Economia: Fall 2009

economi Journal of the Latin American and Caribbean Economic Association Volume 10 Number 1 2009 Fall Llach, Adrogu...

Author: Roberto Rigobon

30 downloads 977 Views 1MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

economi

Journal of the Latin American and Caribbean Economic Association

Volume 10 Number 1

2009

Fall

Llach, Adrogué, and Gigaglia on Effects of Longer School Days

Hernández-Trillo and Smith-Ramírez on Credit Ratings in the Presence of Bailout

Kaminsky, Mati, and Choueiri on Currency Crises in Argentina

Nieto-Parra on the Sovereign Bond Market and Debt Crises

Volume 10 Number 1

economía

Journal of the Latin American and Caribbean Economic Association

2009

Fall

EDITOR

Roberto Rigobon

LATIN AMERICAN AND CARIBBEAN ECONOMIC ASSOCIATION BROOKINGS INSTITUTION PRESS Washington, D.C.

Articles in this publication were developed by the authors for the biannual Economía meetings. In all cases the papers are the product of the authors’ thinking alone and do not imply endorsement by the staff members, officers, or trustees of the Brookings Institution or of LACEA, or of those institutions with which the authors are affiliated. Copyright © 2010 LATIN AMERICAN AND CARIBBEAN ECONOMIC ASSOCIATION www.lacea.org

Published by BROOKINGS INSTITUTION PRESS 1775 Massachusetts Avenue, N.W., Washington, D.C. 20036 www.brookings.edu ISSN 1529-7470 ISBN-13 978-0-8157-0467-6

For information on subscriptions, standing orders, and individual copies, contact Brookings Institution Press, P.O. Box 465, Hanover, PA 17331-0465. Or call 866-698-0010. E-mail [email protected]. Visit Brookings online at www.brookings.edu/press. Brookings periodicals are available online through Online Computer Library Center (contact the OCLC subscriptions department at 800-848-5878, ext. 6251) and Project Muse (http://muse.jhu.edu). Authorization to photocopy items for internal or personal use or the internal or personal use of specific clients is granted by the Brookings Institution for libraries and other users registered with the Copyright Clearance Center Transactional Reporting Service, provided that the basic fee is paid to the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. For more information, please contact CCC at 978-750-8400 and online at www.copyright.com. This authorization does not extend to other kinds of copying, such as copying for general distribution, or for creating new collective works, or for sale. Specific written permission for such copying must be obtained from the Permissions Department, Brookings Institution Press, 1775 Massachusetts Avenue, N.W., Washington, D.C. 20036; fax: 202-536-3623; e-mail: [email protected].

Volume 10 Number 1

economía

Journal of the Latin American and Caribbean Economic Association

2009

Fall

Editor’s Summary

vii

JUAN LLACH, CECILIA ADROGUÉ, AND MARÍA GIGAGLIA

Do Longer School Days Have Enduring Educational, Occupational, or Income Effects? A Natural Experiment in Buenos Aires, Argentina Comment by Catherine Rodriguez Orgales

1

34

FAUSTO HERNÁNDEZ-TRILLO AND RICARDO SMITH-RAMÍREZ

Credit Ratings in the Presence of Bailout: The Case of Mexican Subnational Government Debt Comments by Eduardo Cavallo and Tito Cordella

45

71

GRACIELA KAMINSKY, AMINE MATI, AND NADA CHOUEIRI

Thirty Years of Currency Crises in Argentina: External Shocks or Domestic Fragility? Comment by Carlos Winograd

81

120

SEBASTIÁN NIETO-PARRA

Who Saw Sovereign Debt Crises Coming? Comment by Alicia Garcia Herrero and Enestor Dos Santos

125 165

LATIN AMERICAN AND CARIBBEAN ECONOMIC ASSOCIATION

The Latin American and Caribbean Economic Association (LACEA), or Asociación de Economía de América Latina y el Caribe, is an international association of economists with common research interests in Latin America. It was formed in 1992 to facilitate the exchange of ideas among economists and policymakers. Membership in LACEA is open to all individuals or institutions professionally concerned with the study of Latin American and Caribbean economies. For membership information, please visit the LACEA website at www.lacea.org and click on “join LACEA.” OFFICERS President Mauricio Cárdenas, Fedesarrollo Vice President Ricardo Hausmann, Harvard University Past Presidents Andrés Velasco, Harvard University and Minister of Finance, Chile Mariano Tommasi, Universidad de San Andrés, Buenos Aires Sebastián Edwards, University of California–Los Angeles Guillermo Calvo, University of Maryland Nora Lustig, George Washington University Albert Fishlow, Columbia University Secretariat Fedesarrollo, Department of Economics, Universidad de los Andes, Bogotá Secretary Monica Parra Torrado, Fedesarrollo, Bogotá Treasurer Sergio Schmukler, World Bank

EXECUTIVE COMMITTEE Laura Alfaro, Harvard University Fernando Alvarez, University of Chicago Orazio Attanasio, University College London Raquel Bernal, Universidad de los Andes Sebastián Galiani, Washington University in Saint Louis Arturo Galindo, Universidad de los Andes and Banking Association of Colombia Alejandro Gaviria, Universidad de los Andes Graciela Kaminsky, George Washington University Santiago Levy, Inter-American Development Bank Roberto Rigobon, Massachusetts Institute of Technology Pablo Sanguinetti, Universidad Torcuato Di Tella, Buenos Aires Ernesto Schargrodsky, Universidad Torcuato Di Tella, Buenos Aires ECONOMÍA

Editor Roberto Rigobon, Massachusetts Institute of Technology Editorial Associate Jennifer Hoover Managing Editor Catherine Mathieu-Canuto Editorial Board Raquel Bernal, Universidad de los Andes Tito Cordella, World Bank Kevin Cowan, Central Bank, Chile Rafael Di Tella, Harvard University Marcela Eslava, Universidad de los Andes Eduardo Fernandez-Arias, Inter-American Development Bank Samuel Freije, World Bank Adriana Kugler, University of Houston Catherine Mathieu-Canuto, Economía Juan Pablo Montero, Pontificia Universidad Católica de Chile Carmen Pagés-Serra, Inter-American Development Bank Ugo Panizza, United Nations Conference on Trade and Development Claudio Raddatz, World Bank Roberto Rigobon, Massachusetts Institute of Technology

Francisco Rodriguez, United Nations Development Program Andrés Rodríguez-Clare, Pennsylvania State University Rodrigo Soares, Pontifícia Universidade Católica do Rio de Janeiro Ernesto Stein, Inter-American Development Bank Federico Sturzenegger, Banco Ciudad, Argentina Andrés Velasco, Ministerio de Hacienda, Chile, Harvard University, and Universidad de Chile

ECONOMÍA PANEL FOR VOLUME 10, n.1

Eduardo Cavallo, Inter-American Development Bank Laura Chioda, World Bank Guillermo Cruces, Universidad de la Plata/CEDLAS Marcio Garcia, Pontifícia Universidade Católica do Rio de Janeiro Luis Felipe López-Calva, United Nations Development Program Hugo Ñopo, Inter-American Development Bank Daniel Ortega, Instituto de Estudio Superiores de Administración, Caracas Norbert Schady, World Bank Pablo Serra, Comisión Nacional de Energía, Chile, and Universidad de Chile Felix Vardy, World Bank Eric Verhoogen, Columbia University AUTHORS AND DISCUSSANTS

Cecilia Adrogué, Universidad de San Andrés Eduardo Cavallo, Inter-American Development Bank Tito Cordella, World Bank Nada Choueiri, International Monetary Fund Enestor Dos Santos, BBVA and IE Business School Alicia Garcia Herrero, Inter-American Development Bank María Gigaglia, Universidad Austral Fausto Hernández-Trillo, Centro de Investigación y Docencia Económicas Graciela Kaminsky, George Washington University Juan Llach, Universidad Austral Amine Mati, International Monetary Fund Sebastián Nieto-Parra, Organization for Economic Cooperation and Development Catherine Rodriguez Orgales, Universidad de los Andes Ricardo Smith-Ramírez, Centro de Investigación y Docencia Económica Carlos Winograd, Paris School of Economics, University of Evry

ROBERTO RIGOBON

Editor’s Summary his volume of Economía consists of four papers. One paper studies a quasi-natural experiment of lengthening the school days in Argentina. The other three papers examine sovereign debt crises, currency crises, and pricing issues in subnational debts. The first paper is titled “Do Longer School Days Have Enduring Educational, Occupational or Income Effects?” by Juan J. Llach, Cecilia Adrogué, and María Elina Gigaglia. They study a very interesting event in which school days were increased for about a half of the students in the city of Buenos Aires. The paper starts by describing the context in which the policy reform takes place. For most of the twentieth century, Argentina had a public system of education in which primary school was compulsory and access was universal. To improve the quality of the education system during the 1960s, about 50 percent of the students participated in schools that increased their school day to a double shift. This paper studies the consequences for educational quality and subsequent occupation and income among participants in this program. The paper interviews 380 individuals forty years after their graduation (1977). A key aspect of the paper— indeed the identifying assumption—is that individuals were selected into the different school systems mostly by their location, and only after that criteria had been fulfilled did the schools choose students by their socioeconomic status. This is important because they compare the outcomes of interest between those that were in the different systems; hence if selection occurred on the basis of ability, their results might be biased. The other weakness of the paper is that the data comes from a survey, a methodology that is always subject to collection problems. In the end, the policy experiment is very interesting, the question of extending the school days is of crucial relevance to the region, and the results are quite surprising. They show that graduation rates of the double-shift schools increase. However, their data for the effects of double shifts on tertiary education are inconclusive, and they find almost no impact on income, employment,

T

vii

viii

E C O N O M I A , Fall 2009

and other longer-term outcomes. In the words of the authors, their results “suggest that the content and learning quality in DS [double shift] schools was not too good.” In the end, as with most policies related to educational attainment, the double shift was very effective in with regard to participation but less so with regard to quality. Our second paper, “Credit Ratings in the Presence of Bailout: The Case of Mexican Subnational Government Debt” by Fausto Hernández-Trillo and Ricardo Smith-Ramírez, studies the puzzling behavior of credit rating agencies in rating subnational debts in Mexico. In Mexico some of the subnational governments are in a permanent state of default, and some have been continuously bailed out. Surprisingly, this debt is usually rated investment grade. In this context, it is fair to ask, what are the credit rating agencies really rating? Are they truly assessing the financial soundness of the subnational government, or rather are they just computing the probability of repayment—which includes the bailout. This is the question addressed by the paper. The authors study how political and economic factors affect the rating. The motivation for studying financial and economic factors determining the rating is obvious. The reason why political factors were included as well is because bailouts are the outcome of a political negotiation. Therefore, they study whether the rating is affected by political factors that are likely to influence the likelihood of a bailout, such as the political party in power and the size of the subnational government. The paper not only presents an interesting question but also makes a methodological contribution. Estimating these regressions for several rating agencies at the same time is not a trivial problem. Interested readers will find the details clearly explained in this paper. Let me concentrate on the results here. The paper shows that the larger the size of the subnational government, as measured by population, the more likely it is to receive a positive rating. This can be interpreted as the “too-big-to-fail” syndrome. Second, political affinity also improves the rating, that is, if the party of the subnational or federal government is the same as the one in the central government, the likelihood that the state is bailed out increases. These two findings, as expressed by the authors, “challenge the purpose of rating subnational debt in LDCs with a bailout tradition, since the market may assess the risk of these entities as equivalent or superior to that of sovereign instruments.” From the policy point of view, in recent years, Mexico has implemented new laws to improve transparency and reduce the likelihood of bailouts. This paper shows that “it seems that bond rating agencies are not yet convinced of the success of such legislation.”

Roberto Rigobon

ix

The third paper in this volume, “Thirty Years of Currency Crises in Argentina: External Shocks or Domestic Fragility?” by Graciela Kaminsky, Amine Mati, and Nada Choueiri, studies thirty years of crises in Argentina, from 1970 until 2001. The main question it asks is what has caused the crises in Argentina: external or internal factors? Or more precisely, what proportion of the variance can be explained by external versus domestic factors? During this period, Argentina had eight currency crises, four financial crises, and two defaults. (Clearly, this should lead to a change of most thesaurus entries for volatility: instability, unpredictability, explosive nature, Argentina.) The authors use a structural vector autoregression (VAR) to estimate the decomposition between the external and internal factors. Three external factors are considered: degree of world monetary policy stance, a variable for financial contagion, and a measure of world risk appetite. The domestic factors are measures of credit booms, shocks to real activity, domestic risk aversion, exchange rate controls, price controls, and bank deposit confiscations. The paper starts with a detailed description of the crises. This section alone is a masterful summary of the economic history of Argentina. For those interested in making sense of the economy of one of the most prominent countries in Latin America, and its tumultuous history, this will be a key section. The authors then estimate a VAR where the identifying restrictions come from the description of behavioral equations and accounting relationships. For example, they assume a money demand equation, but they use the fact that local prices are equal to the nominal exchange rate times the real exchange rate. The first equation could be questioned; the second one is mostly a definition. Using these equations they arrive at the restrictions that structural shocks must satisfy within their system. These restrictions are summarized by equation 24. To me, the most telling results are the variance decompositions in tables 2 and 3. They also present results on impulse responses and simulations. Those results are important. Still, I prefer the results from the variance decompositions. It is easy to see that each crisis is driven by a different factor. The only common denominator, though, is shocks to the money demand, which has been the only constant in Argentina during those thirty years. What is also revealing is the proportion explained by the money shocks: in the short run, it is always about 80 percent of the variation, and two years after, it still continues to explain about 20 percent of the variance. The second result, which I particularly like, is that during the convertibility plan, monetary supply shocks play a minimal role. In both the 1991–98 and 1998–2002 periods, the supply shocks have very little explanatory power. This is in contrast with the other

x

E C O N O M I A , Fall 2009

periods where monetary shocks explain about a quarter of the volatility, especially in the long run. Furthermore, two years ahead, monetary supply and monetary demand shocks are about equally important. Finally, our fourth paper, “Who Saw Sovereign Debt Crises Coming?” by Sebastián Nieto-Parra, examines the market microstructure of sovereign debt. The paper explores how fees (underwriting fees) are affected before crises. The author finds that three years before a crisis, the countries that ultimately experienced a crisis paid twice the underwriting fees of those that did not enter a crisis. It is important that the identification assumption of this paper comes from the panel comparison. Of course, this presents the problems that it is possible that unobservable characteristics are explaining the differences. The author argues that those unobservable characteristics are affecting the probability of a debt crisis and the underwriting fees, but they have no impact on other fundamentals, such as the sovereign spreads. This finding has two interesting features. First, the fees tend to anticipate crises. Second, fees provide information that is beyond the sovereign bond spreads. This last one is perhaps the most surprising of all the results: it implies that investment bankers price default risk better than investors. Underwriters price the default risk on the fee whereas investors price the risk in the sovereign spread. The paper also studies the behavior of investment bankers’ fees with regard to the type of crisis suffered by countries: sovereign debt crisis, currency crisis, twin crises, defaults, and International Monetary Fund rescue packages. Interestingly, those crises that are driven by liquidity have much smaller fees (three years prior to the crisis) than those countries that collapse because they had bad fundamentals. This is perhaps the weakest section of the paper (in terms of the econometrics and the identification assumptions required to make these conclusions), but certainly the question and the results are fascinating. Finally, this paper asks important questions regarding the predictability of crises and the pricing of sovereign bonds. If fees provide additional information to the spreads regarding the medium-run likelihood of facing a sovereign default event, why aren’t they included in the pricing of the spreads? This seems like a good topic for additional research. Economía is the result of a collaborative effort, and as usual, thanks are due to many people who made that process successful. Associate editors worked hard to guide papers to publication, members of the panel contributed valuable insights and spirited discussions, and Economía staff, in particular Catherine Mathieu and Myriam Bahiman, helped put it all together. Without their steady and abundant effort, this volume would have not have come to fruition. I thank them for their tremendous support. Finally, in the last meeting of associate

Roberto Rigobon

xi

editors, we decided to expand the number of editors for Economía, adding Raquel Bernal, Rodrigo Soares, and Ugo Panizza to the Economía panel. We four now are responsible for the editorial aspects of the journal. I thank them personally for their continued friendship and the tremendous effort they have already put into producing this journal and that I know they will continue to show. This volume also sees the departure of great associate editors who were with us for a long time. Carmen Pages has been an incredible force for our journal, and we will miss her greatly. Federico Sturzeneger worked without stop, and now that he is taking care of saving the poor in Argentina, he has less time for the journal. We will miss him as well. Finally, we bid farewell to Kevin Cowan; although his span with the journal was short, his work and commitment had no parallel.

JUAN LLACH CECILIA ADROGUÉ MARÍA GIGAGLIA

Do Longer School Days Have Enduring Educational, Occupational, or Income Effects? A Natural Experiment in Buenos Aires, Argentina n 1971 longer school days were decreed for around half of the primary schools in the city of Buenos Aires, Argentina. The policy covered all the city neighborhoods, and the schools were chosen probably at random. An unusual opportunity for a natural experiment was thus created. In 2006 and 2007 we interviewed a sample of 380 alumni of the 1971 cohort, thirty years after their 1977 graduation from schools with and without longer days. We tried to identify how the length of their school days affected their education, occupation, and income. The next section provides a fuller description of the aforementioned policy. The subsequent section, devoted to a review of the literature, is longer than usual. We thought it was important to review and to compare both the older literature on the relationship between the length of school schedules and academic results and the newer literature devoted to renewing the educational production function approach using random or natural experiments. Cross-references between different literatures are rare, but from our point of view, they can promote a better understanding of the issues dealt with here.

I

Juan J. Llach and María E. Gigaglia are with the IAE Business School, Universidad Austral, Argentina. Cecilia Adrogué—now at Universidad de San Andrés and the Consejo Nacional de Investigaciones Cientíﬁcas y Técnicas (Argentina)—was a research assistant at IAE when most of this research was performed. The authors are very grateful to both IAE-Universidad Austral and Fundación Ethos, whose support to this project was decisive. The authors also wish to thank Walter Sosa Escudero, Ernesto Schargrodsky, Flavia Roldán, Milagros Nores, and the participants in seminars at IAE-Universidad Austral and the Argentine National Academy of Education for their helpful comments.

1

2

E C O N O M I A , Fall 2009

The third section presents the methodology and the characteristics of the database, and the fourth section shows the main results of the experiment. We then conclude with a discussion of the results and some of their policy implications.

The Policy and Its Context In this section we describe the policy that gave origin to the natural experiment and the context in which it took place.

Educational System in Argentina and Buenos Aires in the 1970s Since the end of the nineteenth century, the Argentine educational system has been governed by the principles of free and universal access, included laity in the public schools, and, up to the late 1970s, provided for seven years of compulsory primary education.1 Although primary education was constitutionally in the hands of the provinces, the federal government continued running some primary schools in most provinces until the late 1970s and early 1980s. The private sector—both religious and secular—was also authorized to run primary and secondary schools. The case of the city of Buenos Aires was peculiar. As the capital of Argentina, until 1996 its administration was in the hands of the federal government, and the same happened with its schools. Enrollment rates in Argentina have been traditionally high when compared to other Latin American countries. In 1970 school expectancy from primary to tertiary education was 10.3 years at the federal level, and even higher and close to the average of developed countries (11.5 years) in the city of Buenos Aires (Llach 2005).

