. Suppose that two identical and closed economies are at their balanced growth paths. There are two types of manufacturing outputs: consumption goods and capital goods. Inputs are human capital, labour and a continuous set of capital goods. Research and Development (R&D) creates designs for new types of capital goods. In the knowledgedriven (KD) R&D model, human capital and knowledge are the only inputs for producing new designs. On the other hand, their lab-equipment (LE) model uses the same inputs in R&D and the manufacturing technology in the same proportions, that is, human capital, unskilled labour, and capital goods. Knowledge does not have any productive value in this model. Three cases of trade between the economies are considered: (i) free trade of goods but not ideas (no international knowledge spillover and complete patent protection in the world) in the knowledge driven (KD) model, (ii) free trade in goods and ideas (perfect international knowledge spillover) in the KD model, and (iii) free trade in ideas in the lab-equipment model. In the knowledge-driven model without international knowledge spillovers, opening trade in goods has no permanent effect on growth since the return to human capital doubles in both sectors and exactly offsets each other, and hence free trade in goods does not affect the composition of human capital between the sectors. In effect, the balanced growth rate does not change. However, in the case where the countries

32 David Greenaway and M. Emranul Haque

that have integrated start from completely non-intersecting sets of capital goods, both countries would enjoy higher levels of capital goods that would lead to higher output levels. As a result, free trade in goods in this model can affect the level of output and welfare, but not the long-run growth rates. For the KD model with free flows of ideas where free flows of goods is already in place, it is shown that two countries that engage in the exchange of ideas will experience an initial doubling of the stock of knowledge. This increased availability of ideas (or knowledge) for use in research will raise the productivity of human capital in research without having any effect on its productivity in the manufacturing sector. Even if there is no expansion of employment in the research sector, the growth rate of both countries will double. But because of higher profitability there will be expansion of employment in the research sector that will more than double the growth rate. Hence, with a free flow of ideas with free trade in goods, the model is able to generate a permanent growth effect.3 In the lab-equipment model, however, it is possible to generate a growth effect only through trade in goods. The reason is that the intraindustry trade increases the profitability of research, thus attracting more labour into the research sector and raising the saving rate. In a series of papers, Grossman and Helpman have studied the growth effects of trade (openness) through technological spillovers by considering two countries that are not identical, which is in contrast with RiveraBatiz and Romer. For example, in one of their models (Grossman and Helpman, 1991, ch. 6), they assume that there are two factors of production, skilled labour (human capital) and unskilled labour, which are supplied at a fixed rate in a small open economy. One of the final goods is produced with human capital and a fixed amount of differentiated, nontraded intermediate inputs, while the other is produced with unskilled labour and the same bundle of intermediate inputs. The nontraded intermediate inputs are produced under monopolistically competitive conditions with both factors of production. Constant returns to scale prevail for both final and intermediate goods. Human capital is also used in the R&D activities to create new varieties of intermediate goods. As each final good requires a fixed number of intermediate goods, more and more intermediate goods created by R&D bring about higher output. In addition to producing new varieties of intermediate goods, R&D also adds to the stock of knowledge, which is nonappropriable. The higher this knowledge stock, the less the quantity of human capital needed to produce each new variety of intermediate

Liberalisation and Productivity Growth 33

goods. Thus, the growth process is endogenous with R&D raising aggregate productivity by creating new intermediate inputs and adding to the stock of knowledge, which in turn, reduces the amount of human capital required for producing new varieties of inputs. The equilibrium outcome is a constant rate of output growth and a constant rate of growth of factor productivity in the final goods sectors. 3.2.1.3

Physical capital accumulation, trade and growth

Fischer (1995) uses the two-sector AK model of Jones and Manuelli (1990) in an overlapping generations economy, where the investment good is infinitely capital intensive (i.e. it employs no labour). In his model, individuals born with one unit of labour endowment live for two periods; work, save and consume when young but consume only when old. He considers two countries with identical technology and preferences, except with different time preferences. In particular, they have different savings rates. Under free trade, the country with higher savings would experience a higher capital–labour ratio and hence would tend to have a comparative advantage in the investment good. The less thrifty country (with less savings) would be completely specialised in the consumption good, while the other country would be diversified. In this situation, the less thrifty country will have a lower wage–rental ratio and all investment would be attracted to this country. This leads to an increasing capital–labour ratio in the less thrifty country, while the ratio would remain constant in the thrifty country. After a certain period, gradually both countries will diversify, when their capital–labour ratios would be close. When the countries reach an integrated world equilibrium, both will have the same capital–labour ratio that would grow at the same rate. But the savings rate would be the weighted average of the two countries, and hence their growth rate would be in between their autarkic growth rate, that is, the lower savings country would experience higher growth, while the higher savings country would experience lower growth. 3.2.2

Trade liberalisation and growth: theory

It is useful to start by defining trade liberalisation and openness to realise how they differ. Trade liberalisation implies ‘moving away from highly restrictive trade regimes’ while openness implies ‘permitting imports and exports at world prices, and going well beyond simply the easing or removal of restrictions on import’. In this sense, trade liberalisation may be thought of as a step towards opening up the economy.

34 David Greenaway and M. Emranul Haque

Rivera-Batiz and Romer (1991b) study the effects of imposing a tariff on all imported intermediate goods in two identical countries. They find that the growth rate is always lower under restrictions than under free trade. A tariff has two effects: a trade distortion effect and an R&D resource reallocation effect. When the two effects work in opposite directions, the size of tariff rate may determine their relative strength. As a result, the growth rate declines when the tariff rises from zero, but after some critical value of the tariff, the growth rate rises, but never reaches the free trade level. Needless to say both countries experience the same rate of decline initially and then the same increase in growth rate, since they assumed the two countries to be identical. On the other hand, if we consider the effects of tariff reduction (or trade liberalisation) on the imported goods within the model of Grossman and Helpman (1991, ch. 6) that we have described in Section 3.2.1, we find no definitive answer as to whether liberalisation (protection) increases (decreases) or decreases (increases) the growth rate. For example, if the country is importing the good that only uses human capital as a direct input and exporting the good intensively using unskilled labour, the reduction in import tariff will reduce the relative domestic price of the human capital intensive good and reduce the relative wage of skilled labour. This decline in the price of human capital will increase the level of R&D activity by reducing its costs and thus lead to a higher equilibrium growth rate. In contrast, if the country imports the unskilled labour intensive goods, import liberalisation will increase the relative wages of skilled labour and reduce the growth rate. Hence, the growth effect of a trade policy depends on how it affects the amount of resources (skilled labour) used in the R&D sector. In another model of comparative advantage, Grossman and Helpman (1990) consider two countries, where one has got a comparative advantage in R&D over the other. If the country with comparative disadvantage in R&D imposes a tariff on the other country’s exports, more labour will be driven to the R&D sector, thus improving the latter country’s growth and vice-versa. But in the presence of international knowledge spillovers, both countries would grow at the same rate in the long run, and the tariff can improve this growth rate. In a recent interesting study, Ben-David and Loewy (2003) modify the neoclassical growth model by considering an open-economy that makes the growth process endogenous and allows for both level and growth effects to arise from trade liberalisation. The model, adding knowledge as an input to production along with capital and labour, replicates the behaviour of the neoclassical model with exogenous technological

Liberalisation and Productivity Growth 35

progress under autarky. But under trade, the model implies that in addition to the expected level effects, liberalisation also brings about a shared growth effect; all countries grow faster in the steady state. As in the standard neoclassical growth model, the model implies that in a steady state there will be conditional convergence among countries that equalise their tariffs both internally and externally. This in turn implies that those countries maintaining high trade taxes will not see their incomes converge to those with low trade taxes unless and until they lower their taxes as well. Furthermore, tariff reductions increase trade, which in turn increases the common steady-state growth rate of knowledge accumulation in all countries as well as the common steady-state savings rates and marginal products of capital. Consequently, all countries grow faster following the implementation or expansion of trade agreements. Also, as additional countries liberalise trade, growth rates increase further.

3.3 Empirical analysis of liberalisation and growth: methodological issues As we saw in the previous section, theory does not offer unambiguous predictions regarding the effect of liberalisation on economic growth. Notwithstanding this, a very large number of countries have convinced themselves, or have been convinced that liberalisation is good for growth. Later in this chapter we will review the evidence on whether or not it has been good for growth. Before doing so however, we need first of all to consider a number of fundamental modelling and methodological issues. 3.3.1

Modelling liberalisation

In the models described in Section 3.2.2, defining liberalisation is absolutely straightforward, it can be equated with tariff removal. Empirical analysis of liberalisation often starts from this position and tries to proxy actual liberalisation via tariff reduction and one of the most widely used measures of liberalisation in practice is changes in (average) tariffs. It is, however, not quite as straightforward as that, and policy makers and analysts may have one of two other concepts in mind when they discuss trade liberalisation. First, reducing anti-export bias. Protection is a relative concept. If protection is to have real effects, an activity must be protected relative to other activities – some other activity must be ‘disprotected’. In the simple two sector models discussed earlier, import protection results in the export sector being dis-protected.

36 David Greenaway and M. Emranul Haque

In other words, there is anti-export bias. Eliminating the tariff reduces anti-export bias by definition and there is equivalence between the trade liberalisation and reducing anti-export bias concepts. However, they may differ. Anti-export bias could be reduced by introducing an export subsidy and eliminated completely if the subsidy rate is the same as the tariff rate. With this concept therefore, liberalisation can be achieved by leaving the tariff unchanged and introducing an export subsidy. A third concept of liberalisation is what is sometimes called ‘second best liberalisation’, replacing (economically) less efficient instruments with more efficient alternatives. For example, a given degree of protection could be achieved by a tariff or a quantitative restriction (import quota). Theory tells us that tariffs are unambiguously more efficient than quotas. Thus replacing a quota with a tariff which achieves the same degree of protection can be (and in practice often is) thought of as liberalisation. There is an important point to all of this. Different analysts and policy commentators use these different concepts when they are discussing liberalisation and this has obvious relevance for how it is then ‘measured’ for econometric analysis. Here too, different analysts rely upon different measures and as we shall see soon, these can impact on model outputs. Intuitively the most obvious basis for analysis might seem to be a direct measure of proxy for policy change. At the simplest level this can take the form of a policy account. In other words, liberalisation is identified with an announced change in policy. The most thorough and systematic study of this kind is Papageorgiou et al. (1991) which evaluates almost forty liberalisation episodes across 19 countries. In each case they reviewed changes in tariffs, the balance between tariffs and quotas, use of export incentives and so on, and scored liberalisation on a scale of 1 : 20. The information used, country by country is set out in Table 3.1. In terms of detail on individual episodes, these are a very rich source of information, but as Greenaway (1993) showed, not at all helpful for comparative analysis because different criteria are used in scoring liberalisations in different countries. In an ideal world one would like to have a price-based metric which allowed direct comparisons across countries and there have been attempts to develop indices. A good example is the bias index used in Krueger (1978). This is a ratio of the deviation of the domestic price of importables from world prices, relative to the comparable deviation for exportables. It chimes therefore with the notion of anti-export bias discussed earlier. But, although having a number of attractive features, it has not been widely used. This is partly because data requirements are so

37 Table 3.1 Components of liberalisation episodes: the Papageorgiou et al. study Country

Features of liberalisation episode

Argentina 1967–70 1976–80

Average tariff reduced from 94% to 49% Increased variance of tariffs Increase in effective exchange rate Exchange rate devaluation Export promotion

Brazil 1965–73

Tariff reductions Duty drawback scheme Tax exemptions and subsidies for export Exchange rate devaluation

Columbia 1964–66 1968–82

Tariff reductions Prohibited import list abolished Reduced variance of tariffs Differential export subsidies

Peru 1979–80

Maximum tariff reduced from 355% to 60% Some increase in tariff escalation Most imports go on quota free list

Uruguay 1974–82

Tariff reductions QRs eliminated Unification of foreign exchange market

Indonesia 1950–51 1966–72

Rationalised tariff structure Elimination of import licence restrictions Reduced use of QRs, but pervasive controls remained Simplification of export restrictions Exchange rate devaluation and unification of foreign exchange market

South Korea 1965–67

Reduced tariffs to average of 22% Reduced use of QRs, but 1200 items remained controlled Export incentives Exchange rate adjustment

1978–79 New Zealand 1951–56 1964–81 1982–84

Tariff reductions Import licence reductions Tariff compensation Greater exchange rate flexibility

Pakistan 1959–65 1972–78

Exchange rate devaluation Abolition of export bonus scheme

Philippines 1960–65 1970–74

Liberalised foreign exchange system Reduction in peak tariffs Piecemeal reductions in import controls Increase in effective tariffs Continued

38 David Greenaway and M. Emranul Haque Table 3.1 Continued Country

Features of liberalisation episode

Singapore 1968–73

Tariff reductions QR liberalisation

Sri Lanka 1968–70 1977–79

Removal of import licences Increased tariffs Increased export taxes Greater exchange rate flexibility Free trade zone incentives

Turkey 1970–73 1980–84

Tariff reduction QR reduction Export incentives Exchange rate devaluation

Israel 1952–55

Replacement of QRs with tariffs Some tariff reduction to hit target effective protection rate Exchange rate adjustment

1962–68 1969–77 Spain 1960–66 1970–74 1977–80

Tariff reductions Export incentives Exchange rate adjustment

Yugoslavia 1965–67

Average tariffs reduced from 23% to 10.5% Shift of products from restricted to liberal lists Abolition of export subsidies Exchange rate devaluation

Chile 1974–82

Tariff reductions QRs eliminated Unification of foreign exchange market

Source: Adapted from Greenaway (1993).

demanding (the indices have to be built-up from highly disaggregated data) and partly because there remain problems of comparing countries on a like for like basis. In practice therefore, for empirical analysis, researchers date liberalisation episodes either by reference to the commencement of a trade reform programme, or to some discrete and identifiable change in measured trade policy. Table 3.2 taken from Dean et al. (1994) is a good illustrative example. 3.3.2

Modelling strategies

Having identified, or dated, a possible liberalisation episode, there are broadly speaking, two modelling strategies for evaluating its subsequent

39 Table 3.2 Elements of recent trade liberalisations Average nominal tariffa Country

Pre-reform

Taxb Current Ratio dependence

ER

South Asia Bangladesh (1989, 1992) India (1990, 1993) Pakistan (1987, 1990) Sri Lanka (1985, 1992) Average

94 128 69 31 80

50 71 65 25 53

0.53 0.55 0.94 0.81 0.71

0.42 0.30 0.38 0.22 0.40c

East Asia China (1986, 1992) Philippines (1985, 1992) Indonesia (1985, 1991) Korea (1984, 1992) Thailand (1986, 1990)d Average

38 28 27 24 13 29

43 24 22 10 11 25

1.13 0.88 0.81 0.42 0.88 0.82

0.29 0.03 0.17 0.22 0.65c

SSA Cote d’Ivoire (1985, 1989) Ghana (1983, 1991) Kenya (1987, 1992) Madagascar (1988, 1990) Nigeria (1984, 1990) Senegal (1986, 1991) Tanzania (1986, 1992) Zaire (1984, 1990) Average

26 30 40 46 35 98 30 24 41

33 17 34 36 33 90 33 25 38

1.27 0.57 0.85 0.78 0.93 0.92 1.10 1.04 0.94

0.31 0.18 0.23 0.32 0.23 0.43 0.07 0.14 0.29c

CFA 11.1 5.4 11.2 71.2 CFA 145.2 13.1

Latin America Columbia (1984, 1992) Peru (1988, 1992) Costa Rica (1985, 1992) Brazil (1987, 1992) Venezuela (1989, 1991) Chile (1984, 1991) Argentina (1988, 1992) Mexico (1985, 1987) Average

61 57 53 51 37 35 29 29 44

12 17 15 21 19 11 12 10 15

0.20 0.30 0.28 0.41 0.51 0.31 0.41 0.34 0.35

0.13 0.22 0.13 0.02 0.05 0.11 0.05 0.03 0.74c

36.1 106.7 15.8 9.5 0.2 14.5 43.7 3.7

5.3 7.7 11.7 0.5

43.9 1.8 23.2 13.1 0.5

Notes: Years given in parenthesis are pre-reform and current. a Unweighted average nominal tariff (tends to be biased upwards), rounded ratio is current pre-reform – lower ratio implies greater tariff reductions. b Tax dependence is tariff revenue as proportion of tax revenue in 1984. c Figure is the rank correlation between countries ranked in descending order of tax dependence against descending order of tariff ratio; that is, a high positive correlation implies that countries most dependent on tariffs are least able to reduce averaged tariffs. d Import-weighted average nominal tariff. Source: Dean et al. (1994).

40 David Greenaway and M. Emranul Haque

growth effects: computable general equilibrium modelling and econometric analysis. Over the last fifteen years, computable general equilibrium (CGE) modelling has been increasingly used. Its uses are multiple: tax incidence, environmental degradation and regional integration to name a few. Applications to trade reforms is very common indeed. As its name suggests, CGE modelling attempts to capture myriad interactions which result when an economy is subject to a shock, such as trade liberalisation. Essentially it uses computational methods to generate output from general equilibrium models that would otherwise be intractable. To make their potential power more concrete, to solve the simplest kind of 2  2  2 open economy model, we end up with 20 endogenous and four exogenous variables! Imagine then the complexity of trying to solve longer scale models, analytically. Computable general equilibrium modellers attempt to model not only individual sectors of the economy but the interactions between them as well. Basic data is combined with a model specification to generate a benchmark data set. A process called ‘calibration’ is then used to solve the model. Basically this involves running the model until it generates values for unknowns (such as elasticities of substitution between capital and labour). Once solved, the model can then be used to investigate specific shocks, such as liberalisation, and work through the impact of that shock on productivity, factor market adjustment and so on. There are a number of attractions to CGE modelling. First, because it is general equilibrium, interactions between different sectors can be traced. In addition, the approach offers the modeller the opportunity to experiment with different policy shocks. This is potentially very attractive in, for example, exposing the redistributional consequences of different policies. However, as should be apparent from even this short description, there are also some potentially serious shortcomings. The approach is essentially ‘black box’, that is to say, once a given set of outputs has been generated, what is driving those outputs is not always clear. This problem is compounded by the fact that, unlike econometric analysis, no standard, well understood statistical tests exist for gauging the degree of confidence we can have in a particular set of outputs. This leads on to a third problem, namely, the false precision of outputs – the models yield exact numbers. This is attractive to policy makers who generally like to have a single answer! But less attractive to the analyst who is very well aware of the dangers of false precision. The most widely used approach to modelling the consequences of liberalisation is econometric analysis. This might be ex ante or ex post,

Liberalisation and Productivity Growth 41

the latter being the more common. Here analysts work with actual data on outcomes, as opposed to generated data. That data might be used in a cross section context, with a view to exploiting variations in experience across countries. Alternatively, it might be time series, in other words exploiting variation in given series for a specific country through time. Or it might be panel based, in other words combining some variation in outcomes across counties, but also some variation in outcomes through time. Econometric methods are of course widely used in applied economics and relatively well understood. In contrast to CGE modelling, we also have well-developed tests, grounded in statistical theory for judging the degree of confidence we can have in a given set of results. Not surprisingly therefore, econometric analyses of liberalisation episodes are rather more common than CGE analyses. That should not however be read to mean that interpreting the output of econometric models is problem free. That is not the case. There is, for example, the problem of ensuring that we have the appropriate counterfactual, in other words a correct analysis of what would have happened in the absence of liberalisation. We also need to be confident that we are controlling appropriately for other shocks which can impact the economy in question over the period of analysis. Then there is the problem, in comparative work, of allowing for the fact that different countries might react at different speeds, due, for example, to differences in economic structure which manifest themselves in the form of different supply elasticities. All have to be weighed up in any assessment of evidence.

3.4 Empirical analysis of liberalisation and growth: evidence In our discussion of evidence, we focus on outputs from the econometric literature. This is not because we regard CGE analysis as having nothing to say. Rather because it is a methodology best suited in this context to scenario analysis. In other words it is most helpful to asking ‘what if’ questions: what might the potential outcomes be if tariffs are cut by 50 per cent rather than 20 per cent? What if the cross elasticity of substitution between factors is 0.8 rather than 0.4? And so on. Our interest is not on this, but on ‘what has been’ questions, that is, what the impact of actual liberalisation been. The cross-country literature falls into two genres, ‘with–without’ and ‘before–after’. ‘With–without’ has been used by inter alia Mosley et al. (1991). It involves taking a sample of countries subject to trade reforms,

42 David Greenaway and M. Emranul Haque

matching them with non-reform comparators and ascribing any difference to the reform programme. ‘Before–after’ which is again used by Mosley et al. (1991) introduces a time dimension in that it compares ‘with–withouts’ for a few years before and a few years after. In some cases, like Papageorgiou et al. (1991), it is only the ‘withs’ that are examined, in that case for three years before and after. Examples of time series analysis include Harrigan and Mosley (1991), Papageorgiou et al. (1991), Greenaway and Sapsford (1994), Greenaway et al. (1997). Harrigan and Mosley (1991) focus on Structural Adjustment Loans (SALs) as one of a number of possible determinants of growth, export and investment performance in a standard growth model. Greenaway and Sapsford (1994) use a structural break approach in a neo-classical growth model; Greenaway et al. (1997) model growth as a smooth transition process then search for evidence of a coincidence of ‘take off’ and liberalisation. The study which reports the most favourable growth enhancing effects for liberalisation is Papageorgiou et al. (1991) who evaluate 36 liberalisation episodes in 19 countries. Moreover they conclude that more rapid growth of real GDP is secured with minimal transitional costs (in unemployment and fiscal constraints). These conclusions were challenged by Greenaway (1993) and Collier (1993) largely on the grounds that the underlying measure of liberalisation is flawed. Moreover, Greenaway et al. (1997) look specifically at the timing of the Papageorgiou et al. episodes and find no systematic evidence of a connection between trade reforms and growth acceleration. In some cases there is a positive correlation, in some a negative correlation, in others no apparent correlation whatsoever. This seems to be the story where the remaining evidence is concerned. There are many cases where a positive link from liberalisation to growth is reported; equally there are many cases where no association, is reported. Moreover, as Rodriquez and Rodrik (1999) show, many of the reported results are not very robust to changes in specification and/or sample frame. Given the diversity of components of liberalisation programmes, the range of indicators used and the fact that dynamics are rarely modelled, this is perhaps not surprising. The key impediment to modelling dynamics is the lack of good quality long-run data. Recent work has attempted to address this by exploiting the benefits of panel data and more powerful econometric methods, for extracting useful information from panel data. Thus Greenaway et al. (2002) report results for a panel of 73 countries over a 15-year period

Liberalisation and Productivity Growth 43

(1979–93) which encompasses a large number of liberalisations. They also identify liberalisations by reference to three different measures to see whether the way in which liberalisation is proxied makes a difference to their results. What they find is that the measure of liberalisation can make a big difference, with growth effects being stronger for those measures that focus on actual changes in policy rather than intended changes. Second, it was clear that the short- and long-run effects of liberalisation differed. Although in general the long-run effects tended to be positive in their sample, the short-run effects were often negative. In other words it may take time to adjust to the new regime and there may be real output losses in the process of resource reallocation. Yanikkaya (2003) works with an even larger panel, 100 countries spanning the period 1970–97. He also experiments with a range of liberalisation measures, some of which are trade volume based and some trade barrier based. With regard to the former, he reports that long-run results are generally consistent with those reported in the broader literature. He does qualify this however by reporting the conflicting result that trade barriers are positively associated with growth.4

3.5 Conclusions In this chapter we have focused on links between trade liberalisation and growth. We began by reviewing the relevant theory. This is of fairly recent vintage and the predictions from this work are not entirely unambiguous. Some models point to a clear connection between liberalisation and growth, others do not. Notwithstanding this, liberalisation has been a pervasive phenomenon over the last 30 years, with well over 100 countries embarking on trade reform programmes of one form or another. In some cases the liberalisation programmes have been entirely voluntary, in others the programmes have been linked to World Bank structural adjustment loans. Indeed one can plausibly argue that many would not have taken place in the absence of the latter. We then moved on to consider empirical evidence. As we saw a considerable empirical literature has emerged. This is methodologically diverse, ranging from CGE modelling to panel data estimation and policy accounts. Since there is no single unambiguous proxy for trade liberalisation, different analysts deploy different measures. Given this, together with the fact that many liberalisation episodes are relatively recent, it is hardly surprising that results are mixed. What is becoming

44 David Greenaway and M. Emranul Haque

clear however is the importance of distinguishing long- and short-run effects. There is less ambiguity about the long-run relationship between openness and growth than there is about the short-run linkages, pointing up the need for more detailed evaluation of short-run dynamics.

Notes The authors wish to acknowledge financial support from The Leverhulme Trust under Programme Grant F114/BF. 1. See Rodrik (1995) for an overview and criticism of the export-led growth strategy. 2. Authors such as Van and Wan (1997) argue that free trade is a direct channel through which an economy can learn from more advanced economies. 3. There are other studies on international knowledge flows such as Walde (1996), Feenstra (1996) that essentially find the similar growth effects, with either common or varying growth among countries. 4. For an excellent recent survey of the empirical literature on trade liberalisation and growth see Santos Paulino (2005).

References Ben-David, D. and M.B. Loewy (2003), ‘Trade and the Neoclassical Growth Model’, Journal of Economic Integration, 18, 1–16, March. Collier, P. (2003), ‘Higgledy-Piggleday Liberalisation’, The World Economy, 16, 503–12. Dean, J., S. Desai and J. Riedel (1994) ‘Trade Policy Reform in Developing Countries Since 1985: A Review of the Evidence’, World Bank Development Policy Group mimeo. Edwards, S. (1998), ‘Openness, Productivity and Growth: What Do We Really Know?’, Economic Journal, 108, 383–98. Feenstra, R.C. (1996), ‘Trade and Uneven Growth’, Journal of Development Economics, 49, 229–56. Fischer, E.O. (1995), ‘Growth, Trade and International Transfers’, Journal of International Economics, 39, 143–58. Greenaway, D. (1993), ‘Liberalising Foreign Trade Through Rose Tinted Glasses’, Economic Journal, 103, 208–23. Greenaway, D. and C.R. Milner (1993), Trade and Industrial Policy in Developing Countries, London, Macmillan. Greenaway, D. and D. Sapsford (1994), ‘What Does Liberalisation Do for Exports and Growth?’ Weltwirtschaftliches Archiv, 130, 152–74. Greenaway, D., S. Leybourne and D. Sapsford (1997), ‘Modelling Growth (and Liberalisation) Using Smooth Transition Analysis’, Economic Inquiry, 35, 798–814. Greenaway, D., C.W. Morgan and P. Wright (2002), ‘Trade Liberalisation and Growth in Developing Countries’, Journal of Development Economics, 67, 229–44.

Liberalisation and Productivity Growth 45 Grossman, G.M. and E. Helpman (1990), ‘Comparative Advantage and Long Run Growth’, American Economic Review, 80, 796–815. Grossman, G.M. and E. Helpman (1991), ‘Innovation and Growth in the Global Economy’, Cambridge, MA: The MIT Press. Harrigan, J. and P. Mosley (1991), ‘Evaluating the Impact of World Bank Structural Adjustment Lending: 1980–87’, Journal of Development Studies, 27, 63–94. IMF (1998) Trade Liberalisation in IMF-Supported Programs (Washington DC, IMF). Jones, L.E. and R. Manuelli (1990), ‘A Convex Model of Equilibrium Growth’, Journal of Political Economy, 98, 1008–37. Kim, E. (2000), ‘Trade Liberalization and Productivity Growth in Korean Manufacturing Industries: Price Protection, Market Power, and Scale Efficiency’, Journal of Development Economics, 62, 55–83. Krueger, A.O. (1978), Foreign Trade Regimes and Economic Liberalisation, Lexington, MA. Ballinger. Mosley, P., J. Harrigan and J. Toye (eds) (1991), Aid and Power: The World Bank and Policy-Based Lending, Volume 1: Analysis and Policy Proposals, London, Routledge. Papageorgiou, D., M. Michaely and A. Choksi (eds) (1991) Liberalising Foreign Trade, Oxford, Basil Blackwell. Rivera-Batiz, L.A. and P. Romer (1991a), ‘Economic Integration and Endogenous Growth’, Quarterly Journal of Economics, 106, 530–55. Rivera-Batiz, L.A. and P. Romer (1991b), ‘International Trade with Endogenous Technological Change’, European Economic Review, 35, 971–1004. Rodriguez, F. and D. Rodrik (1999), ‘Trade Policy and Economic Growth: A Skeptic’s Guide to the Cross-National Evidence’, Working Paper 7081, NBER. Rodrik, D. (1988), ‘Imperfect Competition, Scale Economies and Trade Policy and Developing Countries’, in R. Baldwin, Ed., Trade Policy Issues and Empirical Analysis, Chicago, University of Chicago Press. Rodrik, D. (1992) Closing the Productivity Gap: Does Trade Liberalization Really Help? Trade Policy, Industrialization and Development: New Perspectives, Oxford, Clarendon Press. Rodrik, D. (1995), ‘Trade and Industrial Policy Reform’, in J. Behrman and T.N. Srinivasan, eds, Handbook of Development Economics, North Holland, Amsterdam. Rodrik, D. (1995), ‘Getting Interventions Right: How Korea and Taiwan Grew Rich’, Economic Policy, 53–108, Cambridge. Rodrik, D. (1997), ‘Globalisation, Social Conflict and Growth’, The World Economy, 21, 143–58. Santos-Paulino, A. (2005), ‘Trade Liberalisation and Economic Performance’, The World Economy, 28, 783–823. Stokey, N.L. (1988), ‘Learning by Doing and Introduction of New Goods’, Journal of Political Economy, 96, 701–17. Stokey, N.L. (1991), ‘Human Capital, Product Quality and Growth’, Quarterly Journal of Economics, 106, 587–616. Tybout, J. (1992), ‘Linking Trade and Productivity: New Research Directions’, World Bank Economic Review, May, 6(2), 189–211. Van, P.H. and H. Wan Jr (1997), ‘Interpreting East Asian Growth’, in B.S. Jansen and K. Wong, eds, Dynamics, Economic Growth and International Trade, The University of Michigan Press.

46 David Greenaway and M. Emranul Haque Walde, K. (1996), ‘Proof of Global Stability, Transitional Dynamics and International Capital Flows in a Two-Country Model of Innovation and Growth’, Journal of Economics, 64(1), 53–84. Yanikkaya, H. (2003), ‘Trade Openness and Economic Growth’, Journal of Development Economics, 72, 55–79. Young, A. (1991), ‘Learning by Doing and the Dynamic Effects of International Trade’, Quarterly Journal of Economics, 106, 369–45.

Part II Cross Border Investment and Productivity

This page intentionally left blank

4 Foreign Direct Investment and Productivity Growth: Theory Ben Ferrett

4.1 Introduction This chapter presents several theoretical perspectives on the relationship between foreign direct investment (FDI) flows and productivity growth, interpreted as growth in total factor productivity (TFP), the joint productivity of a given bundle of inputs (e.g. capital and labour). Two ways in which a firm’s TFP can increase can be distinguished.1 A firm might receive superior technical knowledge from other firms (‘spillovers’), or it might develop better techniques internally, via investment in research and development (R&D). The models outlined in this chapter can be viewed as formalisations of the linkage between FDI flows and these two mechanisms. In Section 4.2 we present some general equilibrium perspectives on the FDI/productivity relationship. An open economy version of Solow’s famous (1956) growth model is developed, where North–South FDI flows equalise the return to capital across countries and simultaneously transfer technical knowledge internationally. We then go on to consider the implications of some recent models of general equilibrium with imperfect competition for the FDI/productivity relationship. These are able to account for the influence of specialisation (‘variety’) in production and endogenous R&D decisions, neither of which can be examined within a perfectly competitive framework. The key advantage of general equilibrium modelling is that it allows the economy-wide effects of FDI flows to be analysed. However, due to the demands of tractability, this is at the expense of detailed study of the impact of FDI at the micro level. Moreover, intuition suggests that the micro effects of FDI flows will be both subtle and important: multinational enterprises (MNEs) typically operate in ‘concentrated’ industries where considerations of strategic 49

50 Ben Ferrett

inter-firm rivalry are likely to exert a significant influence on equilibrium outcomes (Caves, 1996, section 4.1). Therefore, in Section 4.3 a variety of partial equilibrium perspectives are presented. We analyse the strategic effects of spillovers on the firm’s choice between FDI and exporting (Section 4.3.1) and a specific mechanism – worker mobility between firms – through which spillovers occur (Section 4.3.2).2 We also present a model of the interactions between FDI flows and R&D investments (Section 4.3.3). To fix ideas, a comment on how productivity growth is accounted for in these modelling frameworks might be useful. The general equilibrium models typically specify a production function (e.g. the aggregate production function in the Solow model), an explicit component of which is an index of TFP.3 However, in partial equilibrium (industrial organisation) models production functions are rarely explicitly specified; rather, we work with its dual, the cost function. In these models an increase in TFP is generally reflected as a reduction in (constant) marginal cost.4 In Section 4.4 we examine two modelling approaches that disaggregate FDI flows into their components, greenfield-FDI (‘greenfield investment’) and acquisition-FDI (cross-border mergers and acquisitions), to examine their separate relationships with productivity growth. This is an interesting exercise both because there are intuitive industrial–organisation reasons for believing that market structure (at least in the short run) will be differentially affected by greenfield- versus acquisitionFDI, and because neither type of FDI is trivial in real-world flows.5 First, we examine the relationships between the two types of FDI flow and industry R&D performance; and, second, we examine how FDI inflows and outflows (of both kinds) help shape the national productivity distribution across plants in an industry. This second analysis sheds some light on how FDI flows might account for the frequently observed productivity advantages of foreign- over domestically owned firms. Finally, Section 4.5 concludes.

4.2 General equilibrium perspectives In this section we develop a simple model of how FDI flows can arise endogenously as an equilibrium feature of an integrated world capital market.6 In our model FDI flows (in equilibrium) from high to low productivity countries (where productivity is to be interpreted as total factor productivity, TFP), and the productivity advantages of source countries are assumed to be embodied in FDI flows. Both directly (via the import of superior capital) and indirectly (via technological spillovers to local

FDI and Productivity Growth 51

firms), FDI inflows into low-productivity developing countries raise the aggregate TFPs of those economies. Our model thus formalises the process by which North-to-South FDI flows can enable developing countries to catch up (and possibly converge) with industrialised countries. An appealing feature of our analysis is its general equilibrium perspective, but this comes at some cost: for example, we are unable to comment on the ‘strategic’ features of competition that might well be significant in the (typically concentrated) markets in which MNEs operate. The world comprises two regions, North and South. Aggregate production functions take a Cobb-Douglas form: North: YN  ANK
NL1
⇒ yN  ANk
N N South: YS  ASK
S L1
⇒ yS  ASk
S S where upper and lower case letters represent aggregate and per-capita quantities respectively (Y  net output; K  capital; and L  labour). AN and AS index TFP (assumed to be a ‘pure’ public good, both nonrival and nonexcludable, within the country).
 (0, 1) is the share of capital in national income (assuming perfect competition in product and factor markets). Under autarky, the steady-state level of capital per head is determined (as in the Solow growth model) by the requirement that per-capita investment (financed by savings) compensate the capital-diluting effects of depreciation and population growth: 

siyi

(n  )ki, i  {N, S}

冦

冦

Autarky:

realized investment per head (determined by savings)

required investment per head to maintain constant K

The rates of population growth and depreciation (common across regions) are n and  respectively, and s is the constant average propensity to save (different across regions). Therefore, in the steady state y/k  (n  )/s for both regions. Substituting this into the marginal product of capital, y/k 
y/k, we derive the autarky real interest rates: rN 


(n  ) sN ,

rS 


(n  ) sS

The important feature of rN and rS is that they are independent of AN and AS: the immediate rise in the marginal product of capital caused by

52 Ben Ferrett

an improvement in TFP is entirely offset in the steady state by an increase in capital per head (capital deepening). We make the following intuitively appealing assumptions: Technological leadership: AN AS Savings behaviour: sN sS The first assumption means that North is the technological leader, and the second is sufficient to guarantee rS rN in autarky.7 In Figure 4.1 y ≡ real output per head slope = rN

yN

yS

slope = rW slope = rW

(n+d).k

sN.yN slope = rS kX sS.yS

kM

0

kS

kS* kN*

kN

k ≡ capital per head

Figure 4.1 North–South FDI flows in general equilibrium à la Solow Key: kN, kS and rN, rS are autarky capital : labour ratios and real interest rates respectively. Upon integration, the common (world) real interest rate is rW. kN*, kS* and kX, kM are, respectively, capital:labour ratios and capital exports and imports per head in the integrated equilibrium. kX  kM if and only if LN  LS.