Policy of Lengthening School Days in the City of Buenos Aires: Creation of a Natural Experiment The policy introduced a double shift (DS), or full-time attendance, into the primary schools of the city of Buenos Aires.2 This approach began as a pilot 1. At the end of the 1970s, the ﬁrst year of preschool education was also decreed to be compulsory, and in the early 1990s, compulsory education was extended up to the tenth year. 2. The traditional length of a primary school’s schedule in Argentina has been between four and four-and-one-half hours, either in the mornings (more common) or in the afternoons. This system is known as a simple schedule or day (jornada simple). Accordingly, the new system was called a full schedule or day (jornada completa) or double shift (doble jornada). The length of the school day in the new system was nearly eight-and-one-half hours, including around two hours for lunch.

Juan Llach, Cecilia Adrogué, and María Gigaglia

3

experiment in 1957 proposed by the professor Carlos Florit, general inspector of schools of the National Council of Education (Consejo Nacional de Educación).3 During the 1960s, the number of DS schools increased very gradually, but in 1971 it was drastically expanded to encompass almost 50 percent of the primary schools of the city of Buenos Aires.4 The policy was originally conceived to achieve both educational and social purposes (Consejo Nacional de Educación 1968, 1971) and was evenly applied in all the school districts in such a way that, in the early 1970s, the proportion of DS primary schools in every school district was around 50 percent of the total. A D M I S S I O N C R I T E R I A A N D T H E S E L E C T I O N B I A S I S S U E . Even in areas where the middle class predominates in the city, there were, and still are, important socioeconomic differences among the neighborhoods and school districts. From a social perspective, the idea of the new policy was to provide a solution to the uneven consequences of the increasing participation of women in the labor force. While richer households could pay for nurses or other domestic help to look after children before or after their single school shift, the poorer households could not. For that reason, the ﬁrst DS schools were in poorer school districts. This policy was changed in the late 1960s, when educational goals began to supersede social ones. This change was clearly manifested in the parallel modification of admission criteria. In 1968 DS schools were ordered to give priority to —familial, social, and economic needs of the candidates; —proximity of the student’s address to that of the school; and —students with sisters or brothers in the school. Additional changes in criteria occurred in 1971, when it was clearly established that the main and unavoidable condition for admission was living near the school. Only after that was fulfilled could the school consider the following additional social criteria: the family’s unfavorable socioeconomic conditions, both parents working without domestic help, and the number of siblings. The 1971 reform of the admission criteria was critical in reducing the selection bias of our research. In addition, it is very well known that the address and the siblings-in-the-school criteria always have been predominant in the city of Buenos Aires. Taken together, these criteria implied both that the freedom to choose the school was very limited and that the students were not mainly sorted by socioeconomic criteria. Although these factors do not preclude the 3. Feldfeber, Gluz, and Gómez (2003). 4. See Ministerio de Cultura y Educación (1970).

4

E C O N O M I A , Fall 2009

possibility of selection bias—mainly from parents choosing or avoiding DS schools—they reduce it considerably. Academic content was very precisely deﬁned for the new DS schools. Extra time was assigned to the following activities: —reinforcement of the academic content already in place, particularly language and mathematics (35 percent); —one-on-one teacher-assisted studying (25 percent); —learning a foreign language, typically English (12.5 percent); —gym and instruction on health and saving habits (7.5 percent); and —crafts and job training (20 percent). Although originally intended to teach students useful skills for the labor market, much of the craft and job training seemed outdated from the very beginning, so that the quality of the additional school time invested in this activity was not very good. REFORM OF TEACHERS’ CAREERS AND IMPLEMENTATION OF THE EXPERIMENT. In conformance with Law 18.614, an experimental new regime for full-time and part-time teachers was established during the 1970 school year.5 The experiment was implemented in secondary schools and was intended to replace the system of “lesson hours” in full-time and part-time contracts with teachers in some institutions.6 The main purpose was to increase the teacher’s dedication to his or her pupils and avoid “taxi teachers,” who taught classes in many different institutions in order to earn enough money for a decent living. Such geographical dispersion was detrimental to the pedagogical function and educational effectiveness. Resolution 288 from the Ministry of Culture and Education, dated 12 March 1970, established which secondary schools would participate in this “microexperiment.” On 18 March 1970, through Resolution 324, the rules for designating the staff of teachers in such institutions were established, according to article 9 of Law 18.614. In principle, the same teachers that were already in the schools would stay, or they would be relocated if they requested this.7 By the end of the year 1970, it was decided to extend the experiment to other institutions due to the positive results obtained during the ﬁrst year. After the 5. This was known as Proyecto 13, and the teachers would have either a full-time schedule consisting of thirty-six hours or a part-time schedule, which could consist of thirty, twenty-four, or eighteen hours. Teachers had to perform extracurricular tasks for no less than four hours and no more than 50 percent of the teacher’s time in the school. 6. In Spanish these are called horas cátedra. 7. This process applied only to secondary schools, since the problem of “taxi teachers” existed at that level and not with the primary schools.

Juan Llach, Cecilia Adrogué, and María Gigaglia

5

Fourth National Meeting of Ministries of Education, the provinces agreed to enforce in 1971 the new structure and the new curricula and methodology at both primary and secondary levels of the different institutions. In this context, as directed by Resolution 1.885 of the Ministry of Culture and Education (dated 10 September 1970), it was decided that the double shift should be implemented in the different schools. Annex 2 of the resolution designated each of the public institutions where it would be implemented in the city of Buenos Aires, which corresponded to half of the public schools in every school district. As far as the authors know, the schools were assigned to the DS program at random. Since teachers and directors or principals were (and still are) assigned to schools through a very rigid, bureaucratic procedure, once a school was chosen, its teachers and directors or principals were part of the new program by default; they would be relocated and replaced only if they refused to participate. In the city of Buenos Aires, the experiment was implemented in ninetyseven primary schools. Thirty years after the implementation, eight of them do not appear in the list of schools, thirteen are now on a single shift, and the remaining seventy-six continue as DS schools. In some cases, there has been a change in the name; in one instance, a school that had belonged to district 19 had been reassigned to the more recently established district 21.8

Longer School Days, Enduring Effects of Education, and Natural Experiments: Review of the Literature As far as the authors know, no previous research has been done with the same purposes and methods of this paper, that is, to assess the enduring educational, occupational, and income effects of longer school days. For that reason, we separately review the literature on three different issues: the effects of instructional time on educational outcomes; the enduring, lifetime effects of education; and, ﬁnally, natural experiments performed in education. It is a little surprising that the vast majority of the literature on the effectiveness of instructional time was written between 1960 and 1990, as if the issue has disappeared since then. However, as we note at the end of this review, there recently has been a revival of interest in instructional time as a component of educational policy.

8. A complete list of the schools can be found in appendix A.

6

E C O N O M I A , Fall 2009

Longer School Days (Allocated or Scheduled Time) and Educational Outcomes This topic has a long history in the literature of educational science (CIPPEC 2006). Carroll (1963)—perhaps the first to adopt a pedagogically oriented approach—and most of the literature since then have agreed that increasing the allocation of instructional time has positive but small impacts on educational achievements, and that these impacts tend to be higher the lower the countries’ GDP and the students’ socioeconomic status (SES). Cotton (1989)—one of the oldest and most comprehensive reviews of this literature—distinguishes different varieties of allocated time: school time, classroom time, instructional time (the portion of classroom time spent teaching students), engaged time or time on task, and academic learning time (ALT).9 Cotton emphasizes the importance of keeping in mind this taxonomy because only about half of the typical school day is actually used for instruction, and the students are engaged in learning activities for only about half of their in-class time.10 She concludes that —there is a small, positive relationship between allocated time and students’ achievements; —there is a stronger, but still small, relationship between time on task and achievements; —there is a strong and positive association between ALT and students’ achievements and attitudes; —lower-ability students beneﬁt more from increases in allocated or engaged time, whereas higher-ability students only beneﬁt very slightly, if at all; and —beneﬁts are greater in highly structured ﬁelds of study, such as mathematics and foreign languages.11 Only one year after Cotton’s study, Berliner (1990) also emphasized the multidimensionality of instructional time and noted that the popularity of research on scheduled time was largely a consequence of the ease in measuring it.12 Contrary to ﬁndings in the majority of the literature, Berliner concludes 9. ALT refers to that portion of engaged time that students spend working on tasks at an appropriate level of difﬁculty for them and experiencing high levels of success. 10. After reviewing McMeekin (1993), Thrupp (1998), and Martinic (2002), CIPPEC (2006) arrives at the same conclusions regarding the scarcity of effective classroom time, adding that the problem is more serious in regions like Latin America and in poor socioeconomic environments. 11. Cotton (1989) reviewed ﬁfty-seven research studies—mainly from developed countries— concerned with the relationship between the educational time factors cited above and students’ outcomes. 12. He also pointed out that most of the studies on the effects of allocated instructional time on students’ achievements were an outgrowth of the Coleman report (Coleman and others 1966), with its skeptical view of the return of any kind of educational investments on educational outcomes.

Juan Llach, Cecilia Adrogué, and María Gigaglia

7

that in developed countries, the effects of quantity of instruction on achievement are clear and of great relevance for educational policies, and that the effects of quantity and quality of schooling are even clearer in less developed countries.13 Nineteen years after Berliner, Bellei (2009) concludes from his review that most of the American studies on the subject agree —on the existence of a positive and statistically signiﬁcant relationship between instructional time and academic achievements of the students; —on the modest size of that relationship; —that the effect is stronger for students with initially low academic achievement; and —that it tends to be curvilinear, showing diminishing returns to scale with the increase in instructional time.14 Finally, Fuller and Clarke (1994) also conclude that the effects of instructional time on educational outcomes are stronger in developing countries, including those in Latin America (see CIPPEC 2006).15 S C H O O L T E R M L E N G T H . A close family of studies has analyzed the educational, occupational, and income effects of the length of the school year. In his own study, Bellei (2009) uses a natural experimental methodology to evaluate the Chilean “Full School Day Program,” designed to increase the yearly high school instructional time from 955 to 1,216 hours. Every year, an additional group of high schools has been integrated into the program, thereby “potentially establishing a natural experiment.” The selection of the schools was not random but decided by the government according to certain criteria.16 His main ﬁndings indicate that the program had positive effects on students’ achievements, both in language (between 0.05 and 0.07 standard deviation) and mathematics (around 0.07 standard deviation), and had a stronger effect 13. In a shorter review, Pittman, Cox, and Burchﬁel (1986) concur with Berliner’s ﬁrst conclusion. Berliner adds that the study of Hyman, Wright, and Reed (1975) is one of the few to consider the enduring effects of quantity of schooling on overall quality of life. 14. In addition to other already quoted papers, Bellei reviews Jencks and others (1972), Bloom (1976), Wiley (1976), Borg (1980), Fisher and others (1980), Frederick and Walberg (1980), Karweit and Slavin (1981), Brown and Saks (1986, 1987), and Link and Mulligan (1986). He also adds that the methodological limitations of most of the reviewed studies are huge and that it is not clear to what extent they were subject to other factors with the potential to affect the ﬁndings. 15. Latin American cases are analyzed in Cardoso (2004; Uruguay); Cervini (2001; Argentina); Ministerio de Educación de Chile (2003); Administración Nacional de Educación Pública (2003; Uruguay), and Bellei (2009; Chile). 16. Bellei uses a differences-in-differences approach and argues that it provides an unbiased estimate of the causal effect of the program on students’ academic achievement, as measured by standardized tests.

8

E C O N O M I A , Fall 2009

in rural and municipal schools than in urban and private schools. Marcotte (2007) also performed a natural experiment on the effects of instructional time on Maryland’s primary school test scores. He found that the natural variation in snowfall over time, which inﬂuenced the number of effective school days, has a small but signiﬁcant effect on students’ exam performance.17 According to Pischke (2007), most of the studies on the effects of length of the school term, including his own, find that they are positive and significant only regarding educational outcomes like avoiding repetition, but not with regard to test scores, future earnings, or employment.

Enduring Effects of Education The classical reference here is Hyman, Wright, and Reed (1975), who analyzed responses to general knowledge questions in public opinion surveys between 1947 and 1974.18 Based on the fact that the higher the respondents’ level of educational accomplishment, the more often the correct responses were given, the authors conclude that “education produces large, pervasive, and enduring effects on knowledge and receptivity to knowledge” (p. 109). However, Wolﬂe (1980) emphasizes that it is likely that all analyses of educational effects that do not include IQ variables suffer severe, although unknown, specification errors.19 Using a causal model of the enduring effects of education, including the estimated effects of intelligence measures, he concludes that previous studies have seriously overestimated the enduring effects of education.20 More recently, in the age of the methodologies of instrumental variables and natural experiments, Duflo (2001) studies the educational and labor market outcomes of construction of 61,000 primary schools in Indonesia in a very short period (1974–78). Measuring the effects in 1995, twenty years after the program, she ﬁnds that it increased the average years of education by 0.25 to

17. Only 1 to 2 percent fewer students tested in harsh winters performed satisfactorily in math than did students examined after mild winters. 18. In these surveys, people of different ages and educational attainment were polled on their knowledge of a wide variety of issues, from identifying prominent public ﬁgures to responding to questions on vocabulary. 19. His point is quite relevant because it is very uncommon nowadays to include intelligence measures in studies of the determinants of educational outcomes. The studies of Meghir and Palme (2003, 2004) are some of the exceptions. 20. Wolﬂe recognizes, however, that his results are conditional until conﬁrmed by longitudinal studies in which intelligence scores are obtained for a representative sample of children, and their subsequent levels of educational, intellectual, and verbal achievements are measured. He adds the very important point that education does increase general intelligence in such a way that its indirect effect on vocabulary through adult IQ is ﬁve times the size of the direct effect.

Juan Llach, Cecilia Adrogué, and María Gigaglia

9

0.40 of a year; it improved by 12 percent the probability that an affected child would complete primary school; and it raised wages from 3 to 5.4 percent.21 Combining these effects, Duflo estimates economic returns to education ranging from 6.8 to 10.6 percent.22 She also warns about the risks of generalizing her results to other contexts because of a number of factors, such as the strong emphasis on education in Indonesia at that time, the possibility of general equilibrium effects of the program on the returns to education, and the fact that the program induced variations only at the primary school level, while returns to secondary education might have been different.23 Duﬂo recognizes that the program increased the quantity of education and that there is some concern that deterioration in the quality of education might result from this type of program, offsetting any gain in quantity. Also based on a natural experiment, the research of Meghir and Palme (2003, 2004) evaluates, approximately forty years later, the impact on educational attainment and earnings of a major school reform that occurred in Sweden in the 1950s. The reform had many elements in common with those occurring in other European countries at that time, and included an increase in the years of compulsory schooling, a new national curriculum, and the abolition of selection by ability into academic and nonacademic streams at the age of twelve. The authors ﬁnd that the reform increased both the educational attainment and earnings of those whose parents had acquired only the earlier compulsory level of education. However, the earnings of those with more educated parents declined, possibly because of a dilution of quality at the top end of educational levels. Although this study is a benchmark in the research on the long-lasting effects of education, it was not possible to separate in it the quantitative effects from the increase in years of schooling and the qualitative effects from the new curriculum and the elimination of selection by ability at the age of twelve. The effects found were also small (see table 1).24 21. Duﬂo asserts that this wage increase proves that there is a combined effect for quality and quantity changes in education leading to an increase in human capital. 22. She reports that her two-stage least squares estimates are similar to ordinary least squares estimates and also to most estimates reported for developed countries, but they are smaller than those of Psacharopoulos (1994) for developing economies. 23. With regard to general equilibrium effects, because the returns were measured twenty years after the program, in an environment where the educational levels were higher than when the program began, individuals’ returns may be lower than they would be in other developing countries. 24. As a benchmark for the magnitude of these effects, Björklund (2000) estimates the wage premium per additional year of education to be 4.6 percent for Sweden (see Meghir and Palme 2004).

10

E C O N O M I A , Fall 2009

Finally, but perhaps most important, Schweinhart and others (2005) report the results of the High/Scope Perry Preschool Study through Age 40, which identiﬁes “both the short- and long-term effects of a high-quality preschool program on young children living in poverty.” The data come from a randomized experiment in which a sample of 123 low-income African American children, assessed to be at high risk of school failure, was split into a treatment group of 53 and a control group of 68. A variety of outcomes were measured at ages three, eleven, fourteen, fifteen, nineteen, twenty-seven, and forty. The authors find “evidence of positive effects on program-group children’s intellectual performance, school experiences, lifetime earnings and crime rates. Their school achievement was at a higher level, they were more committed to school, and more of them graduated from high school than members of the no-program group. In their adult lives program participants have achieved higher earnings and committed fewer crimes than members of the no-program group” (Schweinhart and others 2005, pp. xv and xvi).

Natural and Randomized Experiments in Education Fortunately, during this century, there has arisen a new class of natural and random experiments on the educational and labor market effects of different educational policies. This development has renewed hope for a better understanding of this very important issue, particularly after the disappointing results of the vast research program on the educational production function following the challenge posed by Coleman and others (1966).25 Although more accurate than the previous research program, the most important common trait of this new crop of research is that, with some relevant exceptions, most of the effects of the measured educational policies on outcomes are positive but modest or very modest, as can be observed in table 1.26 Regrettably, only a few experiments have studied the enduring effects of educational policies. The following is a brief summary of the results shown in table 1. The educational policy (treatment) with more intense and widespread effects on income, educational, and employment outcomes is high-quality preschool education (Schweinhart and others 2005). Also worth mentioning are class size at the primary level, with strong effects on test score gaps (Piketty and Valdenaire 2006); conditional cash transfer effects on access to tertiary 25. To get a balanced view of the educational production function research program, see Glewwe (2002) and Akerlof and Kranton (2002). 26. Piketty (2004) argues that the class size is perhaps the clearest case of assessing the superiority of natural experiments.

Juan Llach, Cecilia Adrogué, and María Gigaglia

11

T A B L E 1 . Compared Results of Natural and Random Experiments on Outcomes of Educational Policiesa Source and experimental type Preschool improvements Schweinhart and others (2005); R, preschool

Treatment, location, and durationb High-quality preschool (High/Scope Perry Project); Chicago, Ill., U.S.; 37 years

Kremer and Vermeersch (2004); R, preschool

Free school meals; Kenya; 1 year

Berlinsky, Galiani, and Gertler. (2006); N, preschool

Vast preprimary classroom construction program; Argentina; 4–5 years

Increase in school resources Duflo (2001); N, P and S

Glewwe, Kremer, and Moulin (2007); R, P

School construction program Indonesia; 20 years

Random provision of textbooks to primary schools; Kenya; 4 years.

Results Graduation. High school graduation rates of 77 percent in treatment group and 60 percent in control group. Income. Sixty percent of the treatment group vs. 40 percent of control group earning ≥US$20,000 a year. Quality of life. The treatment group at age 40 also had lower crime rates, higher employment rates, more fathers assuming child-rearing responsibilities, and higher scores in various intellectual and language tests at very different ages. Attendance. In spite of the increased fees in treatment schools, attendance at them improved by 8.5 pp (31 percent). Attendance gains were for both current students and students who had never attended before. Attendance. One year of preprimary school increased average third grade test scores by 8 percent of a mean or by 0.23 SD of the distribution of test scores. Noncognitive skills. Preprimary school attendance positively affected students’ self-control in the third grade as measured by behaviors such as attention, effort, class participation, and discipline. Graduation. 12 percent increase in the probability of primary school completion. Years of education. Increase of 0.25 to 0.4 of a year. Wages. 3 percent to 7 percent increase in wages. Rates of return. 6.8 percent to 10.6 percent for primary education. Test scores. No increase in test scores, contrary to the results of the previous literature. Textbooks increased scores for students with high initial (continued)

12

E C O N O M I A , Fall 2009

T A B L E 1 . Compared Results of Natural and Random Experiments on Outcomes of Educational Policiesa (Continued) Source and experimental type

Duflo and others (2006); Evans, Kremer, and Ngatia (2008); R, P

Class size Krueger (1999); R, P

Piketty and Valdenaire (2006); N, P and S

Treatment, location, and durationb

Results

Provision of free uniforms with an average price of $5.82; Kenya; up to 5 years.

academic achievement. Students with weaker academic backgrounds did not benefit from the textbooks. Many of them could not read the textbooks, which were written in English, most students’ third language. Attendance. For younger pupils, 6 pp increase (7 percent) in school attendance and 13 pp (15 percent) increase for students without a uniform prior to program. For older pupils, 13.5 percent decline in absenteeism. Years of education. Years of enrollment increased by 0.5 year (13 percent).

Reduction of class size (Tennessee STAR Project); U.S.; 4 years

Class size exogenously determined by policy putting ceiling of 30 students per classroom; France; 6 years

Test scores. Performance of students in smaller classes increased by 4 percentile points the first year and by 1 percentile point per year in subsequent years. Test scores in smaller classes rose by about 0.22 SD. Class size had a larger effect for minority students. Future earnings. A 0.22 SD improvement in test scores resulting from smaller class sizes implied an improvement of 1.7 percent and 2.4 percent average male and female earnings, respectively. Test scores and test scores’ gaps. The reduction of one pupil per primary class allowed an increase in the range of 0.3–0.4 points in math test scores (0.7 in less socioeconomically endowed contexts). These results implied that 5 pupils less per classroom in the poorer zones could lead to closing 46 percent of the test result gap between them and the nonpoor zones. With the same policies, the gap would be closed 22 percent at the college level and only 4 percent at the lycée.

Juan Llach, Cecilia Adrogué, and María Gigaglia

13

T A B L E 1 . Compared Results of Natural and Random Experiments on Outcomes of Educational Policiesa (Continued) Source and experimental type

Treatment, location, and durationb

Results

Dee and West (2008); R, early P

Reduction of class size (Tennessee STAR Project); U.S.; 8 years

Noncognitive skills. Class size reductions in early grades did improve subsequent student initiative, but these effects did not persist into the 8th grade. Smaller classes in the 8th grade led to improvements in measures of student engagement with effect sizes of 0.05 to 0.09, persisting two years later. Internal rate of return was 4.6 percent overall and 7.9 percent in urban schools.

Cash transfers conditional on school attendance and take-up of health services; education grants reduced private costs of going to school by 50 to 75 percent (PROGRESA program); Mexico; 3 years

Attendance. Between 3.4 and 3.6 pp increase in attendance for all children in grades 1 to 8. An 11.1 pp increase (19 percent) in attendance for students who have completed 6th grade, and a 14.5 pp increase for girls who have completed 6th grade. Spillovers to ineligibles in treatment villages of 5 pp increase (7 percent) in secondary enrollment. 1. Direct cash transfers. Attendance, permanence. Increases of 2.8 pp in school attendance; 2.6 pp in school permanence. Years of education. Increases of 2.8 pp in the following year’s enrollment and 23 pp in the probability to matriculate in tertiary studies. Graduation. 4.0 pp increase in graduation rates. 2. Part of the cash transfer postponed. Enrollment in both secondary and tertiary institutions increased over the basic treatment by 3.6 and 3.3 pp, respectively, without reducing the current attendance. Spillovers. Negative spillovers on the nonselected siblings and some positive peer effects on educational outcomes. (continued)

Conditional cash transfers Schultz (2004); R, P and S

Barrera-Osorio and others (2008); R, P and S

Three kinds of conditional cash transfers; Bogotá, Colombia; 1 year

14

E C O N O M I A , Fall 2009

T A B L E 1 . Compared Results of Natural and Random Experiments on Outcomes of Educational Policiesa (Continued) Source and experimental type Additional teaching support Banerjee and others (2005); R, P

Banerjee and others (2005); R, P

Peer effects Goux and Maurin (2007); N, S

Training programs Attanasio, Kugler, and Meghir. (2008); R, training

School reforms Meghir and Palme (2003, 2004); N, S

Treatment, location, and durationb

Results

Additional teaching support to lagging children; India; 2 years

Test scores. 0.14 and 0.28 SD increase in test scores in the first and second year compared to nontreated peers. One year after the program, initial gains faded to about 0.10 SD. Test scores. 0.35 and 0.47 SD increase in math scores in the first and second year of the program, respectively. After that, the increase tended to diminish.

Computer-assisted learning program: 2 hours per week of shared computer time; India; 2+ years

Neighbors’ peer effects; France; cross-section, educational performance up to lower secondary

Repetition. The probability of repetition was found to be 0.20 SD higher for adolescents (15–16 years) living in neighborhoods with a higher proportion of mates that have already been held back a grade at age 15.

Training program with school and on-the-job components; Colombia; 19 to 21 months

Earnings and employment. In the case of women, training increased wage and salaried earnings, probability of being employed, amount of days and hours worked, and probability of having a formal job with a written contract. Smaller effects on men: training only increased wage and salaried earnings and the probability of having a formal job and with a written contract, but not of having employment. Salaried earnings increased 18 percent for women and 8 percent for men. Cost-Benefit. The rate of return of the program emerging from cost-benefit analysis was 13.5 percent for women and 4.5 percent for men. On-the-job training intensity increased the returns of the program.

Vast school reform (1950s): increase in the years of compulsory schooling, new national curriculum, abolition

Years of education. Increase by 0.298 of a year, entirely due to the increase in the educational attainment of those with unskilled fathers.

Juan Llach, Cecilia Adrogué, and María Gigaglia

15

T A B L E 1 . Compared Results of Natural and Random Experiments on Outcomes of Educational Policiesa (Continued) Source and experimental type

Treatment, location, and durationb of selection by ability at the age of 12; Sweden; ∼ 30 years

Hoxby and Rockoff (2005); N, P

Length of the school year Marcotte (2007); N, P and S

Comparison between “lotteriedout” and “lotteried-in” students applying to charter schools; Chicago, U.S.; 2 years (average)

Longer school year as measured by harshness of winters; Maryland, U.S.; cross-section, same year

Bellei (2009); N, S

Full-School-Day Program: increase from 955 to 1,216 hours a year in secondary schools; Chile; 2 years

Sims (2006); N, P

Increase in school days, particularly ones devoted to test preparation; Wisconsin, U.S.; 5 years (average)

Results Earnings. Overall, 1.42 percent increase in earnings. For those with unskilled fathers, the reform increased earnings by 3.4 percent. Rates of return. If all the changes in earnings were due to changes in the quantity of education, the results implied returns of 6.0 percent for low-ability individuals, 11.6 percent for those with high abilities, and 8.4 percent overall. If other variables played a role, those returns were upper bounds. Compared to their lotteried-out fellow applicants, students who applied to and attended charter schools an average of 2 years starting in the elementary grades scored about 6 national percentile rank points higher, both in math and reading. Test scores. Only between 1 and 2 percent fewer students performed satisfactorily after harsh winters than did students examined after mild winters. Test scores. The program had a positive effect on students’ achievement, both in language (between 0.05 and 0.07 SD) and math (around 0.06 SD). Effects were stronger in rural and municipal schools than in urban and private schools. Test scores. Clear and positive relationship between math scores of 4th graders and days of preparation. The implied effect was small in both an absolute sense and relative to other educational reforms.

a. N, natural experiment; R, random experiment; P, primary education; S, secondary education; pp, percentage points; SD, standard deviation. b. Duration refers to the time span between the treatment and the measurement of effects.

16

E C O N O M I A , Fall 2009

education (Barrera-Osorio and others 2008); training program impacts on wages and employment (Attanasio, Kugler, and Meghir 2008); effects of preschool meals on attendance (Kremer and Vermeersch 2004); and the impact of school construction on graduation rates and school attendance (Duﬂo 2001; Berlinsky and Galiani 2007). Surprisingly, and perhaps due to identiﬁcation problems, vast school reforms appear to have more modest effects (Meghir and Palme 2003, 2004; Hoxby and Rockoff 2005). Regarding the lengthening of the school year, the only effects analyzed up to now are test scores, and the three studies reviewed (Marcotte 2007; Bellei 2009; Sims 2006) show smaller impacts on them than other treatments shown in table 1. Finally, some of the studies, such as Banerjee and others (2005) and Hoxby and Rockoff (2005), emphasize the importance of quality over “quantity” of education. This, as well as the critical importance of early childhood education, appears to be clearly proved by Schweinhart and others (2005), as can be seen in its impact on a rich set of dependent variables, from income to intellectual development. As we will show, the results of our experiment are very much in line with the relevant literature, both experimental and nonexperimental: that is, the effects are positive, nuanced, and stronger in lower socioeconomic strata and as regards quantitative outcomes.

Recent Revival of Increased Instructional Time as an Educational Policy The aforementioned recent works of Meghir and Palme (2003, 2004), Banerjee and others (2005), Bellei (2009), and CIPPEC (2006) reveal a sort of revival of increased instructional time as an educational policy, perhaps due to the fact that it is a strategy relatively simple to implement.27 However, impacts of additional instructional time per se seem, up to now, to be modest, validating the point made by Cotton (1989): “Signiﬁcant increases in the quantity of schooling would be required to bring about even modest increases in achievement.”

Methodology and Database In this section we describe the methodology used and the way the database was constructed. 27. CIPPEC (2006) reviews some of the policy-oriented papers that analyze increased instructional time from different (mainly positive) points of view, including Husti (1992), Pereyra (1992a, 1992b), Slavin (1996), Martinic (1998, 2002), Aguerrondo (1998), and, more recently, Feldfeber, Gluz, and Gómez (2003), Boissiere (2004), and Llach (2006). The CIPPEC review also refers to some critical studies, such as those of Karweit (1985) and the National Education Commission (1994).

Juan Llach, Cecilia Adrogué, and María Gigaglia

17

Methodological Approach As stated earlier, the primary purpose of this paper is to estimate the impact of a double-shift educational schedule on schooling and earnings. The parameter of interest can be formally described as follows. For any student observed after implementation of the policy, deﬁne random variables representing what educational level the person would have achieved had he or she attended a double-shift school as opposed to a traditional single-shift school.28 Denote these two potential outcomes by Y1 and Y0 and denote double-shift status by a dummy variable, D. For each person, we observe only Y = Y0 + (Y1 − Y0)D, so Y0 is not observed for students from double-shift schools (D = 1), and Y1 is not observed for students from single-shift schools (D = 0). We still hope to identify certain averages of Y1 − Y0. The effect of treatment on the treated is one such parameter:

)

(

)

(

)

(

E Y1 − Y0 D = 1 = E Y1 D = 1 − E Y0 D = 1 . This would tell us whether, on average, students beneﬁted or suffered from double-shift schooling. Comparisons by type of school shift can be decomposed as follows: (1)

(

)

(

)

(

)

E Y1 D = 1 − E Y0 D = 0 = E Y1 − Y0 D = 1 + ⎡ E Y0 D = 1 − E Y0 D = 0 ⎤ . ⎣ ⎦

(

)

(

)

The term E(Y1 − Y0 冷D = 1) is called the average treatment effect on the treated, and [E(Y0 冷D = 1) − E(Y0 冷D = 0)] is the bias (Lee 2005). If the shift of the school were randomly assigned, then D would be independent of Y0 and Y1, implying that

(

)

)

(

)

)

E Y0 D = 0 = E (Y0 and E Y1 D = 1 = E (Y1 . In this case, the effect of treatment on the treated is also the average treatment effect in the population subject to randomization and can be estimated by simple comparisons. The bias would be zero. We present these estimates by providing the mean differences of the outcome variables for the treated and the control groups. But, as we have seen, predetermined covariates are not

28. Also, we can analyze how much the person would earn in each case.

18

E C O N O M I A , Fall 2009

equal for both groups, which might mean that the probability of assignment to treatment might be inﬂuenced by them. Therefore we make a less restrictive assumption, which is in essence that the shift of the school is ignorable conditional on predetermined covariates, denoted by X. This assumption is also called selection on observables or ignorable treatment (Lee 2005) and can be expressed as follows: (2)

(

)

( )

E Y D, X = E Y X .

With the conditional effect E(Y1 − Y0冷X) identiﬁed, we can get an X-weighted average effect (Angrist 1998). Therefore, in order to avoid the bias due to the existence of confounding factors, we use regression and matching estimations, the kind of natural experiment methodology normally applied to observational studies. In particular, we use propensity score matching to reduce the bias in the estimation of the outcomes, that is, comparing the outcomes using treated and control subjects who are as similar as possible (Becker and Ichino 2002).29 Since this method is not sufficient to estimate the average treatment effect on the treated, we use the kernel matching approach, in which all units of the treatment group are matched with a weighted average of all units of the control group, with weights that are inversely proportional to the distance between the propensity scores of treated and controls.30

Database and Sample The database comes from a randomly sampled survey applied to the 1971 cohort.31 Selection of this cohort is methodologically very relevant because it was the ﬁrst to attend the primary schools of the city of Buenos Aires after the generalization of the DS policy. Although we cannot say that this device elim29. Propensity score matching, as deﬁned by Rosenbaum and Rubin (1983), is the conditional probability of receiving a treatment, given the pretreatment characteristics. The variables included in the estimation of the propensity score were age, gender, school SES, parents’ SES, type of school (boys and girls together or separate), size of the class or section, number of sections in the school, educational level of the mother, and educational level of the father. 30. Both regression and matching estimations were done on the common support, which means that they were limited to covariate values where both treated and control observations were found. In the ﬁrst case, it was done by saturating the model (Angrist and Pischke 2009), and in the second case, by not including in the sample individuals whose propensity score was lower than the minimum probability observed in the treatment group. 31. The survey was conducted by a specialized pooling agency, chosen by a process of competitive selection. For more information about the database, contact Juan Llach at [email protected].

Juan Llach, Cecilia Adrogué, and María Gigaglia

19

T A B L E 2 . Characteristics of the Sample Units as indicated Shift Gender Male No. obs. Percent Female No. obs. Percent Total obs. Percent

Parents’ SES

Students’ SES

Total

Simple

Double

High

Medium

Low

High

Medium

Low

185 48.7

86 50.0

99 47.6

22 53.7

73 50.0

90 46.6

61 56.0

59 41.8

65 50.0

195 51.3 380 100

86 50.0 172 100

109 52.4 208 100

19 46.3 41 100

73 50.0 146 100

103 53.4 193 100

48 44.0 109 100

82 58.2 141 100

65 50.0 130 100

inates the selection bias problems typical of such studies, it very probably helps at least to reduce them. The survey included items such as educational attainment at all levels, information about subjects’ parents, current SES, and labor status. The questionnaire included both closed and open questions. As previously mentioned, DS was implemented in all the school districts of the city of Buenos Aires. To design the sample and obtain a good representation of the whole population (table 2), we took the following steps. First, each school district was assigned a level of SES (low, medium, or high) based on the proportion of households with unsatisﬁed basic needs, and the number of schools where the policy was implemented was calculated for each district (appendix B).32 Then the distribution of the population by level of SES was determined. This exercise showed that either the authorities took into account the stratiﬁcation of the population by SES, or the policy was implemented randomly. This means that if the people with the lowest SES at that time represented 10 percent of the population, only 10 percent of the schools selected for the policy were of low SES. Moreover, to calculate the number of observations by SES for the sample, we used the 1980 primary enrollment of the school districts. In order to build the database, several steps had to be taken. First, to obtain the list of 1977 graduates, we had to acquire an authorization from Alberto Sileoni, the minister of education of the city of Buenos Aires at the time. Together with the request for authorization, we had to submit a list of the schools from which the student sample was to be drawn. Once the 32. Unsatisﬁed basic needs serve as a compound index of social indicators such as housing quality, employment status, and educational attainment.

20

E C O N O M I A , Fall 2009

authorization was obtained, members of the Ministry of Education contacted the supervisors of the school districts, who in turn informed the authorities of the chosen schools about the study. Only then could we request the list of 1977 graduates from the schools.33 Finally, once we had the list of graduates, the telephone numbers had to be searched for in different databases. The people we interviewed who finished primary school in 1977 fell into two categories: those who attended DS schools, where the policy was implemented in 1971 (treatment group [TG]), and those who attended simple shift schools with similar characteristics in pretreatment variables (control group [CG]). Interviews were conducted by telephone, which was possible because in the city of Buenos Aires the penetration of fixed telephony is very high. Almost all homes, even with the lowest incomes, own a phone. Furthermore, the response rate, once the person was located, was very high, regardless of the level of SES.

Results: More Education Does Not Imply Better Education As an initial approach to ﬁnding the effects of the double-shift educational schedule, the mean differences of the outcome variables are presented. Then, for a more accurate estimation, the regression and matching estimators are presented.

Mean Differences The mean differences between treatment and control groups are presented in table 3.34 Some of the relevant variables that cannot be considered to have the same mean in both groups are —conclusion of high school, which is higher in the TG (signiﬁcant at the 1 percent level); —knowledge of a foreign language and the educational level of the spouse, which are higher in the TG (signiﬁcant at the 5 percent level); —pursuit of postgraduate studies, which occurs more often in the CG and their members abandoned the second tertiary study later (both signiﬁcant at the 5 percent level); and 33. Though they had the authorization from the Ministry of Education, some directors of the schools refused to provide the list of graduates in order to protect their private information. This problem was especially likely in high SES schools and delayed the work, since we had to request a new authorization for additional schools. 34. Descriptive statistics on pretreatment and treatment variables are shown in appendix C.

Juan Llach, Cecilia Adrogué, and María Gigaglia

21

T A B L E 3 . Outcome Variables Units as indicated

Outcomes Educational Repetition in primary school (1: yes, 0: no) Conclusion of high school (1: yes, 0: no) Repetition in high school (1: yes, 0: no) Tertiary (postsecondary) studies (1: yes, 0: no) Conclusion of first tertiary study (1: yes, 0: no) Timely conclusion of first tertiary study (1: yes, 0: no) Year in which first tertiary study was interrupted Second tertiary study, if first was finished (1: yes, 0: no) Conclusion of second tertiary study if first one was finished (1: yes, 0: no) Timely conclusion of second tertiary if first one was finished (1: yes, 0: no) Year in which second tertiary study was interrupted if first one was finished Postgraduate studies (1: yes, 0: no) Type of postgraduate studies (1: post degree, 2: master, 3: PhD) Conclusion of postgraduate studies (1: yes, 0: no) Knowledge of foreign language (1: yes, 0:no) Income Net current monthly incomea Others Students’ SES at present (1: low, 2: medium, 3: high) Presence of children in current household (1: yes, 0: no) Number of children in current household Educational level of spouse (2: lowest, 10: highest) Educational level of household head (2: lowest, 10: highest)

No. obs.

Mean

Std. dev.