FDI and Productivity Growth 53

steady-state capital and income per head are lower in the South than in the North under autarky for two reasons: the North’s higher propensity to save and accumulate (sN sS) and its higher level of TFP (AN AS), which increases savings in the North and income per head for any kN  kS, sN  sS.8 So far we have assumed factor immobility between the regions. Now assume that capital becomes mobile internationally (globalisation), which implies rN  rS  rW, the world real interest rate, in equilibrium (otherwise capital could profitably change locations). Because rS rN in autarky, capital will flow from North to South.9 The (steady-state) equilibrium values of kN, kS under perfect capital mobility are characterised by two conditions: rN  rS ⇒

冢 冣

AN kN  AS kS

1/(1
)

(1)

,

which ensures that the marginal products of capital are equalised across regions,10 and

冦

sNyN(n  ) kN = (n  )kS –ssyS , (2)

冦

Capital exports  imports:

Capital exports from Capital imports by North per head South per head which ensures that aggregate capital demand and supply balance. Equation (2) implicitly assumes that the two regions have the same population; the extension to size asymmetries is straightforward, requiring the LHS of Equation (2) to be multiplied by LN/LS. Equations (1) and (2) uniquely determine a (kN, kS)-pair with kN kS where both capital–labour ratios lie inside the autarky range (see Figure 4.1). Two features of the steady-state equilibrium are noteworthy. First, production per head falls in the North and rises in the South as capital migrates.11 Second, capital flows from the North to the South. This creates the possibility of international technology transfer if the North’s TFP advantage is to some extent embodied in its capital outflows. Assume that TFP in the South, AS, evolves according to

冢

冣

dAS AN kM , 0 f AS kS dt

for all

AN kM 1, 0 AS kS

with f(1, ·)  f(·, 0)  0. Therefore, Southern TFP grows if there exists a technological gap between North and South (i.e. AN AS) and the South hosts Northern FDI (i.e. kM  kX 0).12

54 Ben Ferrett

Therefore, in the world depicted in Figure 4.1 perfect capital mobility implies that in the long run AS converges on AN as the South imports Northern FDI and techniques. Note from Equation (1) that AS  AN implies kS  kN for real interest rate equalisation. In the limit, yS  yN for all kS  kN; but the South will continue to host FDI from the North because its propensity to save is lower. There is one especially significant respect in which our analysis differs from convention: the effects of an increase in the Southern propensity to save (sS). Assume that sS rises to the level of sN. Then the two regions’ autarky real interest rates will be equal, and no North-to-South capital movement will occur upon liberalisation. Therefore, South is stuck in a trap and will not catch up over time. This is an instance where saving more can frustrate capital accumulation that would otherwise have occurred in the long run. The reason for this is that increased Southern savings (completely) crowd out FDI inflows from North, so South loses the benefit of technology transfer.13 Of course, this is a polar case because rises in sS that preserve sN sS will be consistent with North-to-South FDI flows and technology transfer upon capital liberalisation. Here, the increase in Southern savings merely retards its convergence with North. These results on the effects of higher Southern savings contrast with the conventional case of fixed national TFPs where increases in the propensity to save will increase the steady-state k if the economy is closed (the Solow case) or large and open (rw will fall); only if the economy is small and open will an increase in saving have no effect on k. In the remainder of this section we briefly review two strands of literature that examine the FDI/productivity relationship in general equilibrium models with imperfect competition. Rodriguez-Clare (1996) examines how the extra demand for locally produced intermediate goods, created by inward FDI in the final-goods sector, enables greater specialisation in the intermediate-goods industry as more varieties are produced (i.e. more firms enter), an issue that cannot be addressed under perfect competition where the number of firms is indeterminately large. Following Ethier (1982) in applying the Dixit-Stiglitz (1977) model of monopolistic competition to the vertical relationship between industries producing intermediate and final goods,14 Rodriguez-Clare showed that inward FDI into the final-goods sector generates a positive externality for other final-good producers via backward linkages: the investing MNE’s demands cause the intermediate-goods sector to expand (more varieties), which raises the TFP of local producers of the final good. (Rodriguez-Clare assumes that domestic firms must buy all their intermediate inputs locally, so inward FDI that represents merely a relocation of production – rather than a global increase – nevertheless confers

FDI and Productivity Growth 55

external productivity benefits. However, MNEs may source their intermediates from abroad; if this occurs to a large extent, inward FDI that displaces local final-goods producers and creates an ‘enclave economy’ within the host country may harm local firms by reducing the number of locally produced intermediate varieties.) Haaland and Wooton (1999) examine the implications of a model similar to Rodriguez-Clare’s for the international agglomeration of final-goods production. Since inward FDI by a single MNE raises the TFP of all firms in the final-goods sector, it increases the incentive for additional MNEs to enter, perhaps leading the production plants of mobile firms to be internationally concentrated in a relatively small number of locations.15 The second group of general equilibrium models with imperfect competition examines relationships between Northern firms’ R&D decisions, undertaken to move up the ‘quality ladder’ and exploit a monopoly position in a niche market, and Southern firms’ rate of imitation. (Imperfect competition is necessary here because it generates the rents firms need to finance sunk R&D investments.) Benchmark models of product-cycle trade, where Northern firms innovate and produce ‘young’ goods before production eventually moves to the cheaper South, are provided by Grossman and Helpman (1991, chs. 11 and 12) and Helpman (1993). A higher rate of imitation by Southern firms of Northern goods, which shifts production Southwards due to its cost advantage, can – somewhat paradoxically – increase R&D spending in the North: although the monopoly profits from a successful innovation are shorter lived, they may be larger because the general Southwards migration of production will depress factor prices in the North. Glass and Saggi (1999) introduce FDI by Northern firms into this framework. There are now two channels of international technology transfer between North and South: Southern firms may imitate either Northern ‘national’ firms or Northern MNEs. The impact of Northern FDI in the South on the level of R&D in the North depends on how aggregate Southern imitation responds to inward FDI. If total imitation rises, then R&D spending in the North will rise via the previous mechanism; however, if imitation of Northern MNEs merely substitutes for imitation of Northern ‘national’ firms, then Northern R&D spending will be essentially unaffected.16

4.3 Partial equilibrium perspectives 4.3.1

Demonstration effects

In this section we simplify from general to partial equilibrium. This has the advantage that we are able to deepen our analysis in certain respects

56 Ben Ferrett

(notably, the inclusion of strategic behaviour) while retaining analytic tractability. We begin with a simple example to highlight some of the issues involved. Assume that a foreign MNE is considering whether to serve a host-country product market by exporting from its domestic production base or by establishing local production facilities (greenfieldFDI). The MNE’s (constant) marginal production cost is cM and exporting incurs a per-unit trade cost of t. The sunk cost of establishing a new plant in the host country is G. The host-country product market contains a single local firm with marginal production cost cL. For the moment we do not need to place restrictions on cL, cM relative to each other. Assume that if both firms produce in the host country there is a probability  (0, 1) that the more productive firm’s technology spills over (via unspecified ‘demonstration effects’) to its rival.17 Denoting the variable profits of firm i in duopolistic competition (perhaps à la Cournot) with firm j by  D (ci, cj), where  is host-country ‘market size’ (population), and assuming cL cM (which fixes – if it occurs – the direction of spillover as M-to-L), the MNE optimally chooses FDI over exporting if and only if

冦

冦

{cL > cM :} [ D(cM, cM)  (1  ) D(cM, cL)]  G >  D (cM  t,cL) Expected variable profits with local production (FDI)

Variable profits under exporting

(3)

A number of comparative-statics results are immediately clear from Equation (3). First, an increase in makes FDI less likely (FDI profits fall because D(cM, cL) D(cM, cM) if cM cL but exporting profits are unchanged) as the risk that the MNE will lose its technological advantage via spillovers increases. Second, an increase in G makes FDI less likely (FDI profits fall but exporting profits are unchanged), but an increase in t makes FDI more likely (exporting profits fall but FDI profits are unchanged).18 This replicates the well-known result that the ‘tariffjumping’ motive for FDI strengthens as tariffs rise (Motta, 1992). Third, the effect of an increase in cL (i.e. in the MNE’s technological lead) on the MNE’s optimal decision depends on whether spillovers are large or small. If is large (⬵1), then exporting profits rise but FDI profits are (approximately) unchanged, so FDI becomes less likely. However, if is small (⬵0), then both exporting and FDI profits increase following a rise in cL but FDI profits will rise by more, making FDI more likely.19 Fourth, as with technological lead, the effect of a rise in market size, , depends on the probability of spillovers. If is small (⬵0), increasing market size

FDI and Productivity Growth 57

makes greenfield-FDI more likely (because D (cM, cL) D (cM  t, cL)). However, if is large (⬵1), then an increase in market size will make greenfield-FDI less likely if and only if D (cM  t, cL) D (cM, cM) (e.g. if t is small and cL is large relative to cM).20 An important assumption underlying the preceding analysis is cL cM. This assumption is a persistent theme in theoretical modelling of competition between MNEs and ‘national’ firms. We cover some of the reasons motivating it in Section 4.4. Note, however, that in the simple framework outlined earlier there is no (logical) reason not to explore the consequences of setting cM cL, ‘multinationals without advantages’ (Fosfuri and Motta, 1999).21 If cM cL, the MNE optimally chooses greenfield-FDI over exporting if and only if

Expected variable profits with local production (FDI)

冦

[ D (cL , cL)  (1  ) D (cM, cL)]  G >  D (cM  t,cL) (4)

冦

{cM > cL:}

Variable profits under exporting

The difference between Equations (3) and (4) lies in the first term on the LHS: if spillovers occur, they flow from the MNE to the local firm when cL cM but in the opposite direction when cM cL. Some of the comparative-statics analysis of Equation (4) (i.e. for t, G) mirrors that of Equation (3). However, if rises when cM cL, greenfield-FDI becomes more likely (because D (cL, cL) D (cM, cL)): such inward FDI has been nicknamed ‘technology-sourcing FDI’ (Driffield and Love, 2003), undertaken in the hope of enabling the MNE to benefit from reverse spillovers from local firms. If the local firm’s technological lead rises, then greenfield-FDI becomes less likely if spillovers are small but more likely if they are large (because the value of catching up is greater, the larger the technology gap). Finally, if cM cL an increase in market size makes greenfield-FDI more likely independent of the probability of spillovers. The two preceding models, both very simple, give some idea of how complex are the links between FDI incentives, spillover possibilities and technological gaps. In particular, I want to emphasise that if we allow MNEs to be strategic players we cannot simply take the magnitude of the FDI flow as fixed when discussing the effects of variations in the degree of spillovers (as we did in the general equilibrium model of the previous section).22 An interesting generalisation of the models presented above would allow for two-way FDI flows (‘cross-hauling’), that is, to give the local

58 Ben Ferrett

firm the option of investing in the MNE’s home country. This would seriously complicate the analysis, however, because spillovers between the two firms can now occur in two countries. Therefore, a given firm’s incentive to invest abroad will typically differ depending on whether or not its rival has undertaken FDI because the possible spillover flows are different. Ferrett (2003b) examines this case.23 The discussion thus far has focussed exclusively on horizontal spillovers (i.e. between firms within a given industry). However, few of the general results appear to carry over to the case of vertical spillovers. Consider a simple extension of the international outsourcing model of Pack and Saggi (2001) where a monopolist has located final-goods production abroad and must purchase intermediate goods from a local monopolistic firm. It is immediately clear that the investing MNE has a strong incentive to encourage vertical spillovers from itself to the local supplier that reduce the latter’s marginal cost, because these will reduce the MNE’s input price. Furthermore, it is also the case that the MNE benefits from horizontal spillovers in the upstream (intermediate-goods) industry that reduce the marginal costs of extra potential suppliers and provoke their entry, stimulating competition in the upstream market and reducing the MNE’s input price. In the remainder of this section we consider two models of the relationships between FDI flows and productivities of host-country firms. The first models a spillover mechanism: trained workers’ mobility. The second analyses the relationships between FDI flows and R&D investment in an attempt to formalise arguments about linkages between inward FDI, competition and R&D spending. 4.3.2

A spillover mechanism: trained workers’ mobility

In this section we present a model of the spillover mechanism: a foreign MNE that establishes a plant in a host country trains local workers, and the spillover occurs when those newly trained workers move to local firms. The model we present is a simplified version of Fosfuri, Motta and Rønde (2001).24 It comprises two stages and analyses equilibrium decisions of firms within a single host country. Figure 4.2 shows the game tree. At Stage 1 a foreign MNE chooses whether to serve the host country via FDI or exports from a pre-existing production base abroad. The MNE’s proprietary technology allows it to produce a good that is not currently produced by host-country firms. If the MNE chooses FDI, its Stage-1 profit is M  G, where M is the monopoly profit from local production and G the sunk cost of a plant. In this case, the MNE must train a local worker, which is assumed to be costless. The worker’s reservation wage

FDI and Productivity Growth 59

Stage 1

Stage 2

vM > vL (Retain)

(2M – G – vL, 0)

Auction FDI and train vL > vM (Lose and train)

(M – G + D, D – vM)

MNE FDI and train

(M – G + MX, 0)

Export MNE

Export

(2MX, 0)

Figure 4.2 Game tree for simplified Fosfuri/Motta/Rønde model – (the MNE’s payoff precedes the local firm’s in brackets; vM  M  D and vL  D)

is zero. If the MNE chooses exports, its Stage-1 monopoly profit is MX  [0, M] because of trade costs. MX is inversely related to trade costs. Stage 2 is the second period of production. If the MNE chose exports in Stage 1, its Stage-2 choices are identical to those in Stage 1 (with corresponding payoffs). We assume MX M  G, so the MNE will optimally choose FDI in Stage 2 if it had previously exported. However, if the MNE undertook FDI in Stage 1 and trained a local worker, it must bid for the (mobile) worker in Stage 2 with/against a single host-country entrant firm.25 We assume that the firm with the higher valuation for the worker wins the auction, paying as a wage its rival’s valuation.26 The local firm will pay at most L  D, where D  [0, M] is its duopoly profit upon entry excluding wage payments. D is inversely related to the intensity of competition between

60 Ben Ferrett

the local firm and the MNE. For example, D  0 if the firms compete in prices (Bertrand) and sell homogeneous goods, whereas D  M if – from a demand-side view – the firms produce in independent markets.27 The MNE will pay at most M  M  D, the difference between its second-period (variable) profits when it keeps and loses the worker.28 Therefore, the MNE wins the auction if and only if M L  D

M , 2

otherwise the local firm could outbid it. Given our assumption that the MNE will undertake FDI in period 2 if it previously exported (i.e. MX M  G), we must now examine cases. First, the MNE will undertake FDI in period 1 if it will win the subsequent auction if and only if MX M  D.

(5)

Second, if it will lose the auction for the trained worker, the MNE will nevertheless undertake FDI in period 1 if and only if MX D.

(6)

Figure 4.3 plots the game’s subgame perfect Nash equilibria in (D, MX)-space under the assumption that M 2G.29 Consider first small MX ( M/2), which relates to large trade costs. Here the MNE will always choose local production in period 1 (tariff-jumping FDI). If L  D is small ( M/2), the MNE will outbid the local firm at auction, and the host country will experience a pecuniary externality: in period 2, the MNE pays the trained worker more than her reservation wage (0). However, if L  D is large ( M/2), the local firm will win the auction, and the host country experiences a technological externality in addition to a pecuniary one: the local firm appropriates the MNE’s technology by hiring away its trained worker. Equilibria are more complicated when MX is large ( M/2), because first-period exporting becomes a viable option. Note that Equations (5) and (6) both embody a trade-off between export profits in period 1 and profits on local production after

FDI and Productivity Growth 61 MX MX = D

MX = M – D M–G Export

FDI

FDI

Pecuniary spillover

Technological and pecuniary spillovers

D 0

MNE wins auction

M/2

MNE loses auction M

Figure 4.3 The MNE’s equilibrium period-1 choices (assuming M 2G)

the auction in period 2.30 For small D, the MNE optimally undertakes FDI in period 1 and wins the auction as before. However, the secondperiod wage of the retained worker increases one-for-one with D, so eventually winning the auction and producing locally in period 2 afterwards becomes unprofitable relative to exporting in period 1 (entry to the exporting region in Figure 4.3 from the left). If the MNE loses the auction (D M/2), then local production following it is more profitable than exporting in period 1 for large D (exit from the exporting region in Figure 4.3 to the right). An important finding here is that two types of spillover are experienced by the host country: a pecuniary spillover because the trained worker commands a higher wage regardless of whether she is poached; and a technological spillover if the trained worker takes her newly acquired skills to a local firm. The degree of competition between MNE and local rival is a key determinant of equilibrium outcomes: if they produce for different markets (D ⬵ M), the MNE will willingly train in the knowledge that its trained worker will be poached but the business activities of the poaching firm will harm the MNE relatively little. We also discovered that the MNE’s ability to export from a foreign country to the host has an important influence on outcomes as a fall-back option.

62 Ben Ferrett

4.3.3

FDI flows and R&D performance

In this section we present a model of the greenfield-FDI and process R&D decisions of rival ‘international duopolists’ taken from Ferrett (2002).31 Some have argued (e.g. Dunning, 1977) that R&D investments cause FDI by providing firms with the capabilities they need to compete in international markets and co-ordinate business activities across national borders, cultures, legal systems and languages. However, we shall see that the FDI/ R&D relationship is more complex than this: two-way linkages exist between these two corporate strategy decisions. Consider a two-firm, two-country world where the firms own preexisting home plants in different countries and compete in a homogeneous good. The firms play the following two-stage international duopoly game: Stage 1: The duopolists simultaneously and irreversibly choose their corporate structures: whether to undertake greenfield-FDI abroad or whether to invest in process R&D. Stage 2: Corporate structures adopted by all firms become common knowledge, as does the success/failure of any R&D investments undertaken. The duopolists compete à la Bertrand to serve the two national product markets. In Stage 1, the duopolists face two discrete choices. By paying a sunk cost of G, they can establish a plant abroad. The benefit of greenfieldFDI is that it reduces the firm’s marginal cost of serving the foreign product market by t, the per-unit trade cost. By paying a sunk cost of I, the firm undertakes process R&D: with probability p, R&D is successful and the firm’s marginal production cost falls from c, its initial level, to 0; with probability 1  p, R&D fails and marginal production cost remains at c. (The probability of R&D success is identical and independent across firms.) I make the following assumption on the marginal cost parameters to limit the taxonomy: 1 c t 0, so the per-unit trade cost lies between the marginal production costs of a successful and an unsuccessful innovator. This assumption on the ordering of marginal cost parameters is important partly because – given the Stage-2 Bertrand competition in homogeneous goods – only the lower-cost firm serves the market. The variables of immediate interest

FDI and Productivity Growth 63

are the probability of R&D success, p, and national market size, , which determines the slope of the demand function in both countries: Qd  (1  price) in both countries. We assume that firms maximise their expected profits and solve the game by backwards induction to isolate its subgame perfect Nash equilibria. Given our assumptions, each duopolist’s corporate structure belongs to {1N, 1R, 2N, 2R}, where the first element of each term indicates whether the firm owns 1 plant or (following greenfield-FDI) 2 plants and the second indicates whether the firm undertakes R&D (R) or not (N). To give a feel for the game’s structure, I briefly present two pairs of specimen payoffs. First, if both firms choose 1N (i.e. to maintain only their home plants and not to invest in R&D), then each earns profits of R(c, c  t), where R(ci, cj) is the variable profit of firm i in Bertrand competition with firm j (ci, cj are their respective marginal costs) within a market of size . In the case where the industrial structure is (1N, 1N), each firm serves only its home market and is afforded some protection from import competition by the trade cost, t. Second, in the industrial structure (1R, 2R) – firm 1 chooses 1R and firm 2 chooses 2R – expected profits are E 1  p(1  p)[R(0, c)  R(t, c)]  I E 2  p(1  p)[R(0, c  t)  R(0, c)]  p2R(0, t)  (1  p)2R(c, c  t)  G  I Because firm 2 has a local plant in country 1, firm 1 must possess a marginal production cost advantage to earn positive variable profits. This occurs with probability p(1  p) when firm 1’s R&D effort succeeds but firm 2’s R&D fails. On the other hand, firm 2 can earn strictly positive variable profits at home when the marginal production costs of the firms are equal because the trade cost affords its domestic plant some protection from foreign competition. The equilibrium industrial structures of the game are plotted in (p, )space in Figure 4.4.32 Several general conclusions on the comparativestatics effects of varying p,  can be drawn from Figure 4.4. The number of R&D investments is increasing in both market size and the probability of R&D success. The equilibrium number of plants is also increasing in national market size.33 The effect of increases in the probability of R&D success on the equilibrium number of plants is less clear. Note that

64 Ben Ferrett m

(2R, 2R)

(1R, 1R) (1R, 1R) (1N, 2R)

(1N, 1N) (1N, 1R)

0

0.5

1

p

Figure 4.4 Equilibrium industrial structures in the international duopoly game Key: The first element of each equilibrium choice indicates whether the firm operates one plant or two, and the second indicates whether (R) or not (N) the firm invests in R&D. See Ferrett (2002) for analytical definitions of the inter-regional boundaries.

the boundaries of the (2R, 2R) equilibrium region are asymptotic to the lines p  0 and p  1, so for sufficiently large p (2R, 2R) is never an equilibrium industrial structure. This is due to our assumption of the Bertand competition in homogeneous goods, which implies that firms will only incur sunk costs if they are likely to generate a marginal cost advantage. Therefore, playing 2R cannot be a best response to 2R for p 艑 1 because the most likely outcome is a market price of 0 in both countries and a loss of G  I for both firms. This underlies the switch in equilibrium industrial structure from (2R, 2R) to (1R, 1R) as we move rightwards for large . From casual inspection of Figure 4.4, it appears therefore that the numbers of FDI and R&D investments are (generally) positively associated in equilibrium. However, FDI and R&D are not causally connected because both are endogenous variables. We can go further than this and investigate how a commitment to undertake either FDI or R&D

FDI and Productivity Growth 65

affects a duopolist’s incentive to undertake the other sunk investment.34 (While this does not relate to causality, it does give a flavour of the relationships between FDI and R&D investment decisions.35) It is immediately clear that a firm committed to investing in R&D is more likely to undertake greenfield-FDI than one that is not. In this model playing 2N is strictly dominated by 1N (because Bertrand competition with the foreign incumbent ensures that rents to cover G can only be earned if R&D is undertaken), so a non-R&D firm will never optimally undertake greenfield-FDI.36 This result captures the FDI/R&D link in OLI (ownershiplocation-internalisation) models (Dunning, 1977; Markusen, 1995, 2002). In order to make greenfield-FDI profitable, the ownership advantages generated by (successful) process R&D are necessary. It can also be shown that a firm committed to undertaking greenfieldFDI is more likely to invest in R&D than one that is not.37 The reason for this is that, following greenfield-FDI and the elimination of trade costs on foreign sales, a firm has a larger output base over which to spread a successful process innovation; therefore, the value of a successful process innovation is larger to a 2-plant than a 1-plant firm. Therefore, two-way (positive) relationships exist between FDI flows and R&D levels. It is also possible to examine how spending on FDI or R&D by one duopolist affects its rival’s incentives to undertake FDI or R&D (see Ferrett, 2003c).

4.4 Does the form of FDI matter? The partial equilibrium models of Section 4.3 all identified FDI in general with greenfield-FDI in particular, while the general equilibrium models covered in Section 4.2 were rather ambiguous about the precise form of FDI (greenfield investment versus cross-border mergers and acquisitions, Mergers and Acquisitions (M&As)) considered.38 In this section we examine two modelling approaches that explicitly disaggregate FDI flows into greenfield- and acquisition-FDI.39 The first (Ferrett, 2003a) examines how the greenfield/acquisition choice interacts with firms’ R&D decisions. Inter alia, this allows a test of the ‘failing firm’ defence of acquisition-FDI inflows (Ferrett, 2003c): that acquisition-FDI is associated with sufficiently enhanced technological performance to offset the adverse welfare effects of increased concentration. Formally, the modelling framework builds on the international duopoly game described in Section 4.3.3 by adding two features: first, a Stage 0 where the two incumbents are able to merge; and second, an intermediate stage between 1 and 2 where a third (potential entrant) firm decides

66 Ben Ferrett

whether to enter the industry at a global level. If no merger occurs at Stage 0, the two incumbents play the international duopoly game, augmented by the possibility of global entry. However, if the two incumbents merge initially, then the integrated incumbent monopolist thus created only has to decide whether to invest in R&D before facing the potential entrant’s decision.40 For our purposes, the welfare conclusions of this analysis are particularly relevant. Comparing the equilibrium industrial structures with and without merger, there is generally a Williamson (1968)-type welfare trade-off between profits and consumer surplus: firms (collectively) benefit but consumers lose from the increased concentration following acquisition-FDI. However, in small markets (where entry by the outside firm never occurs) acquisition-FDI can be Pareto improving: the integrated monopolist created in equilibrium by acquisition-FDI invests in R&D process, whereas the incumbents acting independently would not, and this results in lower prices to consumers despite monopolisation. Consumers have benefited from an increase in TFP, caused by R&D investment, that occurs only when FDI takes the form of acquisition. To understand why industry R&D spending can be higher following acquisition-FDI, consider the incumbents’ R&D incentives in the nomerger threat point of (1N, 1N), where each incumbent maintains only her home plant and undertakes no R&D. Because they must pay trade costs on sales abroad, the return to a successful process innovation (a reduced marginal cost spread over output) is lower than that enjoyed by the integrated monopolist, who jumped the trade cost using acquisition-FDI.41 The second approach that considers FDI/productivity linkages in a model where FDI flows are disaggregated is Ferrett (2003b). The object is to examine the relationships between FDI inflows and outflows (of both forms) and the national productivity distributions across firms (plants) in an industry. The empirical backdrop for this work is the widely documented productivity gaps between foreign- and domestic-owned firms that exist in many industries across numerous countries.42 In the model two characteristics of national productivity distributions are endogenously determined. First, plants can be either high- or low-productivity, depending on which types of technology transfer occur; and, second, the number of plants is endogenously determined at equilibrium (a single potential-entrant firm exists). There are three ways in which firms’ FDI decisions interact with a national productivity distribution in the industry modelled. First, undertaking (either form of) FDI can lead to inter-firm technology transfer (i.e. ‘spillovers’) between the MNE’s newly established branch plant abroad and rival firms located in the host

FDI and Productivity Growth 67

country. In our model spillovers FDI can flow in both directions between a foreign branch plant and local rivals.43 Second, following a flow of acquisition-FDI, intra-firm technology transfer occurs: the highproductivity purchaser is able, at hardly any cost, to install its (superior) technology in the acquired plant abroad. The concept of intra-firm technology transfer is identical to that employed by Long and Vousden (1995) in their model of cross-border mergers, who assume that every plant in a merged firm operates at the minimum marginal cost of its constituent plants before the merger. Third, FDI decisions interact with national productivity distributions through the relationship between the greenfield-FDI/acquisition-FDI choice and the potential entrant’s decision. The world comprises two countries and three firms, two incumbents and one potential entrant. The incumbents initially own one plant each, located in different countries, with different productive efficiencies. The sequence of moves is as follows. At Stage 1 firm M, the highproductivity incumbent, chooses between acquisition-FDI (making a take-it-or-leave-it offer to the rival incumbent), greenfield-FDI (building a plant abroad to jump the trade cost), and exporting. In Stage 2 firm T, the low-productivity incumbent, chooses – if it still exists as an independent player – between greenfield-FDI and exporting. In Stage 3 firm E, the low-productivity potential entrant, chooses between no-entry, 1-plant entry (and the plant’s location), and 2-plant entry. Stage 4 is the market stage. Spillovers (inter-firm technology transfer) occur at the start of Stage 4: with probability , the best-practice technology installed in a country spills over to all local rivals.44 Intra-firm technology transfer also occurs since technology is assumed to be a public good within the firm. Finally, Cournot competition determines market equilibria in both countries.45 Figure 4.5 gives a generic representation of the game’s equilibria.46 As the plant sunk cost rises, the equilibrium number of plants falls. However, there is also a significant non-monotonicity: acquisition-FDI arises in equilibrium on two separated intervals of plant sunk costs. The reason for this is that ‘entry’ (i.e. E’s optimal number of plants) is more likely (i.e. greater) if M chooses acquisition-FDI (and E faces a monopolist) than if M chooses between greenfield-FDI and exporting (and E faces a duopoly). In the upper region where entry is inevitable, acquisitionFDI arises because it substitutes for costly greenfield-FDI. However, in the region where entry is conditional – that is, entry occurs if and only if M chooses acquisition-FDI – then acquisition-FDI is rendered unprofitable by subsequent, rent-dissipating entry.

68 Ben Ferrett Plant sunk cost

M chooses acquisition-FDI. E chooses no-entry.

Entry blockaded (independently of M ’s choice)

M, T choose between exporting and greenfield-FDI. E chooses no-entry. Entry conditional (on M ’s choice)

M chooses acquisition-FDI. E chooses between 1- and 2-plant entry. Entry inevitable (independently of M ’s choice) M, T choose greenfield-FDI. E chooses 2-plant entry.

0

uYProb{spillovers}

1

Figure 4.5 Equilibrium industrial structures Key: Firms M and T are the high- and low-productivity incumbents respectively. Firm E is the (low-productivity) potential entrant.

A number of the features of Figure 4.5 contrast strongly with the implications of Dunning’s (1977) OLI paradigm for the sources of foreign-owned firms’ observed productivity advantages. The OLI paradigm argues that a necessary condition for undertaking FDI is that the potential MNE possesses a proprietary ownership advantage relative to local rivals (e.g. a highly productive technology) to offset the increased costs of co-ordinating business activities across international borders.47 It follows that the observed productivity advantages of foreign-owned MNEs are embodied in their FDI inflows: either a highly productive new plant

FDI and Productivity Growth 69

is established via greenfield-FDI, or the technology in a pre-existing plant is upgraded (intra-firm technology transfer) following acquisitionFDI. In our model, in contrast, the possession of firm-specific ownership advantages is evidently unnecessary for greenfield-FDI: in the bottom three regions of Figure 4.5 the laggard firms, T and E, both build additional plants.48 Moreover, an increase in M ’s technological lead discourages technology-dissipating greenfield-FDI by M (the technological leader) but encourages technology-sourcing greenfield-FDI by T (the laggard). The OLI paradigm draws no strong distinction between greenfieldand acquisition-FDI. However, we found that – through its effect on concentration and thus E’s entry incentives – the greenfield/acquisition choice exerts an important influence on equilibrium industrial structures. Furthermore, although we set the model out by assuming that M is the purchaser, this assumption is not necessary to support our derived equilibria (Figure 4.5). Because the acquisition decision rule is co-operative and the integrated firm’s characteristics are independent of the purchaser’s identity, we could relabel the model with firm T, the lowproductivity incumbent, as the potential acquirer without altering its equilibruim predictions.49 Therefore, whenever incentives for technologyembodied acquisition-FDI exist, so do those for ‘cherry-picking’ acquisition-FDI, and the view that foreign MNEs’ productivity advantages are necessarily embodied in their FDI inflows is without theoretical support from our model.

4.5 Conclusion In this chapter we have surveyed a number of theoretical perspectives on the link between FDI flows and TFP growth. Sections 4.2 and 4.3 discussed, respectively, general and partial equilibrium models. In Section 4.2 we presented an open economy version of the Solow growth model, where North-to-South FDI flows both equalise the return to capital across countries and transfer technical knowledge internationally; and in Section 4.3 we analysed models of (a) how spillovers affect an MNE’s choice between greenfield-FDI and exporting, (b) trained worker mobility as a specific mechanism for spillovers, and (c) the relationship between greenfield-FDI flows and R&D performance. Section 4.4 considered how the form of FDI undertaken affects the FDI/productivity relationship. The most exciting recent development in theoretical modelling of the FDI/productivity relationship is the strategic analysis of firms’ international location (FDI) decisions when TFP is endogenously determined.

70 Ben Ferrett

These game-theoretic models (discussed in Section 4.3) permit consideration of the effects of strategic inter-firm rivalry, which – given that MNEs typically operate in oligopolistic industries – must be important in the determination of real-world outcomes. Three conclusions from this line of research are worth restating. First, the notion that the FDI decision can fruitfully be analysed as prior (i.e. exogenous) to spillover possibilities has been questioned by models showing that a technological leader’s incentive to produce abroad weakens as spillovers become more likely. Second, the assumption that R&D investments are necessarily prior to FDI (as in Dunning’s OLI paradigm) has been challenged by models where MNEs have larger output bases than national firms – because FDI ‘jumps’ trade costs – and, consequently, a stronger incentive to undertake R&D. Third, it appears that the form of FDI undertaken can exert a significant influence on equilibria. For example, if R&D investments are endogenously determined, then both consumers and firms might prefer acquisition-FDI to greenfield-FDI – despite the increase in concentration it implies – because R&D performance improves following acquisition (see Section 4.4). Despite these successes, the strategic approach to analysing FDI/ productivity linkages would benefit from development. New research questions exist that are amenable to examination within (reasonably straightforward) modifications of the current strategic frameworks: for example, analysis of the effect of national institutions – such as the strength of intellectual property rights (IPRs) afforded by the legal system – on firms’ FDI and R&D decisions when they behave strategically vis-à-vis each other and perhaps national governments. Moreover, a key methodological drawback is its partial equilibrium character, which limits the range of issues that can be addressed (e.g. labour market and inter-industry effects that are not well dealt with).50 Examining these and other issues will ensure that theoretical analysis of the FDI/ productivity relationship is an active and productive area of work for the foreseeable future.

Notes For helpful comments on an earlier draft, I am grateful to Nigel Driffield and the editors. All errors are entirely my own. 1. This classification is not meant to be exhaustive. For example, TFP growth due to increased specialisation in the production of intermediates (Adam Smith’s famous pin factory) does not fit well into it. 2. In the literature, spillovers without a well-specified mechanism are referred to as ‘demonstration effects’.

FDI and Productivity Growth 71 3. Note that the empirical studies sometimes interpret productivity as labour productivity, value-added output per worker. From a theoretical viewpoint, it is problematic to think of FDI as affecting labour productivity directly, because output per head is endogenously determined – given market conditions and technology – by profit-maximizing choices of firms. Unless some aspect of technology changes, there is no reason for output per head to change. Therefore, the focus in the theoretical models is on the impact of FDI on TFP. 4. If the firm’s production function exhibits constant returns to scale (and is therefore homothetic) and all factors of production are variable, then average cost is independent of output (and therefore equals marginal cost): factor intensities depend only on relative factor prices. TFP growth, which shifts the unit isoquant inwards in its entirety, must reduce average cost (the optimal, i.e. lowest, isocost line shifts inwards). 5. Note that if acquisition-FDI results in a more ‘concentrated’ market structure than greenfield-FDI, then – to the extent that rents of firms vary with ‘concentration’ – this poses significant problems for using value-added per worker to proxy ‘technology’. 6. Inspiration for the model presented here was drawn from Koizumi and Kopecky (1977), Findlay (1978), Ruffin (1979) and Wang (1990). For simplicity, there is no ‘general’ economic growth in our model (i.e. the production possibilities in the ‘advanced’ countries are unchanging through time), but only ‘catching up’ by developing regions. See Wang (1990) for a model of on-going growth. 7. Of course, sN sS is unnnecessary for rS rN under autarky. We could plausibly assume higher population growth in South, which would achieve the same result. 8. Nothing in our analysis implies (or requires) yS sNyN for all kS  kN. Figure 4.1 assumes this merely for graphical convenience. 9. It is important to note that nothing in the logical structure of our model implies this direction of FDI flow. For example, if AN AS but sS sN (so rN rS under autarky), then capital would flow from South to North upon liberalisation. 10. Therefore, to the extent that national TFPs differ, national capital : output ratios will also differ in equilibrium in an integrated global capital market. Lucas (1990, section II) provides an interesting numerical calibration, which suggests that the return to capital is virtually equalised between the United States of America and India once TFP differences are accounted for. 11. It is important to note that this does not mean that national income per head has fallen in the North. In addition to domestic production, Northern citizens also receive income from their exported capital. Because capital flows Southwards as long as its marginal product in South exceeds that in North (i.e. gain loss at the margin), Northern national income per head rises with capital liberalisation (although real wages fall). 12. Wang (1990) places more structure on the f(·, ·) function by assuming that both of its partial derivatives are strictly positive: AS grows more rapidly, the larger is the technological gap between North and South (AN/AS) and the more important is Northern FDI in the Southern capital stock (kM/kS). The latter hypothesis was first proposed by Findlay (1978). However, this extra

72 Ben Ferrett

13.

14.

15. 16.

17.

structure is unnecessary for our analysis of North-to-South technology transfer via FDI. Indeed, it is conceivable that dAS/dt may not be monotonically increasing in AN/AS. For example, if the South lacks sufficient ‘absorptive capacity’ (or ‘social capability’ à la Abramovitz, 1986), then dAS/dt may decrease in AN/AS for sufficiently large AN/AS (‘small gaps are easier to close than big ones’), perhaps making dAS/dt bell-shaped in AN/AS. Southern savings and FDI inflows from North are perfect substitutes for maintaining the Southern capital stock. The intuition behind this argument is easiest to grasp if one considers a small open economy facing a given world real interest rate. In this case (k is tied down by rW) there is a one-forone inverse relationship between Southern savings and FDI inflows from North. Compared to the Dixit–Stiglitz model of monopolistically competitive firms selling to consumers, Ethier relabels the firm sector ‘intermediate-goods producers’ and the consumer sector ‘final-goods producers’. The love-for-variety utility function of consumers in Dixit–Stiglitz becomes the production function for final goods: ceteris paribus, the output of final goods rises if the number of intermediate varieties rises, which is interpreted as a rise in TFP caused by greater specialisation in intermediates production. For related models of the ‘linkage effects’ of inward FDI, see Markusen and Venables (1999) and Barrios, Görg and Strobl (2005). Two further relevant references are Walz (1997), where FDI spillovers provide the only channel of North–South ‘international technology transfer’, and Glass and Saggi (1998), who examine in detail how the ‘absorptive capacity’ of the South affects its ability to imitate Northern firms. Therefore, the scope of spillovers is geographically bounded (i.e. localised). In our formulation spillovers occur when the more productive firm’s (process) technology becomes common knowledge. This contrasts with the modelling of spillovers in d’Aspremont and Jacquemin (1988), where a proportion, , of marginal cost spills over. In this case, Equation (3) would become

D(cM, cM  (1  )cL)  G D(cM  t, cL).

See Ferrett (2003c) for more discussion of this distinction. 18. ‘Less likely’ and ‘more likely’ here refer to the direction of change in the size of a region in parameter space. For given parameter values, there is no probabilistic element to the MNE’s decision. 19. For example, in a ‘linear’ Cournot duopoly (constant marginal cost, linear inverse demand) a given increase in a rival’s marginal cost causes the same changes in industry price (upwards) and firm production (net decrease) independently of marginal cost levels (see Shy, 1995). Therefore, a lowermarginal cost firm will benefit more from a given increase in the rival’s marginal cost (i.e. given increase in industry price) because its scale of production is larger. 20. Note that this result contrasts with the models of Rowthorn (1992) and Horstmann and Markusen (1992) where spillovers are not considered (i.e.  0 in our terminology) and increases in market size always make FDI more likely in equilibrium. (Those models implicitly set cM  cL.)