Treatment group (TG) mean (I)

Control group (CG) mean (II)

Difference in means (I–II)

380 380 375 343

0.06 0.90 0.21 0.83

0.23 0.30 0.40 0.38

0.05 0.94 0.22 0.83

0.07 0.86 0.19 0.83

−0.02 0.08*** 0.03 0.00

285

0.76

0.43

0.77

0.75

0.02

216

0.67

0.47

0.65

0.70

−0.05

64

2.44

1.28

2.50

2.36

0.14

216

0.26

0.44

0.28

0.24

0.04

57

0.61

0.49

0.69

0.50

0.19*

35

0.66

0.48

0.67

0.64

0.03

22

1.68

0.89

1.27

2.09

−0.82**

67 64

0.96 1.41

0.21 0.61

0.92 1.51

1 1.28

−0.08** 0.23*

64

0.80

0.41

0.74

0.86

−0.12

380

0.87

0.33

0.90

0.84

295

3.23

2.05

3.19

3.27

380

1.94

0.79

2

1.88

0.12*

380

0.78

0.42

0.78

0.77

0.01

380 261

1.55 6.80

1.11 1.98

1.50 7.04

1.60 6.50

−0.10 0.54**

380

7.24

2.07

7.33

7.15

0.18

0.06**

−0.08

*Statistically significant at the 10 percent level; **statistically significant at the 5 percent level; ***statistically significant at the 1 percent level. a. 1: <$323, 2: $324–$484; 3: $485–$635, 4: $636–$806, 5: $807–$968, 6: $969–$1290, 7: > $1,290.

22

E C O N O M I A , Fall 2009

T A B L E 4 . Outcomes of Students from Low SES Households Units as indicated Outcomes Educational Repetition in primary school (1: yes, 0: no) Conclusion of high school (1: yes, 0: no) Repetition in high school (1: yes, 0: no) Knowledge of foreign language (1: yes, 0: no) Others Number of children in current household

No. obs.

Mean

Std. dev.

TG mean (I)

CG mean (II)

Difference in means (I–II)

193 193 189 193

0.08 0.83 0.28 0.83

0.27 0.37 0.45 0.37

0.04 0.91 0.33 0.89

0.11 0.76 0.22 0.78

−0.07** 0.15*** 0.11* 0.11**

193

1.67

1.10

1.52

1.81

−0.29**

*Statistically significant at the 10 percent level; **statistically significant at the 5 percent level; ***statistically significant at the 1 percent level.

—conclusion of the second tertiary study, quality of postgraduate studies, and current SES, all of which are higher in the TG (all signiﬁcant at the 10 percent level).35 Tables 4, 5, and 6 show the main and statistically signiﬁcant mean differences in outcome variables of students originally coming from low, medium, and high SES levels, respectively. In the case of low SES students, conclusion of high school remains higher for the TG and is signiﬁcant at 1 percent; the same happens with knowledge of a foreign language, but that is signiﬁcant at 5 percent; and primary school grade repetition is lower in the TG, signiﬁcant at 5 percent, while high school repetition is higher for the TG, but only at the 10 percent level of signiﬁcance. In the case of medium SES students, the means for both the conclusion of a second tertiary study and the quality of postgraduate studies are higher in the TG, but the members of the CG abandoned the second tertiary study later; all these results are signiﬁcant at the 1 percent level. Contrary to what happened with low SES students, high school graduation rates are higher in the CG, signiﬁcant at the 5 percent level. The middle SES group is the only one in which a signiﬁcant (at 10 percent) mean difference appears with its current SES, which is higher in the CG. 35. The authors also analyzed some of the most important labor market variables (such as unemployment rate and job quality). The CG showed higher unemployment means, but only at the 10 percent level of signiﬁcance. As regards the matching procedure, the only signiﬁcant effect of the DS was that the changes of job were less frequent. These results might be attributable to selection, and since these values showed no relevance to subsequent analysis, they were not systematically presented here.

Juan Llach, Cecilia Adrogué, and María Gigaglia

23

T A B L E 5 . Outcomes of Students from Medium SES Households Units as indicated Outcomes Educational Conclusion of high school (1: yes, 0: no) Conclusion of second tertiary study if first one was finished (1: yes, 0: no) Timely conclusion of second tertiary study if first one was finished (1: yes, 0: no) Year in which second tertiary study was interrupted if first one was finished Type of postgraduate studies (1: post degree, 2: master, 3: PhD) Others Students’ SES at present (1: low, 2: medium, 3: high)

No. obs.

Mean

Std. dev.

TG mean (I)

CG mean (II)

Difference in means (I–II)

146 26

0.97 0.50

0.18 0.51

0.94 0.69

1.00 0.20

−0.06** 0.49***

13

0.62

0.51

0.72

0.00

0.72**

13

1.92

0.95

1.00

2.50

−1.50***

35

1.40

0.65

1.63

1.13

0.50***

146

2.26

0.69

2.20

2.36

−0.16*

*Statistically significant at the 10 percent level; **statistically significant at the 5 percent level; ***statistically significant at the 1 percent level.

Finally, in the case of the high SES group, repetition in high school is higher in the TG, and the quality of postgraduate studies is higher in CG, both results being signiﬁcant at the 10 percent level.

Educational, Occupational, and Income Effects With the exception of students’ nationality and gender, pretreatment variables cannot be considered to have the same mean (appendix C). It follows that mean differences between the TG and CG in the outcome variables cannot be considered as the result of the treatment per se. In order to control for

T A B L E 6 . Outcomes of Students from High SES Households Units as indicated Outcomes Educational Repetition in high school (1: yes, 0: no) Type of postgraduate studies (1: post degree, 2: master, 3: PhD)

No. obs.

Mean

Std. dev.

TG mean (I)

CG (II)

Difference in means (I–II)

41 13

0.10 1.62

0.30 0.65

0.14 1.44

0 2

0.14* −0.56*

*Statistically significant at the 10 percent level; **statistically significant at the 5 percent level; ***statistically significant at the 1 percent level.

24

E C O N O M I A , Fall 2009

T A B L E 7 . Average Effects of DS and Standard Errors (SE)a Units as indicated OLS

SE (robust)

ATTK

SE (bootstrapped)

0 0.05 0.12* −0.08 0.07 −0.02† 0.12† 0.06† 0.35* 0.06† −0.38†

0.03 0.04 0.06 0.05 0.06 0.08 0.44 0.07 0.15 0.25 0.50

−0.10 0.21* 0.03 −0.04 0.06 −0.16* −0.16 0.15* 0.06 0.51* −1.18*

0.09 0.09 0.07 0.06 0.14 0.06 0.30 0.10 0.23 0.24 0.56

−0.07† 0.06 −0.32* 0

0.08 0.20 0.16 0.04

−0.07 0.29 −0.21* 0.09

0.14 0.20 0.11 0.09

Income Net monthly income of person being surveyed at present

−0.39

0.26

0.02

0.63

Others Students’ SES at present Presence of children in current household Number of children in current household Educational level of spouse Educational level of household head

−0.08 0.01 −0.04 −0.03 −0.36

0.09 0.05 0.14 0.28 0.24

0.20 0.02 0.03 0.93* 0.73*

0.18 0.08 0.17 0.33 0.42

Outcomes Educational Repetition in primary school Conclusion of high school Repetition in high school Tertiary (postsecondary) studies Conclusion of first tertiary study Timely conclusion of first tertiary study Year in which first tertiary study was interrupted Second tertiary study if person finished first one Conclusion of second tertiary study if first one was finished Timely conclusion of second tertiary study if first one was finished Year in which second tertiary study was interrupted if first one was finished Postgraduate studies Type of postgraduate studies Conclusion of postgraduate studies Knowledge of foreign language

*Statistically significant at the 5 percent level; †model not significant at the 5 percent level. a. OLS, ordinary least squares regression; ATTK, average effect of treatment with kernel matching.

these preexisting differences, we subjected the data to ordinary least squares (OLS) regressions and kernel propensity score matching estimates, which are presented in tables 7 through 10. Regarding the regression results shown in table 7, we would like to mention that only three of them are statistically signiﬁcant: repetition in high school and conclusion of the second tertiary study if the ﬁrst one was ﬁnished, which are higher for DS graduates, and conclusion of postgraduate studies, which is lower for DS graduates. Kernel propensity score matching estimations of the average effect of the policy on the treatment group (ATTK), as reﬂected in the mean differences attributable to the treatment after controlling them for the differences in pretreatment variables, suggest the following.

Juan Llach, Cecilia Adrogué, and María Gigaglia

25

First, DS is found to have positive and signiﬁcant effects on these outcomes: —conclusion of high school: the secondary school graduation rate is 21 percent higher in the TG, and this is one of the most relevant results of our research; —access to and timely conclusion of a second tertiary study; and —educational level of the spouse and of the household head. On the other hand, the educational variables where the treatment appears to have negative and signiﬁcant effects are the year in which second tertiary study was interrupted if ﬁrst one was ﬁnished, the timely conclusion of the ﬁrst tertiary study, and the conclusion of postgraduate studies. The sign of the effects is the same in the regression and matching estimates for most of the outcome variables, and the cases in which it is not correspond to coefﬁcients that are not statistically signiﬁcant different from zero. However, the fact that the regression estimates and the matching estimates are not the same was expected.36 We also calculated the effects of the treatment for each level of student household SES. Tables 8 to 10 show the statistically signiﬁcant effects for the ATTK estimates; although the OLS estimates are also provided, the following discussion centers on the ATTK estimates. In the group of low SES students, the high school graduation rate is 22 percent higher in the TG—a very relevant result—and pursuit of a second tertiary study once the first one is finished is 25 percent more probable in the TG. However, repetition in high school also is higher in this group, by 13 percent. In the group of students from medium SES households, a great variety of results appear. On the one hand, students at DS schools show a positive and signiﬁcant effect in the quality of their postgraduate study. On the other hand, DS schooling has a small but negative and signiﬁcant impact on the conclusion of high school and on access to tertiary studies. Another negative but stronger effect is on the access to postgraduate studies. The middle SES group is the

36. As Angrist and Pischke (2009) demonstrate, regression estimators differ from matching estimators in the weights used to combine the covariate-speciﬁc effects into a single average effect. In particular, while matching uses the distribution of covariates among the treated to weight covariate-speciﬁc estimates into an estimate of the effect of treatment on the treated, regression produces a variance-weighted average of these effects. Treatment on the treated estimator puts the most weight on covariate cells containing those who are most likely to be treated. In contrast, regression puts the most weight on covariate cells where the conditional variance of treatment status is largest. The regression and matching weighting schemes therefore differ unless treatment is independent on covariates.

26

E C O N O M I A , Fall 2009

T A B L E 8 . Average Effects of Double Shift and Standard Errors, Students from Low SES Householdsa Units as indicated Outcomes Educational Conclusion of high school Repetition in high school Second tertiary study if person finished first one

OLS

SE (robust)

ATTK

SE (boot-strapped)

0.12* 0.17* 0.16

0.06 0.08 0.10

0.22* 0.13* 0.25*

0.09 0.07 0.16

*Statistically significant at the 5 percent level. a. See notes to table 7.

T A B L E 9 . Average Effects of the Double Shift and Standard Errors, Students from Medium SES Householdsa Units as indicated Outcomes Educational Conclusion of high school Tertiary studies Postgraduate studies Type of postgraduate studies Conclusion of postgraduate studies Income Net monthly income of person being surveyed at present Others Students’ SES at present

OLS

SE (robust)

ATTK

SE (boot-strapped)

−0.06 −0.05 −0.13 0.22 −0.11

0.04 0.05 0.12 0.34 0.23

−0.06* −0.05* −0.44* 0.66* −0.21*

0.03 0.03 0.10 0.19 0.11

0.03

0.48

1.04*

0.59

−0.36*

0.14

−0.50*

0.12

*Statistically significant at the 5 percent level. a. See notes to table 7.

T A B L E 1 0 . Average Effects of the Double Shift and Standard Errors, Students from High SES Householdsa Units as indicated Outcomes

OLS

SE (robust)

ATTK

SE (boot-strapped)

Educational Repetition in high school

0.27

0.15

0.14*

0.08

Income Net monthly income of person being surveyed at present

−0.95

0.92

−1.57*

0.85

Others Number of children in current household

−0.07

0.55

1.22*

0.71

*Statistically significant at the 5 percent level. a. See notes to table 7.

Juan Llach, Cecilia Adrogué, and María Gigaglia

27

only one to exhibit a positive impact of DS on income but, surprisingly, a negative impact on the current students’ SES.37 In the case of the high SES group, DS appears to be associated with a higher repetition rate in high school, lower income, and more children in the household. Per level of SES, the regression and matching estimates tend to have the same sign, except for the number of children in the high SES households.

Discussion and Policy Implications We have shown that the introduction of longer school days in half of the primary schools of the city of Buenos Aires in 1971 has significantly improved only one, but very relevant, educational outcome. Students that attended DS primary schools had a secondary school graduation rate 21 percent higher than those that attended single-shift primary schools. Moreover, this result is mainly explained by what happened with the low SES students. Since the introduction of DS implies an increase of around 40 percent in educational expenditures, a back-of-the-envelope calculation shows that the secondary school graduation expenditure elasticity is 0.5, not a low value, particularly considering that the average income of secondary school graduates is 23.7 percent higher than that of nongraduates (Adrogué 2006). As concerns tertiary and postgraduate educational levels, we have found both positive and negative impacts for DS. These last results, taken together with the absence of enduring effects of DS on income and employment and with the fact that DS students do not have a better knowledge of a second language, despite having had it as a subject in the school, suggest that the content and learning quality in DS schools was not too good. This is a very relevant ﬁnding when the extension of DS to other schools or to the whole educational system is being contemplated. Our outcomes emphasize that the content of the additional hours is even more important than increasing the duration of the school day. Just to give an example, the degree of improvement in academic results could be very different depending on whether the extra hours are just an extension of the current curriculum or whether they allow the disadvantaged students to develop their skills and abilities through instruction in and learning of a second language, 37. This is not incompatible because SES is measured only by educational and employment quality variables without including income.

28

E C O N O M I A , Fall 2009

sports, arts, and technologies—the same subjects that their advantaged schoolmates can normally learn and practice. The general meaning of our results coincides with that of most of the literature reviewed here, even the one study performed with the more demanding methodology of natural experiments. A wide variety of educational policies can have large and relevant impacts on the quantity of education and on lowincome populations but not necessarily on the quality of education or other lifelong effects. The main exceptions we found are for high-quality preschool (Schweinhart and others 2005) and, potentially, for class size in primary schools with low SES children (Piketty and Valdenaire 2006). In the ﬁrst case, the effects are not only important, but also widespread to personality, labor, income, and citizenship outcomes. In the second case, the impact is very strong on test scores. All these results (including this paper’s) seem to support the claim made by Piketty and Valdenaire (2006) that a targeted allocation of resources to poorer—or the poorest—schools and students could have a significant impact in reducing educational inequalities, and that the effectiveness of this approach will be much greater if it begins from early childhood onwards. However, it also seems clear that more and better research is still required to understand what speciﬁc policies are needed to improve the quality of education for the poor.

Juan Llach, Cecilia Adrogué, and María Gigaglia

29

Appendix A. Neighborhoods and School Districts

Source: Dirección de Investigación y Estadística, Dirección General de Planeamiento Educativo, Mapa Escolar, Ministerio de Educación del GCBA. Ley Orgánica de Comunas 1.777 of 1 September 2005, limits modified by Act 2.650 of 13 March 2008. School district limits were updated in September 2007 according to territorial modifications and changes in the land value and toponymy of the delimiting streets.

TABLE A-1. Ninety-Seven Schools Where Experimental Double-Shift Policy Was Implemented Name a 1. Juan José Castelli 2. Onésimo Leguizamón 3. Presidente Quintana 4. Bernardino Rivadavia 5. Provincia de Jujuy 6. Presidente Mitre 7. Tomás Manuel de Anchorena 8. Provincia de Catamarca (S) 9. Valentín Gómez 10. María Sánchez de Thompson 11. Carlos Pellegrini 12. Carlos Pellegrini (N, S) 13. Gral. Viamonte 14. Alte. Brown (S) 15. Gral. Aráoz de Lamadrid 16. República Italiana (N) 17. Dr. Guillermo Rawson 18. José Pedro Varela (S) 19. Fray Mamerto Esquiú 20. Juan Enrique Pestalozzi 21. Nieves Escalada de Oromí 22. Manuel de Sarratea 23. Olegario Víctor Andrade 24. Francisco de Gurruchaga 25. Gral. José María Zapiola 26. Dr. Antonio Bermejo (N) 27. República de Colombia 28. Facundo Zuviría 29. Dr. Ernesto E. Padilla 30. Virgen Generala 31. Juan Agustín Maza (S) 32. Presidente Marcelo T. de Alvear 33. Luis Vernet (NF) 34. Antonio A. Zinny 35. Dr. Florián Oliver (N) 36. República de Cuba (N) 37. Juan Crisóstomo Lafinur 38. Capitán Gral. Bernardo O’Higgins (S) 39. Blas Parera 40. Carlos María Biedma 41. Juan Bautista Alberdi 42. Manuel Dorrego 43. Joaquín María Cullen (N) 44. República Dominicana (S) 45. José Alfredo Ferreira (N) 46. Benjamín Zorrilla (N) 47. República Oriental del Uruguay (S) 48. Gral. Juan Galo Lavalle 49. Museo de Bellas Artes Gral. Urquiza 50. Florencio Varela

School district

School number

Zip code

Shift b

1 1 1 1 2 2 2 2 3 3 3 3 3 4 4 4 4 5 5 5 5 5 6 6 6 6 6 7 7 7 8 8 8 8 8 9 9 9 9 9 10 10 10 10 11 11 11 11 12 12

1 3 10 23 11 16 23 24 1 3 21 22 23 1 10 19 22 9 13 18 20 27 7 10 13 18 26 12 14 24 10 11 18 22 24 1 9 15 17 20 5 6 10 16 1 4 15 17 1 2

1112 1060 1052 1117 1182 1196 1170 1215 1099 1071 1133 1133 1219 1161 1166 1168 1103 1279 1277 1264 1276 1275 1242 1247 1215 1210 1256 1406 1405 1405 1424 1424

DS DS DS DS DS DS DS SS: MA DS DS DS SS: MA DS SS: MA DS DS DS SS: MA DS DS DS DS DS DS DS DS DS DS DS DS SS: MA DS

1424 1424 1414 1414 1426 1425 1425 1428 1429 1429 1428 1407 1406 1406 1407 1406 1406

DS DS DS DS SS: MA DS DS DS DS DS SS: MA DS DS SS: MA DS DS DS

TABLE A-1. Ninety-Seven Schools Where Experimental Double-Shift Policy Was Implemented (Continued) Name a 51. Emilio Giménez Zapiola 52. Jorge Newbery (N, S) 53. Remedios E. de San Martín (NF) 54. Ernesto Alejandro Bavio 55. Carlos Calvo (NF) 56. Leandro Nicéforo Alem 57. Prof. José Onaindía 58. Eduardo Ladislao Holmberg 59. Carlos Guido y Spano 60. Dra. Elvira Rawson de Dellepiane 61. Rubén Darío 62. Ing. Álvarez Condarco 63. Carmen Catrén de Méndez Casariego 64. República del Ecuador 65. Cabildo de Buenos Aires 66. NF 67. España 68. Naciones Unidas (S) 69. Félix de Azara 70. Grecia 71. Cnel. Mayor Álvarez Thomas 72. Luis Pasteur 73. Congreso de Tucumán 74. Hilarión María Moreno (S) 75. Dr. Antonio Dellepiane 76. Rafael Ruiz de los Llanos 77. Cap. Juan de San Martín y Gómez 78. Rodolfo Rivarola 79. Abel Ayerza 80. Provincia de Misiones 81. República del Perú 82. Manuel Peña 83. Provincia de Santa Cruz 84. Provincia de Corrientes 85. Juan Andrés de la Peña 86. Maestro Carlos Alberto Carranza 87. Gral. Félix de Olazábal 88. José María Torres (S) 89. República de Filipinas (S) 90. Islas Malvinas 91. Dr. Martiniano Leguizamón 92. Juan Ramón Jiménez 93. Armada Argentina 94. Félix F. Bernasconi (NF) 95. Félix F. Bernasconi (NF) 96. Félix F. Bernasconi (NF) 97. Félix F. Bernasconi (NF)

School district

School number

Zip code

Shift b

12 12 12 12 12 12 13 13 13 13 14 14 14 14 14 15 15 15 15 16 16 16 16 16 17 17 17 17 17 18 18 18 18 19 19 19 20 20 20 20 20 20 21 Ins. Ber.c Ins. Ber. Ins. Ber. Ins. Ber.

3 7 13 16 18 19 1 3 6 18 1 3 8 10 18 5 12 19 22 3 4 6 11 13 1 4 5 17 23 9 11 15 22 5 10 20 4 9 13 14 16 20 5 1 2 3 4

1407 1416

DS SS: MA

1407

DS

1406 1407 1440 1407 1407 1427 1427 1427 1416 1427

DS DS DS DS DS DS DS DS DS DS

1431 1431 1430 1419 1419 1431 1419 1419 1417 1419 1417 1417 1419 1417 1407 1407 1408 1263 1437 1437 1408 1408 1440 1440 1440 1440 1439

DS SS: MA DS DS DS DS DS SS: MA DS DS DS DS DS DS DS DS DS DS DS DS DS SS: MA SS: MA DS DS DS DS

a. N, name changed since policy was applied; S, shift changed since policy was applied; NF, school was not found. b. Double shift (DS), simple shift (SS), morning and afternoon (MA). c. “Ins. Ber.” is the way this school is listed in the document where the policy was announced. It apparently did not belong to a school district.

32

E C O N O M I A , Fall 2009

Appendix B.

Variables Used to Design the Sample

Units as indicated

School district 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Number of schools where policy was implemented

Share of all schools where DS policy implemented (percent)

SES level

Share of 1980 primary enrollment in DS schools (percent)

4 3 4 3 4 5 3 3 4 3 3 5 4 5 2 4 5 4 4 4

5.3 3.9 5.3 3.9 5.3 6.6 3.9 3.9 5.3 3.9 3.9 6.6 5.3 6.6 2.6 5.3 6.6 5.3 5.3 5.3

High Medium Medium Low Low Medium Medium Medium High Medium Medium Medium Medium Medium Medium Medium Medium Medium Medium Medium

10.6 8.4 4.1 3.1 2.9 4.7 5.5 5.0 9.4 8.4 3.7 3.9 3.6 3.3 3.6 3.1 4.7 3.6 5.3 3.1

Juan Llach, Cecilia Adrogué, and María Gigaglia

33

Appendix C. Descriptive Statistics on Pretreatment and Treatment Variablesa Variables Pretreatment Nationality (1: Argentine, 0: foreigner) Age Gender (1: men, 0: women) School SES (1: low, 2: medium, 3: high) Parents’ SES (1: low, 2: medium, 3: high) Type of school (1: boys and girls, 2: girls only, 3: boys only) Number of students Number of sections Father’s educational level (1: lowest, 10: highest) Mother’s educational level (1: lowest, 10: highest) Treatment Foreign language as a subject (1: yes, 0: no) Cultural activities (1: yes, 0: no) Place where cultural activities occur (0: nowhere, 1: at school, 2: at home, 3: at home and at school) Presence of lunch service (1: yes, 0: no) Attendance at lunch service (1: yes, 0: no)

No. obs.

Mean

Std. dev.

TG mean (I)

CG mean (II)

Difference in means (I–II)

380 372 380 380 380 380

1.00 41.52 0.49 2.00 1.60 1.40

0.07 0.70 0.50 0.75 0.68 0.71

1.00 41.44 0.48 2.09 1.69 1.62

1.00 41.62 0.50 1.90 1.49 1.13

0.00 −0.18*** −0.02 0.19*** 0.20*** 0.49***

375 380 380

25.56 1.67 4.47

5.73 1.16 2.10

27.83 1.90 4.68

22.79 1.40 4.20

5.05*** 0.50*** 0.51**

370

5.08

2.37

5.17

4.67

0.48**

380

0.52

0.50

0.93

0.02

0.91***

380 380

0.67 1.31

0.47 0.97

0.67 1.32

0.67 1.31

0.00 0.01

378

0.59

0.49

1.00

0.08

0.92***

221

0.66

0.48

0.70

0.07

0.63***

* Statistically significant at the 10 percent level; **statistically significant at the 5 percent level; ***statistically significant at the 1 percent level. a. Some other characteristics of both groups are: more than half of the students changed their primary school; most of those who attended a DS primary school changed to a single-shift high school; 76 percent of the students who started tertiary studies concluded them, and 70 percent of them did so in time; many surveyed people worked during their studies, and that work was related to what they were studying.

Comment Catherine Rodriguez Orgales: In their paper, Llach, Adrogué, and Gigaglia study the important question of the long-run impact that extending school hours can have on students in several aspects of their adult life. Speciﬁcally, the authors use a policy experiment from Argentina, carried out in the late sixties and early seventies, which doubled the hours spent by students in public primary schools. This policy increased school day length from four to eight hours (including two hours for lunch) per day. To carry out the analysis, the authors designed and applied a random survey to 380 students who graduated from primary schools in 1977, seven years after the policy was generalized. Aside from collecting information on the socioeconomic characteristics of the individuals’ parents while they were in school and on their school experiences, the authors also asked participants about adult outcomes, such as education attained, job status, income, and marriage. The results can be divided into two sets, depending on the methodology used by the authors. Under an OLS approach they ﬁnd mixed results. Even though increasing primary school hours increases the probability of concluding the second tertiary study, it has a negative impact on both high school grade repetition and conclusion of postgraduate studies. The authors ﬁnd no other effect on any of the other outcome variables of interest. The results based on propensity score matching are quite different. Increasing primary school instruction time increases by 21 percent the probability that a student ﬁnishes high school as well as students’ access to and timely conclusion of a second tertiary study. Dividing the sample according to socioeconomic status, the authors argue that the former effect is driven by the effect the program had on low-income students. As in the OLS results, it appears that such policy has a negative effect on the conclusion of postgraduate studies. I believe this paper is a valuable contribution in many ways. First, and probably most important, the ideal school day length is an important and relevant question that has been in the minds of policymakers and education 34

Juan Llach, Cecilia Adrogué, and María Gigaglia

35

researchers in different countries and at various times. However, no consensus has yet been reached, and we still do not know what the appropriate school day length is.1 The second contribution of this paper is its focus on Latin America. With the exception of Bellei (2009), most of the previous research, such as Carroll (1963), Card and Krueger (1992), Marcotte (2007), and Pischke (2007), focused on the effect of school time on students from developed countries. The low quality of education attained by Latin American students, evident in the international tests they have taken, makes it imperative to find new mechanisms to help improve this crucial aspect of these education systems. Though the studies cited in this paper provide mixed evidence regarding the effect of school length on quality of education, the Bellei study actually ﬁnds a small but positive effect of increased school hours in Chile. Given that only this study has concentrated on a Latin American country, additional evidence on the subject is always welcome. The third contribution of the present paper is its analysis of the long-term impact of school day length. Speciﬁcally, Llach and his coworkers evaluate the impact of longer primary school days on adult outcome variables such as high school and college graduation, wages, and job turnover, among others. This set of variables clearly has been underexamined worldwide, and even though Pischke (2007) does analyze some of them, his research is based on a policy implemented in a developed country. To the best of my knowledge, this paper is the ﬁrst to focus on the long-run impacts of school day length in a developing country. Finally, I believe that the fourth contribution of this paper is the construction of the data set, the sample design, and questionnaire elaboration used to empirically evaluate this policy experiment. We deﬁnitely need more of these initiatives in the study of development topics in Latin America. Nonetheless, I have a few comments on the results found, their comparisons with other programs, and their external validity. My first comment is related to the first stage of the matching estimation. Given that this is the authors’ preferred methodology, it would have been important to know which are the variables that the authors believe determine the probability of attending a double-shift school and whether they are indeed good predictors of the latter. The validity of the matching results depends crucially on the relevance of 1. For example, according to the Organization for Economic Cooperation and Development (2008), the difference between the countries with the lowest and highest number of total intended hours is more than 2,500 hours a year.

36

E C O N O M I A , Fall 2009

the control variables in the ﬁrst stage, but the reader has no information about this. It would also have been valuable to know the number of observations that are left in each of the estimations, especially when individuals are divided according to socioeconomic groups. This is important because for some questions, only thirteen people gave an answer in the whole survey, and hence, after dividing the sample according to socioeconomic status and matching individuals, I seriously doubt any true conclusion can be obtained from such a small sample size. Second, it would have been very enriching to have more details about the policy and its implementation. How were the participating schools chosen? The authors mention that the schools were evenly distributed across school districts; it must be stressed, however, that this does not imply randomness in implementation per se.2 Actually, the fact that the pretreatment variables of the control and treatment groups differ signiﬁcantly provides evidence in favor of the hypothesis that the schools were not randomly chosen. Another question that emerges is how the authorities managed to change from a single- to a double-shift program in the participating schools? Was additional infrastructure needed? Were additional teachers hired? If they were, did they have the same level of education and experience as the original teachers did? I believe that these are all crucial details that help the reader get an idea of the impact that the program may or may not have had. Although the authors brieﬂy mention a complementary policy of extending the working period of teachers, many of these doubts remain. For instance, if more but less-prepared teachers were hired and the curriculum did not change, as is suggested by the authors, it is difﬁcult to expect an improvement in the quality of education received by the students. Instead, given that the students had less idle time, one could have expected to observe a reduction in smoking, drug consumption, or criminal involvement, such as that found by Jacob and Lefgren (2003). All the aforementioned could be considered interesting outcome variables in future research. However, without a more careful and detailed description of the policy and its implementation, it is difﬁcult for the reader to know exactly which results are interesting and what to expect about them. Moreover, a detailed description and understanding of the program’s implementation could also give the authors a possible instrument to explain the participation in double-shift schools, one that would be useful for future research. 2. For instance, the fact that 50 percent of the schools in a given district were chosen clearly does not guarantee that they were chosen randomly. For example, the authorities could have chosen the 50 percent poorest or richest schools.

Juan Llach, Cecilia Adrogué, and María Gigaglia

37

There could be many possible candidates, such as whether siblings or other relatives were already attending the school or the distance from students’ homes to the participating schools. Alternatively, from appendix A, it can be observed that even though the share of schools where the policy was implemented was relatively constant, the share of 1980 primary enrollment was not. This could, in principle, serve as an instrument if Argentinean households do not choose neighborhoods according to the type of schools available for their children. Future research on the topic could focus on this alternative estimation methodology given that the main assumption under both OLS and matching estimators is that self-selection of individuals is only based on observable characteristics. This is, of course, a strong assumption, especially if the group of control variables is not complete and detailed. Third, it is important to understand the magnitude of the effects found by the authors and any possible bias from the survey methodology. Speciﬁcally, the authors mention that the survey was conducted among students who graduated from primary school in 1977. These are therefore not necessarily those who started school in 1971, when the authors claim the policy was generally applied. The choice of this cohort could therefore present more problems of self-selection than the authors actually acknowledge and hence could also explain part of the results found. Similarly, of the intended interviews, how many were successful and how many were not? These are all important details that could help in understanding the results found. For instance, I would have expected that a clear effect of the policy, if it was indeed randomly applied, would have been a reduction in repetition rate. This is because even if the policy did not change the curriculum, students would have spent more time in school, and their teachers would have had more opportunity to reinforce the topics learned or help students address difﬁculties. The results, however, do not support this hypothesis. The question that then emerges is whether this is so because some of the sampled individuals were part of the self-selected groups that got into the treated schools before 1971 and thereby naturally increased the repetition rate for the treatment group. Finally, when evaluating the impact of this policy change, the reader must exercise caution when analyzing the results. First, the authors claim that the most important effect is the 21 percent increase in the rate of secondary school graduation. This is a signiﬁcant effect, and one I have not seen before in other supply-side education interventions. I wonder how this effect is compared with the effects obtained through other education programs implemented in Argentina. I have a similar comment for their estimate of a secondary school graduation expenditure elasticity of 0.5. Although further details are needed

38

E C O N O M I A , Fall 2009

on how such a figure is obtained, I believe this is a high number.3 Finally, the authors conclude that given that no effects on tertiary graduation, wages, employment, and knowledge of a second language were found, the content and learning quality of the double-shift schools were not good. I think this is a strong assumption, and one that cannot be directly obtained from such analysis. First, if no effect on quality was achieved, through what channel do the authors attribute the 21 percent increase in the high school graduation rate? Second, the fact that no effect on income and employment is obtained could be consistent with a story in which schooling is simply a signal for the employer, and what is taken into account are the grades approved and not their quality, as Pischke (2007) suggests. Hence such a conclusion should be tested through other channels, such as results in exams or evaluation of the education and experience of participating teachers. As I commented earlier, I believe the line of research taken by these authors is very interesting and relevant. I hope the authors can continue with it and that future work can shed more light on the effect of this very important but understudied policy variable in Latin America.

3. For instance, when estimating the increase in educational expenditures, do they include only public spending in education, or does it include private spending too?

Juan Llach, Cecilia Adrogué, and María Gigaglia

39

References Administración Nacional de Educación Pública. 2003. Resultados en escuelas de tiempo completo y escuelas de áreas integradas. Evaluación nacional de aprendizajes en lenguaje y matemática, 6to año enseñanza primaria, 2002, segundo informe. Montevideo. Adrogué, C. 2006. “Desempleo y retornos a la educación superior en Argentina, 1974–2002.” Universidad Católica de Salta, Salta: XLI Reunión Anual de la Asociación Argentina de Economía Política. Aguerrondo, I. 1998. “América Latina y el desafío del tercer milenio. Educación de mejor calidad con menores costos.” Working Paper 10. Santiago: Programa de Promoción de la Reforma Educativa de América Latina y el Caribe. Akerlof, G. A., and R. E. Kranton. 2002. “Identity and Schooling: Some Lessons for the Economics of Education.” Journal of Economic Literature 40 (4): 1167–201. Angrist, J. 1998. “Estimating the Labor Market Impact of Voluntary Military Service Using Social Security Data on Military Applicants.” Econometrica 66 (2): 249–88. Angrist, J., and J. Pischke. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton University Press. Attanasio, O., A. Kugler, and C. Meghir. 2008. “Training Disadvantaged Youth in Latin America: Evidence from a Randomized Trial.” Working Paper 13931. Cambridge, Mass.: National Bureau of Economic Research. Banerjee, A., and others. 2005. “Remedying Education: Evidence from Two Randomized Experiments in India.” Working Paper 11904. Cambridge, Mass.: National Bureau of Economic Research. Barrera-Osorio, F., and others. 2008. “Conditional Cash Transfers in Education: Design Features, Peer and Sibling Effects: Evidence from Randomized Experiment in Colombia.” Working Paper 13890. Cambridge, Mass.: National Bureau of Economic Research. Becker S., and A. Ichino. 2002. “Estimation of Average Treatment Effects Based on Propensity Scores.” Stata Journal 2 (4): 358–77. Bellei, C. 2009. “Does Lengthening the School Day Increase Students’ Academic Achievement? Results from a Natural Experiment in Chile.” Economics of Education Review 28 (5): 629–40. Berliner, D. C. 1990. “What’s All the Fuss about Instructional Time?” In The Nature of Time in Schools. Theoretical Concepts, Practitioner Perceptions, edited by Miriam Ben-Peretz and Rainer Bromme, pp. 3–35. Columbia University, Teachers College Press. Berlinsky, S., and S. Galiani. 2007. “The Effect of a Large Expansion of Pre-Primary School Facilities on Preschool Attendance and Maternal Employment.” Labour Economics 14 (3): 665–80. Björklund, A. 2000. “Educational Policy and Returns to Education.” Swedish Economic Policy Review 7 (1): 71–105.

40

E C O N O M I A , Fall 2009

Bloom, B. S. 1976. Human Characteristics and School Learning. New York: McGraw-Hill. Boissere, M. 2004. “Determinants of Primary Education Outcomes in Developing Countries.” Washington: World Bank, Operations Evaluation Department. Borg, W. 1980. “Time and School Learning.” In Time to Learn, edited by C. Denham and A. Lieberman, pp. 33–72. Department of Education, National Institute of Education. Brown, B., and D. Saks. 1986. “Measuring the Effects of Instructional Time on Student Learning: Evidence from the Beginning Teacher Evaluation Study.” American Journal of Education 94 (4): 480–500. ———. 1987. “The Microeconomics of the Allocation of Teachers’ Time and Student Learning.” Economics of Education Review 6 (4): 319–37. Card, D., and A. Krueger. 1992. “Does School Quality Matter? Returns to Education and the Characteristics of Public Schools in the United States.” Journal of Political Economy, no. 100: 1–40. Cardoso, M. 2004. Calidad y equidad en las escuelas de tiempo completo: Un análisis de sus resultados en las evaluaciones estandarizadas 1996 y 1999. Prisma. Revista Semestral de Ciencias Humanas, no. 19: 171–90. Carroll, J. B. 1963. “A Model of School Learning.” Teachers College Records, no. 64: 723–33. Cervini, R. 2001. “Efecto de la ‘Oportunidad de aprender’ sobre el logro en matemáticas en la educación básica argentina.” Revista Electrónica de Investigación Educativa 3 (2) (http://redie.uabc.mx/vol3no2/contenido-cervini.html). CIPPEC (Centro de Implementación de Políticas Públicas para la Equidad y el Crecimiento). 2006. Estudio para la implementación de una política nacional de extensión de la jornada escolar. Buenos Aires. Coleman, J., and others. 1966. Equality of Educational Opportunity. U.S. Government Printing Ofﬁce. Consejo Nacional de Educación. 1968. Escuela de jornada completa. Organización y funcionamiento. Buenos Aires. ———. 1971. Reglamento orgánico para escuelas de jornada completa. Cotton, K. 1989. “Educational Time Factors.” School Improvement Research Series (SIRS), Close-Up 8. Portland, Ore.: Northwest Regional Educational Laboratory. Dee, T., and M. West 2008. “The Non-cognitive Returns to Class Size.” Working Paper 13994. Cambridge, Mass.: National Bureau of Economic Research. Duﬂo, E. 2001. “Schooling and Labor Market Consequences of School Construction in Indonesia.” American Economic Review 91 (4): 795–813. Duﬂo, E., and others. 2006. “Education and HIV/AIDS Prevention: Evidence from a Randomized Evaluation in Western Kenya.” Mimeo. MIT. Evans, D., M. Kremer, and M. Ngatia. 2008. “The Impact of Distributing School Uniforms on Children’s Education in Kenya” (http://siteresources.worldbank. org/EXTIMPEVA/Resources/evans_kenya_uniforms.pdf).