FDI and Productivity Growth 73 21. Both Fosfuri and Motta (1999) and Siotis (1999) analyse this case. 22. In Wang and Blomstrom (1992) the degree of spillover from MNE to local rival (the parallel in the framework sketched above is the level of ) is endogenously determined (via MNE investments in technology transfer and local-firm investments in imitation). Das (1987) models the MNE’s international technology transfer decision, but spillovers to local rivals are exogenous. However, both models assume that the MNE’s production location (i.e. use of FDI) is given. 23. Note that the interdependences between strategies of firms have different roots in the models of Horstmann and Markusen (1992) and Rowthorn (1992) compared to the hypothetical model described here. In Horstmann/Markusen and Rowthorn sunk costs on production in their home countries (both firmand plant-specific) create interdependences between the firms’ decisions: a firm may optimally choose to exit the industry in response to inward FDI by a rival. In contrast, in the framework sketched here spillover possibilities, which vary with the two firms’ locations, create interdependences. 24. A key simplification is that the host-country product market is constrained to be of the same size in both periods. I also ignore complications relating to the local firm’s ‘absorptive capacity’. Glass and Saggi (2002) also model spillovers through workers’ mobility, and they consider the host-country government’s optimal policy response given a bidding instrument. 25. Of course, it is possible to imagine the local firm already being in the product market, albeit owning a highly inefficient technology relative to the MNE. 26. The hiring process is a first-price auction: the firms simultaneously and irreversibly make take-it-or-leave-it offers to the mobile trained worker, and we focus on the Pareto dominant equilibrium. We assume that the MNE can only write a one-period contract with the worker. 27. To some extent, it is natural to think of D as determined by the type of training supplied by the MNE in Stage 1. With training in firm-specific technology (e.g. product- and process-specific), D will be ‘small’; but with ‘general’ training – that can readily be used in almost any firm – D will be ‘large.’ 28. If the MNE loses the trained worker at the auction, it must train another (without cost) in period 2. Note that even if it loses the auction, the MNE will continue local production in period 2 because it has already built a plant there, allowing it to ‘jump’ the trade cost. 29. This guarantees that the MNE will optimally choose X in period 1 for some (D, MX)-pairs with MX M  G, our maintained assumption. 30. Independently of its period-1 choice, the MNE will enjoy one period with profits of M  G. If it originally chooses FDI, this occurs in period 1, whereas it occurs in period 2 following exporting. 31. The results we discuss are those for the ‘blockaded entry’ game in Ferrett (2002). Petit and Sanna-Randaccio (2000) analyse a similar game with spillovers, but their analysis does not admit an analytical solution. 32. I show in Ferrett (2002) that sufficient conditions for Figure 4.4 are G  I 0 and c  t ‘sufficiently large’. See Ferrett (2003c) for a complete characterisation of permissible parameter values. 33. In addition to p, , there are four other structural parameters: G, I, t, c. Changing their values will shift the inter-regional boundaries in Figure 4.4. For example, increasing G will move the (2R, 2R) region upwards and will

74 Ben Ferrett

34.

35. 36. 37. 38.

39.

40.

41.

42.

squeeze the (1R, 1R)/(1N, 2R) region from below, making greenfield-FDI ‘less likely’. However, increasing t will have the opposite effects (as well as shifting the upper boundary of the (1R, 1R)/(1N, 2R) region upwards), as ‘tariffjumping’ FDI becomes more attractive. This amounts to analysing how a hypothetical prior commitment to undertake R&D (respectively, greenfield-FDI) affects the incremental profitability of also undertaking greenfield-FDI (respectively, R&D). It is important to note that this does not relate directly to the determination of equilibrium industrial structures, which are determined as Nash equilibria (i.e. mutual best responses) rather than via comparisons of (joint) profit levels. Indeed, the international duopoly game in Figure 4.4 can exhibit Prisoner’s Dilemma characteristics. Petit and Sanna-Randaccio (1998, 2000) undertake a similar analysis with similar results. However, 2R is occasionally chosen over 1R (e.g. whenever 2R arises in equilibrium). See Ferrett (2003c) for a proof. This ambiguity on the form of FDI in general equilibrium models is not surprising. All assume free entry (i.e. perfect or monopolistic competition), so the number of plants is tied down in the long run by a zero-profit condition, making the greenfield/acquisition choice irrelevant for equilibrium ‘concentration’. By way of motivation, there are at least two general reasons for emphasising this distinction. First, the intuitive industrial-organisation response that the greenfield/acquisition distinction affects ‘concentration’ is confirmed by applied work; indeed, UNCTAD (2000) finds that a persistent ‘concentration effect’ is the most significant difference between greenfield- and acquisitionFDI. Second, neither type of FDI is empirically trivial; for example, UNCTAD (2000) estimates that the ratio of acquisition- to greenfield-FDI in aggregate global FDI flows was 4 : 1 in the late 1990s, and since then cross-border M&A flows have collapsed. The merger decision is settled by applying the co-operative decision rule of Salant et al. (1983). The game is solved backwards to Stage 1 to obtain subgame perfect Nash equilibria conditional on whether or not the merger occurs in Stage 0; then, the payoff to the integrated incumbent is compared to the payoffs of the incumbents acting independently in the no-merger equilibrium to assess the profitability of merger. Note that the cause of Pareto improving acquisition-FDI in this model (an ‘output base’ effect) differs from that in Horn and Persson (2001), where mergers are associated with savings in fixed and variable production costs (‘synergies’). For the United Kingdom this ‘productivity gap’ has been documented by Davies and Lyons (1991), Griffith (1999), Girma et al. (2001) and Oulton (2001). In particular, Oulton concludes that the labour productivity of foreign-owned firms has been continuously around 40 per cent higher than in UK-owned firms and that this ‘productivity advantage’ is not entirely due to a concentration of foreign-owned firms in industries with particularly high physical and human capital intensities. International evidence is provided by the Globerman et al. (1994) study on Canada and by Doms and Jensen’s (1998) study of US manufacturing, which found that the significant

FDI and Productivity Growth 75

43. 44. 45.

46.

47.

48.

49. 50.

difference – in terms of ‘productivity gaps’ – is between MNEs and non-MNEs, not between foreign- and domestic-owned firms. This follows Fosfuri and Motta (1999) and Siotis (1999). Therefore, spillovers are geographically bounded in extent. The game’s equilibrium is obtained by solving backwards to Stage 2 for a subgame perfect Nash equilibrium given M ’s choice. M ’s choice between acquisition-FDI and her preferred candidate of {exporting, greenfield-FDI} is determined by the Salant et al. (1983) ‘profitability’ criterion. Owing to the game’s complexity, I am only able to solve backwards to Stage 2 analytically in Ferrett (2003b). M’s equilibrium choices are investigated using numerical simulations. See Markusen (1995) for an introduction to the OLI paradigm. I have argued elsewhere (Ferrett, 2003c) that OLI’s conclusions are consistent with the assumption of monopolistic competition in product markets. Helpman et al. (2004) provide a formal model of the choice between FDI and exporting along these lines. In the ‘conditional entry’ region T’s choice between X and G depends on the level of trade costs (the ‘proximity-concentration tradeoff’ of Brainard, 1997), as does E’s choice between one plant and two in the upper ‘inevitable entry’ region. This would have to preserve M’s ability to move before T if no acquisition occurred. Developing a tractable model of oligopoly in general equilibrium is a problem with a long and distinguished pedigree in economics. See Neary (2003) for a recent analysis.

References Abramovitz, M. (1986), ‘Catching Up, Forging Ahead, and Falling Behind’, Journal of Economic History, 46, 385–406. Barrios, S., H. Görg and E. Strobl (2005), ‘Foreign Direct Investment, Competition and Industrial Development in the Host Country’, European Economic Review, 49, 1761–84. Brainard, S. L. (1997), ‘An Empirical Assessment of the Proximity-Concentration Trade-off between Multinational Sales and Trade’, American Economic Review, 87, 520–44. Caves, R. E. (1996), Multinational Enterprise and Economic Analysis, Cambridge University Press. Das, S. (1987), ‘Externalities and Technology Transfer through Multinational Corporations: A Theoretical Analysis’, Journal of International Economics, 22, 171–82. d’Aspremont, C. and A. Jacquemin (1988), ‘Cooperative and Noncooperative R&D in Duopoly with Spillovers’, American Economic Review, 78, 1133–7. Davies, S. W. and B. R. Lyons (1991), ‘Characterising Relative Performance: The Productivity Advantage of Foreign Owned Firms in the UK’, Oxford Economic Papers, 43, 584–95. Dixit, A. K. and J. E. Stiglitz (1977), ‘Monopolistic Competition and Optimum Product Diversity’, American Economic Review, 67, 297–308.

76 Ben Ferrett Doms, M. E. and J. B. Jensen (1998), ‘Comparing Wages, Skills, and Productivity between Domestically and Foreign-Owned Manufacturing Establishments in the United States’, in R. E. Baldwin, R. E. Lipsey and J. D. Richardson, eds., Geography and Ownership as Bases for Economic Accounting, University of Chicago Press. Driffield, N. and J. H. Love (2003), ‘Foreign Direct Investment, Technology Sourcing and Reverse Spillovers’, Manchester School, 71, 659–72. Dunning, J. H. (1977), ‘Trade, Location of Economic Activity and the MNE: A Search for an Eclectic Approach’, in B. Ohlin, P. O. Hesselborn and P. M. Wijkman, eds, The International Allocation of Economic Activity: Proceedings of a Nobel Symposium held at Stockholm, Macmillan. Ethier, W. J. (1982), ‘National and International Returns to Scale in the Modern Theory of International Trade’, American Economic Review, 72, 389–405. Ferrett, B. (2002), ‘Endogenous R&D and Entry in an International Oligopoly’, Research Paper 2002/22, Leverhulme Centre for Research on Globalisation and Economic Policy (GEP), School of Economics, University of Nottingham. Available at www.nottingham.ac.uk/economics/leverhulme/research_papers/ 02_22.htm Ferrett, B. (2003a), ‘Greenfield Investment Versus Acquisition: Positive Analysis’, Research Paper 2003/02, Leverhulme Centre for Research on Globalisation and Economic Policy (GEP), School of Economics, University of Nottingham. Available at www.nottingham.ac.uk/economics/leverhulme/research_papers/ 03_02.htm Ferrett, B. (2003b), ‘Intra- and Inter-Firm Technology Transfer in an International Oligopoly’, Research Paper 2003/49, Leverhulme Centre for Research on Globalisation and Economic Policy (GEP), School of Economics, University of Nottingham. Available at www.nottingham.ac.uk/economics/leverhulme/ research_papers/03_49.htm Ferrett, B. (2003c), Strategic Decisions of Multinational Enterprises: Foreign Direct Investment and Technology, Ph.D. thesis submitted to Warwick University. Findlay, R. (1978), ‘Relative Backwardness, Direct Foreign Investment, and the Transfer of Technology: A Simple Dynamic Model’, Quarterly Journal of Economics, 92, 1–16. Fosfuri, A. and M. Motta (1999), ‘Multinationals without Advantages’, Scandinavian Journal of Economics, 101, 617–30. Fosfuri, A., M. Motta and T. Rønde (2001), ‘Foreign Direct Investment and Spillovers through Workers’ Mobility’, Journal of International Economics, 53, 205–22. Girma, S., D. Greenaway and K. Wakelin (2001), ‘Who Benefits from Foreign Direct Investment in the UK?’, Scottish Journal of Political Economy, 48, 119–33. Glass, A. J. and K. Saggi (1998), ‘International Technology Transfer and the Technology Gap’, Journal of Development Economics, 55, 369–98. Glass, A. J. and K. Saggi (1999), ‘Foreign Direct Investment and the Nature of R&D’, Canadian Journal of Economics, 32, 92–117. Glass, A. J. and K. Saggi (2002), ‘Multinational Firms and Technology Transfer’, Scandinavian Journal of Economics, 104, 495–513. Globerman, S., J. C. Ries and I. Vertinsky (1994), ‘The Economic Performance of Foreign Affiliates in Canada’, Canadian Journal of Economics, 27, 143–56. Griffith, R. (1999), ‘Using the ARD Establishment Level Data to Look at Foreign Ownership and Productivity in the United Kingdom’, Economic Journal, 109, F416–42.

FDI and Productivity Growth 77 Grossman, G. M. and E. Helpman (1991), Innovation and Growth in the Global Economy, MIT Press. Haaland, J. I. and I. Wooton (1999), ‘International Competition for Multinational Investment’, Scandinavian Journal of Economics, 101, 631–49. Helpman, E. (1993), ‘Innovation, Imitation, and Intellectual Property Rights’, Econometrica, 61, 1247–80. Helpman, E., M. J. Melitz and S. R. Yeaple (2004), ‘Export Versus FDI with Heterogeneous Firms’, American Economic Review, 94, 300–16. Horn, H. and L. Persson (2001), ‘The Equilibrium Ownership of an International Oligopoly’, Journal of International Economics, 53, 307–33. Horstmann, I. J. and J. R. Markusen (1992), ‘Endogenous Market Structures in International Trade (natura facit saltum)’, Journal of International Economics, 32, 109–29. Koizumi, T. and K. J. Kopecky (1977), ‘Economic Growth, Capital Movements and the International Transfer of Technical Knowledge’, Journal of International Economics, 7, 45–65. Long, N. V. and N. Vousden (1995), ‘The Effects of Trade Liberalization on Cost Reducing Horizontal Mergers’, Review of International Economics, 3, 141–55. Lucas, R. E., Jr (1990), ‘Why Doesn’t Capital Flow from Rich to Poor Countries?’, American Economic Review (Papers and Proceedings), 80, 92–6. Markusen, J. R. (1995), ‘The Boundaries of Multinational Enterprises and the Theory of International Trade’, Journal of Economic Perspectives, 9, 169–89. Markusen, J. R. (2002), Multinational Firms and the Theory of International Trade, MIT Press. Markusen, J. R. and A. J. Venables (1999), ‘Foreign Direct Investment as a Catalyst for Industrial Development’, European Economic Review, 43, 335–56. Motta, M. (1992), ‘Multinational Firms and the Tariff-Jumping Argument: A Game-Theoretic Analysis with Some Unconventional Conclusions’, European Economic Review, 36, 1557–71. Neary, J. P. (2003), ‘Globalization and Market Structure’, Journal of the European Economic Association, 1, 245–71. Oulton, N. (2001), ‘Why Do Foreign-Owned Firms in the UK Have Higher Labour Productivity?’, in N. Pain (ed.), Inward Investment, Technological Change and Growth: The Impact of Multinational Corporations on the UK Economy, London: Palgrave, 122–61. Pack, H. and K. Saggi (2001), ‘Vertical Technology Transfer via International Outsourcing’, Journal of Development Economics, 65, 389–415. Petit, M.-L. and F. Sanna-Randaccio (1998), ‘Technological Innovation and Multinational Expansion: A Two-Way Link?’, Journal of Economics (Zeitschrift für Nationalökonomie), 68, 1–26. Petit, M.-L. and F. Sanna-Randaccio (2000), ‘Endogenous R&D and Foreign Direct Investment in International Oligopolies’, International Journal of Industrial Organization, 18, 339–67. Rodriguez-Clare, A. (1996), ‘Multinationals, Linkages, and Economic Development’, American Economic Review, 86, 852–73. Rowthorn, R. E. (1992), ‘Intra-Industry Trade and Investment under Oligopoly: The Role of Market Size’, Economic Journal, 102, 402–14.

78 Ben Ferrett Ruffin, R. J. (1979), ‘Growth and The Long-Run Theory of International Capital Movements’, American Economic Review, 69, 832–42. Salant, S., S. Switzer and R. Reynolds (1983), ‘Losses from Horizontal Merger: The Effects of an Exogenous Change in Industry Structure on Cournot-Nash equilibrium’, Quarterly Journal of Economics, 98, 185–99. Shy, O. (1995), Industrial Organization: Theory and Applications, MIT Press. Siotis, G. (1999), ‘Foreign Direct Investment and Firms’ Capabilities’, Journal of Economics and Management Strategy, 8, 251–70. Solow, R. M. (1956), ‘A Contribution to the Theory of Economic Growth’, Quarterly Journal of Economics, 70, 65–94. UNCTAD (2000), World Investment Report 2000: Cross-Border Mergers and Acquisitions and Development (Overview), New York: United Nations. Walz, U. (1997), ‘Innovation, Foreign Direct Investment and Growth’, Economica, 64, 63–79. Wang, J.-Y. (1990), ‘Growth, Technology Transfer, and the Long-Run Theory of International Capital Movements’, Journal of International Economics, 29, 255–71. Wang, J.-Y. and M. Blomstrom (1992), ‘Foreign Investment and Technology Transfer: A Simple Model’, European Economic Review, 36, 137–55. Williamson, O. E. (1968), ‘Economies as an Antitrust Defense: The Welfare Tradeoffs’, American Economic Review, 58, 18–36.

5 Empirical Evidence on Foreign Ownership and Productivity Growth Sourafel Girma, Steve Thompson and Peter Wright

5.1 Foreign ownership and productivity It has frequently been noted that foreign owned firms appear to have higher levels of productivity than their domestic counterparts. The most common explanation for such a differential is that multinational affiliates enjoy firm-specific proprietary assets, which may be technological or may be intangible factors such as organisational or brand name advantages, which gives the firm a productivity advantage over its domestic rivals. Indeed, the dominant ‘internalisation theory’ primarily considers the multinational firm as a means to transfer firm-specific proprietary assets whilst avoiding the costs associated with international transactions.1 The entry of more efficient foreign firms into the domestic market may have profound implications for the domestic economy. Caves (1974) noted three distinctive potential efficiency consequences of inward foreign direct investment for the host economy: increased technical efficiency; increased allocative efficiency and technology transfer. Increased technical efficiency may arise since market entry by more efficient foreign firms may expose previous sheltered market positions which allowed the survival of inefficient producers. This technical inefficiency may manifest itself in the survival of sub-optimal scale plants2 or as ‘X-inefficiency’ – that is, operations beneath the existing production surface (Caves, 1974). Increased allocative efficiency will occur if entry by foreign firms occurs in concentrated industries. Caves argues that this is likely to be the case since it has been widely observed that FDI tends to occur in industries where high levels of advertising and R&D appear to constitute 79

80 Sourafel Girma, Steve Thompson and Peter Wright

effective barriers to entry since these are precisely the industries where firm-specific advantages, including proprietary technology, international brand-names, and so on, create the potential for international expansion. Effective MNE entry is therefore likely to challenge existing domestic market power and thus produce allocative efficiency gains as monopolistic output restrictions fall.3 Finally, since the ‘internalisation view’ of the MNE (Buckley and Casson, 1976) sees FDI as a mechanism for transferring technology across international boundaries when the transaction costs of using appropriate markets4 are excessive, foreign firms are a potential source of technology transfer between countries. The line of argument advanced by Caves (1974) has received much attention in empirical work since, from a policy context, it is clearly important to distinguish the impact that incoming MNEs have on reducing inefficiency by increasing competition, from the spillover productivity gains that these firms bring to the host economy. If the former effects dominate, a tougher domestic competition policy aimed perhaps at lowering entry barriers and/or reducing anti-competitive conduct, may be a more cost-effective way of increasing productivity than that of seeking to attract in-coming FDI. Conversely, if the spillover benefits from technology transfer (perhaps defined broadly to include best practice arrangements) dominate, then a policy to attract FDI may be justified. While allocative efficiency gains are systemic and are established through inference rather than by direct observation, it should in principle be possible to distinguish between reductions in technical inefficiency and gains via technology transfer. Increased competitive rivalry will remove inefficient units and would thus raise host productivity by reducing the average distance to the production surface. Technology transfer, for example, via the import of process innovations, will push out that surface for the host economy. While the three efficiency effects suggested by Caves (1974) have dominated the subsequent literature, a fourth potential source of efficiency gain concerns lower adjustment costs of labour demand. Whilst institutional factors have been traditionally assumed to work against the MNE, a foreign company may possess some institutional advantages over purely domestic firms. The threat of transferring production elsewhere gives the MNE a powerful bargaining tool with labour which works to the firm’s advantage not just in the setting of wages, a distributional concern, but also when it comes to issues concerning labour utilisation and lay-offs (Cowling and Sugden, 1987). Furthermore, incoming MNEs

Foreign Ownership and Productivity Growth 81

may be able to negotiate exemptions from the existing legal, regulatory and union restrictions on labour usage. Work within the Leverhulme Centre at Nottingham has sought to examine Caves’ (1974) contentions regarding the impact of FDI on the host economy. A number of papers have specifically addressed the issue of whether productivity spillovers occur between foreign and domestic firms and sought to ascertain the route by which such spillovers might occur. These papers are discussed in detail in Chapter 6 of this volume (Görg and Hijzen, 2005). This chapter takes one step back and asks, perhaps the more fundamental question, of whether foreign firms are indeed more productive than domestic firms as Caves presumes. This apparently simple question proves surprisingly difficult to answer. The remainder of this chapter begins by briefly examining the methodological issues associated with measuring productivity differentials between domestic and foreign firms. In this regard it is suggested that compositional effects and omitted variables might explain much of the apparent performance superiority of foreign firms. Section 5.3 then summarises work that has been conducted within the Leverhulme Centre which has sought to circumvent these problems, by using foreign acquisition to isolate the true extent of the productivity differential between foreign and domestic firms. Section 5.4 concludes.

5.2 Empirical work on foreign ownership and productivity A simple comparison of mean productivity levels in domestic- and foreign-owned firms strongly supports Caves’s contention regarding the superiority of the latter. The figures presented in Table 5.1, taken from Conyon et al. (2002b), indicate that mean labour productivity is 29 per cent higher in foreign owned firms.5 However, foreign- and domestic-owned enterprises appear to exhibit significant differences at the plant, firm and industry levels. Table 5.1 indicates for instance that foreign firms are much larger on average than domestic firms. They also pay their workers more than 10 per cent higher wages, an indication that the characteristics of their workforces may well differ. Failure to account for these differences will undoubtedly lead to a biased estimate of the extent of multinational productivity advantages. Capital vintage is also rarely controlled for in cross-sectional work but may generate significant performance differences that are not attributable to multinational ownership per se. More seriously, multinational entrants may also be attracted to more productive and/or more

82 Sourafel Girma, Steve Thompson and Peter Wright Table 5.1 Average employment, wages and labour productivity of domestic- and foreign-owned firms in UK manufacturing industry Variables

1989

1994

Domestic Employment Wage rate Labour productivity

414 (1650) 11.35 (3.87) 72.45 (163.77)

327 (1369) 13.29 (4.31) 74.48 (96.77)

Foreign Employment Wage rate Labour productivity

549 (1641) 12.71 (3.90) 104.68 (137.67)

434 (1098) 15.22 (4.40) 118.28 (154.24)

Notes: (i) Variables are given in real terms. (ii) Wage rate and labour productivity are in £’000. (iii) The wage rate is defined as the wage bill per worker. (iv) Labour productivity is sales per worker. (v) Standard deviations are given in parentheses. Source: Conyon et al. (2002b).

profitable industries leading to simultaneity bias in the cross-sectional evidence (Tybout, 2000). Indeed these contentions are strongly borne out by the data. While it is suspected that the inclusion of plant and industry controls would reduce, if not eliminate, the residual multinational effect, the specification of appropriate control variables is normally problematic (Griffith, 1999), generally requiring information not included in the standard data sets. Davies and Lyons (1991) for instance suggest that 40 per cent of the observed productivity differential arises from the fact that multinationals tend to locate in high-productivity industries. The remainder they attribute to intrinsic advantages. The suspicion remains however that the introduction of more, and better, controls into their estimating equation would reduce this differential further. Their results for instance contrast with the conclusions of Vendrell-Alda (1978) who, on examining Argentinean manufacturing industries, concluded that the entire multinational productivity differential could be explained by differences in market structural characteristics, leaving no residual effect. Inter-firm productivity comparisons strictly require knowledge of physical outputs of a homogeneous good or, equivalently, the monetary value of homogeneous goods sold on perfect markets. However, in practice productivity analysis has to use monetary values of output, usually sales revenue or value-added, generated from the sales of differentiated goods on imperfect markets. This is particularly relevant to any domestic

Foreign Ownership and Productivity Growth 83

versus foreign firm comparison since the latter’s brand name is often precisely the type of intangible asset that MNEs are assumed to transfer to their advantage. However, such an advantage may result in the MNE’s ability to generate a higher price–cost margin. For example, suppose a domestic brewery or pharmaceutical company is acquired by a leading multinational that henceforth uses the same plant to produce its brand or patented product. It is possible that no change at all occurs in the physical quantities of output and inputs involved, but that revenue rises, perhaps substantially, through the brand name effect. Such a productivity gain is real in the sense that it rests on an increase in value, reflecting an increased willingness to pay, but it is removed from conventional notion of movement towards a production surface. The simultaneity problem is intensified when it is recalled that foreign direct investment (FDI) is particularly prevalent in advertising- and R&D-intensive industries; see Caves (1996) for a review of the evidence. Sutton (1998) has demonstrated theoretically and empirically that these features also tend to determine industry concentration. While concentration, in turn, typically exerts a positive influence on profit margins. This inevitably implies that isolating the ‘pure’ multinational effect from the attendant industry effects can never be straightforward.

5.3 Leverhulme Centre work on productivity and wages The suspicion that controlling for more and more differences between foreign and domestic firms might serve to eliminate the productivity differential has motivated much of the Leverhulme work in this area. This issue is tackled directly in work undertaken by Conyon et al. (2002b). They use panel data on firms which allow them to track the productivity of their sample of firms over time and to examine the implication of both domestic and multinational ownership change. An exploration of the short- and medium-term productivity effects of a change in ownership for an existing firm should circumvent many of the control problems inherent in much of the literature. First, it is straightforward to determine whether the acquired firm changes industry. If not, it can be assumed that industry effects will not impact on the analysis. Second, since the firm will initially work with existing physical and human capital, it seems reasonable to assume no changes in factor quality, including major changes in capital vintage. Third, whilst there may be performance changes associated with a change in ownership per se (Brown and Medoff, 1988; Conyon et al., 2002a), a comparison of

84 Sourafel Girma, Steve Thompson and Peter Wright

foreign- and-domestic-acquisitions should reveal the existence, if any, and extent of any distinctive foreign ownership effect. The methodology adopted by Conyon et al. (2002b) therefore proceeds by estimating a labour productivity equation of the following form: (q  l)it 

t  
(k  l)it  f
i  
1Dit  
2Fit  it

(1)

Where i and t index firms and time periods respectively, (q  l) is the logarithm of output per worker and (k  l) is capital intensity.6 In addition, year dummies (

t ) are included to control for aggregate shocks, and firmspecific fixed effects (f
) for permanent differences across firms. Finally, they allow for the possibility that domestic- and foreign-acquisitions may affect productivity in different ways by constructing separate dummies for take-overs by foreign (F) and domestic (D) companies. After controlling for capital intensity, fixed assets and autonomous technical change via the use of time dummies and firm-fixed effects, the results in Table 5.2 indicate that foreign firms have levels of labour Table 5.2 The impact of ownership on labour productivity

Level of productivity (1) Capital intensity Domestic acquisition Foreign acquisition Horizontal Non-horizontal USA EU Others

0.05 (8.59)** 0.00 (1.27) 14.10 (8.31)**

(2)

Endogeneity corrected productivity levels (3)

0.05 (8.59)**

0.054 (4.05)**

2.00 (1.27) 0.00 (0.34) 15.30 (5.23)** 14.10 (6.25)** 13.00 (3.88)**

3.45 (0.84) 1.38 (0.64) 16.44 (2.42)* 11.92 (2.34)* 12.33 (2.51)*

Notes: (i) All estimations are based on 9648 observations corresponding to 1102 firms. Year effects are included. (ii) Absolute values of t-statistics are given in parentheses. * Significant at 5%; ** significant at 1%. (iii) Capital intensity is measured as fixed assets per worker. (iv) Coefficients on acquisition dummies in the level and growth equations are percentage and percentage point differentials respectively. (v) In column C the probability of acquisition in each year is used as an instrument for the acquisition dummies. Source: Conyon et al. (2002b)

Foreign Ownership and Productivity Growth 85

productivity 14 per cent above that of the equivalent domestic firm. There is however considerably heterogeneity of ownership across the sample. Column (2) indicates that whilst a productivity differential is observed with all foreign firms, firms from the United States have the greatest productivity advantage (15.3%), as opposed to firms from the EU (14.1%) and elsewhere (13%). Whilst using acquisition in order to measure the impact of ownership on productivity has a number of distinct advantages over previous studies, the possibility exists that the merger dummy may be endogenous. If the productivity of the firm plays some role in the likelihood of it being taken over, then the impact of ownership change may be biased. Column (3) therefore presents estimates in which the acquisition dummies are instrumented by the probability of acquisition.7 After this correction, the broad thrust of the results are preserved, though serve to emphasise the productivity advantages of the United States over other foreign firms.

5.3.1

Total factor productivity (TFP)

Girma et al. (2002) also examine the impact of ownership change on total factor productivity. These results are reported in Table 5.3. Once again the raw productivity advantage of foreign-owned firms is apparent in the data. After controlling for autonomous technological change and firm-fixed effects, the results indicate that foreign firms have a TFP advantage of 8 per cent over equivalent domestic firms. The IV fixed-effect results, shown in column B, yield remarkably similar estimates.8 Column C investigates the effect on production further, by allowing foreign acquisition to additionally impact on the marginal productivity of the factors of production. The results indicate that the marginal productivity of capital has doubled in the post foreign acquisition period. This is in contrast to the impact of the marginal productivity of labour, which has if anything slightly decreased, though this effect falls short of conventional significance levels. Note that this less constrained equation indicates even larger TFP effects for foreign take-overs of some 14 per cent. Turning to the origin of the foreign acquirer, column E in Table 5.3 indicates that although the increase in TFP is observed across all types of foreign acquisition, as with labour productivity the greatest increase is observed for US firms.

86 Table 5.3 The impact of ownership on total factor productivity Fixed effect with no interaction A TFP effects Horizontal Non-horizontal Foreign

0.011 (0.80) 0.026 (2.77)** 0.082 (6.81)**

Fixed effect IV with no interaction B

0.024 (1.64) 0.016 (1.63) 0.086 (6.28)**

Fixed effect with interaction C

0.113 (1.37) 0.092 (1.94) 0.140 (2.37)*

Fixed effect IV with interaction D

0.165 (1.85) 0.128 (2.52)* 0.146 (2.11)*

USA

0.511 (59.72)**

0.510 (59.64)**

Acquisition interactions Horizontal Non-horizontal Foreign

0.516 (58.99)**

0.517 (58.85)**

0.515 (40.70)**

0.034 (1.79) 0.025 (1.94) 0.050 (2.77)**

0.042 (2.12)* 0.042 (3.01)** 0.040 (1.88)

0.043 (2.17)* 0.043 (3.05)**

USA

0.072 (0.43) 0.001 (0.00)

Non-USA Capital

0.027 (5.69)**

0.027 (5.76)**

Acquisition interactions Domestic

0.025 (5.03)**

0.009 (0.64) 0.001 (0.08) 0.025 (2.21)*

Non-horizontal Foreign

0.025 (4.97)**

0.020 (3.36)**

0.008 (0.53) 0.006 (0.66) 0.017 (1.33)

0.010 (0.69) 0.008 (0.87)

USA

0.10 (2.11)* 0.041 (0.55)

Non-USA Intermediate inputs

0.147 (1.63) 0.112 (2.18)*

0.148 (4.75)** 0.065 (4.09)**

Non-USA Labour

Fixed effect IV with interaction E

0.383 (73.61)**

Hausman test p-value R-squared

0.382 (73.51)**

0.382 (73.57)**

0.7402 0.76

0.382 (73.47)**

0.384 (73.14)**

0.6452 0.76

Notes: (i) All estimations are based on 9648 observations corresponding to 1102 firms. Absolute values of t-statistics are given in parentheses. * Significant at 5%; ** significant at 1%. (ii) Capital intensity is measured as fixed assets per worker. (iii) Coefficients on acquisition dummies in the level and growth equations are percentage and percentage point differentials respectively. Source: Girma et al. (2002).

Foreign Ownership and Productivity Growth 87

5.3.2 The impact of productivity differentials on wages The results of Conyon et al. (2002b) and Girma et al. (2002) offer robust support for the contention that foreign-owned firms have an intrinsic productivity advantage over their domestic rivals. To what extent does this difference impact on other aspects of the firm? Conyon et al. (2002b) were particularly interested in tracing out the labour market ramifications. Table 5.1 shows that foreign firms pay higher average wages than their domestic counterparts. However, rent-sharing arguments might suggest that at least some of this differential might be attributable to the higher levels of productivity observed in foreign firms. This issue is explored in Conyon et al. (2002b), who also examine the impact of ownership change on wage rates. Their results are presented in Table 5.4. Table 5.4 The impact of ownership changes on wage rates by type of acquisition Endogeneity wage levels corrected

Wage levels A

B

C

D

E

F

G

0.05 (4.93)** 0.01 (3.75)** 2.17 (3.79)** 0.00 (0.00)

0.09 (8.11)** 0.03 (8.12)**

0.05 (5.03)** 0.01 (3.91)**

0.111 (4.37)** 0.031 (5.39)**

0.065 (3.77)** 0.019 (3.37)**

2.11 (3.16)** 3.44 (3.46)**

0.09 (7.92)** 0.004 (1.84) 2.10 (3.13)** 3.38 (3.42)**

4.70 (2.78)** 3.90 (2.56)* 3.20 (2.56)* 3.12 (3.45)** 0.00 (0.00)

0.00 (0.00) 1.50 (1.18) 0.10 (.51) 3.50 (4.48)** 0.80 (1.29) 0.29 (52.30)** 0.019 (3.96)**

Industry wage Fixed assets Domestic Foreign USA EU Other foreign Horizontal Nonhorizontal Labour productivity Employment

0.30 (5.42)** 0.07 (12.94)**

6.60 (2.14)* 3.69 (1.43) 3.91 (2.22)* 2.69 (1.25) 1.0 (0.69)

0.074 (5.88)**

1.60 (0.52) 1.28 (0.51) 1.60 (1.05) 3.20 (1.83) 0.40 (0.32) 0.29 (11.44)** 0.020 (1.58)

Notes: (i) All estimations are based on 9648 observations corresponding to 1102 firms. Year effects are included. (ii) The coefficient on acquisition dummies gives the percentage change in wage rates following ownership change. (iii) Absolute values of t-statistics are given in parentheses. * Significant at 5%; ** significant at 1%. (iv) In columns F and G the probability of acquisition in each year is used as an instrument for the acquisition dummies. Source: Conyon et al. (2002b).

88 Sourafel Girma, Steve Thompson and Peter Wright

The crude wage differential paid to workers in foreign owned firms is evident in column A. Workers whose firms are acquired by a foreign company experience wages increases of 3.44 per cent. This is in sharp contrast to those in firms acquired domestically, who experience wage declines of 2.11 per cent. This effect remains apparent when controls are included for the average wage in the industry and firm size in column B. However, after controlling for labour productivity, no measured impact of ownership change is observed (column C). This result suggests that the sole reason why individuals in foreign-owned firms are paid more is because such firms have higher levels of productivity, from which the worker benefits in terms of higher wages. This is in sharp contrast to workers whose firms are acquired domestically since such firms do not experience the same increase in productivity as that associated with foreign acquisition. Indeed it appears that management may use the opportunity occasioned by ownership change to transfer wealth from workers to shareholders, with the average wages in such firms falling by more than 2 per cent. Conyon et al. (2002b) make a number of other interesting observations with regard to the impact of foreign ownership change on wages. First, the increase in productivity does not result from downsizing by the acquiring firm, since no significant employment effects are observed. Second, some heterogeneity is observed in terms of average wages outcomes dependant on the country of origin of the acquirer. Workers whose firms are acquired by US companies experience a 4.7 per cent wage increase compared to 3.9 per cent for EU acquisitions and 3.2 per cent for other foreign acquisitions. They suggest however that these differentials in wages simply reflect the productivity differentials reported earlier, and the differential wage effect disappears once labour productivity is included amongst the controls (column D).

5.4 Conclusions Although the results of Conyon et al. (2002b) and Girma et al. (2002) offer strong support for the contention that foreign-owned firms have a productivity advantage over UK firms on average, a number of interesting questions remain to be addressed. First, does the ‘foreign effect’ decay over time for incoming MNEs? If technology transfer to rivals occurs and intensified competition removes some inefficient domestic producers we might expect to observe convergence. Anecdotal support for this comes in the high-profile case of the US car industry where Japanese and Korean satellite plants sprang up

Foreign Ownership and Productivity Growth 89

in the late 1980s/early 1990s, sometimes as joint ventures with the US carmakers. The latter appear to have successfully adopted the manufacturing and logistics practices of their Asian rivals and subsequently arrested and partially reversed the loss of market share. Second, the evidence reviewed here has overwhelmingly concerned manufacturing industries. Measurement difficulties notwithstanding, is there a comparable differential in the service sector? This is an important policy question given the internationalisation of industries such as banking, insurance and telecoms. Third, how does the productivity differential appear when we compare advertising- and/or R&D-intensive industries with their respective controls? The answers to these and the questions originally posed will have important implications for the design of policy towards corporate governance and mergers within the EU. Within the EU there are substantial differences between countries, such as the United Kingdom that have governance institutions and ownership arrangements which facilitate outside acquisition and countries, such as Germany, where interlocking shareholdings, bank involvement and a relatively illiquid stock market has traditionally restricted acquisition. If foreign acquisition represents a significant mechanism for the transmission of productivity improvements, this suggests that corporate governance reforms within the EU should be such as to encourage mergers and acquisitions. With the emergence of more sophisticated datasets, the prospect of addressing these questions becomes a possibility. The recent linking of the Annual Respondents Database (ARD) and the Annual Foreign Direct Investment register (AFDI) has allowed Criscuolo and Martin (2004) to control for the multinationality of domestic firms and hence to address the question of whether foreign firms have a productivity advantage over UK multinationals. In their study they conclude that the foreign ownership advantage is largely down to multinationality, as might be suspected, though they find that a productivity leadership is held by US firms, largely as a result of their ability to cherry-pick high productivity plants in the United Kingdom. Another interesting aspect of Criscuolo and Martin’s (2004) work is that they are able to examine the causality of the relationship between foreign engagement and productivity. Is it the case that firms are multinational because they are more productive, or does multinationality confer a productivity advantage on a firm because of scale economies and learning effects? In order to examine this question, Criscuolo and Martin (2004) examine whether there is an ex-post increase in productivity of UK firms that become multinational

90 Sourafel Girma, Steve Thompson and Peter Wright

over their sample period. For this they find little robust evidence, though they do offer the caveat that the data period used in their study is short. The issue of the causality between multinationality and productivity is examined further by Greenaway and Kneller (2005), later in this volume.