Juan Llach, Cecilia Adrogué, and María Gigaglia

41

Feldfeber, M., N. Gluz, and M. Gómez. 2003. La jornada completa en la Ciudad de Buenos Aires. Serie Estudios de Base, vol. 3. Buenos Aires: Gobierno de la Ciudad Autónoma de Buenos Aires, Secretaria de Educación, Dirección de Investigación. Fisher, C., and others. 1980. “Teaching Behaviors, Academic Learning Time and Student Achievement: An Overview.” In Time to Learn, edited by C. Denham and A. Lieberman, pp. 7–32. Department of Education, National Institute of Education. Frederick, W., and H. Walberg. 1980. “Learning as a Function of Time.” Journal of Educational Research 73 (4): 183–94. Fuller, B., and Clarke, P. 1994. “Raising School Effects while Ignoring Culture? Local Conditions and the Inﬂuence of Classroom Tools, Rules and Pedagogy.” Review of Educational Research 64 (1): 119–57. Glewwe, P. 2002. “Schools and Skills in Developing Countries: Education Policies and Socioeconomic Outcomes.” Journal of Economic Literature 40 (2): 436–82. Glewwe, P., M. Kremer, and S. Moulin. 2007. “Many Children Left Behind? Textbooks and Test Scores in Kenya.” Working Paper 13300. Cambridge, Mass.: National Bureau of Economic Research. Goux, D., and E. Maurin. 2007. “Close Neighbours Matter: Neighbourhood Effects on Early Performance at School.” Economic Journal 117 (523): 1193–215. Hoxby, C. M., and J. E. Rockoff. 2005. “The Impact of Charter Schools on Student Achievement.” HIER Working Paper. Harvard University. Husti, A. 1992. “Del tiempo escolar uniforme al tiempo escolar móvil.” Revista de Educación (Madrid), no. 298. Hyman, H. H., C. R. Wright, and J. S. Reed. 1975. The Enduring Effects of Education. Chicago University Press. Jacob, B. A., and L. Lefgren. 2003. “Are Idle Hands the Devil’s Workshop? Incapacitation, Concentration, and Juvenile Crime.” American Economic Review 93 (5): 1560–77. Jencks, C., and others. 1972. Inequality. New York: Basic Books. Karweit, N. L., 1985. “Should We Lengthen the School Term?” Educational Researcher 14 (6): 9–15. Karweit, N., and R. E. Slavin. 1981. “Measurement and Modeling Choices in Studies of Time and Learning.” American Educational Research Journal 18 (2): 157–71. Kremer, M., and C. Vermeersch. 2004. “School Meals, Educational Attainment, and School Competition: Evidence from a Randomized Evaluation.” Policy Research Working Paper WPS3523. Washington: World Bank. Krueger, A. B. 1999. “Experimental Estimates of Education Production Functions.” Quarterly Journal of Economics 114 (2): 497–532. Lee, M. 2005. Micro-Econometrics for Policy, Program, and Treatment Effects. Oxford University Press. Link, C., and J. Mulligan. 1986. “The Merits of a Longer School Day.” Economics of Education Review 5 (4): 373–81.

42

E C O N O M I A , Fall 2009

Llach, J. J. 2005. “The Challenge of International Educational Gaps in the Context of Globalizations.” Vatican City: Pontiﬁcal Academy of Social Sciences. ———. 2006. El desafío de la equidad educativa. Diagnóstico y propuestas. Buenos Aires: Granica. Marcotte, D. E. 2007. “Schooling and Test Scores: A Mother-Natural Experiment.” Economics of Education Review 26 (5): 629–40. Martinic, S. 1998. “Tiempo y aprendizaje.” LCSHD Paper Series 26. Washington: World Bank, Human Development Department. ———. 2002. “El tiempo y el aprendizaje en América Latina.” Serie Políticas. Santiago: Programa de Promoción de la Reforma Educativa de América Latina y el Caribe. McMeekin, R. W. 1993. “La investigación al servicio de la educación: Tiempo y aprendizaje.” Proyecto Principal de Educación en América Latina y el Caribe, boletín 30 (abril): 71–76. Meghir, C., and M. Palme 2003. “Ability, Parental Background and Education Policy: Empirical Evidence from a Social Experiment.” Working Paper 03/05. London: Institute for Fiscal Studies. ———. 2004. “Educational Reform, Ability and Family Background.” Working Paper 04/10. London: Institute for Fiscal Studies. Ministerio de Cultura y Educación (Argentina). 1970. “Establecimientos que aplicarán la nueva estructura del sistema Educativo, resolución no. 1885.” Boletín de Comunicaciones 15 (21): 1–5. Ministerio de Educación de Chile. 2003. “Factores que inciden en el rendimiento de los alumnos.” In Prueba SIMCE 4° Básico 2002. Nota Técnica. Santiago: Departamento de Estudios y Estadísticas. National Education Commission on Time and Learning. 1994. Prisoners of Time. Report of the National Education Commission on Time and Learning. U.S. Department of Education (www.ed.gov/pubs/PrisonersOfTime/index.html). Organization for Economic Cooperation and Development. 2008. Education at a Glance 2008: OECD Indicator. Paris. Pereyra, M. A. 1992a. “La construcción social del tiempo escolar.” Cuadernos de Pedagogía, no. 206, CD-ROM. ———. 1992b. “La jornada escolar en Europa.” Cuadernos de Pedagogía, no. 206, CD-ROM. Piketty, T. 2004. “L’impact de la taille des classes et de la ségrégation sociale sur la réussite scolaire dans les écoles françaises: Une estimation à partir du panel primaire 1997.” Unpublished paper (http://www.jourdan.ens.fr/piketty/ﬁchiers/ public/Piketty2004b.pdf). Piketty, T. and M. Valdenaire. 2006. “L’impact de la taille des classes sur la réussite scolaire dans les écoles collèges et lycées français. Estimations à partir du panel primaire 1997 et du panel secondaire 1995.” Les Dossiers 173. Paris: Ministère de l’Éducation Nationale, de l’Enseignement Supérieur et de la Recherche.

Juan Llach, Cecilia Adrogué, and María Gigaglia

43

Pischke, J. 2007. “The Impact of the Length of the School Year on Student Performance and Earnings: Evidence from the German Short School Years.” Economic Journal 117 (October): 1216–42. Pittman, R., R. Cox, and G. Burchﬁel. 1986. “The Extended School Year: Implications for Student Achievement.” Journal of Experimental Education 54: 211–15. Psacharopoulos, G. 1994. “Returns to Investments in Education: A Global Update.” World Development 22 (9): 1325–43. Rosenbaum, P. R., and D. B. Rubin. 1983. “The Central Role of Propensity Score in Observational Studies for Causal Effects.” Biometrika 70 (1): 41–55. Sims, D. P. 2006. “Strategic Responses to School Accountability Measures: It’s All on the Timing.” Economics of Education Review 27 (1): 58–68. Schweinhart, L. J., and others. 2005. Lifetime Effects. The High/Scope Perry Preschool Study through Age 40. Ypsilanti, Mich.: High/Scope Press. Schultz, T. P. 2004. “School Subsidies for the Poor: Evaluating the Mexican PROGRESA Poverty Program.” Journal of Development Economics 74 (1): 199–250. Slavin, R. E. 1996. “Salas de clases efectivas, escuelas efectivas: Plataforma de investigación para una reforma educativa en América Latina.” Working paper. Santiago de Chile: Programa de Promoción de la Reforma Educativa de América Latina y el Caribe. Thrupp, M. 1998. “The Art of the Possible: Organizing and Managing High and Low Socio-Economic Schools.” Journal of Education Policy 13 (2): 197–219. Wiley, D. E. 1976. “Another Hour, Another Day: Quantity of Schooling, a Potent Path for Policy.” In Schooling and Achievement in American Society, edited by W. Sewell, R. Hauser, and D. Featherman. New York: Academic Press. Wolﬂe, L. M. 1980. “The Enduring Effects of Education on Verbal Skills.” Sociology of Education 53 (April): 104–14.

FAUSTO HERNÁNDEZ-TRILLO RICARDO SMITH-RAMÍREZ

Credit Ratings in the Presence of Bailout: The Case of Mexican Subnational Government Debt ond ratings have existed for nearly a century, and they have become a matter of public policy concern (Cavallo, Powell and Rigobón 2008). Debt issued by ﬁrms, sovereign countries, and subnational governments (SNGs) are regularly rated in industrial countries (Cantor and Packer 1995, 1996).1 The rating history for less developed countries (LDCs) is shorter. International raters turned their attention to LDCs in the 1980s when agencies started rating LDC sovereign bonds in reaction to several international debt crises. As a result, literature on grading SNGs and sovereign bonds in industrial countries abounds, while for LDCs it is scarce. Rating agencies have come under scrutiny in regard to their grading of industrial countries and LDCs. For example, the Wall Street Journal (2004) reported that credit ratings in China could be merely guesswork. In the case of sovereign credit ratings, there is a growing body of literature that casts doubt on their role (see, for example, Reinhart 2001 and 2002), especially after the Asian and Argentinean crises of 1997–98 and 2001, respectively. Others have attempted to reﬁne the measurement of risk (Remolona, Scatigna, and Wu 2008; Alfonso 2003). More recently, the credibility of rating agencies has

B

Hernández-Trillo and Smith-Ramírez are with Centro de Investigación y Docencia Económicas (CIDE), Mexico City. For helpful comments, we would like to thank participants in the seminars held at the Universidad de las Américas, Puebla; Banco de México; Ohio State University; University of Chicago Harris School Public Policy; Pontiﬁcia Universidad Católica del Perú, Lima; CIDE; University of California, Los Angeles; and especially those at the Economia—Latin American and Caribbean Economic Association LACEA seminar held in Bogotá, Colombia. We also would like to thank our discussants and the editors of this volume, Tito Cordella, Eduardo Cavallo, and Roberto Rigobón. 1. See Carleton and Lerner (1969) for pioneering work on SNGs.

45

46

E C O N O M I A , Fall 2009

been challenged due to the role of their evaluations in losses associated with the U.S. mortgage crisis. In this paper, we target the rating technology used by agencies to grade SNGs in LDCs. By using data from the SNG bond market of Mexico, a country with a tradition of bailouts, we analyze how political and ﬁnancial factors are weighted in the construction of ratings. This case exempliﬁes the situation in most Latin American countries. Latin American governments have a long tradition of bailing out SNGs; Bevilaqua and Garcia (2002) document this phenomenon in Brazil, and Sanguinetti and others (2002) do likewise for Argentina. A high bailout probability raises at least two issues: the adequacy of the bond rating process and its purpose. Rating principles should take into account the many differences between industrial and LDC countries (Laulajainen 1999). Typically, developing countries have serious institutional and legal shortcomings (see InterAmerican Development Bank 1997). Most relevant, they are very centralized, law enforcement is deﬁcient (La Porta and others 1998), and most of them have just started ﬁscal decentralization reform, which in many cases has responded more to political pressure than to efﬁciency-enhancing purposes (see Díaz 2006; Giugalle, Korobow, and Webb 2001). These characteristics are important when rating bonds in their local currencies. For instance, Mexican SNGs are not allowed to issue debt denominated in foreign currency. Such differences call for different rating technologies than those used when rating entities within industrial countries, where many of the aforementioned shortcomings are not present. Surprisingly, one of the largest states in Mexico, the State of Mexico, has been continuously bailed out since 1995; though this SNG is virtually bankrupt, it still has been assigned an investment grade.2 Likewise, Sanguinetti and others (2002) report that the provincial government of La Rioja, Argentina, was bailed out several times previous to the 2001 crisis, and it still continues to receive an investment grading.3 Bond ratings are meant to indicate the likelihood of default (Bhatia 2002).4 If SNGs are to be bailed out anytime they 2. Reported by Black (2003). The grade assigned by Fitch in 2003 was BBB. 3. Sanguinetti, among others, argues that the Argentinean crisis was in part due to the ﬁscal permissiveness of SNGs in that country. For this reason, raters were questioned in Argentina. 4. It has been shown that these agencies specialize in gathering and processing financial information and are certiﬁed by screening agents who, in turn, are able to diversify their risky payoffs. In this setting, raters solve, at least in part, the informational asymmetry in capital markets, involving insiders possessing more accurate information about the true economic values of their ﬁrms or governments than outsiders. In turn, rating agencies gain from sharing their information. See Millon and Thakor (1985).

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

47

face ﬁnancial problems, then their risk is passed on to the federal government. Thus SNG rates eventually would become similar to those of sovereign debt.5 Is this happening in LDCs? If so, then the purpose of rating SNG debt may be arguable. Rating agency results are puzzling in LDCs since, as pointed out before, they often give high rates to ﬁnancially bankrupt SNGs. What, then, are agencies actually rating? Are they rating ﬁnancial soundness or just probability of bailout? Do rating agencies foster market discipline in the presence of implicit guarantees, or do they tend to exacerbate moral hazard problems?6 In this article, we try to answer these questions. Bailout events are most frequently the result of political negotiations. Therefore, we focus our analysis on the relevance, if any, that political factors have in the rating technology of three agencies: Standard and Poor’s (S&P), Fitch, and Moody’s. More speciﬁcally, we want to know whether the number of voters and political party in power matter; given Mexico’s long bailout tradition, we expect they do. Additionally, we analyze how ﬁnancial factors inﬂuence rates. Finally, unlike previous studies, we study the determinants of choosing a speciﬁc grading agency. Our econometrics extend and modify the seminal methodology of Moon and Stotsky (1993) in that we consider data from three rating agencies instead of two, and we use a novel formulation of the Monte Carlo expectation maximization algorithm (see Wei and Tanner 1990) to circumvent the estimation of multidimensional integrals instead of using probability simulators. Our results indicate that rating agencies differ slightly in how they weight relevant variables to assess the risk. Most notably, we found a strong negative correlation between SNG population, our proxy for number of voters, and debt risk. We interpret this as a “too big to fail” situation, that is, rating agencies consider that large entities, because of their political power, are more likely to be bailed out when facing ﬁnancial problems (Hernández, Díaz, and Gamboa 2002). A second strong determinant of ratings is whether the party governing the country is also governing the entity under evaluation. When the parties are the same, debt risk decreases significantly. This is evidence that raters take into account the bailout phenomenon based on both the population of the SNG and political afﬁnity between SNG and federal governments.7 5. In many cases, grades ran counter to the “sovereign ceiling” rule, as we will show later. See Durbin and Ng (2005). 6. The “market discipline” approach to subnational ﬁnance requires that moral hazard derived from the possibility of a central government bailout be made insigniﬁcantly small (Londero 2005). 7. Population has been interpreted as a political variable in the U.S. system of federal transfers under the New Deal. For a discussion, see Wallis (1998, 2001) and Fleck (2001).

48

E C O N O M I A , Fall 2009

These results are, to the best of our knowledge, novel in the bond rating literature. The paper is organized as follows. The first section provides a brief description of Mexican intergovernmental relations and reviews the SNG debt environment in Mexico. The second section presents a discussion about the opacity of Mexican SNGs. In the third part, we present the model, describe the variables, and examine some descriptive statistics. The fourth section discusses the empirical results, followed by final remarks in the conclusion.

A Brief Overview of Mexico’s Intergovernmental Relations and SNG Debt Regulation Mexico is a federal republic composed of three levels of government: the central or federal government; 32 local entities, which consist of 31 states and the Federal District; and 2,477 municipalities (hereafter referred to as SNGs). Like many countries in the Latin-American region, Mexico is characterized by strong regional and state disparities. While the Federal District and the states of Mexico and Nuevo Léon produce about 40 percent of total GDP, Chiapas, Guerrero, Hidalgo, and Oaxaca generate a subtotal of only 6.8 percent of total GDP. Clearly, the southern region of the country is by far the poorest. Mexico follows a revenue sharing system where the federal government collects main taxes, namely corporate and personal income taxes, value-added tax, and most excise taxes. These constitute 95 percent of total public sector tax revenue. Through a formula, 20 percent of this revenue is redistributed among states and municipalities. These net block transfers are known as participaciones. The main deﬁciencies identiﬁed in the system have been the local governments’ lack of tax independence and the formula itself. Recently, decentralization efforts have been undertaken. However, this decentralization has not included the revenue side and instead concentrates on expenditures. Moreover, the process has been anarchic and has responded to political pressures and not to efﬁciency purposes (Hernández 1998). The way SNG debt is regulated perhaps provides one of the most important explanations for its behavior (Ter-Minnasian 1999). For this reason, we now explain the Mexican case in more detail. First, SNG borrowing is regulated by the national constitution, which speciﬁes that states can only borrow in

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

49

pesos and solely for productive investment. The details for guaranteeing state credits are contained in the National Fiscal Coordination Law, which stipulates that these entities can borrow from commercial and development banks and by writing bonds to ﬁnance investment projects, subject to the previous authorization of the state congress. Before the tequila crisis of 1994–95, when a unique political party dominated the country, SNG debt was virtually decided by the federal government in a unilateral manner by direct control of state governments (Díaz 2003). Later, as a consequence of the rapid democratization of the country, this control ended. The new situation allowed states to take advantage of the federal government’s concerns about both the banking system—nearly bankrupt as a result of the tequila crisis—and states’ abilities to deliver public services (Hernández 1998). Bailouts were common before the tequila crisis, though the largest in Mexican history was extended in 1995.8 As a consequence, virtually no commercial bank developed an institutional capacity to assess subnational lending. When the tequila crisis erupted, most states had high debt ratios and federal bailout occurred. To correct the situation, the Mexican federal government faced the challenge of guaranteeing that bailouts would not occur in the future. This would allegedly be solved by imposing an ex ante market-based mechanism. So a new regulatory framework for debt management by local governments was introduced in 1997.9 States and creditors were induced to make their own trust arrangements in the collateralizing of debt with the block transfers and assuming all the legal risks involved, thus providing recourse for the federal government. A link was established between the risk of bank loans to SNGs and government credit rating. Currently, credit ratings performed by reputable international agencies are published on a global scale. Bank regulators use these ratings to assign capital risk weight for loans provided to states and to municipalities. To control 8. For a review of bailout events in Mexico, see Hernandez, Díaz, and Gamboa (2002). 9. Firms or governments benefit from obtaining a good rating by lowering the cost of servicing the debt. Many studies of industrial countries have demonstrated empirically that this is generally the case, as they have gained greater acceptance in the market. Ratings also have been used in financial regulation because they simplify the task of prudential regulation (Cantor and Packer 1995). Thus, as in the Mexican case, regulators have adopted ratingsdependent rules.