Notes 1. Since foreign firms must be expected to face some additional institutional costs associated with language, laws, custom and culture, the internalisation view suggests that there must be efficiency advantages from the transfer of firm-specific resources at least sufficient to offset such costs. 2. Tybout (2000) suggests that this may be particularly the case in developing economies. 3. The longer-term impact on industry concentration may be less favourable. For example, if an indigenous soft drinks market is entered by Pepsi and/or Coca Cola the immediate effect may be an intensification of competitive rivalry, but the longer-term outlet would be that of domination by a global duopoly. 4. For example, via licensing know-how. 5. These results concur with those found in earlier studies. Davies and Lyons (1991), for example, report that foreign-owned firms in the United Kingdom enjoy a 30 per cent productivity advantage over their domestic-owned equivalents. 6. Capital (fixed assets) is defined as net (of depreciation) book value of equipment, plant and machinery, fixtures and fittings and vehicles. 7. The probability of acquisition is modelled as a function of lagged wages, lagged profits, lagged size, lagged industry concentration, and the share of foreign employment in total 4-digit industry employment. For detailed results see Conyon et al. (2002b). 8. Indeed, Hausman tests find no systematic differences in the coefficient estimates between the standard and IV fixed effect models suggesting that foreign companies do not systematically acquire higher/lower total factor productivity firms.

References Buckley, P. and M.C. Casson (1976), The Future of the Multinational Enterprise, Macmillan. Brown C. and J.L. Medoff (1988), ‘The impact of firm acquisition on Labor’, in A.J. Auerbach (ed.) Corporate Take-overs: Causes and Consequences. University of Chicago Press, pp. 9–25. Caves, R.E. (1974), ‘Multinational Firms, Competition and Productivity in Host-Country Markets’, Economica, 41, 76–93. Caves, R.E. (1996), The Multinational Enterprise and Economic Analysis, 2nd edition, Cambridge: Cambridge university Press.

Foreign Ownership and Productivity Growth 91 Conyon, M., S. Girma, S. Thompson and P. Wright (2002a), ‘The Impact of Mergers and Acquisitions on Company Employment’, European Economic Review, 46(1), 31–49. Conyon, M., S. Girma, S. Thompson and P. Wright (2002b), ‘The Impact of Foreign Acquisition on Wages and Productivity in the United Kingdom’, Journal of Industrial Economics, 50(1), 85–102. Cowling, K.G. and R. Sugden (1987), Transnational Monopoly Capitalism, Wheatsheaf Books. Criscuolo, C. and R. Martin (2004), ‘Multinationals and US productivity Leadership: Evidence from Great Britain’, CeRiBA Discussion Paper. Davies, S.W. and B.R. Lyons (1991), ‘Characterising Relative Performance: The Productivity Advantage of Foreign Owned Firms in the UK’, Oxford Economic Papers, 43, 584–95. Girma, S., S. Thompson and P. Wright (2002), ‘Why are Productivity and Wages Higher in Foreign Firms?’, Economic and Social Review, 33(1), 93–100. Greenaway D. and R. Kneller (2005), ‘Empirical Evidence on Exporting and Productivity Growth’, chapter 8, in H. Görg, D. Greenaway and R. Kneller, Globalisation and Productivity Growth, eds. Basingstoke: Palgrave. Griffith, R. (1999), ‘Using the ARD Establishment Level Data to Look at Foreign Ownership and Productivity in the UK’, The Economic Journal, 109, F416–F442. Görg, H. and A. Hijzen (2005), ‘Multinational Enterprises and Spillovers’, chapter 6 in H. Görg, D. Greenaway and R. Kneller, Globalisation and Productivity Growth, eds., Basingstoke: Palgrave. Sutton J. (1998), Technology and Market Structure, MIT Press. Tybout, J. (2000), ‘Manufacturing Firms in Developing Countries: How Well Do they Do and Why?’, Journal of Economic Literature, 38(1), 11–44. Vendrell-Alda, J.L.M. (1978), Comparing Foreign Subsidiaries and Domestic Firms: A Research Methodology Applied To Argentine Industry, New York: Garland.

6 Multinational Enterprises and Spillovers Holger Görg and Alexander Hijzen

6.1 Introduction It has been estimated that the British government provided grants worth £50.75 million to Motorola to locate a production facility in Scotland in 1991, providing around 3000 jobs. Also, Siemens received around £50 million in order to attract it to build a plant employing 1000 workers in the North East of England in 1996.1 It seems reasonable to ask how such active policy can be justified. Apart from regional concerns in particular about direct job creation, the argument frequently centres around the possibility that foreign multinationals bring with them new technologies which may spill over to the local economy, benefiting not only the region but the economy as a whole. This argument has become particularly important given policy makers’ concerns that the United Kingdom is lagging behind its European and North American partners in terms of technology and productivity (e.g. DTI, 2001). Hence, an influx of foreign direct investment (FDI) associated with an inflow of new knowledge and technologies is seen as particularly advantageous. Specifically, the inflow of foreign knowledge may benefit domestic firms as they may learn from the multinationals, allowing them to upgrade their own production process and as a result increase productivity. The theoretical argument for why one may expect such productivity spillovers, as they are usually referred to, from foreign multinationals is straightforward. Given the multinationals’ limited knowledge of the local market, and distance from their parent firm, they are generally at a disadvantage compared with local firms in the host country. Hence, multinationals will only be able to locate profitably abroad if they have some sort of offsetting advantage. This takes the form of a firm specific asset (FSA), such as superior production technique, know-how or 92

Multinational Enterprises and Spillovers 93

management strategy, which has at least to some extent the characteristics of a public good and enables the firm to locate profitably abroad (Caves, 1996; Markusen, 2002). These FSAs can be transferred at low or zero cost between subsidiaries of the same firm. The possibility of productivity spillovers arises because multinationals may find it difficult to protect a leakage of an FSA to other firms in the host country. The public good characteristics imply that once the FSA is out on the external market it can be used by other firms as well, due to it being to some extent non-rival and non-excludable. The inability of the multinationals to protect the asset is due to a number of reasons. First, labour may move from multinationals to domestic firms, taking with them some of the knowledge of the FSA. Second, domestic firms supplying to or purchasing inputs from multinationals may be exposed to the superior technology used in the foreign firm. Third, domestic firms may be in competition with multinationals on the final product market, hence being able to learn from the foreign competitor. The purpose of this chapter is to discuss the evidence on productivity spillovers. We do this by starting in Section 6.2 with a brief overview of the literature, which has exploded in the last ten years or so. We then go on to discuss in more detail selected studies dealing in particular with the United Kingdom. This literature, in line with the overall trend, has also grown exponentially in the last five years. We then move on to an empirical analysis which illustrates the arguments developed in the paper. We first look at total spillovers, as in Girma et al. (2001). Then we turn to examining the role of exporting, both by multinationals and domestic firms, for spillover benefits (somewhat similar to the analysis by Girma et al. 2004). Furthermore, we analyse the role of absorptive capacity – measured in terms of intangible assets – and the role of nationality of foreign multinationals for our understanding of productivity spillovers.

6.2 International evidence on productivity spillovers Over the last thirty years, a large empirical literature has developed, starting with Caves (1974), Globerman (1979) and Blomström (1986) using data for Australia, Canada and Mexico, respectively. Since then, their empirical models have been extended and refined although the basic approach is still, by and large, similar. Most econometric analyses are undertaken in a framework in which labour productivity or total factor productivity (TFP) of domestic firms is regressed on a range of

94 Holger Görg and Alexander Hijzen

independent variables. To measure productivity spillovers from multinationals a variable is included which proxies the extent of foreign firms’ penetration, usually calculated as the share of employment or sales in multinationals over total industry employment/sales in a given sector. In other words, the regression allows for an effect of FDI on productivity of domestic firms in the same industry. If the regression analysis yields a positive and statistically significant coefficient on the foreign presence variable, this is taken as evidence that spillovers have occurred from MNEs to domestic firms.2,3 Many papers, in particular early contributions, use cross-sectional data which may lead to biased results, as pointed out by Görg and Strobl (2001). They argue that panels, using firm or plant level data are the most appropriate estimating framework for two reasons. First, panel data allow one to investigate the development of domestic firms’ productivity over a longer time period, rather than relying on one data point. Second, they allow one to investigate spillovers after controlling for other factors. Cross-sectional data, in particular if they are aggregated at the sectoral level, fail to control for time-invariant differences in productivity across sectors which might be correlated with, but not caused by, foreign presence. Thus, coefficients on cross section estimates are likely to be biased. For example, if productivity in the electronics sector is higher than, say, the food sector, multinationals may be attracted into the former. In a cross section, one would find a positive and statistically significant relationship between the level of foreign investment and productivity, consistent with spillovers, even though foreign investment did not cause high levels of productivity but rather was attracted by them. A large body of evidence has been amassed in terms of studies of productivity spillovers for many developing, transition and developed countries. Much econometric work has been completed that provides, at best, mixed results as to the importance of spillovers. There is some supportive evidence from case studies of spillover benefits to domestic firms (e.g. Moran, 2001) although there is, even at that level, disagreement in particular instances.4 A number of explanations have been offered to explain these mixed results, including methodological differences (Görg and Strobl, 2001) and country characteristics (Lipsey and Sjöholm, 2004). Rather than reviewing all of these papers we focus on three particular econometric studies, which can serve to highlight the main arguments: Aitken and Harrison (1999) for Venezuela, Keller and Yeaple (2003) for the United States and Smarzynska-Javorcik (2004) for Lithuania.5

Multinational Enterprises and Spillovers 95

Aitken and Harrison (1999) use plant level panel data for Venezuela covering the period 1976–89. Estimating an augmented Cobb–Douglas production function and controlling for plant level fixed effects they find some evidence that the presence of foreign multinationals in the same industry has had negative effects on the productivity of domestic firms. They attribute this to a negative competition effect. Domestic firms compete with multinationals on domestic product markets. When multinationals enter, they capture business from domestic firms which due to increasing returns to scale reduces their output and forces them up their average cost curve, reducing productivity. They argue that these effects seem to have more than outweighed any potentially positive productivity spillovers. In what appears to be the only study for the United States to-date, Keller and Yeaple (2003) provide evidence that even in a high-income developed country, domestic firms are able to gain in terms of productivity improvements from the presence of foreign multinationals. They use firm level panel data for the years 1987–96 and find evidence for substantial intra-industry spillovers from multinationals. One of their explanations for such large effects is their measurement of FDI activity in an industry, which is based on the industry classification of the activity of the affiliates’ employees, rather than the classification of the affiliate as a whole (by its main line of business). The paper by Smarzynska-Javorcik (2004) extends the standard approach by developing the idea that spillovers are more likely to occur through vertical relationships, rather than horizontally as has been the predominant view in the literature. Using firm level panel data for Lithuania for 1996–2000 she finds evidence consistent with her conjecture. Domestic firms in sector j increase their productivity following the establishment of multinationals in industries which are being supplied by j. She refers to this as spillovers through backward linkages. While the evidence on such backward linkages is robust to a number of amendments, there is no robust evidence that domestic firms benefit from horizontal spillovers from multinationals.

6.3 Evidence for the United Kingdom The issue of productivity spillovers has attracted a large number of research papers over the last five years or so in the United Kingdom. One of the reasons may be policy interest: it is frequently argued that the United Kingdom is lagging behind its European and North American partners in terms of productivity performance (see DTI, 2001), and the

96 Holger Görg and Alexander Hijzen

influx of FDI is seen as one potential mechanism to catch-up with other countries. A second reason is the recent availability of microlevel datasets, which have undoubtedly facilitated interest in analysing the effects of FDI in micro data. In what follows, we therefore focus our attention on studies using firm or establishment level data.6 Two main data sources have been used in microlevel studies of productivity spillovers in the United Kingdom. The first is OneSource, a commercial database based on accounts that companies are legally required to deposit at Company’s House. It provides data at the firm level, including information on all public limited companies, all companies with employees greater than 50 and the top companies based on turnover, net worth, total assets or shareholders funds (whichever is largest) up to a maximum of 110,000 companies.7 Amongst others, OneSource provides information on turnover, value-added, employment, wages, physical capital and intangible assets in a consistent way across firms and time. Furthermore, the dataset provides information on exporting activity at the firm level. These data have been used by Girma et al. (2001), Girma and Wakelin (2002) and Girma et al. (2004). The second is the Annual Respondents Database (ARD), which consists of individual establishment’s records that underlie the Annual Census of Production and is available from the Office for National Statistics (ONS) under controlled conditions. Information in the ARD includes detailed production data and is available at the level of the establishment, which is defined as the smallest unit deemed capable of providing information on the Census questionnaire. An establishment can consist of one or more plants (or ‘local units’ in the parlance of ARD).8 This database has been used in a number of recent spillover studies (e.g. Girma and Wakelin, 2001; Girma and Görg, 2002; Haskel et al., 2002; Harris and Robinson, 2004). Which database is better in the sense of being more appropriate for the study of productivity spillovers? As with many issues, the answer is: it depends. The ARD is at a more disaggregated unit of observation than OneSource, although even the ARD is not, strictly speaking, at the plant, but at establishment level. However, if one measures productivity spillovers that is in essence the diffusion of a technological asset in a local enterprise it is not clear why it should be more appropriate to look at the establishment rather than the firm level. After all, the knowledge would be assumed to transfer easily within the same firm, benefiting all plants within the same firm equally. One advantage of the ARD is, however, that it has a clear sampling frame, the details of which are available to the researcher. Hence, it is

Multinational Enterprises and Spillovers 97

fairly straightforward to get an idea about the population upon which the sample is based. This is not as clear-cut with OneSource, where one would need to revert back to published aggregate data in order to gauge the representativeness of the sample. The countervailing bonus of One Source is that it includes more data that may be potentially relevant – in particular on exporting activities, as well as on the financial situation of firms. Such information is not given in the ARD. A number of studies have used OneSource to examine productivity spillovers from FDI in the United Kingdom. Girma et al. (2001) provide the first detailed micro level analysis of intra-industry productivity spillovers from FDI in the United Kingdom, for all manufacturing industries over the period 1991–96. They find that foreign firms have higher productivity and wages than domestic firms, both in absolute levels and in growth rates. Second, in a pooled estimation they find no statistically significant evidence for productivity spillovers from FDI. In that estimation, foreign presence is defined at the 4-digit level and the coefficient on the spillover variable is constrained to be the same for all firms. They subsequently relax this assumption, allowing the level of spillovers to vary according to firm and industry characteristics. They find that the impact of FDI on the productivity of domestic firms increases with higher levels of import competition and skills in the industry. Also, they find evidence that firms with low initial productivity levels, which also have a high productivity gap relative to the industry leader, have a slower productivity spillover rate. Hence, one can conclude from their results that a firm’s absorptive capacity is important. That is, a firm needs a certain level of expertise in order to be able to utilise the knowledge that spills over from multinationals usefully. Girma et al. (2001) measure this absorptive capacity in terms of the productivity gap between the individual firm and the industry leader (defined as the 90th percentile TFP in the two digit industry).9 Girma and Wakelin (2002) extend this analysis by focussing on the regional dimension to spillovers. They calculate two measures of foreign presence, one being the share of employment in multinationals in the same region as the domestic firm, and the other – the foreign presence outside the region. Their findings point to an important regional component: domestic firms benefit from productivity spillovers from FDI within the same region, but are negatively affected by FDI outside their own region. This latter result may be evidence of a negative competition effect between foreign and domestic firms, while this negative effect may be outweighed by positive spillovers from FDI located close to the domestic firm.10

98 Holger Görg and Alexander Hijzen

In a recent paper, Girma et al. (2004) broaden the scope of the previous analyses in two dimensions. First, they allow for inter-industry (i.e. vertical) spillovers in addition to intra-industry spillovers. This is done in a manner similar to Smarzynska-Javorcik (2004) by calculating foreign presence indicators for industries which have forward or backward linkages with domestic firms. Second, they investigate what role export activity plays in determining spillovers. The export activity of domestic firms is seen as being an indicator of the firms’ absorptive capacity, with exporters being expected to be better able to benefit from spillovers due to their being linked into foreign networks through exporting activities. They also distinguish foreign presence of multinationals in domestic market and exporting presence (measured as the share of output in foreign firms devoted to the domestic market and exported, respectively). This is seen as a way of distinguishing competition effects from technology spillovers. The assumption is that the competition effect would manifest itself more strongly from domestic market oriented FDI than from export oriented FDI. Using OneSource data for the period 1992–1999 they indeed find these distinctions to be important. There is evidence that inter-industry spillovers appear more important than intra-industry spillovers, although there is substantial heterogeneity in the effects across domestic exporters and non-exporters and depending on whether FDI is domestic market or export oriented. There have also been a number of microlevel studies on productivity spillovers using the ARD. Girma and Wakelin (2001) use data for the electronics industry for 1980–92.11 They re-examine the regional dimension of spillovers, employing the Olley-Pakes (1996) approach to deal with selectivity and endogeneity in the production function. The results are in line with the earlier paper by the same authors based on OneSource for all manufacturing industries: there are positive spillovers from FDI, but these are mostly confined to the region in which the multinationals locate. A further interesting result regards the nationality of foreign multinationals: spillovers appear to be highest from non-EU, in particular Japanese firms. The main justification for focussing on particular industries is the recognition that there is substantial heterogeneity in productivity across industries and even across firms within the same industries. Hence, to avoid inappropriate pooling over heterogeneous industries, Girma and Wakelin (2001) focus on a narrow set of sectors instead. Girma and Görg (2002) also confine their analysis to data for particular sectors, namely the electronics industry, as well as mechanical and instrument engineering, using data covering the same time period. The emphasis is on studying in detail the role of domestic firms’ absorptive capacity

Multinational Enterprises and Spillovers 99

(ABC) for spillover benefits. Similar to Girma et al. (2001) they measure absorptive capacity in terms of firms’ relative productivity levels (relative to the industry leader). Using a quantile regression estimator they find clear evidence that absorptive capacity of domestic firms matters for productivity spillover benefits. Specifically, they find a u-shape relationship between absorptive capacity and spillovers from FDI. In order to determine in more detail the importance of absorptive capacity they determine the exact turning point for the quadratic relationship and evaluate the marginal effects of changes in ABC on productivity holding the FDI variables constant. Haskel et al. (2002) study spillovers from FDI for the total sample of manufacturing establishments using ARD data for the period 1973 to 1992. Their paper is, hence, most akin to Girma et al. (2001), which is based on OneSource, covering all manufacturing industries. One of the extensions of the paper is the focus on more lags of the foreign presence indicator. Most studies generally tend to use only a contemporaneous foreign presence indicator or at most one lag. They include up to three lags in their reported results in the paper. Based on their estimations they find evidence for positive productivity spillovers to domestic firms over a number of specifications and show that these benefits are not confined to contemporaneous definitions of the foreign presence variable, but also to longer lags. They also distinguish FDI according to its nationality and find the highest spillover benefits from US and French firms, while there are negative spillovers from Japanese firms. Note that this is in contrast with the findings by Girma and Wakelin (2001) for the electronics industry, who find that Japanese firms tend to bestow the highest spillover benefits for domestic firms. In contrast to Haskel et al. (2002) who pool data over all manufacturing industries, Harris and Robinson (2004) allow for heterogeneity by estimating production functions for individual sectors. They also allow for inter-industry spillovers, by calculating foreign presence in industries that are linked via input–output linkages to the industry in which the domestic firms operate. Their econometric analysis, using ARD data for 1974–95, suggests that the incidence and magnitude of spillovers differ substantially across industries. Also, they find that inter-industry spillovers are generally more important than intra-industry spillovers.

6.4 Empirical analysis for the United Kingdom In what follows, we extend the empirical literature on spillovers for the United Kingdom, focussing on the role of exporting and intangible

100 Holger Görg and Alexander Hijzen

assets for horizontal – that is, intra-industry – spillovers. We use OneSource, as it allows us to investigate the nexus between exporting and productivity spillovers from FDI, and also includes data on intangible assets at the level of the firm. We concentrate on firms in UK manufacturing over the period 1988–96. A firm is defined as foreign if the country of origin of their ultimate holding company is not the United Kingdom. Information on foreign ownership is only available for 1996, while annual information specifies whether a firm is a subsidiary or independent. For present purposes it is, therefore, assumed that ownership was constant over the sample period.12 It is possible to increase the variability of the foreign ownership measure by using the subsidiary indicator. When a firm changed from being independent to a subsidiary and when its ownership status in 1996 was foreign, the year of the switch in the subsidiary indicator is interpreted as a takeover by a foreign firm. The firms in the dataset were scrutinised for data availability on output, factors of production, exports and foreign ownership. All firms for which this information was incomplete were removed from the sample. Furthermore, firms that did not survive until 1996 were dropped. In order to take account of outliers the bottom and top percentile of the distribution of the growth in turnover of each year and (broadly defined) industry were also removed. This left an unbalanced panel of 19,598 observations. This includes 3207 firms out of which 670 were foreign-owned in 1996 (62 of those were taken over during the sample period).13 6.4.1

Descriptive statistics

Table 6.1 provides summary statistics on a number of selected variables whilst distinguishing between domestic firms and foreign-owned multinationals (MNEs). Multinationals tend to be larger in terms of both employment and the value of output. They are also generally more capital intensive, have higher labour productivity and pay higher wages than domestic firms. Moreover, MNEs also appear to have faster output growth, wage growth and productivity growth suggesting that the gap between domestic firms and MNEs increases over time. These summary statistics are in line with a large number of studies which examine in detail productivity and performance differences between domestic firms and foreign multinationals – an issue that is discussed in detail in Chapter 5 (Girma et al.). Finally, the export behaviour of MNEs appears to differ importantly from domestic firms. Not only do multinationals export more in absolute

Multinational Enterprises and Spillovers 101 Table 6.1 Summary statistics Domestic firms

Turnover* Employment Hourly wage* Labour productivity* Capital intensity* Exports* Export propensity Export share % Turnover % Hourly wage % Labour productivity

Foreign-owned firms

Observation

Mean

Standard deviation

Observation

Mean

Standard deviation

15,209 15,209 15,209 15,209 15,209 15,209 15,209 15,209 13,568 13,565 13,567

193.229 243.223 0.150 0.240 0.159 38.561 0.198 0.467 0.050 0.026 0.012

631.026 585.571 0.045 0.166 0.272 196.681 0.233 0.033 0.218 0.159 0.321

4,389 4,389 4,389 4,389 4,389 4,389 4,389 4,389 4,133 4,164 4,130

421.957 352.469 0.166 0.272 0.266 153.088 0.289 0.533 0.072 0.033 0.038

1,294.356 783.760 0.050 0.187 0.452 786.974 0.257 0.033 0.246 0.180 0.451

Note: * Times 100,000. Source: Own calculations using OneSource data.

terms due to their being larger, they also export considerably more in terms of the value of their output: MNEs export on average 29 per cent of their output, while the mean export ratio for domestic firms is 20 per cent. In our sample, multinationals account for more than half of all exports (53%). Table 6.2 provides similar summary statistics but distinguishes firms not only by ownership status (domestic and foreign) but also by export activity. Three types of export behaviour are considered: firms that never export during the sample period, firms that always export and firms that sometimes export. Comparing exporters with non-exporters reveals that the former are larger in terms of employment but smaller in terms of turnover. This is only partially in line with the literature pointing out that exporters are usually larger and more productive than non-exporters (see Chapter 8 by Greenaway and Kneller).14 Capital-intensity, labour productivity and hourly wages as well as their respective growth rates do not seem to be significantly different. These variables, however, do appear to differ importantly between domestic firms and multinationals. Conclusions regarding performance differentials between different types of firms on the basis of the present summary statistics should, of course, be drawn with caution. An econometric analysis is necessary in order to gain a more in-depth understanding. Furthermore, it is interesting to decompose foreign ownership by nationality as in Table 6.3. We distinguish four groups: North America (mainly the United States), East Asia (dominated by Japan), Europe and

102 Holger Görg and Alexander Hijzen Table 6.2 Descriptive statistics by export behaviour, 1988–96 Domestic

Foreign

Observation

Mean

Standard deviation

Observation

Mean

Standard deviation

Never exporter Turnover* Employment Hourly wage* Labour productivity* Capital intensity* Intangible assets* % Turnover % Hourly wage % Labour productivity

1,496 1,496 1,496 1,496 1,496 1,281 1,343 1,345 1,343

224.720 221.384 0.147 0.240 0.189 7.580 0.054 0.021 0.008

1,009.918 705.538 0.049 0.185 0.474 91.893 0.215 0.167 0.363

285 285 285 285 285 113 271 274 271

249.265 389.597 0.157 0.267 0.236 1.973 0.078 0.024 0.037

683.235 922.390 0.041 0.123 0.247 9.089 0.225 0.142 0.317

Always exporter Turnover* Employment Hourly wage* Labour productivity* Capital intensity* Intangible assets* Exports* Export propensity % Turnover % Hourly wage % Labour productivity

10,526 10,526 10,526 10,526 10,526 9,025 10,526 10,526 9,392 9,387 9,392

184.586 250.143 0.151 0.240 0.148 1.741 49.097 0.262 0.043 0.027 0.012

523.918 569.580 0.043 0.121 0.213 26.213 208.601 0.238 0.214 0.154 0.310

3,583 3,583 3,583 3,583 3,583 885 3,583 3,583 3,370 3,388 3,367

453.885 371.473 0.167 0.272 0.246 4.972 182.272 0.332 0.065 0.035 0.037

1,392.245 816.761 0.051 0.181 0.371 26.950 867.675 0.250 0.241 0.178 0.448

Sometimes exporter Turnover* Employment* Hourly wage* Labour productivity* Capital intensity* Intangible assets* Exports* Export propensity % Turnover % Hourly wage % Labour productivity

3,187 3,187 3,187 3,187 3,187 2,002 3,187 3,187 2,833 2,833 2,832

206.994 230.618 0.150 0.239 0.183 6.213 21.863 0.077 0.072 0.028 0.012

717.255 575.247 0.050 0.259 0.310 66.665 198.890 0.162 0.228 0.171 0.333

521 521 521 521 521 110 521 521 492 502 492

296.846 201.463 0.164 0.278 0.418 0.992 36.128 0.152 0.119 0.031 0.047

700.867 312.301 0.048 0.249 0.847 3.025 87.816 0.219 0.286 0.213 0.532

Note: * Times 100,000. Source: Own calculations using OneSource data.

the rest of the world. While all groups tend to be larger, more productive, pay higher wages and grow faster in those categories than their domestic counterparts, accounting for nationality also reveals a significant degree of heterogeneity among foreign-owned multinationals. The most striking difference is perhaps between European and non-European firms. NonEuropean multinationals in the United Kingdom appear not to produce just for the local market but use the United Kingdom as an exportplatform. Although no data are available on this in the present dataset it seems plausible that US and Japanese firms use the United Kingdom as a

Multinational Enterprises and Spillovers 103 Table 6.3 Summary statistics by nationality of ownership

Turnover* Employment Hourly wage* Labour productivity* Capital intensity* Intangible assets* Exports* Export propensity % Turnover % Hourly wage % Labour productivity

North America

East Asia

Europe

Other

(n ⬍ ⫽ 375)

(n ⬍ ⫽ 108)

(n ⬍ ⫽ 541)

(n ⬍ ⫽ 113)

736.053 429.963 0.170 0.242 0.225 5.224 397.408 0.327 0.071 0.033 0.012

2,654.551 757.093 1,185.506 411.393 879.426 681.157 547.345 328.553 0.050 0.156 0.049 0.165 0.180 0.286 0.149 0.224 0.212 0.351 0.553 0.249 24.592 0.178 1.303 1.691 1,954.557 269.150 422.450 123.373 0.251 0.381 0.285 0.229 0.267 0.100 0.280 0.070 0.182 0.051 0.237 0.035 0.332 0.057 0.398 0.052

854.269 231.676 252.289 430.161 284.699 300.894 0.052 0.172 0.050 0.123 0.282 0.178 0.259 0.290 0.303 9.593 8.881 40.842 524.391 46.121 65.003 0.243 0.246 0.292 0.233 0.069 0.281 0.180 0.019 0.211 0.468 0.013 0.404

Note: * Times 100,000, n refers to number of observations. Source: Own calculations using OneSource data.

base to supply the European market. By contrast, EU firms in the United Kingdom maintain much smaller production levels. 6.4.2

Econometric methodology

To estimate intra-industry productivity spillovers due to the presence of foreign multinationals we choose, in line with the literature, a Cobb–Douglas specification of a production function for firm i in industry j at time t,

ln yijt 
0 

M

兺

m1

m ln

zijt 

F

兺

f1

f

FPIjt  dj  dt  it

(1)

We assume three factors of production z: labour (L), capital (K) and materials (M).15 Labour is measured by the number of employees, capital by fixed assets, and materials by the difference between turnover and value-added. The regression includes a full set of industry, region and time dummies. The error term consists of a time-invariant firm specific effect and a remaining white noise error term. The first error component is purged in a within transformation of Equation (1). The regressions are only conducted for domestic firms. The regression is extended with relevant indicators of foreign presence, constructed at the 2-digit level of disaggregation. The Foreign

104 Holger Görg and Alexander Hijzen

Presence Index (FPI) is obtained by dividing the sum of turnover produced by multinationals over total turnover in industry j. F

FPIjt 

兺y

i1 N

兺

f ijt

(2)

yijt

i1

The overview in the previous section concluded that the evidence on intra-industry spillovers is ambiguous. A potential explanation could be that foreign presence is associated with offsetting effects. In an effort to disentangle the different effects we construct a measure for foreign presence in the domestic market and one for foreign presence in the foreign market. The assumption is that a negative competition effect is strongest from domestic market oriented FDI, while export oriented FDI may be more likely to lead to positive spillovers. The Foreign Presence Index in the domestic market (FPID) is given by F

FPIDjt



兺y

f ijt

 xfijt

兺y

ijt

 xijt

i1 N i1

(3)

where y is total output and x is total exports at the level of firm i. Similarly, the Foreign Presence Index in the export market (FPIE ) is calculated as F

FPIFjt



兺x

i1 N

兺x

i1

6.4.3 6.4.3.1

f ijt

(4)

ijt

Econometric results Intra-industry spillovers and exporting

The results of the regression analysis of Equation (1) using OneSource data are reported in Table 6.4. We present the regression pooled over all domestic firms (ALL) in column (1). We then estimate the model for different subsamples, according to domestic firms’ export activity: those that never export (DNE), always export (DPE) or sometimes export (REST). One reason for doing so is that export activity of domestic firms

Multinational Enterprises and Spillovers 105 Table 6.4 Basic regression results by export activity ALL (1) lnL lnK lnM FPI Constant Obs. R-squared

DNE (2)

DPE (3)

REST (4)

0.224 (0.020)*** 0.013 (0.004)*** 0.722 (0.022)*** 0.031 (0.049) 0.047 (0.019)** 14681

0.163 (0.034)*** 0.005 (0.008) 0.785 (0.040)*** 0.238 (0.094)** 0.094 (0.027)*** 1427

0.230 (0.019)*** 0.017 (0.006)*** 0.707 (0.016)*** 0.006 (0.055) 0.027 (0.023) 10211

0.237 (0.044)*** 0.012 (0.011) 0.737 (0.053)*** 0.062 (0.072) 0.083 (0.026)*** 3043

0.88

0.87

0.87

0.89

Notes: (i) Robust standard errors in parentheses. (ii) statistically significant at 10%, 5% and 1%. (iii) Regressions include full set of industry, region and time dummies. (iv) FPI indices at 2-digit industry. (v) Error terms are clustered around 2-digit industries.

can be seen as an indicator of firms’ level of absorptive capacity (Girma et al., 2004). Given that exporters have to compete on international markets, they may be expected to be more effective in absorbing the knowledge that spills over from foreign multinationals located in the United Kingdom. Overall, we find no statistically significant evidence for spillovers to all domestic firms (ALL) from column (1). However, the impact of multinational presence on domestic firms appears to differ across types of domestic firms. Whilst firms that always or sometimes export are not affected by foreign presence, domestic firms that never export are affected negatively. An explanation for this finding may be that foreign multinationals ‘crowd out’ domestic firms thereby moving domestic firms back up the average cost curve (Aitken and Harrison, 1999). This effect may be stronger for non-exporters which compete with multinationals on the domestic market. It is therefore interesting to decompose the measure of foreign presence into its presence in the domestic and export market respectively. We are implicitly assuming that the competition effect is stronger from domestic market oriented multinationals and less so from export platform FDI (see also Girma et al., 2004, for a discussion of this point). The results, which are reported in Table 6.5, indicate that foreign presence in the domestic market is harmful to all types of domestic firms,

106 Holger Görg and Alexander Hijzen Table 6.5 Regression results by export and domestic market orientation MNEs

lnL lnK lnM FPID FPIX Constant Obs. R-squared

ALL (1)

DNE (2)

DPE (3)

REST (4)

0.223 (0.020)*** 0.013 (0.004)*** 0.723 (0.022)*** 0.105 (0.066) 0.078 (0.045)* 0.044 (0.018)** 14681

0.162 (0.033)*** 0.005 (0.008) 0.785 (0.040)*** 0.256 (0.098)*** 0.030 (0.084) 0.089 (0.027)*** 1427

0.23 (0.019)*** 0.017 (0.006)*** 0.707 (0.016)*** 0.109 (0.081) 0.111 (0.046)** 0.022 (0.022) 10211

0.237 (0.044)*** 0.011 0.011 0.737 (0.053)*** 0.07 (0.051) 0.002 (0.062) 0.086 (0.027)*** 3043

0.88

0.87

0.87

0.89

Notes: (i) Robust standard errors in parentheses. (ii) *, **, *** statistically significant at 10%, 5% and 1%. (iii) Regressions include full set of industry, region and time dummies. (iv) FPI indices at 2-digit industry. (v) Error terms are clustered around 2-digit industries.

although it is only statistically significant for non-exporters. The presence of foreign firms in export markets has a positive impact on domestic firms within the same 2-digit industry, but is only statistically significant for permanent exporters. In conclusion, the ambiguous effect of foreign presence on the productivity of domestic firms in the literature on intra-industry productivity spillovers may be due to different offsetting effects related to the presence of multinational firms. Competition in the domestic market by foreign multinationals appears to affect negatively the productivity of domestic firms, especially if they do not export. This may happen because domestic firms are forced to reduce output thereby increasing average costs as suggested by Aitken and Harrison (1999). However, multinational presence in domestic export markets appears to benefit domestic firms, in particular if they are exporters. It is worth highlighting that intra-industry spillovers do not only depend on the export behaviour of multinationals, but also on the export orientation of domestic firms so that spillovers are unevenly distributed across domestic firms. Thus, whilst the aggregate spillovers on the domestic economy may be neutral, there are some firms that are likely to gain and some to suffer from multinational presence in their industry.

Multinational Enterprises and Spillovers 107

6.4.3.2

Intangible assets and the role of absorptive capacity

Much of the recent literature has stressed the importance of firms’ absorptive capacity in order for them to benefit from spillovers from foreign multinationals (see Greenaway and Görg, 2004). Common proxies for this are firms’ R&D expenditure and relative productivity levels of exporting activity. An interesting feature of OneSource is that it contains information on intangible assets. Intangible assets can be taken as a reflection of FSAs and hence, are a crucial aspect of multinationals’ maintaining an international production network. Hence, intangible assets may also be a useful proxy for absorptive capacity, as it indicates how well-endowed domestic firms are with FSAs. In order to analyse the significance of intangible assets for firms’ absorptive capacity we construct a measure of absorptive capacity somewhat similar to Girma and Görg (2002) as16 ABCijt 

intijt max i j

intjt

(5)

Hence, absorptive capacity of firm i is measured by the distance in terms of intangible assets of firm i from the highest level of intangible assets in the industry. This distance is measured as the ratio of a firm’s intangible assets to the industry’s leader. In order to analyse to what extent spillover effects are related to absorptive capacity the foreign presence indicator is interacted with absorptive capacity in a quadratic specification to allow for additional flexibility. From the results in Table 6.6 we observe an inverted U-shaped relationship between absorptive capacity and productivity spillovers for domestic exporters. This suggests that for a given level of foreign presence in a sector its impact is positive up to a threshold and then turns negative. An explanation for this finding could be that domestic firms are in closer competition with foreign firms as they build up their level of intangible assets. In order to gain more insight in these results it is useful to control again for the market orientation of multinationals. The recurrence of an inverted U-relationship between the domestic market orientated of multinationals and their absorptive capacity reinforces the conjecture that domestic firms with higher levels of absorptive capacity are in closer competition with multinationals. By contrast, however, a U-shaped relationship is observed between the presence of multinationals in export markets and absorptive capacity; a result that is consistent with the findings by Girma and Görg (2002). While the results are somewhat different the qualitative patterns are the same

Table 6.6 Regression results with absorptive capacity and intangible assets

LnL lnK lnM FPI FPI*abc1 FPI*abc2

0.226 (0.020)*** 0.010 (0.005)** 0.718 (0.022)*** 0.043 (0.051) 0.196 (0.108)* 0.189 (0.123)

DNE (2) 0.159 (0.033)*** 0.010 (0.009) 0.798 (0.039)*** 0.255 (0.095)*** 1.253 (1.290) 1.305 (1.288)

DPE (3) 0.234 (0.020)*** 0.015 (0.006)** 0.703 (0.016)*** 0.012 (0.058) 0.248 (0.106)** 0.225 (0.119)*

REST (4) 0.243 (0.053)*** 0.006 (0.014) 0.730 (0.059)*** 0.129 (0.081) 0.240 (0.139)* 0.378 (0.159)**

FPID FPID*abc1 FPID*abc2 FPIX FPIX*abc1 FPIX*abc2 Constant Observations R-squared

0.048 (0.023)** 11,763 0.87

0.105 (0.029)*** 1,210 0.86

0.032 (0.023) 8,695 0.87

0.090 (0.027)*** 1,858 0.88

ALL (5)

DNE (6)

DPE (7)

REST (8)

0.225 (0.020)*** 0.010 (0.005)** 0.719 (0.022)***

0.158 (0.032)*** 0.010 (0.009) 0.798 (0.039)***

0.234 (0.020)*** 0.015 (0.006)** 0.703 (0.016)***

0.243 (0.053)*** 0.005 (0.014) 0.731 (0.059)***

0.122 (0.068)* 0.745 (0.191)*** 0.623 (0.218)*** 0.080 (0.048)* 0.415 (0.147)*** 0.322 (0.177)* 0.044 (0.021)** 11,763

0.241 (0.100)** 0.758 (1.459) 0.807 (1.403) 0.000 (0.076) 0.411 (0.942) 0.421 (1.084) 0.101 (0.028)*** 1,210

0.127 (0.080) 0.712 (0.265)*** 0.583 (0.308)* 0.121 (0.044)*** 0.353 (0.216) 0.267 (0.258) 0.027 (0.021) 8,695

0.105 (0.066) 1.328 (0.421)*** 1.368 (0.608)** 0.058 (0.084) 0.698 (0.260)*** 0.627 (0.429) 0.101 (0.028)*** 1,858

0.87

0.87

0.87

0.89

Notes: (i) Robust standard errors in parentheses. (ii) *, **, *** statistically significant at 10%, 5% and 1%. (iii) Regressions include full set of industry, region and time dummies. (iv) FPI indices at 2-digit industry. (v) Error terms are clustered around 2-digit industries.