50

E C O N O M I A , Fall 2009

agency shopping, two ratings are mandatory for regulation. In case of a large discrepancy, the capital risk weight of the lower rate applies. The National Securities Commission recognizes three rating agencies: Moody’s, Standard and Poor’s, and Fitch. The main purpose of the regulation is to discipline SNG debt markets, especially in a new framework characterized by the absence of federal intervention. Financially weaker states and municipalities are likely to be priced or rationed out of the market, while stronger ones would see interest rates on their loans fall (Giugalle, Korobow, and Webb 2001). Another important element in the new regulation is the registration of SNG loans with the federal government. Registration is conditioned on the borrowing state or municipality being current on the publication of its debt, the related fiscal statistics from the preceding year’s final accounts, and on all of its debt service obligations toward the government’s development banks. At the same time, in order to make that registration appealing, unregistered loans are automatically risk weighed by the regulators at 150 percent. Several elements need to be considered to ensure the success of this type of regulation, including the market credibility of the federal government’s commitment to not bail out defaulting SNGs, the quality of the enforcement of capital rules, and the quality and reliability of SNG ﬁscal information, as well as homogeneity in accounting standards. As we pointed out previously, the largest state in Mexico has been continuously bailed out in the past. Furthermore, states and municipalities currently differ in their accounting standards, and not all of them publish their ﬁnancial statements (ARegional 2004). These elements cast some doubt on the success of the new regulation. No new SNG default and therefore no bailout has occurred so far; however, unless more stringent oversight is exerted over SNGs, there is no guarantee that they will not occur in the future.

Are Mexican SNGs Opaque? SNG ﬁscal information is like a black box in Mexico, mainly due to lack of an adequate institutional and legal framework and lack of accounting standards.10 In general, rule of law in Mexico is poor (La Porta and others 1998). This 10. For example, for some municipalities the service of paving roads is registered in current expenditures, whereas for others it is treated as an investment (Hernández 1998).

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

51

T A B L E 1 . Relative Opacity Kappa Index Entity Banks Other sectors States and municipalities

United States

Mexico

0.30 0.45

0.27 0.36 0.13

Source: Morgan (2002) for U.S. values; authors’ calculations for Mexican values.

problem is greater at state and municipal levels, where transparency is nonexistent since governments are not required to publish their ﬁnancial statements (Ugalde 2003). Transparency issues should be taken into account when rating SNG bonds. Were SNGs transparent, there would be no need for a lender of last resort since fully transparent states could borrow at market rates that fairly reﬂected their risk. However, SNG transparency—and thus ﬁnancial soundness—is more a matter of faith than fact in Mexico. To discuss this point, we use Morgan’s deﬁnition of relative opacity, which is framed in terms of disagreement between the major bond rating agencies— Fitch, S&P, and Moody’s—when grading an entity and is used as a proxy for uncertainty (see Morgan 2002). The argument is this: if SNG risk is harder to observe, raters in the business of judging risk should disagree more over SNG bond issues than over other entities. As table 1 demonstrates, this is the case with Mexican SNGs. This table presents kappa statistics, which are used as a measure of disagreement in biometrics (Cohen 1968).11 Kappa essentially locates raters along a spectrum between complete disagreement (kappa = 0) and complete agreement (kappa = 1). Kappa is 0.13 for the whole set of Mexican SNGs—states and municipalities— rated by the three agencies, which suggests a strong disagreement. This ﬁgure worsens to 0.05 if only state governments are included. Some SNGs have applied only for two ratings. In this case, when the agencies are Fitch and S&P, the kappa is 0.24; when they are Fitch and Moody’s, the ﬁgure is 0.17; and ﬁnally, when they are Moody’s and S&P, the kappa indicator is 0.04. These ﬁgures suggest that SNGs are opaque according to Morgan’s deﬁnition. U.S. SNGs rated by Moody’s and Fitch have a kappa of 0.61, which suggests that these entities are not as opaque as those in Mexico. 11. Kappa = (po − pe)/(1 − pe), where po is the observed percentage of graded bonds equally, and pe is the expected percentage, given the current distribution of grades.

52

E C O N O M I A , Fall 2009

Ederington, Yawitz, and Roberts (1987) suggest three reasons for differing bond ratings. First, agencies may agree on the creditworthiness of a bond but apply different standards for a particular rating. Second, they may differ systematically in the factors they consider or the weights they attach to each factor. And third, due to the inherent subjectivity of the process, they may give different ratings for random reasons. In this article, we expect to shed some light on which of these three explanations for disparities predominates when subnational entities in Mexico are rated.

Empirical Model and Estimation A selectivity problem arises during the analysis of the determinants of SNG bond rating. This follows from the fact that ratings are observed only for those municipalities that have chosen to be rated rather than for all entities in the sample with outstanding debt. As in Moon and Stotsky (1993), we treat this self-selection problem by developing a model in which we jointly analyze the determinants of the bond rating and the determinants of the decision to obtain a rating. Due to the short history of SNG bond rating in Mexico and in order to gather enough information for our study, we collected ratings from three agencies (Moody’s, S&P, and Fitch) instead of the two (Moody’s and S&P) used by Moon and Stotsky. Although estimating a three-agency model is more challenging, it has the advantage of expanding the scope of our conclusions, as we are now able to compare the rating technologies of more agencies. Additionally, by controlling for trivariate self-selection, we can consider in the analysis not only SNGs with three ratings but also those with only one or two ratings, as well as SNGs with no ratings but with outstanding debt. We also examine jointly the determinants of the bond ratings for the three rating agencies. A joint estimation enables more efﬁcient estimates by allowing free correlation between selection and rating equations. Allowing free correlation is important since rating decisions are not necessarily independent. Recall that an SNG needs at least two ratings in order to issue a bond registered with the Mexican treasury department, and we are considering three measures of credit risk obtained by the three agencies. Thus SNG administrators may show preferences for certain agencies if they believe these agencies have a less stringent rating procedure. Additionally, an entity may have incentives to obtain more than one or two ratings if by doing so it lowers the cost of its debt. The literature shows evidence that not only value of ratings but also the number

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

53

of them inﬂuence the cost of debt (Ederington Yawitz, and Roberts 1987). Hence a multivariate framework applies.

The Model Following the discussion above, the equation system to solve is: (1)

y*s = X s␤s + ␧s

propensity to obtain S&P’s ratings

w*s = Z s ␥s + ␩s

S&P’s perceived riskiness

y*f = X f ␤f + ␧f

propensity to obtain Fitch’s ratings

w*f = Z f ␥ f + ␩ f

Fitch’s perceived riskiness

y*m = X m␤m + ␧m

propensity to obtain Moody’s rating

w*m = Zm ␥m + ␩m

Moody’s perceived riskiness,

where index k = s, f, m refers to S&P, Fitch, and Moody’s, respectively; matrices Xk and Zk are matrices of explicatory variables; and ␤k and ␥k are vectors of parameters to be estimated. The disturbance vector is assumed to be i.i.d. over entities according to the following six-dimensional normal distribution: ⎛ ⎡ 1 ⎛ ␧s,i ⎞ ⎡0⎤ ⎢ ⎜ ⎟ ⎜ ⎜ ⎢ ⎥ ⎢ρε η ⎜ ␩s,i ⎟ ⎜ ⎢0⎥ ⎢ ⎟ ⎜ ⎜ ⎢ ⎥ ⎢ρ ⎜ ␧f ,i ⎟ ⎢0⎥ ⎢ ε ε ⎟ ∼ N6 ⎜ ⎢ ⎥ , ⎢ ⎜ ⎜ ⎜ ␩ f ,i ⎟ ⎜ ⎢ 0 ⎥ ⎢ρε η ⎟ ⎜ ⎜⎢ ⎥ ⎢ ⎜␧ ⎟ m,i ⎜ ⎢ 0 ⎥ ⎢ρε ε ⎟ ⎜ ⎜ ⎢0⎥ ⎢ ⎜␩ ⎟ ⎜⎝ ⎣ ⎦ ⎢ρε η ⎝ m,i ⎠ ⎣

ρε η

ρε ε

ρε η

f

ρε ε

1

ρη ε

ρη η

f

ρη ε

s f

ρη ε

1

ρε η

f

ρε ε

s

f

ρη η

s m

s m

s s

s s

(2)

s f

s f

s f

ρε η

ρη ε

s m

ρε ε

ρη η

ρε η

s m

f

f

f m

f

s

f

f

s

s

m

s m

s m

f m

1

ρη ε

ρη ε

1

f m

f m

ρη η f

m

ρε

m ηm

ρε η ⎤ ⎞ ⎥⎟ ρη η ⎥ ⎟ ⎥⎟ ρε η ⎥⎥ ⎟ ⎟ ⎥⎟ , ρη η ⎥ ⎟ ⎥⎟ ρε η ⎥ ⎟ ⎥⎟ 1 ⎥ ⎟⎠ ⎦ s m

s m

f

m

f

m

m m

where i = 1, . . . , N and N is the sample size. Note that all observations contribute to the estimation of the correlation terms ρε ε j, k = s, f, m. However, due to self-selection, only those SNGs that have received ratings from the * is not respective agencies contribute to the estimation of terms ρε η . Variable y k,i j k

j k

54

E C O N O M I A , Fall 2009

* > 0 and takes observable, but a binary counterpart, yk,i, takes the value of 1 if yk,i the value of 0 otherwise. The observable counterpart of w*k,i is categorical ordered so that ⎧lk,1 ⎪ ⎪⎪lk,2 wk,i ⎨ ⎪ ⎪ ⎪⎩lk,r

(3)

α k,1 < wk*,i ≤ α k,2 if

α k,2 < wk*,i ≤ α k,3

,

α k,r < wk*,i ≤ α k,r +1

where lk,1 < lk,2 . . . < lk,r are consecutive integer values, αk,1 = −∞, αk,r+1 = ∞, and thresholds αk,2 < αk,3 < . . . < αk,r are extra parameters to estimate. In our analysis, we have six categories for all agencies, that is, r = 6 with lk,1 = 0 and lk,6 = 5 ∀ k (see table 2). If yk,i = 0, then wk,i does not exist, in accordance with the self-selection mechanism discussed above. Given the binary and categorical ordered nature of the observed counterparts of the dependent variables, parameter identiﬁcation requires normalization of the diagonal elements in the disturbance covariance matrix as it is presented in equation 2. Additionally, identiﬁcation of the coefﬁcients γk in the perceived riskiness equations requires either setting to zero one of the thresholds in equation 3 for each equation or setting the intercept parameter in these equations equal to zero. We chose the ﬁrst alternative and set αk,2 = 0, k = s, f, m.

Model Specification, Data, and Description of Variables In theory an entity decides to obtain a credit rating because it expects to save enough interest costs to outweigh the agency fee. Thus the level of debt may T A B L E 2 . Equivalence between Ordinal and Qualitative Rates Qualitative rating by institution Ordinal rate 0 1 2 3 4 5 Source: Authors’ determinations.

S&P

Fitch

Moody’s

AA+, AA AA− A+ A A− BB+, BB−

AA AA− A+ A A−, A3 BBB+, BBB

Aa2 Aa3 A1 A2 A3 Baa1, Bba1

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

55

be a good determinant of the propensity to be rated since the higher the debt, the greater the savings in interest costs (see Moon and Stotsky 1993). Likewise, as in most previous literature, we include the total revenue of the entity, as it may represent a good proxy for the local income tax base, and it allows controlling for the size of the entity in terms of economic importance. If large municipalities, in terms of population, perceive that they will be bailed out, they will then have strong incentives to be rated and obtain debt. We use population as a proxy for size importance in political terms, after controlling for economic size, since Hernández, Díaz, and Gamboa (2002) have shown that in the past more populated entities have been bailed out more favorably than less populated ones. This variable also has been discussed in the U.S. case. Wallis (1998, 2001) and Fleck (2001) maintain a debate about the political motive of using population during New Deal transfers to states. Since we use a log speciﬁcation, the inclusion of total revenue and population rules out the possibility of adding revenue per capita as a regressor to avoid perfect collinearity. Nonetheless, during the estimation process, we tried using total revenue and revenue per capita alternatively; we detected neither signiﬁcant qualitative nor quantitative differences in the ﬁnal results except, of course, in the coefﬁcients of the two regressors. Finally, we control for political party, hypothesizing that the left-wing party has either less ﬁnancial culture or dismisses market-based approaches with respect to obtaining debt. Thus dummies for the main political parties were included in the propensity equation. Regarding the risk assessment equations, the major categories include political factors, some indicators of ﬁnancial soundness including contingent liabilities, indicators of debt level, and economic indicators such as gross state product and its composition. Next we describe the variables considered in our analysis. Again, population size in political terms is a variable that may affect rating behavior, in two ways in particular. First, as previously mentioned, political decisionmaking varies with the size of population. Hernández, Díaz, and Gamboa (2002) have shown that this variable is a good proxy for the “too big to fail” hypothesis for state bailouts. In this sense, the larger the entity, the higher the number of political votes it has. Second, population is important as a measure of tax base in Mexico. This may be different in advanced economies where smaller municipalities tend to be mostly residential, while larger municipalities tend to have a more substantial industrial base and a more diverse population. In contrast, in LDCs—and Mexico is no exception—

56

E C O N O M I A , Fall 2009

small municipalities tend to be more rural and thus less subject to being taxed. Complementarily, to control for economic size, we include the entity’s total revenues. This is necessary since there may be municipalities that are small in terms of population but large in terms of economic importance.12 We include a dummy with the value one when the political party in control of the municipal government is the same as that controlling the federal government, and with the value zero otherwise. As already mentioned, we expect that raters assign a greater probability of bailout to entities that share political afﬁnity with the central government, which therefore translates to a better risk grade. For ﬁnancial soundness, we choose several variables previously used in the literature (see, for example, Ederington, Yawitz, and Roberts 1987; Cantor and Packer 1996). We include the ratio of an entity’s own revenues to total revenue for two reasons. First, it reﬂects the ﬂexibility an entity has to absorb a shock; and second, federal transfers to total revenue reﬂect how compromised the transfer is beforehand. With respect to debt, we use debt-to-income ratio. Mexican law requires that all new debt must be used in public investment. Thus one would expect that higher levels of ﬁscal responsibility imply larger amounts of investment; for this reason, we also include the investment-tototal-expenditure ratio. Regarding the functional form assumed for the model, we follow Moon and Stotsky (1993) and use the log form of all the continuous regressors. The data set contains information from 149 urban municipalities for the year 2001, 148 municipalities for the year 2002, and 147 municipalities for the year 2003.13 Descriptive statistics for the data are presented in table 3. We obtain the ﬁnancial and political variables from the Municipal Information System of the National Institute of Statistics (2003).

Estimation Approach In contrast to the selection problems involving continuous or limited dependent response variables, the responses to the debt risk according to the different agencies are not observed at all in our problem. All that we know about these responses is a discrete ordinal manifestation in the ratings. Thus a selectivity12. The typical example of this in Mexico is San Pedro in the state of Nuevo León. 13. Remember that the regulation is biased toward the largest 150 municipalities in Mexico, and that the remaining municipalities were virtually excluded from credit markets, as argued above (see Hernández 1998).

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

57

T A B L E 3 . Descriptive Statistics of Dependent and Explanatory Variables Binary dependent variables S&P Fitch Moody’s

Sum

Entity rated by S&P in the period 2001–03 (yes = 1) Entity rated by Fitch in the period 2001–03 (yes = 1) Entity rated by Moody’s in the period 2001–03 (yes = 1)

96 74 40

Dummy explanatory variables PRD PRI COA PAN A

Sum

Entity administered by the Partido Revolucionario Democrático (yes = 1) Entity administered by the Partido Revolucionario Institutional (yes = 1) Entity administered by a coalition party (yes = 1) Entity administered by the Partido Acción Nacional (yes = 1) Entity administered by the same party as federal government (yes = 1)

Continuous explanatory variables a POP TI O_T D_I Debt P_D I_G

45 148 60 191 191 Mean

5

2000 population (×10 ) Total annual income (U.S.$ × 108) Own-to-total revenue ratio Debt-to-income ratio Total debt (U.S.$ × 106) Per capita debt (U.S.$ × 103) Investment-to-total-expenditure ratio

3.3 25.7 0.23 0.12 20.8 0.58 0.23

Std. deviation 3.1 31.7 0.12 0.17 44.8 0.90 0.13

Source: National Institute of Statistics (2003). a. The log10 form of the continuous explanatory variables was used in the estimation.

corrected Heckman-type estimator cannot be calculated since least squares cannot be applied on an unobserved variable in the second stage of Heckman’s procedure. Therefore we use a full information maximum likelihood (FIML) approach. It is well known that the main problem when using FIML to estimate equation systems involving latent variables is the presence of high dimensional integrals in the likelihood function, the highest possible order of integration being equal to the number of latent variables in the system. In contrast to Moon and Stotsky (1993), who use the probability simulator of Börsch-Supan and Hajivassiliou (1993), we deal with this issue by formulating a Monte Carlo expectation maximization (MCEM) algorithm. The main advantages of the MCEM approach are its robustness both to the selection of starting values and to fragile identiﬁcation (Natarajan and others 2000). To get a feel for how the MCEM method works, consider the following many-to-one mapping, z ∈ Z → y = y(z) ∈ Y. In other words, z is only known to lie in Z(y), and the subset of Z is determined by the equation y = y(z), where y is the observed data variables yk and wk in our case, and z is the unobserved

58

E C O N O M I A , Fall 2009

information, our y*k and w *k variables. Thus the complete data is x = (y, z), and the log-likelihood of the observed information is (4)

( )

( )

( )

ᐉ θ y = ln L θ y = ln ∫Z y L θ x dz. ()

Hence the multidimensional integration problem appears when we try to exclude the unobserved information by integration. Instead of trying to solve equation 4 directly, the expectation maximization (EM) algorithm focuses on the complete information log-likelihood ᐉc(θ冷x), and maximizes E[ᐉc(θ冷x)] by executing two steps iteratively (Dempster, Laird, and Rubin 1977). The ﬁrst one is the so-called expectation step (E-step), which computes Q(θ冷θ(m), y) = E[ᐉc(θ冷x)] at iteration m+1. The term E[ᐉc (θ冷x)] is the expectation of the complete information log-likelihood conditional on the observed information, provided that the conditional density f (x冷 y, θ(m)) is known. The E-step is followed by the maximization step (M-step), which maximizes Q(θ冷θ(m), y) to ﬁnd θ(m+1). Then the procedure is repeated until convergence is attained. The Monte Carlo version of the EM algorithm avoids troublesome computations in the E-step by imputing the unobserved information by Gibbs sampling (Casella and George 1992), conditional on what is observed and on distribution assumptions. In this approach, the term Q(θ冷θ(m), y) is approx1 K imated by the mean ∑ Q(θ, z(k)冷 y), where the z(k) are random samples from K k =1 f(x冷θ(m), y). The formulation of an MCEM algorithm for estimating equation system 1 is presented in appendix A, and the information matrix was obtained using Louis’s identity (Louis 1982; see appendix B).