108

ALL (1)

Multinational Enterprises and Spillovers 109

across different types of domestic firms. Domestic firms seem to be better equipped to benefit from foreign presence in exports markets when they have higher levels of intangible assets. 6.4.3.3

The role of nationality

In this section we analyse to what extent intra-industry productivity spillovers are related to nationality of ownership. As before we distinguish four regions of ownership: North America (US), Europe, East Asia ( Japan) and others. Our results in Table 6.7 suggest that there are positive spillovers from US, Japanese and EU multinationals for most groups of firms, although domestic non-exporters are affected negatively by the presence of Japanese MNEs. These results contrast somewhat with the evidence provided by Haskel et al. (2002) and Girma and Wakelin (2001) based on ARD data, who do not find much evidence for spillovers from European firms. However, the findings in these two papers are also at odds with each other, as we discussed in Section 6.3. This suggests that more research is needed to pin down the effects of multinationals from different countries in more detail. Table 6.7 Regression results by nationality

lnL lnK lnM FPI FPIUS FPIJP FPIEU Constant Observations R-squared

ALL (1)

DNE (2)

DPE (3)

REST (4)

0.222 (0.012)*** 0.018 (0.005)*** 0.695 (0.020)*** 0.299 (0.108)*** 0.835 (0.098)*** 0.180 (0.050)*** 1.510 (0.614)** 0.065 (0.031)** 6721 0.88

0.201 (0.052)*** 0.020 (0.014) 0.736 (0.072)*** 0.788 (0.357)** 1.090 (0.475)** 0.629 (0.173)*** 3.110 (1.215)** 0.109 (0.130) 528 0.84

0.217 (0.016)*** 0.019 (0.004)*** 0.695 (0.016)*** 0.266 (0.110)** 0.878 (0.104)*** 0.305 (0.069)*** 1.531 (0.657)** 0.049 (0.027)* 4968 0.88

0.252 (0.028)*** 0.030 (0.013)** 0.686 (0.043)*** 0.563 (0.394) 0.947 (0.352)*** 0.170 (0.240) 1.535 (0.821)* 0.233 (0.179) 1225 0.90

Notes: (i) Robust standard errors in parentheses. (ii) statistically significant at * 10%, ** 5% and *** 1%. (iii) Regressions include full set of industry, region and time dummies. (iv) FPI indices at 2-digit industry. (v) Error terms are clustered around 2-digit industries.

110 Holger Görg and Alexander Hijzen

6.5 Conclusions The British government has been quite pro-active in attempting to attract foreign multinationals to locate in particular regions of the United Kingdom. While there may be many reasons for doing so, one argument voiced quite frequently is that domestic firms may benefit from FDI in terms of productivity spillovers. This chapter reviews the evidence and presents some empirical work dealing with this issue. While the international evidence on productivity spillovers is far from conclusive on whether or not these benefits actually accrue to domestic firms, recent evidence based on microlevel data for the United Kingdom is quite encouraging. All studies reviewed here do find evidence for productivity spillovers, although these may depend on the characteristics of domestic firms, in particular their levels of absorptive capacity, and whether or not they are located close to multinationals. Our empirical analysis, based on OneSource for the period 1988–96, is in line with that evidence. We find evidence for spillovers over a number of specifications of our empirical model. However, there are a number of qualifications. First, spillovers depend on the market orientation of FDI, with export oriented FDI being more likely to generate positive spillovers, while domestic market oriented FDI seems to crowd out domestic firms and reduce their productivity. Second, the export orientation of domestic firms matters; in general, exporters appear to benefit most from spillovers. In extensions of our model we also find some evidence that spillovers appear to differ across firms of different nationalities, and that measuring absorptive capacity in terms of intangible assets may be a fruitful direction for further research.

Notes 1. These figures are taken from Haskel et al. (2002). 2. The interpretation of this coefficient of course hinges on the assumption that the FDI variable does not merely pick up the effect of other correlated factors on productivity, i.e. one needs to assume that there is a full vector of productivity augmenting activities included in the empirical model. 3. This approach, of course, treats the mechanism through which spillovers take place as a ‘black box’. Hence, one does not know the channels through which spillovers actually occur. Görg and Strobl (2005) present a first attempt at looking empirically at one of the channels in more detail, by measuring productivity spillovers through movements of workers from multinationals to domestic firms. Görg and Strobl (2002, 2003) and Barrios et al. (2005) also provide alternative approaches to estimating the beneficial effects of FDI on local development, by looking at the impact of FDI on entry and survival of domestic plants.

Multinational Enterprises and Spillovers 111 4. For example, Larrain et al. (2000) conclude that the location of Intel in Costa Rica has had positive effects on the local economy, Hanson (2001) argues that there is little evidence for spillovers from Intel on domestic firms. Hanson (2000) also argues that the location of Ford and General Motors in Brazil have failed to show the expected spillover benefits. 5. A more detailed discussion of a long list of spillover studies is provided by Görg and Greenaway (2004). See also Görg and Strobl (2001) for a metaanalysis of papers on productivity spillovers. 6. For example, Liu et al. (2000), Driffield (2001) and Driffield et al. (2002) use industry level data to analyse the effect of FDI on domestic productivity in the United Kingdom. 7. See Hart and Oulton (1995) for a detailed description of the dataset. 8. Barnes and Martin (2002) provide a very detailed and useful introduction to the ARD. 9. Girma (2002) also examines the importance of absorptive capacity for FDI spillovers using a threshold regression technique. 10. It is interesting to note that, in contrast to UK studies, Sjöholm (1999) and Aitken and Harrison (1999) find a significant regional element to spillovers in their studies for Indonesia and Venezuela, respectively. 11. 1992 is a natural cut-off point as the United Kingdom’s sectoral classification (SIC) changed after 1992. It is therefore complicated to link the microdata between 1992 and 1993. 12. In cases where no ownership information was available for 1996 but there was for 1999, its ownership status in 1999 was assumed to apply to the whole sample period. 13. Nominal aggregates were deflated using 5-digit level industry deflators used from the ONS. 14. For the full sample of non-exporters (NE) mean real turnover is 19.377 and for permanent exports (PE) 33.392. Controlling for nationality reduces the difference considerably to 19.491 for domestic non-exporters (DNE) and 21.403 for domestic permanent exporters (DPE). The sign reverses when only concentrating on firms that are present in 1996 the year for which our foreign ownership measure is available (22.472 for DNE and 18.459 for DPE). 15. In alternative regressions we estimated production functions using value added, capital and labour. Results of these estimations are largely similar to those reported later. 16. Note, however, that the ABC measure used by Girma and Görg (2002) is calculated using total factor productivity rather than intangible assets.

References Aitken, B.J. and A.E. Harrison (1999), ‘Do Domestic Firms Benefit from Direct Investment? Evidence from Venezuela’, American Economic Review, 89, 605–18. Barnes, Matthew and Ralf Martin (2002), ‘Business Data Linking: An Introduction’, Economic Trends, 581, 34–41. Barrios, Salvador, Holger Görg and Eric Strobl (2005), ‘Foreign Direct Investment, Competition and Industrial Development in the Host Country’, European Economic Review, 49, 2005, 1761–84. Blomström, Magnus (1986), ‘Foreign Investment and Productive Efficiency: The Case of Mexico’, Journal of Industrial Economics, 35, 97–112.

112 Holger Görg and Alexander Hijzen Caves, Richard E. (1974), ‘Multinational Firms, Competition, and Productivity in Host-Country Markets’, Economica, 41, 176–93. Caves, Richard E. (1996), Multinational Enterprise and Economic Analysis. Second Edition. Cambridge: Cambridge University Press. Driffield, Nigel (2001), ‘The Impact on Domestic Productivity of Inward Investment in the UK’, The Manchester School, 69, 103–19. Driffield, Nigel, Max Munday and Annette Roberts (2002), ‘Foreign Direct Investment, Transactions Linkages, and the Performance of the Domestic Sector’, International Journal of the Economics of Business, 9, 335–51. DTI (2001), UK competitiveness indicators, second edition. Department of Trade and Industry, London. Girma, Sourafel (2002), ‘Absorptive Capacity and Productivity Spillovers from FDI: A Threshold Regression Analysis’, GEP Research Paper 02/08, University of Nottingham. Girma, Sourafel and Holger Görg (2002): ‘Foreign Direct Investment, Spillovers and Absorptive Capacity: Evidence from Quantile Regressions’, GEP Research Paper 02/14, University of Nottingham. Girma, Sourafel, Holger Görg and Mauro Pisu (2004), ‘The Role of Exports and Foreign Linkages for FDI Productivity Spillovers’, mimeo, University of Nottingham. Girma, Sourafel, David Greenaway and Katharine Wakelin (2001), ‘Who Benefits from Foreign Direct Investment in the UK?’, Scottish Journal of Political Economy, 48, 119–33. Girma, Sourafel and Katharine Wakelin (2001), ‘Regional Underdevelopment: Is FDI the Solution? A Semiparametric Analysis’, GEP Research Paper 01/11, University of Nottingham. Girma, Sourafel and Katharine Wakelin (2002), ‘Are There Regional Spillovers from FDI in the UK?’, in David Greenaway, Richard Upward, Katharine Wakelin eds, Trade, Investment, Migration and Labour Markets, Basingstoke, Macmillan. Globerman, Steven (1979), ‘Foreign Direct Investment and “Spillover” Efficiency Benefits in Canadian Manufacturing Industries’, Canadian Journal of Economics, 12, 42–56. Görg, Holger and David Greenaway (2004), ‘Much Ado about Nothing? Do Domestic Firms Really Benefit from Foreign Direct Investment?’, World Bank Research Observer, 19, 171–97. Görg, Holger and Eric Strobl (2001), ‘Multinational Companies and Productivity Spillovers: A Meta-Analysis’, Economic Journal, 111, F723–F739. Görg, Holger and Eric Strobl (2002), ‘Multinational Companies and Indigenous Development: An Empirical Analysis’, European Economic Review, 46, 1305–22. Görg, Holger and Eric Strobl (2003), ‘Multinational Companies, Technology Spillovers, and Plant Survival’, Scandinavian Journal of Economics, 105(4), 581–95. Görg, Holger and Eric Strobl (2005), ‘Spillovers from Foreign Firms through Worker Mobility: An Empirical Investigation’, Scandinavian Journal of Economics, 107, forthcoming. Hanson, Gordon H. (2001), ‘Should Countries Promote Foreign Direct Investment’, G-24 Discussion Paper No. 9, UNCTAD, Geneva. Harris, Richard and Catherine Robinson (2004), ‘Productivity Impacts and Spillovers from Foreign Ownership in the United Kingdom’, National Institute Economic Review, 187, 58–75.

Multinational Enterprises and Spillovers 113 Hart, Peter E. and Nicholas Oulton (1995), ‘Growth and Size of Firms’, NIESR Discussion Paper No. 77, National Institute of Economic Research, London. Haskel, Jonathan E., Sonia C. Pereira and Matthew J. Slaughter (2002), ‘Does Inward Foreign Direct Investment Boost the Productivity of Domestic Firms?’, NBER Working Paper 8724. Keller, Wolfgang and Stephen Yeaple (2003), ‘Multinational Enterprises, International Trade, and Productivity Growth: Firm-Level Evidence from the United States’, GEP Research Paper 03/03, University of Nottingham. Larrain, Felipe B., Luis F. Lopez-Calva and Andres Rodriguez-Clare (2000), ‘Intel: A Case Study of Foreign Direct Investment in Central America’, Center for International Development Working Paper No. 58, Harvard University. Liu, Xiaming, Pamela Siler, Chengqi Wang and Yingqi Wei (2000): ‘Productivity Spillovers from Foreign Direct Investment: Evidence from UK Industry Level Panel Data’, Journal of International Business Studies, 31, 407–25. Lipsey, Robert E. and Fredrik Sjöholm (2004), ‘Host Country Impacts of Inward FDI: Why Such Different Answers’, in M. Blomström, E. Graham and T. Moran, eds, The Impact of Foreign Direct Investment on Development, Washington: Institute for International Economics. Markusen, James R. (2002), Multinational Firms and the Theory of International Trade, Cambridge, MA: MIT Press. Moran, Theodore (2001), Parental Supervision: The New Paradigm for Foreign Direct Investment and Development, Washington DC: Institute for International Economics. Olley, S. and A. Pakes (1996), ‘The Dynamics of Productivity in the Telecommunications Equipment Industry’, Econometria, 64, 1263–97. Sjöholm, Fredrik (1999), ‘Productivity Growth in Indonesia: The Role of Regional Characteristics and Direct Foreign Investment’, Economic Development and Cultural Change, 47, 559–84. Smarzynska-Javorcik, Beata K. (2004), ‘Does Foreign Direct Investment Increase the Productivity of Domestic Firms? In Search of Spillovers through Backward Linkages’, American Economic Review, 94(3).

This page intentionally left blank

Part III Exporting and Productivity

This page intentionally left blank

7 Exporting and Productivity Growth: Theory Rod Falvey and Zhihong Yu

7.1 Introduction Traditional treatments of the gains from trade were primarily concerned with the static gains realisable from realigning production in accordance with comparative advantage. In an open economy, resources would be (re)allocated more ‘productively’, taking advantage of the opportunities to import those products in whose production the country is relatively less efficient by exporting those in whose production it is relatively more efficient. These gains are available even if autarky production is technically efficient (i.e. on the production possibility frontier), and in a world where technologies exhibit constant returns to scale, firms are largely irrelevant and the effects of trade on the efficiency of resource use within the industry cannot be explored. Developments within the industry were never entirely neglected of course. But the natural focus when considering inter-sectoral resource reallocations in line with comparative advantage was what was happening in the import-competing industries, since these were the sectors that were contracting and thus the presumed losers in the process. How would these industries respond to increased competition from imports? Would new production techniques be adopted, and if so what would be the change in factor intensities? One of the earliest arguments for protection was explicitly concerned with the effects of import competition on industry productivity. The venerable infant-industry argument calls for temporary protection for a (potential) import-competing industry to allow it to work its way down the learning curve sufficiently to be competitive without further support. The logical strengths and weaknesses of this argument are well known, as is its poor record of success when implemented. This argument was frequently misused and the inefficiency 117

118 Rod Falvey and Zhihong Yu

of most heavily protected industries later became one strong argument for trade liberalisation. The relative success of those countries that adopted outward oriented trade strategies prompted enquiries into why export-led development strategies might improve an economy’s technical efficiency. It is now recognised that import-substituting strategies sacrifice more than the lost opportunities for allocative efficiency. Indeed, Romer argues that the true costs of trade restrictions come not from their static reallocation effects on already existing activities, but from ‘the stifling effect distortions have on the adoption of new technologies, the provision of new types of services and on imports of new types of capital goods and produced inputs’ (1994, p. 25). At the same time, the development of the ‘new’ trade theory, brought the firm to centre stage. In these models monopolistically competitive firms produce differentiated products using technologies that involve fixed costs. To maintain tractability in a general equilibrium context, a ‘representative firm’ framework is emphasised. Now the focus is on product specialisation by firms and on intra-industry trade, where all firms both export to foreign markets and compete with imports in their domestic markets. Opening up to trade could affect both the size and number of firms. If firms become larger, industry productivity improved, but even if they did not trade increases the variety of products available to consumers. Gains from trade are not dependent on firm productivity improvements. While the wider costs and benefits of export-oriented policies were subject to considerable debate from the 1970s, it is only in the last decade that the effects of trade on exporting firms have fallen under a similar spotlight. Our aim in this chapter is to review this recent theoretical literature. In Section 7.2 we consider the potential links between exporting and the productivity of the individual exporting firm. In Section 7.3 we take individual firm productivity as given and examine how opening the export market can affect average industry productivity through differential effects on different firms. The final section presents some concluding comments.

7.2 Exporting and intra-firm productivity growth Inspired by empirical evidence and field studies, recent theoretical work has demonstrated various channels through which exporting may serve as a stimulant to firm productivity growth. A large and growing body of

Exporting and Productivity Growth 119

firm- and plant-level empirical evidence has documented the exceptional performance, especially in terms of productivity, of exporters compared to their domestic counterparts in the same industry (see Chapter 8). Exporters are on average more productive than non-exporters, whether productivity is measured by total factor productivity, labour productivity or average variable costs. Why is this the case? 7.2.1 Self-selection and learning by exporting As Bernard and Jensen (1999) observe in their pioneering study of the exporting-productivity nexus, this correlation is consistent with two alternative hypotheses. One is that firms within an industry differ in their productivity for reasons unrelated to the world market but that only the more productive firms become exporters – the export market self-selection hypothesis. The other is that those firms that choose to become exporters are initially no more or less productive than other firms, but benefit from productivity improvements as a result of their foreign contacts – the learning-by-exporting hypothesis. Of course these hypotheses are not mutually exclusive, and their degree of applicability is likely to vary across countries and, particularly, across industries. But they do have quite different policy implications. If relatively productive firms become exporters, but exporting itself conveys no productivity benefits, then little is achieved by assisting existing exporters. But if firms become more productive in order to export, then rewarding export performance will encourage such productivity improvements. If exporting itself conveys productivity benefits, then firms will anticipate these and the appropriate policy approach would be to avoid placing barriers to exporting. The ‘only good firms export’ (self-selection) hypothesis will seem quite intuitive because the business press is full of stories where success in the domestic market is followed by success on the global stage. The economic modelling of this hypothesis is not without its subtleties, however. One cannot claim that in any model where firms differ in productivity only the ‘good’ firms will export. Consider a monopolistically competitive industry characterised on the demand side by symmetric Dixit–Stiglitz preferences, in which consumers love variety and firms produce differentiated goods under increasing returns to scale. Firms differ in their constant marginal costs of production (i.e. their productivity). If we assume that exporting incurs an ad valorem trade cost, then all firms, regardless of productivity, will export. This is because a foreign demand exists for each variety, and the ad valorem trade costs simply eat

120 Rod Falvey and Zhihong Yu

up a proportion of a firm’s revenue from foreign sales, but do not make exporting unprofitable for any firm. This changes if we make the more realistic assumption that exporting incurs both ad valorem and fixed costs. The latter will include market entry costs such as market research, product modification to meet local tastes and regulations, setting up a distribution and service network, advertising and so on. Empirical studies have already identified the significance of such costs and their importance in the export decision (Roberts and Tybout, 1997). Only the more productive firms, with lower marginal costs, will earn sufficiently high profits to cover such fixed costs (Clerides et al., 1998). The less productive firms find the fixed export costs exceed potential export profits and therefore remain purely domestic even if variable trade costs are zero. A firm’s productivity level can vary over time for a range of reasons, and such variations will impact on its exporting status. Clerides et al., build a dynamic model with forward-looking firms and sunk export entry costs. In this model the export market participation decision depends on a firm’s previous export status as well as its productivity level. They show that each firm will observe a two threshold strategy in making this decision. Non-exporters will break into the export market if their productivity levels grow beyond the upper productivity threshold. Current exporters, who have incurred the sunk entry/re-entry costs, will exit only if their productivity levels fall below a lower productivity threshold, but prefer to continue exporting if their productivity levels are between the thresholds. This ‘hysteresis’ effect (Baldwin and Krugman, 1989), arises since a forward-looking exporter will choose to stay in the export market even if it experiences a bad productivity shock, to avoid the re-entry costs which will be incurred if the firm exits at the current stage and re-enters in the future. This suggests that the firms that actually exit the export market should have significantly lower productivity than those entering. While the self-selection story explains the exporting-productivity nexus by high productivity being necessary to cover the fixed costs of exporting, the causal link could alternatively run in the opposite direction. Firms may undertake the fixed costs of exporting in the expectation of productivity improvements that will offset them. One commonly cited mechanism through which exporters can enhance productivity post-entry is the ‘learning-by-exporting’ hypothesis, which emphasises vertical knowledge diffusion from firms’ international buyers. Firms in developing countries in particular, may acquire better practices and technologies from their clients in more advanced countries. The latter may offer technical expertise and other assistance to their suppliers to

Exporting and Productivity Growth 121

improve efficiency. Pack and Saggi (2001), for example, show that a developed country (DC) importer has a strong incentive to transfer technology to its less developed country (LDC) supplier, even if this technology can eventually diffuse to rival DC importers via other LDC suppliers. The existence of such transfers is supported by anecdotal evidence from case studies in the Asian NICs (see, e.g., Blalock and Gertler, 2002; and Rhee et al., 1984). Clerides et al., develop perhaps the most comprehensive model, which incorporates the learning-by-exporting forces as an upward shift in the stochastic process governing firms’ productivity induced by exporting. They report simulation results of the productivity trajectories of firms when both self-selection and learning-by-exporting effects are present. Firms that always export are the most productive, whereas firms that never export are the least productive. Firms that switch to exporting (‘entrants’) exhibit the fastest productivity growth several years before entry, since it is this productivity growth that takes them above the export entry threshold. Firms that start as exporters but then exit export markets (‘exiters’) suffer from ongoing negative productivity shocks which eventually push them below the lower threshold. The distinguishing feature of the existence of the learning force is that entrants continue their productivity growth after entry. New entrants continue to increase their productivity advantage relative to non-exporters, and converge to the constant exporters. Further, the entry threshold is lower than that in the no-learning case, since the expectation that breaking into the export market will bring ongoing post-entry productivity gains will attract non-exporters to start exporting at a lower productivity threshold. As Clerides et al., point, these patterns imply that learning forces may increase the cross-sectional productivity dispersion of exporting firms and shift downward their average productivity level, and that self-selection will lead to faster productivity growth among some nonexporters who are future entrants. Hence, a cross section of exporters versus non-exporters productivity levels or their growth rate gap may underestimate the extent of learning forces. 7.2.2 Exporting, productivity and innovation The learning-by-exporting hypothesis emphasises the productivity gains through vertical technology spillovers, implying such effects are more likely to be found in technologically backward industries or economies. However, this is not the only channel through which exporting might boost productivity. Other channels include (a) increased incentives to

122 Rod Falvey and Zhihong Yu

innovate; (b) increased competition; and (c) increased incentives to upgrade product quality. We discuss each in turn. First, exposure to the export market may increase a firm’s incentives to innovate. The access to a larger market size brought by a reduction of foreign trade barriers can raise the relative return to technological innovation. Holmes and Schmitz (2001) argue that, in many oligopolistic industries, firms engage in both socially productive and socially unproductive entrepreneurial activities. The former produce innovations, while the latter aim at protecting the firm from its (potential) rivals, often by blocking competition through taxes or regulations. The balance between these activities will depend, inter alia, on the state of trade, since we are unlikely to see successful blocking on world markets due to both the number of competitors and the incentives for firms to free ride on each others’ blocking activities. Participation in export markets shifts the balance of incentives towards the innovation alternative. Likewise, there is little point in blocking domestic rivals if they will be replaced by imports. Opening up the domestic market provides increased incentives for innovation also. Second, in a world where trade impediments (natural and policy imposed) are still significant, exporters must be able to overcome these barriers and compete with each other in foreign markets. Exporters therefore have stronger incentives to cut managerial slack and X inefficiency and to adopt new and more efficient technologies, compared to their purely domestic counterparts which are protected by such barriers. This ‘competition facilitates efficiency’ argument has traditionally been applied to the effects of trade liberalisation on import-competing firms, but a similar line of reasoning can also be applied to exporters. Of course, as Bernard and Jensen (1999) point out, one needs to explain why profit-maximising firms would not already improve their productivity regardless of exporting status.1 In industries where products are vertically differentiated (i.e. differ in ‘quality’), the exports for many countries will be dominated by higher quality products. Such an outcome is suggested by the well-known transport cost effect. If unit trade costs are approximately equal for all qualities, they form a lower proportion of the cost, insurance and freight (cif) price of higher quality products. The latter then have a lower relative price in export markets, and the composition of demand shifts in their favour. A similar outcome can result from certain types of trade policies. This effect may also induce those firms intent on breaking into export markets to select technologies that produce higher quality products, and hence are more productive per unit of output. Alvarez and

Exporting and Productivity Growth 123

Robertson (2004) assume that higher quality requires a higher skillintensity and consider an industry equilibrium where firms are divided between high quality exporters and low quality domestic producers. A reduction in export barriers sees an increase in the share of exporters and consequently in the share of higher quality output. Where these products are also ‘experience’ goods (i.e. their quality is not observable by consumers prior to purchase), then the costs of establishing an appropriate reputation are an important barrier to entry in export markets. Absent a readily identifiable brand name, the quality reputation of a product entering a new market may solely reflect its country of origin. Mayer (1984) discusses the potential positive externalities that high quality exports by a firm or industry can have on others through such reputational effects. Of course once a country or industry has a reputation for exporting in a certain quality range, this will influence its firms’ quality decisions and can therefore become self-fulfilling. Chisik (2003) considers policies that might mitigate the adverse effects of a negative stereotype on quality selection. 7.2.3

Summary

In this section we have reviewed a number of models predicting a positive link between exporting and productivity at the firm level, which is consistent with the stylised facts revealed by recent firm level trade studies. Existing trade models also provided different, but not mutually exclusive explanations of the causality underlying this relationship. The self-selection models stress the importance of fixed costs associated with exporting. If more productive firms earn higher export profits, it follows that only the more productive firms can afford the fixed export costs and self-select into export market. Furthermore, since entry is costly, incumbent exporters would be reluctant to leave unless they experience a significantly adverse productivity shock. This implies that export status, as well as switching into or out of export markets, can be a good indicator of firm success, since ‘good’ firms export and ‘bad’ firms do not. On the other hand, the causality between productivity improvement and exporting could run the other way. Entry into export markets may boost firm productivity, and existing trade theory provided various reasons for this causal link. The ‘learning by exporting’ hypothesis is perhaps the most commonly cited of these, and relies on vertical knowledge transmission from downstream foreign buyers. Under different assumptions, various other channels through which exporting may raise firm productivity emerge. These include cases where export market

124 Rod Falvey and Zhihong Yu

entry promotes innovation, quality upgrading or intensified competition. If any of these mechanisms is at work, one would expect exporting itself to help make a firm ‘good’. To summarise, theory predicts causal links between exporting and firm productivity running both ways, leaving the relative importance of the causalities involved as ultimately an empirical question.

7.3 Exporting and industry productivity growth To this point, we have concentrated on the intra-firm productivity effects of dynamic adjustment into (and out of) export markets. The analysis has occurred at the individual firm level, and the theoretical models are within a partial equilibrium framework. However, a growing empirical literature has revealed the important role of trade-induced inter-firm market share reallocation in facilitating overall industry productivity growth (Tybout and Westbrook 1995; Bernard and Jensen 2000b; Pavnik 2001). Clearly productivity gains due to cross-firm reallocations cannot be explained within a representative firm trade model such as Helpman and Krugman (1985), and are difficult to encompass in the partial equilibrium models described in the last section. Recently a new generation of general equilibrium models has brought firm productivity heterogeneity into a central place in trade theory. This new approach reveals a second linkage between exporting and productivity growth, which will be the focus of this section. Simply put, increased exposure to trade can lead to resource and output share reallocations towards the more productive firms, which raises overall industry productivity through a ‘Darwinian evolution process’ (Melitz 2003). 7.3.1 7.3.1.1

Inter-firm reallocation and productivity growth A dynamic intra-industry trade model

Building on the work of Hopenhayn (1992) and Krugman (1980), Melitz (2003) develops a dynamic industry model with heterogeneous firms to show how opening up the export market can have a positive impact on aggregate industry productivity growth. He considers a world comprised of a number of identical countries, and a differentiated goods industry characterised by Dixit–Stiglitz monopolistic competition across firms, each of which produces a unique variety under increasing returns to scale. Departing from the representative firm trade models, Melitz models firm productivity heterogeneity as a result of pre-entry productivity uncertainty, and incorporates it into intra-industry trade.

Exporting and Productivity Growth 125 Expect positive profits Sink irreversible investment to enter

Potential Entrants

Randomly draw productivity

High productivity entrants

Survivors Survivors

Low productivity entrants

Exiters Exiters

Entrants

Expect negative profits

Figure 7.1 Flow chart showing productivity uncertainty and firm entry/exit

Figure 7.1 depicts a flow chart describing the process whereby heterogeneity is generated. There exists an infinite number of identical potential entrants for the industry. To enter, each has to incur an irreversible investment, like paying an entrance ticket, and then draws its productivity level from an exogenous distribution that is common to all entrants. Once its productivity level is known, the entrant will decide whether to pay a further fixed production cost or leave immediately, depending on whether its productivity level is above the threshold necessary to generate a positive operating profit. Firms drawing a productivity level above the zero profit productivity threshold will find it profitable to serve the domestic market, while lower productivity firms will leave immediately without producing. Firm turnover is assured by assuming that all producing firms face an exogenous probability of death in each period, which is independent of a firm’s productivity. In equilibrium, the free entry condition will drive the expected profit of entry to zero, and thus endogenously determine the productivity threshold required to survive. The steady state equilibrium is characterised by a constant productivity distribution across all producing firms and constant number of producing firms, though there are simultaneous flows of new entrants and exiters which offset each other and have no net effect on the steady state aggregate variables. Melitz then demonstrates how the opening up of the export market will lead to industry rationalisation and reallocation effects, both of which will raise average industry productivity. In an open economy, due to love of variety, all countries are involved in intra-industry trade. One impact of the exposure to the export market is that the expected profit of entry into the domestic market will rise, as potential entrants add an extra export profit into their expectation of operating profit. This

126 Rod Falvey and Zhihong Yu

attracts more entrants into the market, and leads to the expansion of more productive firms who export, which raises the demand for the labour and bids up the real wage. Hence the productivity threshold required to survive will be pushed up to a new level. Since the new threshold is greater than the old one, those firms with productivity between these two thresholds will be forced to leave. This rationalisation effect purges firms at the low productivity end, which generates an average industry productivity increase. Assuming there are some fixed costs associated with exporting, as discussed earlier, opening up the export market will also lead to expansion of the most productive firms who export, but will make the less productive firms who remain purely domestic, shrink. Melitz shows that the export market and domestic market will remain segmented if the trade costs are relatively high, hence there exists an export productivity threshold, above the survival threshold, which partitions producing firms into low productivity non-exporters and high productivity exporters. These thresholds are illustrated in Figure 7.2. Since opening up the export market induces more entrants, the domestic sales of all nonexporters and exporters will decline. However, for exporters the revenue gain from the export market more than compensates for their loss in the domestic market, and thus the more productive firms (i.e. the exporters) expand their total sales. Non-exporters, on the other hand, contract absolutely. This reallocation effect shifts resources and output share to the more productive firms who export, and away from the less productive Probability density

Exiters

Purely domestic firms

Leave

Contra

Domestic threshold (autarky)

Domestic threshold (open)

Exporters Expand Export threshold

Firm level productivity

Figure 7.2 Productivity heterogeneity and with-in industry reallocation

Exporting and Productivity Growth 127

firms who remain domestic. As a result, aggregate industry productivity, defined as the output share weighted average industry productivity, increases. 7.3.1.2

A static firm-level Ricadian model

These trade-induced inter-firm rationalisation and reallocation effects are not confined to the framework used by Melitz. Bernard, Eaton, Jensen and Kortum (2003) (BEJK hereinafter), employ a very different approach, constructing a static Ricardian model of heterogeneous firms with Bertrand competition. A continuum of goods is produced in a world comprised of a given number of countries. Firm heterogeneity is modelled as differences in firm efficiency levels, with the efficiency distribution of the producer with the best technology across goods given exogenously. In the closed economy, under a Bertrand pricing rule2 only the most efficient domestic producer of each good will supply the market, in an open economy, domestic firms will compete with foreign firms in domestic and foreign markets. A home firm will serve its domestic market if it can produce with the lowest cost, compared not only to its domestic rivals but also with all existing foreign firms once trade costs are taken into account. A home firm will serve a given export market, if it can not only survive in its domestic market but also become the lowest cost (including variable trade costs) producer in that market. Assuming iceberg3 transportation costs and imposing a triangle inequality restriction4 on these geographic barriers, BEJK show that in equilibrium every country imports some products, produces some products consumed purely domestically and exports a subset of the goods it produces. Most significantly, they demonstrate that exporters are more efficient than purely domestic firms. A fall in trade costs implies more firms at the high efficiency end of the productivity distribution will expand their share of resource employment as they become able to occupy a greater number of export markets. The least efficient firms exit as they lose their position as the lowest cost supplier in favour of foreign firms. Hence aggregate productivity increases because of a combination of rationalisation and reallocation effects across different goods. To summarise, the following predictions emerge from both the Melitz and BEJK models. Increased exposure to trade: ●

●

forces the least productive firms to leave and thus leads to average productivity gains. allows the most productive firms to expand and export, while less productive firms remain purely domestic.

128 Rod Falvey and Zhihong Yu ●

reallocates resources towards more productive exporting firms and thus leads to aggregate productivity gains.

However, it is noteworthy that the trade-induced industry rationalisation occurs through different channels in the two models. In Melitz (2003), the productivity threshold is raised by the entry of forwardlooking firms, rather than import competition from foreign firms. This is due to the assumption of Dixit-Stiglitz competition with a large number of varieties, where the price elasticity of demand for each variety does not respond to changes in the number of competing varieties. In contrast, in the static BEJK model which assumes Bertrand competition across foreign and domestic firms, it is the import competition that forces the least efficient firms to leave, since more efficient foreign firms replace less efficient domestic producers by charging a lower price for the same good. Table 7.1 summarises these and other differences between these two models. Melitz and Ottaviano (2004) show that the demand structure of the Melitz model can be modified to allow the intensified competition from imports to affect firm mark ups in the domestic market. They develop a

Table 7.1 Summary of model differences Melitz (2003)

BEJK (2003)

Modelling framework

Dynamic industry model

Static Ricardian model

Modelling of firm heterogeneity

Firms face pre-entry productivity uncertainty when entry requires an irreversible investment

Specific distribution of frontier efficiencies for producing each good

Why exporters are more productive than non-exporters

Existence of fixed export costs

Existence of per unit trade costs

Why trade forces the least productive firms to leave

Increase of productivity threshold required for survival due to increase in real wage

Betrand competition from more productive foreign firms producing the same variety

Why trade expands the most productive firms

More productive firms export, and their increased sales more than compensate for their loss of domestic sales due to import competition

Only more productive firms can displace foreign firms in export markets and expand their sales

Exporting and Productivity Growth 129

heterogeneous firm model with an endogenous distribution of mark ups across firms.5 These mark ups are affected by the ‘toughness of competition’ in the market, as measured by the number of competing firms and their average productivity. They find that competition in a larger market is actually ‘tougher’, in that the productivity cut off for survival is higher, the ex ante probability of survival of entrants is lower, and price mark ups are lower.6 As a consequence, firms in larger markets are on average more productive and charge lower prices, and this result holds in both closed and open economies. Bilateral trade liberalisation then increases average productivity in both countries, although by more in the larger country. This is not unexpected, since greater openness will induce disproportionately more firms to locate in the larger market, thereby increasing competitive pressure, which eventually forces a larger proportion of the more unproductive firms to exit the market. 7.3.2 Comparative advantage and industry productivity growth 7.3.2.1

Cross-country efficiency gaps

One limitation of the Melitz model is that it is restricted to a world consisting of identical countries. This implies that all trade is intraindustry trade and that the productivity and welfare gains from trade are symmetrically distributed across countries. But what happens if countries are not identical? Do these rationalisation and reallocation effects work differently in countries with a comparative advantage or disadvantage in the differentiated product industry? Heterogeneous firm models with two asymmetric countries have been developed to examine these questions. Interestingly, the answer one finds is sensitive to one’s assumptions on firms’ knowledge of their production technology prior to entry. Montagna (2002) focuses on the impact of free trade on average industry productivity in two countries which produce two goods, one homogeneous and one differentiated. The only difference across countries is their average efficiency level in the production of the differentiated good. Unlike Melitz, she assumes a given distribution of production technologies and that firm (owners) know their productivity prior to entry. For any given level of demand, only those firms whose productivity is above a threshold will find production profitable, and only those firms will enter. An expansion of demand will lead to a price increase and a reduction in this threshold, inducing the entry of firms with productivities between the old and new thresholds. All entrants are less productive than the incumbents. This implies that, for a given productivity

130 Rod Falvey and Zhihong Yu

distribution, average industry productivity is determined by the least efficient surviving firm. At the international level, the two countries are assumed to have productivity distributions of the same ‘shape’, but the ‘best’ firm in the more efficient country has a higher productivity level. In autarky the productivity threshold of a country will be endogenously determined by its own country-specific firm productivity distribution. Due to its superior productivity distribution, the more efficient country will have a higher productivity threshold and average industry productivity. But these differences in productivity threshold levels across countries will be eliminated in an open economy where there are no trade barriers or transport costs. If the markets of two countries are fully integrated by free trade, firms in each will face identical demand and competition conditions, which lead to the equalisation of the productivity thresholds across countries. This common productivity threshold will be lower (higher) than the autarky threshold of the more (less) efficient country, however. That is, the opening up to trade will reduce (raise) the minimum productivity level in the more (less) efficient country, which will lead to the entry (exit) of less productive firms and therefore a fall (rise) in the average industry productivity of the more (less) efficient country. In this case, trade will have an adverse effect on industry productivity in the more efficient country, which has the comparative advantage in this product and whose industry expands with free trade. Only the less efficient country, whose industry contracts, will enjoy the productivity gains as a result of market integration. Furthermore, the larger the crosscountry efficiency gap, the greater the magnitude of the industry productivity gain (reduction) in the less (more) efficiency country. It is notable that these results rely heavily on two key assumptions: first, that firm’s productivity levels are known before entry; and second, that free trade fully integrates the markets of different countries. However, in other models in which trade costs are assumed to be sufficiently high that the markets remain segmented in the open economy, the above results may not hold. Like Montagna, Jean (2002) considers a two country two sector model with a cross country efficiency gap.7 However, like Melitz (2003), he assumes a random draw of firm-specific productivity after entry, as well as the existence of relatively high trade costs which segment the domestic and the export markets. Jean then makes the point that trade opening will raise productivity thresholds in both countries through two mechanisms. One is import-driven, and refers to the import competition from foreign firms and will always occur in the less efficient country. The other is export-driven, and reflects the new entry

Exporting and Productivity Growth 131

of domestic firms attracted by export opportunities and always occurs in the more efficient country. Both effects will exacerbate market competition and push up the minimum productivity level required to survive. Jean hypothesises that in the case of ‘partial’ inter-industry trade, where the more efficient country exports the differentiated good to the less efficient country (which has positive domestic production of differentiated goods), only the export-driven (import-driven) logic applies in the more (less) efficient country. However, when the efficiency gap is sufficiently small there will be intra-industry trade and both forces are at work in both countries. Hence trade opening unambiguously leads to average productivity improvements in both countries, not only in intraindustry trade but also in partial inter-industry trade. Similar results are derived by Falvey, Greenaway and Yu (2004) (FGY), who adapt the Melitz model to two countries in which the shape of the distribution from which entrants draw their productivities is the same, but the lowest possible productivity draw in the more efficient country is higher than in the less efficient country. In this framework free trade will see the differentiated good industry concentrated in the more efficient country where entrants can expect a better productivity draw. There will then be one-way trade, rather than intra-industry trade in the differentiated good sector. Where trade costs are relatively high, there will be intraindustry trade, and the mechanism whereby trade opening induces new entry and raises the domestic productivity threshold, as described in Melitz, is at work in both countries. But entrants in the more efficient country will have a higher ex-ante probability of exporting, leading a greater number of firms to enter, which pushes up the productivity threshold as well as the death rate among the least productive firms compared to autarky. Hence, the country with an absolute advantage in the differentiated good sector will reap the higher productivity gains. 7.3.2.2

Cross-country factor endowment differences

The above models are restricted to a one factor framework. But what is the role of firm productivity heterogeneity when trade is driven by both factor endowment based comparative advantage and horizontal product differentiation? Based on Helpman and Krugman (1985) and Melitz (2003), Bernard, Redding and Schott (2003) integrate heterogeneous firms into the relative endowment framework of Heckscher and Ohlin. They assume that countries have identical technology and productivity distributions across firm8 and only differ in their relative factor endowments. One feature of the model is that they show that opening to costly trade will lead to industry productivity improvement in all

132 Rod Falvey and Zhihong Yu

industries in all countries, but with greater improvements in a country’s comparative advantage industry relative to its comparative disadvantage industry. The underlying reasoning is quite intuitive: firms in the comparative advantage industry will be exposed to less competition in the export market, therefore the ex-ante probability of exporting will be higher in the comparative advantage industry. This will result in relatively greater domestic entry, which pushes up the productivity threshold by more. Hence the trade-induced rationalisation effect on industry productivity will be stronger in the comparative advantage industry compared to the comparative disadvantage industry.

7.3.3

Summary

In this section we reviewed a series of new trade models incorporating firm level productivity heterogeneity, which contribute to our understanding of the impact of international trade on industry productivity dynamics. One important aspect of the heterogeneous firm trade theories is that they provide plausible explanations for how trade may raise industry productivity through within industry reallocation, as has been documented in plant- and firm-level empirical trade studies. They also generate a rich set of predictions on the interaction between the trade-induced productivity improvements and a country’s comparative advantage, which have yet to be empirically tested. However, despite their consensus on the potential benefits of trade on productivity through within industry rationalisation, the existing models differ in a number of important respects: ●

●

●

Modelling productivity heterogeneity. Static models assume fixed productivity among entrants, whereas dynamic models rely on preentry productivity uncertainty. The role of import competition and exposure to export markets. Is it pressure from the entry of foreign firms or from more domestic entrants that force the least productive firms to exit? Some models (Melitz, BRS, FGY) rely on the latter, others the former (BEJK, Montagna) and some both ( Jean), depending on their assumptions on the structure of competition. The sign and magnitude of the productivity gains. Will all countries and all industries gain, and which gain the most? While some models extending Melitz (e.g. FGY, BRS) imply that the gain occurs in all industries and countries and is stronger in industries with comparative advantage, Montagna shows the possibility of an adverse productivity effect of trade.

Exporting and Productivity Growth 133

Employing different approaches to deal with the above issues will lead to different empirical predictions. Future research may be directed at reconciling and testing these approaches, and integrating cross-firm productivity heterogeneity with other main-stream trade models.

7.4 Conclusions Our discussion of the theoretical literature that examines the effects of exporting on firm and industry productivity, separated the effects that occur within the firm from those that differentially affect heterogeneous firms within the industry. Obviously this is a somewhat artificial separation in practice, but it was useful to distinguish the ‘within firm’ and ‘within industry’ effects in our analysis. Our discussion presented an array of circumstances and channels through which opening an economy to trade can improve the average productivity of its exporting (and in most cases also its import-competing) firms. Determining which of these channels, if any, is relevant in any particular case will likely depend on the characteristics of the industries and countries concerned, and is ultimately an empirical matter. But where theory can help is by identifying the full implications of particular sets of assumptions so as to differentiate between explanations that have many predictions in common. Clearly there are many important issues worthy of further investigation. For example, learning by exporting appears to be an important potential source of productivity improvements for developing countries in particular. But attempts to formulate the most supportive policy environment for such productivity transfers need to know what characterises the firms best able to take advantage of the opportunities to learn by exporting, and which domestic market characteristics generate the maximum productivity spillovers from these learners to their domestic competitors. The increasing integration of national goods and factor markets through the globalisation process, has been facilitated by reductions in international trade costs (broadly interpreted). Most of the models that we have reviewed have included costs of exporting, and in many cases the distinction between fixed and variable exporting costs has been important. In the Melitz model, for example, it is the fixed costs that create the exporting productivity threshold. While any trade cost reduction will encourage exporting, it would be useful to know what has happened to the relative sizes of fixed and variable export costs during the globalisation process. Finally, one key feature of many of the models considered above is heterogeneity in firms’ productivities within an industry. It is this

134 Rod Falvey and Zhihong Yu

heterogeneity that leads some firms to become exporters, some to confine their activities to the domestic market and others to shut down. Identifying the sources of this heterogeneity is basic to explaining which firms will end up in which category. This is clearly an important area for further research, if for no other reason than it is only when these sources have been pinned down that appropriate policy action can be formulated.

Notes 1. Whether increased import competition encouraged firms to become more efficient is a debate of long standing. Balassa (1967) and Leibenstein (1966) provide early informal discussions, while Horn et al., (1995) treat trade policy in environments with principal agent problems. 2. Betrand competition here means that the producer charging the lowest price obtains the whole market for that variety, and the upper bound of this price is the cost of the second most efficient (potential) producer. 3. Iceberg trade costs means that transportation costs are proportional to production costs. In the presence of efficiency heterogeneity, an immediate implication is that more efficient firms have both lower production costs and lower transportation costs. 4. The upper bound on the cost of moving goods directly from country A to country B is the cost of transporting them via a third country C. 5. This is achieved by using a linear demand system with horizontal product differentiation based on quadratic utility which is developed by Ottaviano, Tabuchi and Thisse (2002). The assumptions on firm heterogeneity etc., are carried over from Melitz (2003 ). 6. This is in contrast to some other heterogeneous firm models in which a larger market can accommodate more unproductive firms. See for example, Montagna (2003). 7. In Jean (2002) in the more efficient country both the highest and the lowest bound of the productivity distribution are higher than those in the less efficient country. 8. In their two factor model, firm level productivity is defined as the physical output per input bundle.

References Alvarez, R. and R. Robertson (2004), ‘Exposure to Foreign Markets and Firm-Level Innovation: Evidence from Chile and Mexico’, Journal of International Trade and Economic Development, 13, 57–87. Balassa, B. (1967), Trade Liberalization Among Industrialized Countries: Objectives and Alternatives, New York: McGraw-Hill. Baldwin, R. and P. Krugman (1989), ‘Persistent Effects of Large Exchange Rate Shocks’, Quarterly Journal of Economics, 104, 635–54. Balock, G. and P. Gertler (2004), ‘Learning from Exporting Revisited in a Less Developed Setting’, Journal of Development Economics, 75, 397–416. Bernard, A., J. Eaton, B. Jensen, and S. Kortum (2003), ‘Plants and Productivity in International Trade’, American Economic Review, 93(4), September, 1268–90.

Exporting and Productivity Growth 135 Bernard, A. and B. Jensen (2004), ‘Exporting and Productivity’, Mimeo, Hanover: Tuck School of Business, Dartmouth College. Bernard, A. and B. Jensen (1999), ‘Economics, Exceptional Exporter Performance: Cause, Effect, or Both?’, Journal of International Economics, 47, 1–25. Bernard, A., S. Redding and P. Schott (2004), ‘Comparative Advantage and Heterogeneous Firms’, Mimeo, Hanover: Tuck School of Business, Dartmouth College. Chisik, R. (2003) ‘Export Industry Policy and Reputational Comparative Advantage’, Journal of International Economics, 59(2), 423–51. Clerides, M., S. Lach and J. Tybout (1998), ‘Is Learning by Exporting Important? Micro-Dynamic Evidence From Colombia, Mexico, and Morocco’, Quarterly Journal of Economics, 113, 903–47. Falvey, R., D. Greenaway and Z. Yu (2004), ‘Intra-industry Trade Between Asymmetric Countries with Heterogeneous Firms’, Working Paper 04/2004, Leverhulme Centre for Research on Globalisation and Economic Policy, School of Economics, University of Nottingham. Holmes, T. and J. Schmitz (2001), ‘A Gain From Trade: From Unproductive to Productive Entrepreneurship’, Journal of Monetary Economics, 47(2), 417–46 Hopenhayn, H. (1992), ‘Entry, Exit and Firm Dynamics in Long Run Equilibrium’, Econometrica, 60, 1127–50 Horn, H. H. Lang and S. Lundgren (1995), ‘Managerial Effort Incentives, X-inefficiency and International Trade’, European Economic Review, 39, 117–138. Jean, S. (2002), ‘International Trade and Firms’ Heterogeneity under Monopolistic Competition’, Open Economies Review, 13, 291–311. Krugman, P. (1980), ‘Scale Economies, Product Differentiation, and the Pattern of Trade’, American Economic Review, 70, 950–959 Leibenstein, H. (1966), ‘Allocative Efficiency vs. “X-efficiency” ’, American Economic Review, 56, 392–415. Montagna, C. (2001), ‘Efficiency Gaps, Love of Variety and International Trade’, Economica, 68, 27–44. Melitz, M. (2003), ‘The impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity’, Econometrica, 71, 1695–726. Melitz, M. and G. Ottaviano (2003), ‘Market Size, Trade and Productivity,’ Harvard University, Department of Economics. Ottaviano, G.I.P., T. Tabuchi and J.-F. Thisse (2002), ‘Agglomeration and Trade Revisited’, International Economic Review, 43, 409–36. Pavnik, N. (2002), ‘Trade liberalization, Exit and Productivity Improvement: Evidence from Chilean Plants’, The Review of Economic Studies, 69, 245–76. Rhee, Ross-Larson and Pursell (1984) Korea’s Competitive Edge: Managing the Entry into the World Markets, Baltimore, MD: Johns Hopkins University Press. Roberts, K. and J. Tybout (1997), ‘The Decision to Export in Colombia: An Empirical Model of Entry with Sunk Costs’, American Economic Review, 87, 545–564. Tybout J. and D. Westbrook (1995), ‘Trade Liberalisation and the Dimensions of Efficiency Change in Mexican Manufacturing Industries’, Journal of International Economics, 39, 53–78.

8 Empirical Evidence on Exporting and Productivity Growth in the UK David Greenaway and Richard Kneller

8.1 Introduction One important, but almost invisible outcome of the Uruguay Round of multilateral trade negotiations was the creation of the Trade Policy Review Mechanism (TPRM). What this established was a regular audit of trade policies of Members, with its frequency depending upon country size – from every two years for the United States, Japan and EU through to every seven years for countries like Vanuatu and Barbados. One of the key purposes of the TPRM is to bring greater transparency to border and non-border protectionist measures. But, interestingly, every Trade Policy Review includes a chapter on ‘Measures directly affecting exports’. Moreover, this is always a substantive chapter and always reports on export promotion measures. Export promotion is pervasive, not as pervasive as import protection, but pervasive nonetheless. Putting to one side high profile but rare ‘strategic trade policy’ cases such as wide bodied jets, it is also less controversial and all governments provide some kind of support. For example, as in other industrialised countries, the United Kingdom has an interdepartmental agency, namely UK Trade and Investment (UKTI). This ‘offers support to companies based in the UK to achieve their export potential … [by helping] develop export capabilities and provide expert advice, reliable data and professional research’.1 UKTI does not provide direct export subsidies but does offer a range of information and facilitation services designed to reduce the sunk costs of export market entry. This kind of intervention, as well as the provision of direct export subsidies which are such a common feature of Export Processing Zones in developing countries, occurs because policy makers see promoting export as a good thing. They connect it with exploiting comparative 136

Exporting and Productivity Growth 137

advantage (probably unconsciously) and linked to export-led growth (almost certainly consciously). But what is the evidence base for this? For the most part it is macroeconomic and macroeconometric. In the main it derives from time series and cross-country correlations between growth in real exports and growth in real output or total factor productivity (TFP) growth. It also derives from a perception that export promoting developing countries outperformed inward orientated developing countries in the second half of the twentieth century. The ‘macro’ evidence is controversial with some arguing that links between exporting and growth are clear and causal (e.g. Edwards, 1998) and others arguing that the reality is rather more complicated and the role of export promotion overstated (e.g. Rodrik and Rodriguez, 2000). Be that as it may, what is interesting is the absence of any substantive microeconomic or microeconometric evidence base to support intervention. This is interesting because it is firms rather than countries, or industries that export and most active export promotion policies tend to be targeted at firms. But that is changing and quite rapidly. Over the last seven or eight years there has been enormous interest in links between entry to export markets and firm level performance. Initially this was empirically driven, in particular by the work of Bernard and Jensen (1995) on US data, quickly followed by a range of studies on other countries. This work seems to point to several regularities in the data. For example, firms which export tend to be larger than those which do not. They also tend to be more productive, exhibiting higher productivity before they enter export markets. In turn these regularities have stimulated the development of models of firm level adjustment by Melitz (2003), Helpman et al. (2004) and Bernard et al. (2003) to provide some theoretical underpinning to what we seem to be seeing in the data. In this paper we focus on the United Kingdom, on which a number of studies have now been completed. We review that evidence, but also report on new evidence relating to factors which are related to export market entry and firm level performance once entry has taken place. We begin in Section 8.2 with an analysis of how exporters differ from non-exporters and focus in Section 8.3 on the determinants of entry. Section 8.4 evaluates the consequences of entry. In Section 8.5 we discuss the implications of our results for economic policy and conclude.

8.2 How exporters differ In Table 8.1 we compare UK export firms and non-exporters across a range of performance characteristics.2 The first column reports how

138 David Greenaway and Richard Kneller Table 8.1 Percentage difference between exporters and non-exporters and their statistical significance

Exporters vs. non-exporters Employment Output Wages Labour productivity TFP

Export premium (%)

t-test of difference in means

Conditional export premium (%)

t-statistic

12.6 20.8 0.5 2.2

43.25* 50.11* 8.55* 15.72*

6.2 3.9 2.3 11.4

9.17* 9.21* 7.16* 20.17*

9.7

20.81*

8.3

15.56*

Notes: * Denotes significance at 5% level.

much larger exporters are than non-exporters (measured at the mean). This evidence is consistent with the stylised facts from other countries. At the mean, employment and output levels in export firms are 12.6 per cent and 20.8 per cent greater than those in non-export firms. Exporters also pay on average higher wages, although the difference here is small at just 0.5 per cent. Given these output and employment figures it is no surprise that they have higher labour productivity (a premium of 2.2 per cent). This is not just confined to a difference in capital intensity, exporters also have higher TFP. The TFP of non-exporters lies on average 4.3 per cent below the industry mean and that for exporters 5.4 per cent above, a net difference of 9.7 per cent. The second column reports test statistics from a t-test of whether these differences are significant or not. They suggest that all of the differences reported in column one are significant at standard levels; exporters and non-exporters do appear to have different characteristics. In the final columns of Table 8.1 we condition the export premium reported in column one on other covariates that might affect firm performance and may bias the result from a simple comparison of means. For example, the export premium may be biased upwards if exporters are concentrated in industries where economies of scale are important, and therefore output of the average firm is greater than in industries where economies of scale are less important and exporting less likely. To control for additional covariates we estimate the regression, ln Yit 
0 
1EXPDUMit 
2 ln Zit1 

101

13

兺
IND  兺
T

j1

j

j

t1

t

t

 it

(1)

Exporting and Productivity Growth 139

Where, Y is the firm characteristic under test, EXPDUM is a dummy variable indicating the export status of the firm, the matrix Z controls for other firm level characteristics (chosen from amongst measures of total employment, average wages, capital stock and TFP), IND controls for fixed industry effects (measured at the 3-digit SIC level) and T fixed time effects. The subscript i indexes firms, t time and j industries. We report only the estimated parameter value on the export dummy and its t-statistic (4th column) to conserve space. Conditional on these the premium in performance characteristics between export and non-export firms remains, although in some cases the differences are noticeably smaller than in column 1. Export firms are larger, pay higher wages and have higher productivity and these differences are all significantly different from zero. In terms of the size of the firm, the premium is between 3.9 per cent and 6.2 per cent, when measured by either output or employment, wages are 2.3 per cent higher and productivity is between 8.3 per cent and 11.4 per cent higher (measured by TFP or labour productivity, respectively). While the differences between exporters and non-exporters are large they do not provide us with information about causality; do good firms become exporters or does exporting make firms good? Which characteristics fashion the probability that a firm will export? Roberts and Tybout (1997) find that firm level characteristics such as plant size and age are important for entry in Colombian plants, while Bernard and Jensen (2004) report that in addition productivity and average wage levels are important for US firms. We estimate a probit regression, where export status is regressed on a series of firm level and industry characteristics, P(EXPDUMit 0/1)  f(EXPDUMit1,TFPit1,sizeit1, wagesit1,Tt,REGk,INDjt)

(2)

The fixed time and industry effects are included as in Equation (1). In addition we add a measure of industry agglomeration at the 5-digit level from Duranton and Overmans (2002). The firm level variables are all lagged one period; we are interested whether ex-ante success helps predict export status (Bernard and Jenson, 1999). To provide some interpretation of the estimated coefficients we report the marginal effect calculated at the mean of each right hand side variable, except where the determinant of exporting is a dummy, where the reported marginal effect is the effect of the change from zero to one.

140 David Greenaway and Richard Kneller

From this it seems that the most important determinant of exporting is experience. Exporting yesterday is a good predictor of exporting today. Export experience in the previous period raises the probability of exporting in this period by 0.83, in line with the marginal effects of experience reported in Bugamelli and Infante (2002). This matches evidence of persistence noted later on, where entry and exit rates from export markets were relatively low. The importance of past behaviour is usually interpreted as evidence of sunk-costs. All of the other determinants have a positive effect on the probability of exporting, but only size and wages are statistically significant.3 These characteristics also appear to be quantitatively far less important than experience. Increasing TFP, size or wages by one unit raises the probability of exporting by 0.7, 2.0 and 2.5 percentage points, respectively. Even when one allows for the sample variation in these measures the effect is small. At the mean an increase by one standard deviation for each of these variables raises the probability of exporting by 2.6 percentage points for size, 0.4 percentage points for TFP and 1 percentage point for wages. The results from Table 8.2 suggest that agglomeration is important. To explore this further two types of regional variables are included. The first measures fixed regional effects, whereas the second is a measure of agglomeration of industries. We might expect that because of higher transaction costs, the costs of exporting rise the greater the distance from continental Europe. There is some evidence for this. Relative to Central London, the probability of a firm located in Northern Scotland, Southern Scotland and Wales exporting is significantly lower. The probability is also lower for Outer London, suggesting that these dummies in part capture the industrial mix of regions. More important is the agglomeration of industries, where there is strong evidence that industrial agglomeration is significantly correlated with the probability a firm will export. Whilst consistent with evidence of spillover effects from the co-location of firms within the same industry, caution should be applied before interpreting this as evidence of a causal relationship from industrial agglomeration to exporting. Industrial agglomeration and exporting may both be determined by some third omitted variable. We consider this further later on. According to the estimated marginal effects, at the mean an increase in industry agglomeration by one standard deviation increases the probability of exporting by close to 3.5 percentage points. This industry specific factor therefore has a similar effect on the probability of exporting to firm characteristics such as size and TFP.

Exporting and Productivity Growth 141 Table 8.2 Probability model of exporting Estimated coefficient (z-stat)

Variable EXPDUMt  1 Log(EMP)t  1 Log(Wage)t  1 TFPt  1 Industry agglomeration

Observation Observed P

0.830 (146.52)** 0.020 (9.69)** 0.025 (3.28)** 0.007 (1.45) 0.650 (11.09)**

54,731 0.683

Variable

Estimated coefficient (z-stat)

East Midlands Home Counties North East North Scotland North West

0.006 (0.56) 0.001 (0.10) 0.013 (1.00) 0.018 (1.18) 0.008 (0.78) 0.041 (1.84) 0.015 (1.34)

Pseudo R2 Predicted P

0.699 0.833

Central Southern East Anglia

Variable Northern Ireland Outer London South East South West Southern Scotland Wales West Midlands

Estimated coefficient (z-stat) 0.071 (0.90) 0.069 (5.45)** 0.006 (0.47) 0.005 (0.39) 0.032 (2.21)* 0.033 (2.09)* 0.009 (0.83)

Notes:  denotes significance at the 10% level; * significance at the 5% level; and ** significant at the 1% level.In a probit model the marginal effect of a regressor xi on the probability y is given

by P(y  1)  F(x
)  f(x
)*i where the normal density function f(.) is conventionally evaluated xi

xi



at the predicted probability x
where x stands for the sample average of the regressor and  the estimated coefficient.

8.3 New entrants – determinants of entry The preceding results show that firms operating in the same industry can differ substantially in their performance, and these differences help explain why some become exporters and others do not. Ex-ante success is an important determinant of participation in export markets (Bernard and Jensen, 1999). Simply put, the best firms self-select into becoming export firms. Yet within the data we also find firms that transit between the two states of exporting and non-exporting. Over the sample period 15 per cent of firms that were not exporting in t1 start to export in t, while 2 per cent of firms that were exporting stop (Table 8.3). Since past experience is such an important factor in predicting participation we might ask what enables a firm to overcome the barriers to entry into export markets? Is it the result of endogenous improvements in firm

142 David Greenaway and Richard Kneller Table 8.3 Transitions in and out of exporting

Not-exportingt  1 Exportingt  1

Not-exportingt

Exportingt

19,605 (85%) 884 (2%)

3,461 (15%) 43,311 (98%)

Table 8.4 Characteristics of new exporters versus established exporters and non-exporters New exporters versus non-exporters TFP (t-test) TFP growth (t-test) Average firm sales (t-test) Average number of employees (t-test)

6.6% (7.94)* 1.9% (2.52)* 18% (22.90)* 26.3% (22.48)*

New exporters versus exporters 3.2% (1.41) 2.8% (4.18)* 1% (0.26) 17.0% (0.85)

Notes: * denotes significance at the 5% level.

performance in the periods leading up to entry? Does the experience gained by other new and existing exporters spill over, effectively reducing sunk-costs of entry? And what happens to the firm in the years following entry? The basic characteristics of new relative to ‘old’ exporters and nonexporters are set out in Table 8.4. On average the productivity of firms that become exporters during the sample period lies between that of exporters and non-exporters. TFP in new export firms is 6.6 per cent above that of firms that do not export and 3.2 per cent below that of firms that do. Using a t-test of differences in mean we find that the difference with respect to non-exporters is statistically significant, whereas it is not so with respect to established exporters. Bernard and Jensen (1999) suggest that in the United States new export firms have faster TFP growth than both non-exporters and established exporters. Similar evidence can be found for the United Kingdom. Average TFP growth is 1.9 percentage points per annum faster in new exporters than nonexporters and 2.4 percentage points faster than established export firms (both statistically significant).

Exporting and Productivity Growth 143 Table 8.5 Probit model of export market entry

Variable Log(EMP)t  1 TFPt  1 Industry agglomeration Central Southern East Anglia

Observation Observed P

Estimated coefficient (z-stat) 0.020 (9.34)** 0.011 (2.62)** 0.785 (10.30)** 0.017 (1.59) 0.006 (0.48)

16,646 0.144

Variable East Midlands Home Counties North East North Scotland North West Northern Ireland Pseudo R2 Predicted P

Estimated coefficient (z-stat) 0.008 (0.72) 0.035 (2.11)* 0.005 (0.47) 0.024 (1.23) 0.019 (1.88) 0.008 (0.16) 0.232 0.099

Variable Outer London South East South West Southern Scotland Wales West Midlands

Estimated coefficient (z-stat) 0.057 (5.68)** 0.003 (0.24) 0.021 (1.43) 0.024 (1.97)* 0.035 (2.62)** 0.023 (2.01)*

Notes:  denotes significance at the 10% level; * significance at the 5% level; and ** significance at the 1% level.

The remaining variables in Table 8.4 confirm that new export firms are closer in their characteristics to established exporters than non-exporters. Evidence suggests a significant difference in the size of new exporters relative to non-exporters, but not between new and established exporters. Given the differences that exist between new exporters and nonexporters it is perhaps no surprise that these result in some leaving the pool of non-export firms. In Table 8.5 we report the results from a probit regression of export market entry on the level of TFP and size of the firm and a series of industry level characteristics. Again the reported coefficients are estimated marginal effects, calculated at the mean. Comparing the marginal effects between Tables 8.5 and 8.2 shows little heterogeneity in the effect of the determinants of entry versus exporting more generally. The marginal effect of an increase in firm size is identical in the two models, whereas the marginal effect of TFP is not identical.4 However as the average predicted value serves as the guide through which to interpret the reported marginal effects, and the average predicted value in Table 8.5 is considerably lower than in Table 8.2 we can conclude that firm level changes play an important role in determining entry. In Section 8.4 below we examine the productivity performance of firms that enter export markets in the sample period across time. As in

144 David Greenaway and Richard Kneller

Bernard and Jensen (1999) one characteristic of these firms is their noticeable improvement in productivity in the periods leading up to entry. Using the productivity growth that occurs in the 5-year period before entry, one might ask how much does additional productivity growth increase the probability of entry. Applying regression analysis on the unmatched sample of exporters and non-exporters we find that productivity growth of new exporters was 2 per cent per annum faster than non-export firms five years before entry (but not statistically significant), 2.4 per cent the next year (significant), 3 per cent the next (significant), then 1 per cent (not significant) and 1.7 per cent (significant) faster in the period before entry. Cumulatively this raises the probability of entry of these firms relative to non-exporters from 0.2 of a percentage point 5 years before entry, to 0.8 of a percentage point three years before entry and 1.1 percentage points in the year before entry. This is about 10 per cent of the predicted probability of export market entry. The fixed regional factors also have similar effects to those reported in Table 8.2, although the North West and West Midlands are now significant and North Scotland insignificant. We return to the effect of agglomeration below. The specification of the above regression allows export market entry to be determined solely from changes in firm level characteristics (and fixed time effects). Greenaway and Kneller (2004a and b) explore whether industry level determinants also matter. Agglomeration may encourage entry through a number of different channels. If information about export market opportunities and costs are an important barrier then we might expect that the co-location of firms within the same industry and region leads to sharing of information across potential entrants, reducing these costs. Or the concentration of exporters may lead to improvement in the infrastructure necessary to provide access to foreign markets, or improve information about the ‘tastes’ of foreign consumers (Aitken et al., 1997). Whatever their source, we might expect that entry by non-export firms is more likely the greater is the existing concentration of export activity in the industry. In Table 8.6 we report results from Greenaway and Kneller (2004b) from a similar specification to that used to generate the results in Table 8.5, except where additional measures of agglomeration of export firms are included.5 Agglomeration of exporters is measured by counting the number of export firms within various industry/regional combinations. Industries are measured at the 3-digit division of the Standard Industrial Classification (SIC), which gives 101 industries, whereas regional categorisation relies on the NUTS classification at the 3-digit level (yielding

Exporting and Productivity Growth 145 Table 8.6 Probit model of export market entry: agglomeration effects Same region same industry 0.003 (6.29)** New same region New same industry 0.014 (8.56)**

Diff. region same industry 0.001 (4.53)** New diff. region new same industry 0.005 (6.59)**

Same region diff. industry 0.000 (2.78)**

Diff. region diff. industry 0.000 (2.68)**

New same region New diff. region new diff. industry new diff. industry 0.003 0.000 (5.54)** (3.23)**

Notes:  denotes significance at the 10% level; * significance at the 5% level; and ** significant at the 1% level.

65 identifiable regions). Of course, the possibility exists that the gains from information or demonstration spillovers decay through time. The experiences of more recent entrants may be of more value than older entrants. One way of investigating this is to distinguish between the presence of existing exporters in a given region or industry and other contemporaneous entrants. The reported regression includes information from both. To control for possible correlation of the agglomeration of new exporters with macroeconomic shocks that have an industry and regional dimension the regression includes fixed time effects that vary across regions and industries. Finally we also control for fixed regional and industry effects and the industrial agglomeration variable included in Table 8.5. Of these additional covariates it is worth noting the potential importance of including the measure of industrial agglomeration. If the concentration of export activity is a consequence of agglomeration of economic activity more generally, and there are no spillover effects from agglomeration of exporters, we would expect that the addition of information about industrial agglomeration will remove the significance of the export agglomeration variables. The regression therefore controls as much as possible for other factors that might lead us to spuriously conclude in favour of agglomeration amongst export firms. We recalculate the estimations made in Greenaway and Kneller (2004b) and report instead estimated marginal effects. Two patterns emerge: first, the effects of agglomeration decline as physical and industrial distance between firms increases. The effect of being in the same region and industry has the greatest impact on the probability of entry by non-export firms. The industrial dimension of agglomeration would appear to be more important, the effect of being

146 David Greenaway and Richard Kneller

in the same industry but a different region is greater than being in the same region but a different industry. Second, there is evidence of decay in the usefulness of this information across time. Contemporaneous entry of other export firms in the same period has a larger impact on the estimated probability of entry than established export firms. At the sample mean the estimated marginal effect of an additional export firm has a strict ordering according to when the firm became an exporter, where it is located and in which industry it operates. The effect of an additional firm in the current period located in the same region and industry raises the probability of exporting by 1.4 percentage points, whereas if the same firm were located in a different industry the effect is just 0.5 of a percentage point. The effect of an additional established export firm located in the same region and industry has a smaller marginal effect, estimated to be just 0.3 of a percentage point. This is the same as the effect of an additional export firm located in the same region but a different industry and which begins to export in the current period. The beneficial impact of co-location of firms on the export decision has also been explored by Greenaway et al. (2004) for the United Kingdom, Aitken et al. (1997) for Mexico and Bernard and Jensen (2004) for the United States. The first two studies explore the role played by multinationals in encouraging entry amongst domestic firms, finding that their presence has a statistically positive and significant effect. Bernard and Jensen (2004) as well as Aitken et al. (1997) also explore agglomeration effects from the concentration of export firms more generally. The evidence presented there is mixed on this point, the results depending upon the inclusion of particular industries. Using somewhat more aggregated measures of agglomeration of industries and regions (regions are measured by states and industries at the 2-digit level) than Greenaway and Kneller (2004a and b) and Aitken et al. (1997), Bernard and Jensen (2004) find no effect from agglomeration on the export entry decision.

8.4 New entrants – consequences of entry While there would appear to be general agreement of a causal relationship from productivity to exporting, there is less agreement whether exporting brings any additional benefit to the firm. Bernard and Jensen (1999) for example, find that exporting has a significant effect on the survival of firms. Even conditioning on plant level characteristics, they have higher employment growth but not faster productivity growth.

Exporting and Productivity Growth 147

Girma et al. (2004) in contrast find a significant effect from export market entry to productivity growth. In part this disagreement reflects a difference in methodology, or more specifically how new export firms are compared to non-export firms. The effect of entry on firm performance is most commonly evaluated by comparing the performance of the firm in the time periods before with those after entry. Ceteris paribus this can be known from the information contained in the cross-time performance of a sample of new entrants. If as seems likely such an assumption does not hold, then this approach will tend to over/understate the effect of export market entry according to the direction of any shocks or firm specific changes. The effect of entry can then only be known relative to what would have happened to the firm had it not entered the export market. This counterfactual is of course unobservable. The question therefore arises as to what constitutes a ‘valid’ control group. The literature diverges on this point. Bernard and Jensen (1995) and others assume that all non-export firms (or a random sample of them) are capable of providing the counterfactual. One objection to this might be the heterogeneous nature of productivity levels between export and non-export firms suggested by recent contributions to the theoretical literature (see in particular Melitz, 2003; Helpman et al., 2004; and Head and Ries, 2003). In support of this the differences between new export market entrants and non-exporters in Table 8.4 were found to be statistically significant across a number of dimensions of firm performance. If there are substantial differences in the distribution of the observed covariates between exporters and non-exporters, as the theory and Table 8.4 suggest, then a pooled regression will yield unreliable results (Rubin, 2001). Given that new export market entrants appear to come from the upper end of the distribution of non-export firms, we might expect that such an approach will tend to bias upward the observed cross-time performance between exporters and non-exporters. Wagner (2002), Girma et al. (2004), Greenaway and Kneller (2003) and Greenaway et al. (2003) chose instead to generate a control group of non-export firms using information on observable firm characteristics.6 That is, to select from the reservoir of non-exporters those firms in which the distribution of the right hand side variables is as similar as possible to that for exporting firms in the period before the latter become exporters. There are several alternative methods of generating this comparison group. For brevity we describe the propensity score matching method used in Girma et al. (2004) and refer the reader to Blundell and Costa Dias (2000) for others.

148 David Greenaway and Richard Kneller

Using information from a probit regression, each firm is assigned a probability score (or propensity score) that they will become an export firm. Using this, actual export market entrants are then matched to a non-export firm so as to minimise the difference in the propensity scores. This is then combined with difference-in-difference analysis to control for changes in other observable determinants of firm performance in the post-entry period. While this method of generating the comparison group on the basis of observable firm level characteristics appears preferable on a-priori grounds, there are a number of questions that might reasonably be asked about its effectiveness and the conclusions drawn from its application. First, does it have an important effect on the results? That is, should we be concerned more by heterogeneity of the right hand side variables in the difference-in-difference regression (as the matching methodology suggests we should) or by the loss of efficiency that comes from removing information from the sample (if we implement matching). Second, how sensitive are the results to changes in the matching process? This might operate either across changes to the probit regression or the closeness of the ‘match’ allowed in the matching procedure. Finally, do unobservable firm characteristics, such as managerial ability, determine both the selection into export markets and the effect on firm performance once it does? We deal with the last of these first; consider what impact matching has relative to not matching; and refer the reader to Greenaway and Kneller (2004b) for the second. In that study the effect of export market entry was not found to depend on the choice of matching procedure or on the probit regression. The procedure typically adopted matches firms based on their observable characteristics. Yet, it would seem likely that unobservable characteristics, such as managerial ability or product characteristics, play an important role in the participation decision, despite predictions from recent theoretical innovations such as Melitz (2003). In the heterogeneous firm model, entry into export markets within an industry is determined strictly on the basis of the productivity of the firm. The marginal exporter should have higher productivity than the marginal nonexporter. In practice there are some firms that, based on their observable characteristics, could export but do not. In the data controlling for productivity, as well as a range of other firm and industry characteristics, we find that at the mean of the sample the model (presented in Table 8.2) predicts that 83 per cent of firms will export, against an observed incidence of 69 per cent. Another way of looking at the same point is to generate a measure of the degree of uncertainty in the choice that firms

Exporting and Productivity Growth 149

make. Conditional on the industry and time period we find that of 1064 observations on (101) 3-digit industries over the sample period, there are just 24 observations (across 18 industries) in which the marginal exporter strictly has greater TFP than the marginal non-exporter. Of these, 17 observations fall either in the first or the last three years of the sample where the number of observations is somewhat less. In parlance of the evaluation literature the conditional independence assumption is violated. While we might expect that the effect of some unobservables, such as managerial ability, on the selection process might be lessened, because they are also likely to affect observable characteristics such as productivity, the importance of unobservables in the decision to become an exporter should not be seen as an argument against matching. Matching on observables is still preferable to using the entire sample of nonexport firms as the counterfactual if it brings the comparison group closer to the treatment (Blundell and Costa-Dias, 2000). In addition to their effect on the decision to become an exporter we might think of two other effects of managerial ability on the estimated impact of exporting. First, managerial ability or product attributes are likely to affect firm performance independent of the export entry decision. Firms with high managerial ability for example are likely to consistently outperform those with low ability. This persistence can be removed by combining matching with difference-in-differences and is unlikely to bias the estimated effect of export market entry. The error term is assumed to comprise a firm specific, time invariant component and a random component. Firm fixed effects can therefore appropriately control for the persistent element of firm performance in the estimated regression. Second, the unobservable factor may induce heterogeneity in the outcome. Some firms may benefit more from export market entry than others because they are high managerial ability firms. These timevarying effects of unobserved characteristics are not possible to eliminate through either matching or difference-in-differences. In this case we observe in our results not the effect of export market entry on performance where any firm randomly becomes an exporter, but the effect of export market entry on those who chose to become exporters. In the parlance of the literature, we observe the treatment effect on the treated not the treatment effect on the population. Knowing the effect of the treatment on the treated is a valid research question however, and its impact is not on the plausibility of the results but on interpretation and policy conclusions. In terms of policy, it limits our ability to say from our results that encouraging additional export market participation

150 David Greenaway and Richard Kneller Table 8.7 Sample characteristics of exporters and nonexporters in a matched sample of firms

Matched new exporters TFP TFP growth Log(Emp) Log(Output) Matched non-exporters TFP TFP growth Log(Emp) Log(Output)

Obs.

Mean

Standard deviation

1,271

0.010 0.011 4.64 8.92

0.61 0.31 1.43 1.48

1,154

0.006 0.020 4.54 8.77

0.60 0.30 1.35 1.39

amongst the pool of non-export firms will yield the same returns to productivity as we observe in the data. In order to generate our propensity scores we apply the probit model in Table 8.5 earlier. In Table 8.7 we report the characteristics of the matched and un-matched samples of new exporters and non-export firms. Matching has the expected effect of reducing the export premium of new exporters found in Table 8.4. Using either a t-test of difference in means or estimating a regression that additionally controls for fixed time and industry effects we find that the export premium in the level of TFP between new exporters and non-exporters is statistically insignificant. In the matched sample the TFP premium of new export firms is between 1.6 and 1.7 per cent depending on the method of estimation and statistically insignificant. This contrasts with the evidence presented in Table 8.4. The results for the effect of entry on performance are reported for the matched and un-matched samples in Table 8.8. Obviously it is not possible to generate periods before and after entry for the un-matched sample. For this reason we estimate regressions separately for each period and test for the difference in the growth of TFP for the two sets of firms, controlling for fixed firm, time and industry effects as well as the lagged level of TFP, employment and wages. We report results for the matched sample in the same way, but add a regression that compares TFP growth across all periods.7 In the unmatched sample TFP growth was 2.9 per cent faster before entry, 4.4 per cent faster in the year of entry and still significantly faster

Exporting and Productivity Growth 151 Table 8.8 Effect of export market entry on firm performance for a matched and unmatched sample of firms All time periods Unmatched Export Premium Observations R-squared Matched Export Premium Observations R-squared

0.024 (3.95)** 11,580 0.13

Pre-entry

Entry period

Entry t⫹1

0.029 (4.56)** 18,106 0.12

0.044 (5.02)** 19,266 0.14

0.036 (5.21)** 18,047 0.12

0.018 (2.36)* 15,423 0.09

0.002 (0.16) 2,417 0.17

0.036 (5.16)** 3,470 0.23

0.015 (1.41) 3,074 0.07

0.001 (0.07) 2,619 0.09

Entry t ⫹ 2

Notes: * denotes significance at the 5% level; ** significance at the 1% level.

two years after entry. As reported by Bernard and Jensen (1999) and Clarides, Lach and Tybout (1998) firms deciding to export for the first time improve their performance in the periods leading up to entry (relative to non-export firms). Whether this is a consequence of the decision to enter export markets at some future point, or the decision to export is a consequence of this productivity improvement is not evident. Firms could, for example, improve their productivity in the periods leading up to entry in anticipation of tougher competition in international markets than domestically. Nevertheless, as found in Table 8.5 ex-ante productivity improvements have been found to be an important determinant of export market entry across studies for a large number of countries. Comparing these results to the matched sample it is clear that matching has two principal effects. First, the estimated TFP growth premium accorded to new entrants relative to the unmatched sample is in every period lower. For example, in the entry period the growth premium to new exporters is 4.4 per cent in the unmatched sample and 3.6 per cent in the matched sample. The second effect is to remove the significance of the premium variable for all except the entry period, although the entry t  1 effect is reasonably close to significance. Bernard and Jensen (1999) argue against a causal effect from exporting to productivity on the basis of the type of pre-entry effects found in Table 8.8. The results from the matched sample however are consistent with a causal effect from export market entry, albeit one that is relatively small and short-lived. Comparing non-export firms with similar

152 David Greenaway and Richard Kneller

productivity trajectories to new entrants in the period leading up to the decision of one group to enter export markets shows no difference in their performance in the pre-entry period and some difference in the post-entry period. Firms that were on similar growth paths diverge in their performance only when one group becomes exporters. To answer the first question, matching has both an important effect on the results achieved and the conclusions that can be drawn. Although we do not report the results, it is also worth noting that a positive effect of entry can be found for a range of different indicators of firm performance in the matched sample. Finally, the co-presence of exporters in the same region and industry was found to increase the probability of entry. As argued above, this is consistent with demonstration effects between potential entrants and existing exporters. To investigate whether similar information spillovers between existing exporting firms and new entrants maximises productivity benefits after entry, Greenaway and Kneller (2004b) interact the entry variables with industry and regional agglomerations. This does not appear to have any statistically significant impact on the effect of entry.

8.5 Conclusions What are the policy conclusions that might be drawn from the evidence based on exporting at the firm level? First, not all firms have the ‘right’ characteristics to become exporters. Entry into export markets is costly and not all firms within an industry are large enough or productive enough to meet these costs. Firms are therefore likely to self-select into any export-promotion policies. Second, there are a pool of firms that do not export, but which have similar characteristics to firms that do. Whether intervention to encourage the entry of these potential exporters is worthwhile depends upon the explanation of why, given their positive underlying characteristics, these firms choose not to export. The existing evidence base provides no unequivocal answer to this question, but does suggest some interpretations. If the explanation lies in unfavourable unobservable characteristics such as low managerial ability or product attributes then policy intervention to encourage entry may be seen as a waste of resources. If the failure to become exporters is due to information asymmetries, for example because the costs of entry are perceived to be too high or too uncertain or the profitable opportunities perceived to be too low or uncertain then intervention may be of some benefit. The evidence found in favour of spillover effects from the co-location of export firms (either generally or

Exporting and Productivity Growth 153

from multinationals) may be used to suggest that information asymmetries form at least part of the problem. Finally, with reference to the entry effects found in studies of UK exporters. Evidence of post export market entry effects might be thought of as the area in which the case for policy intervention is most persuasive. As discussed earlier, the existing empirical evidence does not lead us to this conclusion. Firms self-select into export markets. It is therefore likely that the evidence of entry effects found in the data represent the effects of export market entry on that group of self-selecting firms (the effect of treatment on the treated) and not what the effect of entry might be on other non-exporting firms (the effect of treatment on the population). Further research into the question of post-entry improvements in firm performance offers potentially high returns. There are, for example, methodologies that might be applied that allow an understanding of the effect of export market entry on the population of firms. Given that the current Government is interested in raising average rates of productivity growth in the economy the question might be raised as to how much does exporting contribute to this. We might think of two effects here. First, the evidence of a positive effect on productivity growth from export market entry, which would appear to be relatively small and short-lived. A second effect is from the reallocation of resources from low productivity firms to firms with higher productivity. Unfortunately it is not possible to assess the contribution of exporters to this for the United Kingdom owing to data limitations but Disney et al. (2003) show that reallocation of the latter type is the major source of productivity growth in the UK manufacturing industry.8

Appendix: data sources The United Kingdom is a relatively large industrialised economy and an important exporter of manufactures; in fact the fifth largest globally. Unfortunately its production census (the Annual Respondents Database) does not collect information on firms’ exporting activity. However, two other firms’ level surveys do, namely OneSource and FAME.9 Our sample frame does not encompass the full dataset from either source however, for three reasons. First, we are only interested in manufactures since, in general, export data on service providers is not available. Second, we exclude foreign-owned companies since they will have different motives for exporting than indigenous firms and face different costs. Third we exclude firms for which there is incomplete information on output and factor inputs. Our final data set therefore contains comprehensive information on 11,225 firms for the period 1989–2002 yielding a total of 78,606 observations. On average there are six years of data on each firm.

154 David Greenaway and Richard Kneller

Notes The authors acknowledge helpful feedback from an Editorial Seminar at Oxford and detailed comments from Christopher Allsop. Financial support from The Leverhulme Trust under Programme Grant F114/BF is also gratefully acknowledged. This paper was first published in the Oxford Review of Economic Policy, Vol. 20, 2004, pp. 358–71. The permission of the Editors of the journal to reprint the paper in this volume is also acknowledged. 1. www.uktradeinvest.gov.uk 2. We briefly outline the data sources used in the Appendix, although further detail can be found in Girma et al. (2004) and Greenaway and Kneller (2004). 3. The TFP variable is significant when the wage variable is removed from the regression. This would suggest that differences in TFP within 2-digit industries are less important for explaining the export status than across regions. 4. When the wage variable is removed from the regression used to generate the results in Table 8.2 is also identical to that found in Table 8.5, the marginal effect of TFP is 0.011 and statistically significant. 5. The regression used to generate the results reported in Table 8.5 differs in that both average wages and TFP are included and regions are measured at NUTS level-3 (not NUTS level-3 as in Table 8.5). 6. For a comprehensive review on the microeconometric evaluation literature see Blundell and Costa Dias (2000). 7. This approach modifies the presentation but not the conclusions for the matched sample compared to those found in Girma et al. (2004) and Greenaway and Kneller (2004). 8. See also Bernard and Jensen (2004) (the importance of reallocation) for an analysis of US firms. 9. Both Onesource and FAME are non-stratified samples with an oversampling of large firms. This in part provides a motivation for the matching analysis of later sections, although throughout we consider the robustness of the results to the use of a sample of small firms (employment less than 50).

References Aitken, B., G.H. Hanson and A.E. Harrison (1997), ‘Spillovers, Foreign Investment, and Export Behavior’, Journal of International Economics, 43, 103–32. Bernard, A., J. Eaton, B. Jensen and S. Kortum (2003), ‘Plants and Productivity in International Trade’, American Economic Review, 93(4), 1268–90. Bernard, A. and J.B. Jensen (1995), ‘Exporters, Jobs and Wages in US Manufacturing: 1976–1987’, Brookings Papers on Economic Activity, Microeconomics, 67–119. Bernard, A. and J.B. Jensen (1999), ‘Exceptional Exporters Performance: Cause, Effect or Both?’, Journal of International Economics, 47, 1–25. Bernard, A. and J. Jensen (2004), ‘Why Some Firms Export’, Review of Economics and Statistics, 86, 561–69. Blundell, R. and M. Costa Dias (2000), ‘Evaluation Methods for NonExperimental Data’, Fiscal Studies, 21, 427–68. Bugamelli, M. and L. Infante (2002), ‘Sunk Costs to Exports’, Bank of Italy Research Papers.

Exporting and Productivity Growth 155 Castellani, D. (2002), ‘Export Behaviour and Productivity Growth: Evidence from Italian Manufacturing Firms’, Weltwirtschaftliches Archiv, 138, 605–28. Clerides, S., S. Lach and J. Tybout (1998), ‘Is Learning by Exporting Important? Micro-Dynamic Evidence from Columbia, Mexico and Morocco’, Quarterly Journal of Economics, 113, 903–48. Disney, R., J. Haskel and Y. Heden (2003), ‘Restructuring and Productivity Growth in UK Manufacturing’, Economic Journal, 113, July, 666–94. Duranton, G. and H. Overmans (2002), ‘Testing for Location Using MicroGeographic Data’, CEPR Discussion Paper 3379. Edwards, S. (1998), ‘Openness, Productivity and Growth: What Do We Really Know?’, Economic Journal, 108, 383–98. Girma, S., D. Greenaway and R. Kneller (2004), ‘Does Exporting Increase Productivity? A Microeconometric Analysis of Matched Firms’, Review of International Economics, 12, 855–66. Greenaway, D., J. Gullstrand and R. Kneller (2003), ‘Exporting May Not Always Boost Firm Level Productivity’, GEP Research Paper 03/26, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham. Greenaway, D. and R. Kneller (2004a), Participation in Export Markets and Productivity in UK Manufacturing, HMSO. Greenaway, D. and R. Kneller (2004b), ‘Exporting, Productivity and Agglomeration: A Matched Difference in Difference Analysis of Matched Firms’, GEP Research Paper 03/45, Leverhulme Centre for Research on Globalisation and Economic Policy, University of Nottingham. Greenaway, D., N. Sousa and K. Wakelin (2004), ‘Do Domestic Firms Learn to Export from Multinationals?’, European Journal of Political Economy, 20, 1027–44. Head, K., and J. Ries (2003), ‘Heterogeneity and the Foreign Direct Investment Versus Exports Decision of Japanese Manufacturers’, Journal of the Japanese and International Economies, 17, 448–67. Helpman, E., M. Melitz and S. Yeaple (2004), ‘Export Versus FDI’, American Economic Review, 94(1), 300–16. Melitz, M. (2003), ‘The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity’, Econometrica, 71, 1695–725. Roberts, M. and J.R. Tybout (1997), ‘The Decision to Export in Colombia: An Empirical Model of Entry with Sunk Costs’, American Economic Review, 87, 545–64. Rodrik, D. and F. Rodriguez (2000), ‘Trade Policy and Economic Growth: A Skeptic’s Guide to the Cross-National Evidence’, Macroeconomics Annual 2000, eds, B. Bernanke and K. S. Rogoff, MIT Press for NBER, Cambridge, MA. Rubin, D. B. (2001), ‘Using Propensity Scores to Help Design Observational Studies: Application to the Tobacco Litigation’, Health Sciences and Outcomes Research methodology, 2, 169–88. Wagner, J. (2002), ‘The Causal Effects of Exports on Firm Size and Productivity: First Evidence From a Matching Approach’, Economics Letters, 77, 287–92.

9 Multinationals, Exporting and Overseas Production Sourafel Girma, Richard Kneller and Mauro Pisu

9.1 Introduction Domestic firms serve overseas markets either by exporting their products or by establishing foreign production facilities. The rising integration of the world economy in the last half-century and as in previous periods of integration is characterised by the simultaneous increase in both arms-length trade (exporting) and foreign direct investment (FDI). Table 9.1, using data from Obstfeld and Taylor (2003) and Maddison (2001), reports the value of world merchandise exports and the stock of foreign assets (both as a percentage of world gross domestic product (GDP)) for various time periods from 1820 to 1998. Increased integration through both trade and FDI is evident up until the inter-war period. Trade increased from 1 per cent of world output in 1820 to 4.6 per cent in 1870 to 9 per cent in 1929 and foreign assets from 6.9 per cent in 1870 to 17.5 per cent in 1913. The protectionist trade stance and capital controls of many countries during the inter-war years led to an end to this process, although the greater collapse was in foreign ownership. From the end of the Second World War the progressive liberalisation of trade, and following the collapse of the Bretton Woods international monetary system, capital controls have seen a rebound to well above pre-war highs. By the end of the sample period the ratio of exports and the stock of foreign assets to world GDP was 17.2 per cent (1998 figure) and 56.8 per cent respectively (1995 figure). In response to this rising trend in both exports and FDI a research question has arisen of whether they are related; exports and FDI are complementary production choices and an increase in one leads to an increase in the other, or whether they substitute for each other, and if so what leads firms to make these alternate choices. This chapter reviews 156

Multinationals, Exporting and Overseas Production 157

the theoretical and empirical literature on this question and provides for the first time, empirical evidence on the export behaviour of UK multinational firms. From the empirical literature on location choice two stylised facts have emerged. First, the same firm often does both, it is both an exporter and a multinational.1 Second, there are many industries in which FDI and exports are bi-directional.2 Countries both import and export the same products and there are home and foreign multinationals within that industry. Until recent theoretical developments the first of these was somewhat easier to explain than the second. In the early theoretical literature on location choice (see, e.g. Brainard, 1993) exporting and FDI were seen as substitutes, and therefore unable to explain either of the two stylised facts above. The decision whether to export or to undertake FDI was determined by the balance between proximity and concentration. The firm chose to export when there were cost advantages of concentration (economies of scale), and chose foreign production when proximity to local markets was more important. The firm therefore chose either exports or FDI not both, and given that costs were identical for all firms in the industry none were likely to differ in this choice. Removing the assumptions of single-product firms and fixed demand enriches the predictions from the model (Lipsey and Weiss, 1984) sufficiently to explain the first stylised fact; firms export and undertake FDI contemporaneously. Adding the production of multiple products allows firms to make alternative location choices across their product range. Given the relative costs of proximity and concentration the firm could feasibly choose to become an exporter for some products and a multinational in others. Exports and FDI become positively correlated if there are horizontal or vertical complementarities across these products. For example, in the case of horizontal complementarities, increased demand for the good supplied by foreign production may lead to increased demand for all goods produced by that firm, some of which may be supplied through arms-length trade. In a special case the model extended to multiple products and cross-product complementarities is also capable of explaining the coexistence of exports and FDI within the same industry. In the unusual case where the costs of proximity are exactly equal to those of concentration, firms are indifferent how they serve foreign markets and a proportion become exporters and a proportion multinationals. Unsatisfied with this explanation of the co-existence of exporters and multinationals the theoretical literature has developed along three

158 Sourafel Girma, Richard Kneller and Mauro Pisu

lines: strategic choice, firm level heterogeneity and what has been labelled the complex integration strategies of multinationals (viz. strategies that are a hybrid of vertical and horizontal integration). In models of strategic choice the single product firms’ decisions on how to serve markets is determined with reference to the choices made by their competitors using a game-theoretic framework. For example, in some versions of these models firms establish foreign production facilities and export in order to restrict competition within the industry and raise profits (Choi and Davidson, 2003). The heterogeneous firm model (e.g. Helpman et al., 2004) builds on the idea that exports and FDI are substitutes at the firm level, but allows firms within the same industry to make alternative choices based on their different productivity. Only the most productive firms within an industry find it profitable to meet the higher costs associated with FDI; the next set find it profitable to serve foreign markets through exporting; while the least productive firms find it profitable to serve only the domestic market. In theories trying to explain complex integration strategies of multinationals, the co-existence of exports and FDI can arise in different model frameworks. At their simplest they can be seen as an extension of the proximity-concentration models with cross-product complementarity to a setting in which there are multiple locations (countries), multiple stages of production and multiple products. In equilibrium, multinationals may find it profitable to undertake horizontal FDI in some countries and vertical FDI in others, involving exporting intermediates and/or the final product. The addition of more countries also allows for some to be used as export platforms. At the industry level the location decision can be made even more complex if firms differ in their productivity. In what remains of the chapter we develop these points more formally. Section 9.2 develops a simple model of location choice based on the proximity-concentration trade-off. The development of this adds greater complexity to the outcomes and is detailed in Section 9.3. These models allow complementarity between exports and FDI. Section 9.4 summarises the available empirical evidence. As will be seen, because data is generally available only at the level of the firm, evidence of complementarity is stronger than that of substitution. Section 9.5 then provides empirical evidence for the United Kingdom while Section 9.6 concludes.

9.2 Exporting and FDI as substitutes 9.2.1

Proximity and concentration

To understand the development of the literature on location of production we begin with a simplification of the model of proximity versus

Multinationals, Exporting and Overseas Production 159

concentration in Brainard (1997). In this single firm, single product setting export and FDI are substitutes, the correlation between them is therefore negative. We assume there are two countries, home and foreign (labelled H and F), each of which are potential markets for the sale of the good and also potential locations for the single firm (and its single product line) to establish its production facilities. The two countries can differ in their relative size such that there are Q H potential consumers in the home market and Q F in the foreign market. Production in the home market incurs no fixed costs but a variable cost that depends (solely) on labour input, which for simplicity is assumed to equal the wage in the economy, wH. The total cost of production in the home country is therefore equal to variable costs multiplied by the amount of the product sold, Q H. Foreign markets can be served through one of two means. The firm can conduct arms-length trade (it can export) or it can establish production facilities in the foreign territory (it can become multinational). In addition to the costs of home production, exporting incurs two types of costs, fixed and variable. The fixed costs of exporting, FX, are typically thought to include the costs of research into product compliance, distribution networks, advertising, and so on. Variable costs, TX, are assumed to increase with physical distance between countries. These variable costs include the costs associated with geography, such as transport costs, as well those imposed by policy, for example, tariffs. The structure of costs incurred from foreign production is similar to that from home production, that is, it depends on the wage in the foreign country, to which we add fixed costs, FF. These represent the difference between the sunk-costs of establishing a home production facility versus establishing a foreign production facility. These costs might be thought to include the duplication of domestic costs as well as the building of new production facilities or acquisition of an existing firm. They are also assumed to represent any economies of scale in production. The potential for economies of scale increases the difference in the cost of establishing foreign rather than domestic production facilities. Total costs associated with serving the home market and foreign markets either through exporting, CX, or foreign production, CM, are given by the following expressions. CX  wHQ H  FX  (wH  TX)Q F

(1)

CM  wHQ H  FF  wFQ F

(2)

Exporting relative to serving the home market incurs additional fixed costs FX and additional variable costs TX, whereas foreign production

160 Sourafel Girma, Richard Kneller and Mauro Pisu

relative to home production incurs additional fixed costs FF and variable costs differ as home and foreign wages differ. The firm’s decision will depend upon the balance of costs expressed in Equations 1 and 2. CX  CM  [wHQH  FX  (wH  TX)QF]  [wHQ H  FF  wFQ F]

(3)

Exporting will become preferable when Equation 3 is negative; becoming multinational when Equation 3 is positive; and firms are indifferent when they are equal. Rearranging Equation (3) we can show that exporting is preferred to foreign production when, FX  FF  wH  wF  TX 0 QF

(4)

Equation (4) is less likely to be negative, as the size of the foreign market, Q F, increases. Foreign production also becomes more likely as the costs of exporting increase, this effect holds for both the fixed and variable costs of exporting. Together these are known as the advantages of proximity on location choice. Foreign production becomes less likely in contrast, as the costs of FDI versus exporting increase. In this model as FF increases the fixed cost of setting up a foreign production facility grows and it is better for the firms to concentrate production in a single location and export. The empirical predictions of this model are clear, there is substitution between export and FDI at firm level. In practice, finding data sufficient to test its predictions is difficult, at the very least because firms, especially multinational firms, produce many different products.

9.3 Exporting and FDI as complements 9.3.1

Complementarity within the firm

To the extent that there is cross-industry variation in the costs of exporting versus establishing foreign plants and the size of markets, it follows that if the firm produces a number of different products then it might make different choices about where to locate production across this product range. Removing the assumption of fixed demand (Lipsey and Weiss, 1984) to allow complementarity across the products produced by the firm then generates positive co-movement in exports and FDI. This effect is sometimes labelled a proximity advantage (Brainard, 1993, 1997) or demand complementarities. Two sorts of complementarity

Multinationals, Exporting and Overseas Production 161

have generally been considered. First, across products that are vertically integrated in the production process (vertical complementarity); and second, products that have no such vertical integration (which we label horizontal complementarity). Under horizontal complementarity the increased demand for one product will positively effect the demand for other products produced by the firm. Lipsey and Weiss (1984) and Rugman (1990) develop this point more formally, arguing that an increase in demand might occur for the brand as a whole as a result of improved provision of after-sales services and more efficient distribution. Vertical complementarity will occur if, as the name suggests, one good produced by the firm is used as an intermediate input in the product of another. 9.3.2

Complementarity within the industry

Given that the parameters of the model might reasonably be expected to be the same for all firms within a single industry why some firms choose to become exporters and some multinationals is unclear. All firms should either choose to become exporters or multinationals. The coexistence of exports and FDI then depends on the special case where the cost of overseas production is exactly matched by the costs of serving foreign markets through arms-length trade. In such an outcome a proportion of firms,
, within the industry will choose to become exporters and the remainder (1 
) multinationals. A dissatisfaction with the knife-edge nature of this result has led to the development of three alternatives. We discuss briefly the literature on strategic choice below before providing more detail on the recent developments of the heterogeneous firm model and complex integration strategies. 9.3.3

Strategic choice

This literature explains complementarity between exporting and FDI within an industry by allowing the single product firm to both export and establish foreign production facilities for strategic reasons. There are two main lines of research. The first considers the firms’ location choice in the face of uncertainty, either in demand or the costs of establishing production facilities. The second, adds an element of strategy; the firms location decision is dependent on the location decisions of its rivals. We begin with the models of uncertainty. Rob and Vettas (2003) generate complementarity between exports and FDI using uncertainty in foreign market demand. The firm uses a combination of exporting and FDI to hedge against this uncertain demand when investment is irreversible. The firm will establish a

162 Sourafel Girma, Richard Kneller and Mauro Pisu

foreign production plant to meet proven demand and use exports to satisfy the uncertain elements of demand. Choi and Davidson (2003) argue that such a model is difficult to justify when markets are mature. Uncertainty again plays a role in Kogut and Kulatilaka (1994), although this time it is in relation to the costs of setting up a plant. The firm builds two plants, one in each country. If costs in one are then higher than expected the firm moves production to the low cost plant and exports to the foreign country. Complementarity of a different sort is generated in the models of Head et al. (2002) and Choi and Davidson (2003). In Choi and Davidson (2003) the nature of competition generates complementarity. Firms choose to establish foreign production facilities because this raises marginal costs. This has the effect of reducing competition in the market helping to raise profits for both firms. A game-theoretic approach is also used by Head et al. (2002). They show in a model of oligopolisitic competition that uncertainty about costs in the foreign market will lead a sufficiently risk averse oligolpolist to establish production facilities abroad once its rivals have made such an investment. 9.3.4

Firm level heterogeneity

Heterogeneous firm level models assume industries populated by firms with different productivity levels. The complementarity between exporting and FDI at the industry level derives from these productivity differences even when exports and FDI are substitutes at the level of the firm. Helpman et al. (2004), building on Melitz (2003), formulate a model in which the most productive firms will become multinational; firms whose productivity falls within an intermediate range export, whereas the least productive companies will sell in the domestic market only. Using the simplification of this model in Head and Ries (2003) we assume constant marginal costs equal to, wH/Ai and wF/Ai – depending on the market being served. The term Ai represents the productivity level specific to firm i. Adding these terms to the model outlined in Section 9.2 and re-arranging gives us a slight variation on Equation 4. CX  CM 

FX  FF wH  wF   TX 0 Ai QF

(5)

Exports and FDI are substitutes (at the firm level) as before, but now the likelihood that the firm will become an exporter is decreasing as the firm’s productivity increases.3 There is self-selection into exports and FDI. We summarise these relationships in Figure 9.1, where we plot productivity against its probability density function. For the productivity

Multinationals, Exporting and Overseas Production 163

Probability density

Domestic only

aD

Export

aX

FDI

aM

Figure 9.1 Probability density of productivity and firm choice

probability density, Figure 9.1 suggests that firms with a productivity level to the right of each critical level of productivity (aD, aX , or aM) will participate differently in the international market. Producers with a productivity level between aD and aX will produce for the domestic market only; while firms with a productivity draw between aX and aM will find it profitable to sell in the domestic market and in the foreign one through exporting (since they will find it profitable to pay the transportation cost involved in shipping the goods to another market). The most productive firms will sell goods in the foreign market by producing therein (i.e. given the sunk cost of setting up production facilities they will find it profitable to pay such costs instead of the transport costs).4 9.3.5 The complex integration strategies of multinationals As emphasised by the World Investment Report (1998) the integration strategies of multinational firms are more complex than those discussed above, and increasingly so over time. From a modelling perspective greater complexity can be generated by adding more countries, product lines and stages of production. For example, adding a third country generates complementarity between exports and FDI by allowing one country to be used as an export platform. If the third country is located close to one of the other countries but is small in size, then given the

164 Sourafel Girma, Richard Kneller and Mauro Pisu

right parameter values the firm may choose to locate foreign production in the larger country and export to its smaller neighbour. In a model with two countries and two stages of production (e.g. manufacturing and headquarter services) the integration strategies of MNEs have been broadly classified as vertical and horizontal (e.g. Markusen 2002). Vertical FDI occurs when different stages of the production process occur in more than one country; and horizontal FDI when the same stage of production occurs in more than one country. When there are more than two countries and more than two stages of production, multinationals are likely to undertake more complex choices that involves intra-firm trade and export platform FDI.5 The issue has been analysed in different settings by Yeaple (2003), Ekholm et al. (2003) and Grossman et al. (2003). Yeaple (2003) employs the new trade theory framework of Helpman and Krugman (1985), a representative firm operating in monopolistic competition. Ekholm et al. (2003) uses a strategic choice model, a duopoly, whereas Grossman et al. (2003) use the heterogeneous firms model introduced by Melitz (2003). The main common characteristics of these models are that there are three economies, two are identical and located in the North and one different and located in the South, and increasing returns to scale in one sector. Firms are constrained to provide headquarter services from their home country in the North but can choose where to produce intermediates and assemble the final product. The predictions each model generates are robust to different competitive settings. All three models show that in equilibrium firms may undertake FDI strategies that are neither purely horizontal nor vertical, and which, in many instances, involve the export of intermediates or the final product. Yeaple (2003) considers two industries, one producing a homogeneous good sold in a perfectly competitive market and the other producing a differentiated good sold in a monopolistically competitive market. The latter is consumed only in the North. The final good requires two intermediate inputs, one of which is cheaper in the North (skill) and the other in the South (labour). The profitability of the four integration strategies are compared: (1) no FDI: companies concentrate all production in the country of origin (and therefore may serve the other northern markets through export); (2) vertical FDI: one component is produced in the South and the other in the (northern) home country; (3) horizontal FDI: firms conduct the entire production in both northern countries; (4) complex FDI: one component is produced in the South and the other in both countries in the North (a hybrid horizontal and vertical case).

Multinationals, Exporting and Overseas Production 165

Export platform FDI is explicitly analysed by Ekholm et al. (2003). In a duopoly model with symmetric trade costs and a small cost advantage for the South the model again predicts four types of FDI. Lowering these two kind of costs leads to the following series of integration strategies: (1) no FDI: firms produce in their home country only and export the good in the other northern country; (2) horizontal FDI: each firm headquartered in the North sets up a subsidiary in the other northern country to serve this market; (3) purely export-platform FDI: each firm produces in the home country the final good to be sold therein; the final good to be sold in the other northern country is produced in the South; (4) purely export-platform FDI and vertical FDI (hybrid MNE): each firm manages a plant in the South to export the good in both markets in the North. This last alternative gives rise to the hybrid MNE since vertical FDI is undertaken toward the home country and purely exports platform FDI towards the other markets. Grossman et al. (2003) tackle the same problem but follow Yeaple (2003) in allowing firms to differ in productivity levels (as in the heterogeneous firm model described above). Their analysis allows for the coexistence in the same sector of a rich array of profitable FDI strategies. In brief, this chapter validates the fact that the least productive firms will not undertake FDI. More productive firms will become different types of MNEs depending on their exact productivity level and other parameters of the model. The strategies are in general complex involving complementarity between FDI and export.

9.4 Empirical evidence Owing to data limitations separately testing the within firm theoretical predictions and within industry predictions has proved difficult. Instead the literature has concentrated on net-complementarity and net-substitutability between exports and FDI at various levels of aggregation. Empirical evidence of complementarity and substitution between exports and FDI has depended to a large extent on the level of aggregation used. As perhaps suggested from Tables 9.1a and 9.1b, at the aggregate level strong evidence of complementarity exists. Lipsey and Weiss (1984) for example, estimate the substitution and complementarity effects between exports and FDI using a regression in which export sales are a function of production and sales in the foreign country. Data are for bilateral trade to the United States for 14 industries in 1970. The authors find strong evidence of complementarity. Exports tend to

166 Sourafel Girma, Richard Kneller and Mauro Pisu Table 9.1a World merchandise exports/GDP (per cent) 1820 1870 1913 1929

1.0 4.6 7.9 9.0

1950 1973 1998

5.5 10.5 17.2

Source: Maddison (2001).

Table 9.1b Foreign assets/world GDP (per cent) 1870 1913 1930

6.9 17.5 8.4

1960 1980 1995

6.4 17.7 56.8

Source: Obstfeld and Taylor (2003).

rise with sales by foreign affiliates. These results hold strongly for exports of intermediate products (vertical complementarity) but also for exports of finished products. Using a similar approach within an extended data set (in terms of country and time coverage) Clausing (2000) produces similar results, where exports are positively correlated with sales of overseas affiliate firms. Lipsey and Weiss (1984) and Graham (2000) also find net complementarity again using aggregate data. Amiti and Wakelin (2003) focus on the trade FDI relationship from a somewhat different perspective. They test whether reductions in investment costs stimulates different types of FDI, which in turn impacts on trade. Estimating export gravity equations for bilateral trade flows between 36 countries they find that relative country characteristics are important. When countries are different (they have different factor endowments) and trade costs are low investment liberalisation stimulates exports. Whereas if countries are similar and trade costs are high investment liberalisation reduces the flow of exports. Complementarity or substitutability therefore depends importantly on cross-country differences. From the use of more disaggregated data a separate effect of substitution between exports and FDI is found. Several studies relate to disaggregated industry level data and one to firm level evidence. Of previous studies on location choice, Blonigen (2001) uses perhaps the most disaggregated data, that of product level data for trade between the United States and Japan. The most interesting of the regressions estimated by Blonigen

Multinationals, Exporting and Overseas Production 167

(2001) relate to the estimation of import demand equations for Japanese automobile products into the United States as a function of auto part production in the United States (substitution effect) and separate terms covering Japanese and US automobile production (vertical complementarity). The results display clear substitution effects in nine of the ten product lines examined and vertical complementarity effects from Japanese automobiles in nine product lines. The effect of US automobiles was less clear, being significant and positive in three product lines, negative in one and insignificant in the remainder. Swenson (1999) takes a similar approach to Blonigen (2001) albeit for a much larger number of product lines. Estimating import demand equations Swenson finds evidence of a substitution effect from foreign investment at the product (3-digit) and industry level (2-digit) but complementarity at the overall manufacturing level. The negative substitution effect at the industry level might be thought of as contradicting results for the same variable found in Blonigen (2001). Finally, of the industry level studies Brainard (1997) is the only one that attempts to directly measure the determinants of the proximityconcentration trade-off. Data are available for 1989 on bilateral exports and imports along with sales of multinational firms (both domestic and foreign). Brainard finds that the share of affiliate sales is increasing in the size of trade barriers, transport costs and corporate scale economies and decreasing in barriers to FDI and production scale economies. Head and Ries (2003) take a different approach, estimating an export demand equation for the firm. They find that the effect of overseas production on firm exports is positive no matter whether the overseas facilities are used for production or sales. Interacting this term with the degree of vertical integration of the firm they find a positive effect; the complementarity effect is positive in the most vertically integrated firms whereas evidence of substitution can be found for the least integrated firms (it is positive average). There are no direct tests within the literature on the more complex models of location choice outlined in Section 9.3. There is however indirect evidence on the heterogeneous firm model. A large body of evidence exists on productivity differences across exporters, non-exporters and multinational firms. The higher productivity level of exporters versus non-exporters is an established stylised fact within the literature that has been found to hold across countries and across time. (See among others Bernard and Wagner (1997), Clerides et al. (1998) Bernard and Jensen (1999), Kray (1999), Castellani (2002), Girma et al. (2004). Similarly others have investigated whether multinational firms (usually

168 Sourafel Girma, Richard Kneller and Mauro Pisu

foreign multinationals) have higher mean productivity than domestically owned firms (e.g. Girma et al., 2001; Conyon et al., 2002). Finally, evidence has also been found to establish that multinational firms have higher productivity than exporters (Girma et al., 2005; Head and Ries, 2003; Girma et al., 2004).

9.5 Empirical methodology, data and main findings In this section we present empirical evidence on the relationship between overseas production and the exporting decision of UK firms. We adopt a similar approach to Head and Ries (2003) and estimate firm level export demand equations that include indicators of multinationality as an explanatory variable. We use disaggregated data in order to maximise the possibility of finding evidence of substitution. We begin however by outlining the empirical methodology and the data. 9.5.1

Empirical methodology

In order to isolate the impact of multinationality (MULT) on the export performance of UK multinationals the following equation is estimated EXPit  
Xit  f(MULT)  Di  Dt  Dsic  it

(9)

In the above equation i and t index firms and time periods respectively; the dependent variable is log of exports, X is a vector of factors that are hypothesised to impact on firms’ export performances. It consists of TFP, the ratio of intangible assets to fixed assets and firm size, proxied by log of employment. This study estimates firm level TFP using the semiparametric approach of Levinshon and Petrin (2003). The major advantage of this approach over more traditional production function estimation techniques is its ability to control more effectively for the correlation between unobservable productivity shocks and inputs.6 The inclusion of TFP in the export equation is motivated by the robust finding in the literature that exporting and productivity are highly correlated (e.g. Bernard and Jensen, 1999; Girma et al., 2004). In the same literature firm size is also found to have significant effect on exporting, presumably because large firms are able to exploit economies of scale. The intangible assets variable is an estimate of the firms’ investments in R&D and marketing, and the value of patents, copyrights and goodwill. Braunerhjelm (1996) argues that intangible assets more closely correspond

Multinationals, Exporting and Overseas Production 169

to the theoretical notion of ‘firm specific assets’, and firms with a higher level of intangible assets tend to export more. Following Head and Ries (2003) the variables in the X vector are lagged by one period to mitigate potential problems of endogeneity. In the above model Dt, Di, Dsic2 represent time dummies, firm level fixed effects and 2-digit industry dummies respectively. Time dummies account for aggregate macro shocks and business cycles effects common to all firms, while fixed effects control for firm level heterogeneity that may be correlated with the determinants of exporting. It is assumed that the random error is distributed independently of the explanatory variables. In Equation (2), MULT is a measure of the degree of multinationality measured by the number of countries the UK MNE has foreign subsidiaries in. From the discussion in the previous section we infer that this variable may have a positive effect on the export share of multinationals. Indeed, according to the theories concerning the complex integration strategies of multinationals, the higher the number of production stages and the higher the number of countries where these stages can be relocated the more likely firms will undertake FDI strategies that are not purely horizontal nor purely vertical. This will involve exports of intermediates and/or final products, thus implying complementarity between FDI and exports. We estimate two versions of the model in which we postulate that the relationship between the log level of productivity and the degree of multinationality (MULT) is either linear or quadratic. In the quadratic model we have f(MULT)  d1MULT  d2MULT2

(10)

where the ds are parameters to be estimated. Setting d2  0 gives the linear model, which implies that exports either increase or decrease with the degree of multinationality monotonically. The quadratic specification is more flexible in that it allows for the rate at which exports increase to vary with the degree of multinationality, as measured by the geographic dispersion of the firm. For example with d1 0 and d2 0, the initially positive impact of multinational diversification on exports will start to diminish once the degree of multinationality gets past the critical level   (d1/2d2), assuming that there are degrees of multinationality greater than . The asymptotic variance of this turning point can be constructed via the ‘delta’ method, given consistent estimates of

170 Sourafel Girma, Richard Kneller and Mauro Pisu

d1 and d2 (dˆ1 and dˆ2, say), and this can be expressed as ˆ1)  4 ˆcov(d ˆ1,d ˆ2)  4 ˆ2var(d ˆ2)] ˆ)  1 [var(d Var( ˆ22 4d

(11)

The parameters of the model are estimated using the Hausman–Taylor (1981) panel instrumental variables fixed effects estimator. This method uses deviations of time-varying variables from their means as instruments for explanatory variables that may be correlated with individual effects, and is more efficient than the standard fixed effects estimator because it exploits more variation in the data to identify the parameters. 9.5.2

Data

The primary source of information for this study is the OneSource database of private and public companies, which is derived from the accounts that companies are legally required to deposit at Companies House. All public limited companies, all companies with employees greater than 50, and the top companies based on turnover, net worth, total assets or shareholders funds (whichever is largest) up to a maximum of 110,000 companies are included in the database. The database provides information on employment, physical capital and output in a consistent way both across firms and across time.7 It is also one of the very few databases with firm level export data. However OneSource does not provide any information on the multinational activity of UK-owned firms either. Fortunately enough we were able to merge it with a newly created database of foreign multinational activity: the European Linkages and International Ownership Structure (ELIOS) database built at the University of Urbino.8 In the final analysis we identified 409 UK manufacturing multinationals (and 3252 firm-year observations) for which the necessary information to estimate our econometric model is available. Table 9.2 reports summary statistics for the main variables of interest. The average UK multinational in the sample operates in about five countries and four industries, although there is substantial heterogeneity in the sample. The average export intensity is quite high, standing at more than 84 per cent. 9.5.3

Main findings

Our results are reported in Table 9.3. The estimated coefficients on TFP and firm size are positive in all specifications, consistent with findings reported in the literature. By contrast the effect of intangible assets is found to be negligible. This may be because differences in TFP and firm

171 Table 9.2 Summary statistics

Number of five-digit industries firm operates Export intensity Employment Real sales (£’000) Intangible/fixes assets Log of total factor productivity Number of countries firm operates in

Mean

Standard deviation

3.51

1.76

0.844 5,194.7 46,5429.3 0.199 0.070 4.62

0.313 20,048.9 1,981,385 1.378 0.492 6.53

Note: The number of firms is 409 and the above figures are based on 3252 firm-year observations.

Table 9.3 The determinants of MNEs’ exports (UK) All subsidiaries Linear model (1) TFP Intangible/ fixed assets Size Number of countries MNE operates

Linear model (3)

Quadratic model (4)

Manufacturing subsidiaries Linear model (5)

Quadratic model (6)

0.229 (3.43)** 0.016

0.229 (3.43)** 0.014

0.228 (3.42)** 0.012

0.201 (2.84)** 0.008

0.207 (2.58)** 0.007

0.211 (2.76)** 0.001

(0.46) 0.834 (12.50)** 0.111

(0.41) 0.843 (12.78)** 0.132

(0.34) 0.831 (12.45)** 0.172

(0.22) 0.772 (10.75)** 0.968

(0.17) 0.778 (9.58)** 1.567

(0.02) 0.805 (10.48)** 1.279

(3.86)**

(3.58)** 0.037

(3.75)**

(3.84)** 0.078

(3.76)**

(3.15)** 0.079

Number of countries squared Observations Number of firms

quadratic model (2)

Service subsidiaries

2828 390

(3.43)** 2828 390

2828 390

(3.46)** 2828 390

2828 390

(2.75)** 2828 390

Notes: (i) Results are based on the Hausman–Taylor estimator. (ii) All specifications include year dummies and two-digit industry dummies. (iii) TFP, intangible/fixed assets and size (log of employment) are all lagged by one period. (iv) Absolute value of z-statistics in parentheses. (v) significant at 10%; ** significant at 5%; *** significant at 1%.

172 Sourafel Girma, Richard Kneller and Mauro Pisu

size already explain most of the variation in intangible assets. Alternatively it may be that while the possession of firm specific intangible assets influences the decision of whether to export or not (Braunerhjelm, 1996), it does not explain the decision of how much to export, once the firm has entered the export markets. The results from the linear model (Table 9.3, column 1) show that the degree of multinationality exerts a positive influence on exports, suggesting that exporting and overseas production are net complements. Setting up subsidiaries in a new country leads to an 11 per cent increase in the total amount of production shipped abroad. This is indeed an economically important effect. The evidence from the quadratic model provides evidence of an inverted-U shape relationship between the level of exports and the degree of multinationality, as measured by the number of countries the firms operate in. Multinationality and total exports are positively related up to a threshold beyond which a further increase in the number of countries the firm operates in is associated with a declining level of exports. There is therefore net substitution when the level of multinationality is high. The results in column 2 of Table 9.3 suggest that this threshold value is around 13 countries, and the 95 per cent confidence interval of this turning point lies between 12 and 14 countries. Inspection of the raw data reveals that only 10 per cent of the firms in our sample operate in more than 13 countries. Combined with our econometric results, this would appear to suggest that for the vast majority of firms in our sample exports and overseas investments are complements at the margin. We also explored whether overseas subsidiaries are manufacturing or service firms. To this end, separate models are estimated based on two definitions of multinationality: number of countries in which foreign subsidiaries are engaged in manufacturing/non-manufacturing activities. The results from these experiments are reported in the last four columns of Table 9.3, and they indicate that the curvilinear relationship between overseas production and exports is not sensitive to the activities of the foreign subsidiaries.9 However the marginal effect of overseas production on the amount exported by the parent company starts to decline at a lower level (the 95 per cent confidence lies between 7 and 9 countries) when manufacturing subsidiaries only are considered.

9.6 Conclusions Are exports and overseas production complements or substitutes? The answer would appear to depend on the level of aggregation used. At the

Multinationals, Exporting and Overseas Production 173

most disaggregated level all of the theoretical models suggest that they are substitutes, dependent on the relative cost of proximity versus concentration. Production in a single location and exporting is more likely when there are cost advantages to concentration, and the duplication of production facilities in more than one country is more likely when there are advantages to being close to markets. Complementarity is generated as we use higher levels of aggregation. Within the firm complementarity is possible once we allow for multiple products and cross-product dependence of demand. This might occur because of a linkage through the supply chain or through greater brand awareness. Complementarity within the industry has been generated through two methods: there is a strategic link between exports and FDI, or firm heterogeneity in productivity. The predictions of the model become even richer when we allow for multiple countries, industries and stages of production. A similar pattern of substitution at low levels of aggregation (albeit with net-complementarity) and complementarity at higher levels would also summarise the empirical evidence. At the lowest levels of disaggregation, such as the product and the firm, there is net complementarity, but for certain types of firms and products some evidence of substitution is found. The evidence would appear to be similar for the United Kingdom. Detailing evidence for UK multinationals we find that when the number of markets a firm sets up foreign production facilities in increases beyond a critical point, net complementarity is reversed and further increases in multinationality leads to a decline in exports. Around 10 per cent of firms in our sample lie beyond this threshold.

Notes 1. Blomstrom et al. (1988), Bloningen (2001), Head and Ries (2003) have all found evidence on net complementarity. 2. Evidence on this have been provided by Brainard (1997) and Julius (1990) 3. We assume that FM FX and wF wH in order to generate the result that at least some firms will chose exports rather than become multinational. 4. Falvey et al. (2004) and Melitz and Ottaviano (2003) have provided extensions of the Melitz (2004) model considering asymmetric countries. Falvey et al. (2004) have scrutinised the case of two countries with different efficiency levels while Melitz and Ottaviano (2003) have considered countries of different size. The main result is that self-selection into the export market holds true; however trade liberalisation will affect the two countries differently and the export cut-off will be different in the two economies. 5. See for example Feinberg and Keane (2003), Hanson et al. (2001) 6. Estimation details are available from the authors upon request. 7. Detailed industry level price indices are used to deflate nominal values.

174 Sourafel Girma, Richard Kneller and Mauro Pisu 8. Girma wishes to express his gratitude to Davide Castellani and Antonello Zanfei for allowing him to use some information from this database. 9. Strong collinearity between these two indicators of multinationality (manufacturing and non-manufacturing activities) prevented their separate use in the same regression.

References Amiti, M. and Wakelin, K. (2003), ‘Investment Liberalisation and International Trade’, Journal of International Economics, 61, 101–26. Bernard, A.B. and C.I. Jones (1996), ‘Comparing Apples to Oranges: Productivity Convergence and Measurement Across Industries and Countries’, American Economic Review, 86, 1216–52. Bernard, A. and J.B. Jensen (1999), ‘Exceptional Exporters Performance: Cause, Effect or Both?’, Journal of International Economics, 47, 1–25. Bernard, A. and J. Jensen (2001), ‘Why Some Firms Export’, NBER Working Paper No. 8349. Bernard, A. and Wagner J. (1997), ‘Exports and Success in German Manufacturing’, Weltwirtschaftliches Archiv, 133, 134–57. Blonigen, B. (2001), ‘In Search of Substitution between Foreign Production and Exports’, Journal of International Economics, 53, 81–104. Brainard, S.L. (1993), ‘A Simple Theory of Multinational Corportations and Trade with a Trade-off between Proximity and Concentration’, NBER Working Paper No. 4269. Brainard, S.L. (1997), ‘An Empirical Assessment of the Proximity-Concentration Trade-off between Multinational Sales and Trade, American Economic Review, 87, 520–44. Braunerhjelm, P. (1996). ‘The Relation between Firm-Specific Intangibles and Exports’. Economic Letters, 53, 213–19. Castellani, D. (2002), ‘Export Behaviour and Productivity Growth: Evidence from Italian Manufacturing Firms’, Weltwirtschaftliches Archiv, 138, 605–28. Choi, J.P and C. Davidson (2003), ‘Strategic Second Sourcing by Multinationals’, International Economic Review, 45, 579–600. Clausing, K.A. (2000), ‘Does Multinational Activity Displace Trade?’, Economic Inquiry, 38, 190–205. Clerides, S., S. Lach and J. Tybout (1998), ‘Is Learning by Exporting Important? Micro Dynamic Evidence from Columbia, Mexico and Morocco, Quarterly Journal of Economics, 113, 903–28. Conyon, M.J., S. Girma, S. Thompson and P.W. Wright (2002), ‘The Productivity and Wage Effects of Foreign Acquisition in the United Kingdom’, Journal of Industrial Economics, L, 85–102. Ekholm, K., R. Forslid and J. R. Markusen (2003), ‘Export-Platform Foreign Direct Investment’, NBER Working Paper No. 9517. Falvey, R., D. Greenaway and Zhihong Yu (2004), ‘Intra-Industry Trade between Asymmetric Countries with Heterogeneous Firms’, Globalisation and Economic Policy Research Centre, University of Nottingham, Working Paper No. 04/05. Feinberg, S.E. and M.P. Keane (2003), ‘Accounting for the Growth of MNC Based Trade Using a Structural Model of U.S. MNCs’, University of Maryland, manuscript.

Multinationals, Exporting and Overseas Production 175 Girma, S., D. Greenaway, and K. Wakelin (2001), ‘Who Benefits from Foreign Direct Investment in the UK?’, Scottish Journal of Political Economy, 48, 119–33. Girma, S., H. Görg and E. Strobl (2004), ‘Exports, International Investment, and Plant Performance: Evidence from a Non-Parametric Test, Economics Letters, 83 (3), 317–24. Girma, S., D. Greenaway and R. Kneller (2004), ‘Does Exporting Lead to Better Performance? A Microecometric Analysis of Matched Firms, Review of International Economics, 12, 855–66. Girma, S., R. Kneller and M. Pisu (2005), ‘Exports versus FDI: An Empirical Test’, Review of World Economics, 141(2), 193–218. Graham, E.M. (2000), ‘The Relationships between Trade and Foreign Direct Investment in the Manufacturing Sector: Empirical Results for the United States and Japan’, in D. Encarnation (ed.), Does ownership matter? Japanese multinationals in East Asia, Oxford University Press, London. Grossman, G.M., E. Helpman and Adam Szeidl (2003), ‘Optimal Integration Strategies for the Multinational Firm’, NBER Working Paper No. 10189. Hanson, G.H., R.J. Mataloni and M. J. Slaughter (2001), ‘Expansion Strategies of US Multinationals Firms, NBER Working Paper No. 02138. Hausman, J. A. and W. E. Taylor (1981), ‘Panel Data and Unobservable Individual Effects’, Econometrica 49(6) 1377–98. Head, K., T. Mayer and J. Ries (2002), ‘Revisiting Oligopolistic Reaction: Are FDI Decisions Strategic Complements?’, Journal of Economics and Management Strategy, 11(3), 453–72. Head K. and J. Ries (2003), ‘Overseas Investment and Firm Exports’, Review of International Economics, 9, 108–22. Helpman, E. and P. Krugman (1985), Market Structure and Foreign Trade, Cambridge, MA: MIT Press. Helpman E., M.J. Melitz and S.R. Yeaple (2004), ‘Export Versus FDI’, American Economic Review, 94(1) 300–16. Julius, DeA. (1990), Global Companies and Public Policy, London, Royal Institute of International Affairs. Kraay, A. (1999), ‘Exports and Economic Performance: Evidence from a Panel of Chinese Enterprises’, mimeo World Bank. Levinsohn, J. and A. Petrin (2003), ‘Estimating Production Functions Using Inputs to Control for Unobservables’, Review of Economic Studies, 70(2), 317–42. Lipsey, R.E. and M.Y. Weiss (1984), ‘Foreign Production and the Exports of Individual Firms’, Review of Economics and Statistics, 66, 340–07. Maddison, A. (2001), ‘The World Economy: A Millennial Perspective’, Paris, OECD. Markusen, James R. (2002), Multinational firms and the theory of international trade, Cambridge: MIT Press. Melitz M.J. (2003), ‘The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity’, Econometrica, 71, 1695–725. Melitz, M.J. and Gianmarco I.P. Ottaviano (2003), ‘Market Size, Trade and Productivity, presented at the conference Exporting and Firm Level Adjustment, held the University of Nottingham in October 2003. Obstfeld, M. and A. Taylor (2003), Global Capital Markets: Integration, Crisis and Growth, Cambridge University Press, Cambridge. Rob, R. and N. Vettas (2001), ‘Foreign Direct Investment and Exports with Growing Demand’, Review of Economic Studies, 70, 629–48.

176 Sourafel Girma, Richard Kneller and Mauro Pisu Rugman, A.M. (1990), Multinationals and Canada–United States Free Trade, University of South Carolina Press, Columbia. Swenson, D. (1999), ‘Effects of Overseas Production on Home Country Exports: Evidence Based on Swedish Multinationals’, Weltwirtschaftliches Archiv, 132(2), 304–29. UNCTAD (1998), World Investment Report, United Nations, New York. Yeaple, S.R. (2003), ‘The Complex Integration Strategies of Multinationals and Cross Country Dependencies in the Structure of Foreign Direct Investment’, Journal of International Economics, 60, 293–314.

Index ABC measure for spillover benefits, 99, 111 Abramovitz, M., 72 absorptive capacity, 72–3, 107 acquisition-FDI, 50, 65, 67, 70–1 see also FDI advertising- and R&D intensive industries, 83 Aitken, B.J., 94–5, 106, 111, 144, 146 Alvarez, R., 122 Amiti, M., 166 Annual Census of Production, 96 Annual Foreign Direct Investment register (AFDI), 89 Annual Respondents Database (ARD), 89, 96, 153 anti-export bias, 35–6 Argentinean manufacturing industries, 82 arms-length trade, 156, 159 Asian NICs, 121 Australia, 95 Average Black Market Premium, 12, 19, 24 Average Coverage for Non-Tariff Barriers, 12 Average Import Tariff on Manufacturing (TARIFF), 12, 17 backward linkages, 54, 95, 98 Balassa, B., 134 Baldwin, R., 9, 120 Barbados, 136 Barnes, M., 111 Barrios, S., 72, 110 Benchmark models of product-cycle trade, 55 Ben-David, D., 34 Benhabib, J., 23 Bernard, A., 119, 122, 124, 127, 131, 137, 139, 141–2, 144, 146–7, 151, 154, 167–8

Bernard, Eaton, Jensen and Kortum model, 127–8, 132 Bernard, Redding and Schott model, 131–2 Bertrand, J., 60, 62, 134 Bertrand competition, 62, 65, 127 bilateral trade liberalisation, 129 Blalock, G., 121 ‘blockaded entry’ game, 73 Blomstrom, M., 73, 95, 173 Blonigen, B., 166–7, 173 Blundell, R., 147, 149, 154 Bosworth, B., 25 Brainard, S.L., 75, 157, 159–60, 167, 173 Braunerhjelm, P., 168, 172 Brown, C., 83 Buckley, P., 80 Bugamelli, M., 140 Canada, 95 capital import of superior, 50 intensity, 138 physical, accumulation, 33 vintage, 81 Casson, C., 80 Castellani, D., 167, 174 causality multinationality and productivity, 90 productivity improvement and exporting, 123 Caves, R.E., 79–81, 83, 93, 95 CGE (computable general equilibrium) modelling, 40–1, 43 Chisik, R., 123 Choi, J.P., 158, 162 Clausing, K.A., 166 Clerides, S., 120–1, 167 Lach and Tybout, 151 177

178 Index Cobb–Douglas production function, 11, 26, 51, 95, 103 Coca Cola, 90 Collected Trade Taxes Ratio (CTR), 12–13, 15, 19 Collier, P., 42 Collins, S., 25 Colombia, 139 Companies House, 170 comparative advantage, 5, 31, 33–4 and industry productivity growth, 129–32 complementarity between exporting and FDI within industry, 161 relationship, firm level, 5 within firm, 160–1 within industry, 161, 173 complex integration strategies of multinationals, 158 composite index (PCF), 12 computable general equilibrium modelling, see CGE Conyon, M.J., 81–4, 87–8, 90, 168 corporate structures adopted by firms, 62 Costa Dias, M., 147, 149, 154 Cournot, 56 competition, 67 duopoly, 72 Cowling, K.G., 80 Criscuolo, C., 89 cross-border investment and productivity, 3–4 d’Aspremont, C., 72 Davidson, C., 158, 162 Davies, S.W., 74, 82, 90 Dean, J., 38–9 developed country, 121 Disney, R., 153 Dixit–Stiglitz, 54 monopolistic competition, 72, 124, 128 preferences, 119 Dollar, D., 10 domestic- and foreign-owned firms, mean productivity levels in, 81 domestic non-exporters (DNE), 111

domestic permanent exporters (DPE), 111 Doms, M.E., 74 and Jensen, J.B., study of US manufacturing, 74 Driffield, N., 57, 111 DTI (Department of Trade and Industry), 92, 95 Dunning, J. H., 62, 65 OLI paradigm, 68, 70 duopoly game, international, 62, 64–5 Duranton, G., 139 Eaton, J., 127 see also Bernard, Eaton, Jensen and Kortum model econometric analysis, consequences of liberalisation, 40 Edwards, S., 9–12, 15, 18–19, 24, 137 open economy TFP growth model, 12 Ekholm, K., 164–5 entrants, 121, 125 equilibrium industrial structures with and without merger, 66 Ethier, W.J., 54 EU, see European Union European Linkages and International Ownership Structure (ELIOS) database, 170 European and non-European firms, 102 European Union, 85, 136 acquisitions, 88 corporate governance and mergers within the, 89 firms in United Kingdom, 103 exiters, 121, 125 export/ing effect on survival of firms, 146 and FDI, 110, 157, 173 and FDI, complementarity and substitution between, 158–65 industry productivity growth, 124–32 intangibles, 99

Index 179 export/ing – continued intra-firm productivity growth, 118–24 and location, agglomeration of industries, 140 and overseas production, 172 premium in TFP, 150 productivity, 4–5, 117, 120–3, 133, 136 promotion, 137 and stock of foreign assets to world GDP, 156 exporters agglomeration of, 144 and multinationals, co-existence of, 157 and non-exporters firm level characteristics, 139 export market entry effects on firm performance, 151 on productivity growth, 147 Export Processing Zones in developing countries, 136 factor endowment differences, cross-country, 131 see also TFP Falvey, Greenaway and Yu model, 131–2 Falvey, R., 4–5, 117, 131, 173 FAME, 153–4 FDI (foreign direct investment), 3, 83, 156 advertising- and R&D-intensive industries, 83 entry and survival of domestic plants, 110 export, see export form of, 65–9 greenfield and acquisition, 3 host economy, impact on, 81 incentives, 57 inward, linkages, 54–5, 72 mechanism for transferring technology, 80 and productivity relationship, 3, 49, 54, 66 and R&D, 65 spillovers, 72

tariff-jumping, 74 technology-sourcing, 57 FDI flows demonstration effects, 55–8 general equilibrium perspectives, 50–5 North-to-South, 51 partial equilibrium perspectives, 55–65 and productivity growth, 49 R&D performance, 62 Feenstra, R.C., 44 Feinberg, S.E., 173 Felipe, J., 10 Ferrett, B., 3, 49, 58, 65–6, 72–5 Findlay, R., 71 firms complementarity, 173 exiting export markets, see exiters FDI and R&D decision, 70 heterogeneity, 162–3, 173 high managerial ability, 149 international location (FDI) decisions, 69 level models, heterogeneous, 162 production function, 71 productivity, 3 switching to exporting, see entrants firm specific asset (FSA), 92–3, 107 Fischer, E.O., 25, 33 Ford and General Motors in Brazil, 111 foreign assets, 156 foreign direct investment, see FDI foreign firms efficient entry into domestic market, 79 productivity advantages of spillover to domestic firms, 4 foreign ownership and productivity, 79–81 differentials, impact on wages, 87–8 empirical work on, 81–3 foreign ownership by nationality, 101 total factor productivity (TFP), 85 in United Kingdom, 90

180 Index Foreign Presence Index in domestic market, 104 in export market, 104 foreign takeovers and productivity growth, 3 Fosfuri, A., 57–8, 73, 75 Fosfuri/Motta/Rønde model, game tree, 59 French firms, highest spillover benefits in UK, 99 general equilibrium modelling, key advantage of FDI flows, 49 Gertler, P., 121 Girma, S., 3, 5, 74, 79, 86–9, 93, 96–9, 100, 105, 109, 111, 147, 154, 156, 167–8, 174 see also ABC measure Glass, A.J., 55, 72–3 globalisation and aggregate productivity growth, 1, 9 integration of national goods and factor markets, 133 Globerman, S., 74, 95 Görg, H., 1, 4, 72, 81, 92, 94, 96, 98, 107, 110–11 see also ABC measure Graham, E.M., 166 Greenaway, D., 1–2, 5, 22, 28, 36, 38, 42, 90, 101, 107, 111, 136, 144–8, 152, 154 greenfield-FDI, 50, 56–7, 62, 65, 70–1, 74 acquisition-FDI, 67, 74 and exporting, 69 flows and R&D performance, 69 versus cross-border mergers and acquisitions, 65 greenfield investment, see greenfield-FDI Griffith, R., 74, 82 gross domestic product (GDP) per capita, 9 Grossman, G.M., 32, 34, 55, 164–5 growth effects of trade (openness) through technological spillovers, 32 rate gap, 121

in real exports, 137 in real output, 137 theories in open economy context, 29–33 in total factor productivity (TFP), 49 Haaland, J.I., 55 Hansen J (chi-square) statistic, 19 Hanson, G.H., 9, 111, 173 Haque, M.E., 2, 28 Harrigan, J., 42 Harrison, A. E., 9, 10, 26, 94–5, 106, 111 Harris, R., 96, 99 Hart, P.E., 111 Haskel, J., 96, 99, 110 Hausman–Taylor estimator, 171 Head, K., 147, 162, 167–9, 173 Heckscher, E., and Ohlin, B., relative endowment framework of, 131 Helpman, E., 32, 34, 55, 75, 131, 137, 147, 158, 162 Henry, M., 2, 9, 28 Heritage Foundation Index of Distortions in International Trade (HERITAGE), 12–13, 18 Heston, A., 26 Hijzen, A., 4, 81, 92 Holmes, T., 122 Hopenhayn, H., 124 Horn, H., 74, 134 Horstmann, I.J., 72–3 human capital, 29, 34, 74 India and US, return to capital, 71 Indonesia, 111 industry rationalisation and reallocation effects, 125 structures, equilibrium, 68 Infante, L., 140 ‘infant industry’ argument, 30, 117 information spillovers, 152 innovation, 124 intangible assets and absorptive capacity, 107 integration strategies of MNEs, 164

Index 181 Intel in Costa Rica, 111 intellectual property rights (IPRs), 70 inter-firm productivity comparisons, 82 reallocation and productivity growth, 124–9 inter-firm technology transfer, 66 internalisation theory, 79–80 international outsourcing model, 58 international trade costs, 133 trade liberalisation and growth, 28–35 intra-firm technology transfer, 67 intra-firm trade, 164 intra-industry productivity spillovers from FDI in the United Kingdom, 97 and nationality of ownership, 109 intra-industry trade model, dynamic, 124–7 Islam, N., 23 Jacquemin, A., 72 Japan, 136, 166 automobile production, 167 firms in UK, 102 MNEs, 109 satellite car industry plants, 98 Jean, S., 13, 130, 132, 134 Jensen, J.B., 119, 122, 124, 127, 137, 139, 141–2, 144, 146–7, 151, 154, 167–8 see also Bernard, Eaton, Jensen and Kortum model Jones, C.I., 33 Julius, De A., 173 Keane, M.P., 173 Keller, W., 94–5 Kim, E., 29 Kneller, R., 1, 5, 90, 101, 136, 144–5, 147–8, 152, 154, 156 knowledge-driven (KD) R&D model, human capital, 31 Kogut, 162 Koizumi, T., 71 Kopecky, K.J., 71 Korean satellite car industry plants, 98

Kortum, S., 127 see also Bernard, Eaton, Jensen and Kortum model Kray, A., 10, 167 Krueger, A.O., 36 Krugman, P., 124, 131, 164 Kulatilka, R., 162 lab-equipment (LE) model, 31 labour productivity of domestic firms, 93 Larrain, F.B., 111 Leamer’s Openness Index (LEAMER), 12, 17 learning-by-doing, 29–31 learning-by-exporting hypothesis, 119–21, 123 learning technologies with north–south trade, models of, 30 Leibenstein, H., 134 less developed country (LDC) supplier, 121 Leverhulme GEP Centre, 81 work on productivity and wages, 83–8 Levine, R., 9, 22 Levinshon, J.,168 liberalisation and aggregate productivity growth, 28 episode, features of, 37–8 growth, empirical analysis methodological issues, 35–41 and linkage to growth, 2 modelling, 35–8 study, 36–8 Lipsey, R.E., 94, 157, 160–1, 165–6 Lithuania, 94–5 Liu, X., 111 local price deviation from PPP (PRIDEV), 12, 20, 22, 24 Loewy, M.B., 34 Long, N.V., 67 Love, J.H., 57 Lucas, R.E., Jr, 29 Lyons, B.R., 74, 82, 90 Maddison, A., 156, 166 Manuelli, R., 33

182

Index

Markusen, J.R., 65, 72–3, 75, 93, 164 Martin, R., 89, 111 Mayer, W., 123 Medoff, J.L., 83 Melitz, M.J., 124–5, 127–8, 131–2, 134, 137, 147–8, 162, 173 model, 129, 133, 164, 173 model and Bernard, Eaton, Jensen and Kortum models, summary of differences, 128 Mexico, 95, 146 Miller, S.M., 10, 20, 23 Milner, C., 2, 9, 28 MNEs (multinational enterprises), 4, 49 complex integration strategies of, 163–5 decision, 72 domestic firms and foreign-owned, 100 entry, effective, 80 export behaviour of, 100 exporting and overseas production, 156 home country, 58 inter-firm rivalry, 50 oligopolistic industries, 70 production location, 73 productivity advantages, 81 and spillovers, 92 monopolistic output restrictions, 80 Montagna, C., 129–30, 132, 134 Moran, T., 94 Mosley, P., 41–2 Motorola, production facility in Scotland, 92 Motta, M., 56–8, 73, 75 see also Fosfuri/Motta/Rønde model multinational activity of UK-owned firms, 170 multinational enterprises, see MNEs multinationality and total exports, 172 national capital, output ratios, 71 nationality of firms, role of, 109 Nelson, R., 10 ‘new’ trade theory, 118

non-exporters (NE), 111 North and South FDI flows, 49, 51 ‘international technology transfer’, 72 technological gap between, 71 NUTS classification, 144, 154 Obstfeld, M., 156, 166 Office for National Statistics (ONS), 96 Ohlin, B., 131 OLI paradigm, 65, 68–70, 75 OneSource, 97–103, 107, 110, 153–4, 170 openness aggregate, 2 and TFP growth, measuring, 10–24 Ottaviano, G.I.P., 128, 173 Oulton, N., 74, 111 overlapping generations economy, two-sector AK model of, 33 Overmans, H., 139 ownership impact labour productivity, 84 total factor productivity, 86 wage rates by type of acquisition, 87 ownership-localisationinternalisation, see OLI Pack, H., 10, 121 see also international outsourcing model panel data, benefits of, 42 Papageorgiou, D., 36–8, 42 see also liberalisation episodes study Pavnik, N., 124 Pepsi, 90 Persson, L., 74 Pesaran, M.H., 11 Petit, M.-L., 73–4 Petrin, A., 168 Pisu, M., 5, 156 post export market entry effects, 153 Pritchett, L., 9, 24 productivity gaps, 75 growth, aggregate, 2–3 growth, empirical evidence on foreign ownership and, 79

Index 183 productivity – continued heterogenity and with-in industry reallocation, 126 as labour productivity, 71 performance of firms entering export markets, 143 uncertainty and firm entry/exit, flow chart showing, 125 productivity spillovers, 92–3, 110–11 for developing, transition and developed countries, 94 from FDI in United Kingdom, 97 international evidence on, 93–5 ‘profitability’ criterion, 75 protection, 35 proximity and concentration, 5, 75, 158, 167 quality products, 122 upgrading, 124 R&D decisions, endogenous and specialisation, 3 investments causing FDI, 62 rationalisation effect, 126 reallocation effect, 126 Redding, S., 131 see also Bernard, Redding Schott model Renelt, D., 9, 22 Research and Development, see R&D Rhee, R.-L., 121 Ricardian model, static firm-level, 127 Ries, J., 147, 162, 167–8, 169, 173 Rivera-Batiz, L.A., 31–2, 34 Roberts, M., 120, 139 Robertson, R., 123 Robinson, C., 96, 99 Rob, R., 161 Rodriguez-Clare, A., 54–5 Rodriguez, F., 9–10, 12–16, 18–19, 24, 26, 137 Rodrik, D., 9–16, 18–19, 24, 26, 29, 42, 44, 137 Romer, P., 31–2, 34

Rønde, T., 58 see also Fosfuri/Motta/Rønde model Rowthorn, R.E., 72–3 Rubin, D.B., 147 Ruffin, R.J., 71 Rugman, A.M., 161 Sachs, J., 10, 22 Sachs–Warner Openness Index (SWOPEN), 12, 15–16, 19, 24–5 Saggi, K., 55, 72–3, 121 Sala-i-Martin, X., 22 Salant, S., 74–5 Sanna-Randaccio, F., 73–4 Santos-Paulino, A., 44 Sapsford, D., 42 Sargan test of overidentification, 18 Schmitz, J., 122 Schott, P., 131 see also Bernard, Redding, Schott model Second World War end, progressive liberalisation of trade, 156 self-selection, 119–21 see also learning by exporting Shy, O., 72 Siemens, 92 simultaneity problem, 83 Sjöholm, F., 94, 111 Smarzynska-Javorcik, B.F., 94–5, 98 Smith, A., pin factory, 70 Smith, R., 11, 27 Söderbom, M., 10 Solow, R.M., growth model, 49, 51 South East Asian countries, productivity growth rates, 10 Southern savings and FDI inflows from North, 72 Spiegel, M., 23 spillovers (inter-firm technology transfer), 57, 67 ‘demonstration effects’, 70 geographically bounded, 75 mechanism, trained workers’ mobility, 58–61 MNEs to domestic firms, 94 MNE to local rival, 73 modelling of, 72

184 Index Standard Industrial Classification (SIC), 144 stochastic growth accounting methodology, 25 Stokey, N.L., 30 strategic choice, 161–2 Strobl, E., 72, 94, 110–11 Structural Adjustment Loans (SALs), 42 substitute relationship, 5 Sugden, R., 80 Summers, R., 26 Sutton, J., 83 Swenson, D., 167 Tabuchi, T., 134 TARIFF, see Average Import Tariff on Manufacturing Taylor, A., 156, 166 Teal, F., 10 technical efficiency, 79 technological spillovers to local firms, 50 trade and growth, 31–3 TFP (total factor productivity), 2, 137 of domestic firms, 93 growth, 9, 11 higher, of exporters, 138 homogeneity, impact of ownership change on Girma, 85–6 national, 71 of non-exporters, 138 TFPG – Generalised Method of Moments (TFPG-GMM), 11, 16 TFPG-heterogeneity (TFPG-HET), 11, 16–17, 23 TFP intermediate measure (TFPG-AMG), 11, 14, 20 Thisse, J.-F., 134 Thompson, S., 3, 79 total factor productivity, see TFP trade FDI relationship, 166 growth, 29–31 impediments, 122 inter-firm market share reallocation, 124

models, 132 policy, 9, 24 rationalisation effect on industry productivity, 132 regime indicator, 9 trade liberalisation elements of recent, 39 and growth, links between, 43 and growth, theory, 33–5 long- and short-run effects, 44 see also liberalisation trade openness and growth, 9 indices of, 12 measures of, 11–12 Trade Policy Review, 136 Trade Policy Review Mechanism (TPRM), 136 transportation costs, 127 triangle inequality restriction, 127 Tybout, J., 29, 82, 90, 120, 124, 139 UK, see United Kingdom UK Trade and Investment (UKTI), 136 UNCTAD, 74 United Kingdom, 142, 146 domestic firms’ absorptive capacity, 98 export behaviour, descriptive statistics by, 102 exporters, difference, 137–42, 150 exporting, 5, 141–2 information on foreign ownership, 100 manufacturing industry, 82, 100, 153 multinationals, 88–9, 106, 171, 173 new entrants, 141–52 overseas production and exporting decision of firms, 168 ‘productivity gap’, 74 productivity performance, terms of, 95 sectoral classification (SIC), 111 spillovers, 95–109 summary statistics by nationality of ownership, 103

Index 185 United States, 94–5, 136–7, 139, 146, 165–6 car industry, 88, 167 firms, greatest productivity advantage, 85 firms in UK, 99, 102 and India, return to capital, 71 intra-industry spillovers from multinationals, 95 multinationals, 109 new export firms, 142 University of Urbino, 170 Upadhyay, M.P., 10, 20, 23 Uruguay Round of multilateral trade negotiations, 136 US, see United States Vanuatu, 136 Van, P.H., 44 Venables, A.J., 72 Vendrell-Alda, J.L.M., 82 Venezuela, 94–5, 111 Vettas, N., 161 Vousden, N., 67 Wacziarg, R., 9 Wagner, J., 147, 167 Wakelin, K., 96–9, 109, 166

Walde, K., 44 Walz, U., 72 Wan, H., Jr, 44 Wang, J.-Y., 71, 73 Warner, A., 10, 22 Weiss, M.Y., 157, 160–1, 165–6 Westbrook, D., 124 Williamson, O.E., 66 Wolf’s Index of Import Distortions (WOLF), 12 Wooton, I., 55 World Bank STARS database, 26 structural adjustment loans, 43 World Development Report Outward Orientation Index (WDR), 12–13 World Investment Report, 163 world merchandise exports GDP, 166 value of, 156 Wright, P., 3, 79 Yanikkaya, H., 43 Yeaple, S.R., 94–5, 164–5 Young, A., 10, 30 Yu, Z., 4–5, 117 Zanfei, A., 174