Determinants of Credit Ratings Determinants of the Rating Propensity Estimation results for the whole set of parameters in the model are given in tables 4 and 5. We dropped the dummy representing the left-wing political party, Partido Revolucionario Democrático (PRD), from the regression in order to compare the impact of political orientation on the propensity to be rated. Tables 6 and 7 show the marginal effects of the explanatory variables on propensity-to-be-rated and rating equations, respectively. As it is well known, direct discussion of parameter estimates can be misleading in nonlinear models

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

59

T A B L E 4 . Model Estimates a S&P

Fitch

Moody’s

Equation

Variable

Estimate

Std. error

Estimate

Std. error

Estimate

Std. error

Propensity to be rated

Constant PRI Coalition PAN POP TI O_T D_I

−1.4967*** 0.3496 0.5976** 0.8437*** 1.8514*** −0.0251 1.1772*** −0.0072

0.3798 0.2497 0.2941 0.2611 0.3395 0.2197 0.2541 0.0573

−1.0241*** 0.4515** 0.3982 0.9766*** 1.6517*** −0.4905** 0.7546*** 0.1878***

0.3944 0.2033 0.2815 0.2011 0.3202 0.2361 0.2413 0.0512

−4.9557*** 3.2767*** 3.4165*** 3.4630*** 1.0679** 0.0638 −0.0828 0.2601***

0.4660 0.4196 0.4330 0.4176 0.4313 0.2566 0.2801 0.0633

Rating

Constant A POP TI O_T D_I I_G

0.9318 −0.7958*** −2.3515*** 0.2692 −2.5806*** 0.1255 −1.0191**

0.8961 0.2494 0.6811 0.3831 0.8021 0.0925 0.4320

0.8772 −0.7496** −1.7098* 0.1023 −2.8562*** 0.0378 −1.2384**

1.0615 0.3714 0.9723 0.5397 0.8788 0.1409 0.5175

1.9660 −1.9033*** −1.5152 0.4503 −2.3790*** 0.1500 −1.5604**

1.3718 0.3383 1.8990 1.0746 0.9092 0.1873 0.6599

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. Test for model significance (restricted model: slopes are all zero) chi-squared = 710.4315, gl = 39 (p < 0.01). For abbreviations, see table 3.

T A B L E 5 . Thresholds and Covariance Matrix S&P

Fitch

Moody’s

Thresholds

Estimate

αk,3 αk,4 αk,5 αk,6

0.5030*** 1.4025*** 2.0564*** 2.9212***

Covariance matrix

Estimate

Std. error

Covariance matrix

Estimate

Std. error

−0.2675 0.6536*** −0.0745 0.2840*** −0.4171 0.1288 0.6351*** −0.1191

0.1898 0.0387 0.3518 0.0635 0.4097 0.2232 0.1291 0.1832

ρηsηm ρ⑀f ηf ρ⑀f ⑀m ρ⑀f ηm ρηf ⑀m ρηf ηm ρ⑀mηm

0.7370*** −0.1022 0.0442 −0.1963 −0.0662 0.6897*** 0.0218

0.1229 0.5946 0.0778 0.4274 0.1986 0.2519 0.3303

ρ⑀ η ρ⑀ ⑀ f ρ⑀ ηf ρ⑀ ⑀m ρ⑀ ηm ρη ⑀f ρη ηf ρ⑀ ηm s s s s s s

s s

m

Std. error

Estimate

Std. error

0.1242 0.1205 0.1146 0.2337

0.8303*** 1.8727*** 2.4665*** 3.5228***

0.1732 0.1200 0.1385 0.2850

***Statistically significant at the 1 percent level.

Estimate 2.6975*** 3.0433*** 3.8627*** 4.5533***

Std. error 0.3094 0.2239 0.3222 0.2924

E C O N O M I A , Fall 2009

60

T A B L E 6 . Marginal Effects for the Propensity-to-Be-Rated Equationsa S&P

Fitch

Moody’s

Variable

Estimate

Std. error

Estimate

Std. error

Estimate

Std. error

PRI Coalition PAN POP TI O_T D_I

0.0553 0.1052** 0.1633*** 0.3873*** −0.0053 0.2462 −0.0015***

0.0354 0.0482 0.0400 0.0703 0.0460 0.0542 0.0120

0.0611** 0.0521 0.1771*** 0.3378*** −0.1003** 0.1543*** 0.0384***

0.0249 0.0387 0.0290 0.0658 0.0485 0.0508 0.0107

0.0815*** 0.1021*** 0.1097*** 0.1498** 0.0090 −0.0116 0.0365***

0.0164 0.0288 0.0168 0.0633 0.0360 0.0393 0.0094

**Statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. For abbreviations, see table 3.

since they measure the impact of the regressors on latent dependent variables, which might have an intuitive meaning but not a deﬁnite one (Greene 2000). Therefore we focus our discussion on marginal effects, which estimate the effect of regressors on the observed counterparts of the dependent variables for the sample under study. For the particular case of the propensity-to-be-rated equation, the marginal effect accounts for the change in the probability that an entity requests to be rated as a result of a change in the respective regressor. Marginal effects are calculated for each observation; we report sample averages and standard errors calculated by the delta method. It turns out that political orientation is important. As observed, the propensity to request a rate increases as with the shift from the left- to the right-wing preferences. Thus it is the Partido Acción Nacional (PAN), the rightist party, that demonstrates the highest propensity. According to table 6, ceteris paribus, a municipality governed by the PAN shows a probability to be rated by S&P 16 percentage points (pp) higher than one ruled by the PRD, the leftist party. This ﬁgure climbs to approximately 18 pp for Fitch and decreases to 11 pp for Moody’s. This result indicates that entities governed by the PAN are the most willing to obtain a grade, a ﬁnding that makes sense since the PAN is associated with local entrepreneurs, a group with more ﬁnancial culture than other constituencies (Cabrero 2004). Another signiﬁcant variable that explains propensity to be rated is municipality size, measured in population terms. According to table 6, if municipality A has twice the population of municipality B, then the probability that A asks for the services of S&P would be about 11 pp higher than it would be for B.14 14. To get this ﬁgure, multiply the corresponding marginal effect by log10 (2) ≈ 0.3.

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

61

The respective ﬁgures for Fitch and Moody’s are 10 pp and 4 pp, all of them signiﬁcant at the usual levels of signiﬁcance. It can be noted that, aside from political preferences, population is the most important variable in explaining the decision to be rated. This suggests the ex ante existence of a self-selection mechanism, where larger municipalities select themselves into the rating process. Regarding ﬁnancial factors, ratio of own to total revenue is important among those that choose S&P and Fitch, while ratio of debt to total income is important among those that choose Fitch and Moody’s.

Determinants of the Rating Overall, the estimates support the arguments presented in this article, namely, that population, political afﬁnity with the federal government, the ratio of own to total revenues, and the investment variable inﬂuence the grade positively. Coefﬁcient signs are negative because we assign a lower risk to higher grades (see table 2). Conversely, the ratio of debt to total income affects the grade negatively. It can be seen that political variables are important for rating agencies. On the one hand, as discussed earlier, population size is important probably because it is politically more costly not to rescue a large entity. On the other hand, the high signiﬁcance of the dummy for political afﬁnity is evidence that raters allocate a higher rate to those entities having a higher bailout probability, that is, entities administrated by the party holding federal ofﬁce.15 For a discussion based on probabilities, table 7 presents the marginal effects of regressors on the probabilities of receiving a given grade 0 to 5 as described in table 2, conditional on the SNG requesting a rating. The regressors that provide statistically significant marginal effects in the S&P rating equation are political affinity, population, ratio of own to total revenue, and ratio of debt to total revenue. Marginal effects for population indicate that a rise in population size shifts the probability distribution from lower to higher grades. In particular, a 10 percent average rise in population brings a 2 pp average increase in the probability of receiving an AA or AA+ grade and 0.4 pp increase in the probability of receiving an AA− from S&P, with simultaneous reductions of 1.0 pp and 0.8 pp in the probability of being rated with A− or BB, respectively. Although an increase in population favors 15. Although, for the period covered in our analysis, no changes in the federal government occurred, local governments did change. This happens because federal elections may take place at different dates than municipal ones.

62

E C O N O M I A , Fall 2009

T A B L E 7 . Marginal Effects for the Rating Equationsa S&P Variable TI

O_T

D_I

I_G

Rate 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5

b

Fitch

Moody’s

Estimate

Std. error

Estimate

Std. error

Estimate

Std. error

−0.0627 −0.0149 0.0000 0.0209 0.0318 0.0250 0.5662*** 0.1315*** −0.0055 −0.1885** −0.2831*** −0.2207** −0.0294 −0.0070 0 0.0098 0.0149 0.0118 0.2409** 0.0577** 0.0005 −0.0801** −0.1224* −0.0966*

0.0904 0.0216 0.0061 0.0302 0.0461 0.0372 0.1826 0.0483 0.0547 0.0750 0.1098 0.1051 0.0216 0.0052 0.0028 0.0074 0.0115 0.0097 0.0987 0.0281 0.0232 0.0390 0.0626 0.0526

−0.0072 −0.0079 −0.0053 0.0026 0.0094 0.0084 0.2955*** 0.3337*** 0.2398** −0.0999 −0.3961*** −0.3730** −0.0053 −0.0061 −0.0046 0.0017 0.0073 0.0071 0.1305** 0.1476** 0.1063* −0.0440 −0.1751** −0.1652*

0.0470 0.0530 0.0374 0.0165 0.0627 0.0583 0.0991 0.1175 0.1102 0.0655 0.1494 0.1482 0.0126 0.0143 0.0106 0.0044 0.0170 0.0162 0.0620 0.0650 0.0610 0.0285 0.0867 0.0866

−0.0178 −0.0829 0.0002 0.0102 0.0172 0.0732 0.0943 0.4389*** −0.0011 −0.0537 −0.0911 −0.3873*** −0.0058 −0.0269 0.0001 0.0033 0.0056 0.0237 0.0619 0.2880** −0.0007 −0.0353 −0.0598 −0.2542*

0.0456 0.1928 0.0044 0.0248 0.0393 0.1749 0.0689 0.1583 0.0224 0.0542 0.0756 0.1488 0.0093 0.0343 0.0014 0.0063 0.0078 0.0299 0.0386 0.1439 0.0145 0.0311 0.0565 0.1317

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. For abbreviations, see table 3. b. For equivalence between ordinal and qualitative rates, see table 2.

the probability of obtaining a better grade from Fitch as well, the changes in the distribution of that probability differ from S&P. Thus a 10 percent increase in population implies a 0.9 pp reduction in the probability of getting an A− or BB rating and, similarly, 0.6 pp increase in the probability of obtaining an A+, AA−, or AA+. In other words, changes in population tend to have a more uniform impact across the rates for Fitch, while for S&P they tend to affect the lowest and highest rates preferentially. On the other hand, population tested not signiﬁcant in the Moody’s rating equation. Regarding political afﬁnity, entities governed by the same party as the executive branch have a 14 pp higher probability of getting an AA+ rating, and a 6 pp and 8 pp lower probability of obtaining an A− and BB rating, respectively, from S&P than those governed by a different party. Corresponding results

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

63

for Fitch are 6 pp, 9 pp, and 10 pp, and for Moody’s, 3 pp, 14 pp, and 36 pp. Again, although the impacts of these determinants may seem to affect agencies in a similar qualitative way, their effects differ quantitatively as they relocate rate probabilities differently across agencies (see table 7).

Opacity We have already demonstrated that agencies seem to take into account the same group of variables when constructing a grade. However, this condition is not sufﬁcient to ensure that different agencies will grant the same grade to a single municipality. Agencies might consider the same factors, but they could weight them differently. In what follows, we test for SNG opacity by examining whether raters weight the factors in the same way when constructing a grade. Direct examination of the sample indicates that among those entities rated by both S&P and Fitch, in only 60 percent of the cases did the two agencies grant the same grade to a particular entity. The proportion is smaller, 44 percent, among those rated by Fitch and Moody’s, and only 38 percent among those rated by both S&P and Moody’s. In order to perform a statistical test to detect weighting differences across raters, we compare the marginal effects obtained for the rating equations. Three Wald tests comparing the estimates of the rating agencies by pairs showed high statistical differences between S&P and Moody’s chi squared value— 148.1, p < 0.01—and between Fitch and Moody’s chi squared value—78.80, p < 0.01. Smaller but still signiﬁcant differences were detected between S&P and Fitch’s chi squared: 19.98, p < 0.05. Overall, these results indicate that raters weigh factors differently during their rating process, which implies there is a high likelihood that they generate different rates even for the same municipality. This is consistent with the kappa analysis presented earlier.

Is There a Violation of “Sovereign Ceiling” Rule? Figures 1 and 2 show histograms for the differences between SNG and sovereign ratings for S&P and Moody’s, respectively. The differences were calculated as sovereign grade minus local grade.16 As it may be noted, grades ran counter to the sovereign ceiling rule. Some have argued that this damages 16. In contrast to table 2, where we use a 0–5 rating scale, for illustration purposes we use a 0–7 scale to construct these graphs.

F I G U R E 1 . S&P Distribution of Sovereign Ceilings, 2001–04a No. of times 30 Sovereign Ceiling 25 20 15 10

5 0

-6

-5

-4

-3

-2 -1 0 1 2 Sovereign grades minus local grade

3

4

5

6

Source: Authors’ calculations. a. In contrast to table 2, where we use a 0–5 rating scale, for illustration purposes we use a 0–7 scale to construct these graphs.

F I G U R E 2 . Moody’s Distribution of Sovereign Ceilings, 2001–04a No. of times 14 Sovereign Ceiling 12 10 8 6 4 2 0

-6

-5

-4

-3

-2 -1 0 1 2 Sovereign grades minus local grade

3

4

5

Source: Authors’ calculations. a. In contrast to table 2, where we use a 0–5 rating scale, for illustration purposes we use a 0–7 scale to construct these graphs.

6

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

65

the credibility of raters since no domestic ﬁrm should get a better rate than its government (see Durbin and Ng 2005). Our econometric analysis showed that bailout probability is heavily weighted in the construction of SNGs’ rates. On this basis, we would have expected the sovereign ceiling rule to hold for our sample since, in a country with a bailout tradition, the lender of last resort is the federal government. Therefore, the failure to conform to the sovereign ceiling rule, as illustrated in ﬁgures 1 and 2, is surprising. This contradiction indicates a disconnection between how SNG and sovereign rates are generated, something that should not happen in a country with a long history of bailout. In our opinion, this disconnect challenges the credibility of a market-based regulation. In the introduction of this article, we mentioned that bond raters have been under scrutiny, especially after the crises in the nineties. Additionally, we noted that the operation of this market in less developed countries has not been studied, despite the fact that some doubts about its performance have been expressed (see the Chinese example at the beginning of this paper). Our results suggest that the puzzling grades often observed in LDCs could be explained if one considers not only financial factors in the construction of a rating but also political issues. We have proved that in a country with a history of bailout and opacity, such as Mexico, political variables become important in explaining the grade assigned to the debt of subnational governments, a fact that may undermine the credibility of a market-based regulatory framework.

Conclusions In this paper, we studied both the determinants of the decision to be rated and the ratings for SNG debt in Mexico, a prominent LDC. One of the main ﬁndings is that not only ﬁnancial but also political factors matter. We showed that population size is a strong determinant of debt rating. In a country with a long bailout history, this result supports our “too big to fail” hypothesis. First, large entities select themselves to be rated and so to obtain new debt because they know that they have political power; second, raters know that the probability that the federal government will bail out large entities is high. We also showed that political closeness between local and federal governments is important: rating agencies give a better grade to those entities being governed by the same party as the national executive branch. These outcomes challenge the purpose of rating subnational debt in LDCs with a bailout tradition, since the market may

E C O N O M I A , Fall 2009

66

assess the risk of these entities as equivalent or superior to that of sovereign instruments. Mexico has implemented new legislation for the SNG debt market, with the goal of increasing the transparency of the market and ruling out debt bailouts. According to our results, which show a high relevance of the bailout probability on ratings, it seems that bond rating agencies are not yet convinced of the success of such legislation. It is apparent that ratings methodologies take time to evolve, and, for the Mexican case at least, they continue echoing the market opacity and bailout tradition of the country. Mexican regulation in this sense needs to be revised to foster its credibility.

Appendix A. The MCEM Algorithm Let y be a matrix containing all the observed information. The complete information log-likelihood function for the six-equation system discussed in the text is standard and can be written as the sum of the contributions from eight different regimes. The regimes are represented by the subsample receiving no grading; the potential three subsamples being graded by a single agency k = s, f, or m; the potential three subsamples being graded by two agencies; and the subsample receiving grades from the all three agencies. The corresponding contributions from the j = 1, . . . , 8 regimes to the likelihood are —regime j = 1: ym,i = ys,i = yf,i = 0

)

(

ᐉcj ␪j , ⍀j y = −

3n j 2

)

ln ( 2 π −

nj 2

ln ⍀j −

1 ⎛ −1 ⎞ tr ⎜ ⍀j ∑ ␧ji ␧′ji ⎟ ⎠ 2 ⎝ i

—regimes j = 2: ym,i = 1; ys,i = yf,i = 0; j = 3: ys,i = 1; ym,i = yf,i = 0; and j = 4: yf,i = 1; ym,i = ys,i = 0

(

)

)

ᐉcj ␪j , ⍀j y = −2n j ln ( 2 π −

nj 2

ln ⍀j −

1 ⎛ −1 ⎞ tr ⍀ ∑ ␧ ␧′ 2 ⎜⎝ j i ji ji ⎟⎠

j = 2, 3, 4

—regimes j = 5: ym,i = ys,i = 1; yf,i = 0; j = 6: ym,i = yf,i = 1; ys,i = 0; and j = 7: yf,i = ys,i = 1; ym,i = 0

(

)

ᐉcj ␪j , ⍀j y = −

5n j 2

)

ln ( 2 π −

nj 2

ln ⍀j −

1 ⎛ −1 ⎞ tr ⍀ ∑ ␧ ␧′ 2 ⎜⎝ j i ji ji ⎟⎠

j = 5, 6, 7

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

67

—regime j = 8: ym,i = ys,i = yf,i = 1

)

(

)

ᐉcj ␪j , ⍀j y = −3n j ln ( 2 π −

nj 2

ln ⍀j −

1 ⎛ −1 ⎞ tr ⎜ ⍀j ∑ ␧ji ␧′ji ⎟ ⎠ 2 ⎝ i

j = 8.

Thus

(

)

ᐉc ␪, ⍀ y =

(5)

∑ ᐉcj (␪ j , ⍀j 8

j =1

)

y,

where ␪ = (␤m ␥m ␤s ␥s ␤f ␥f)′, ␪j contains the components of ␪ present in the equations solved for entities in regime j, ⍀j is the covariance matrix of the disturbance terms associated to those equations j, nj is the number of observations in regime j, and ∑ n j = N, the sample size. j

E-Step The expectation of expression 5 above, conditional on observed information and distribution assumptions, can be written as ⎡3 ⎤ 5 E ⎡ ᐉc ␪j , ⍀j y ⎤ = − ⎢ n1 + 2 ( n2 + n3 + n4 + ( n5 + n6 + n7 + 3n8 ⎥ ln ( 2 π ⎣ ⎦ 2 ⎣2 ⎦ 1 1 ⎛ −1 ⎞ − ∑ n j ln ⍀j − ∑ tr ⎜ ⍀j ∑ E ⎡⎣␧ji ␧′ji ⎤⎦ ⎟ . ⎠ 2 j 2 j ⎝ i

)

(

)

)

The E-step at iteration m + 1 requires the calculation of

(

)

(m (m Q ji ␪ ␪( m ) , ⍀j ) , y = E ⎡␧ji ␧′ji ␪( m ) , ⍀j ) , y ⎤ ⎦⎥ ⎣⎢

⎛ μ(m) y ⎜ (m) ⎜ μw ⎜ (m) μy 2( m ) = σ ji + ⎜ (m ) ⎜μ ⎜ wm ⎜ μ(y ) ⎜ (m) ⎜⎝ μ w * m,i

* m ,i

* s ,i

* s ,i

* f ,i

* f ,i

− X m ,i␤m ⎞ ⎛ μ y ) ⎟ ⎜ (m − Z m ,i ␥m ⎟ ⎜ μ w ) ⎟ ⎜ (m − X s,i␤s ⎟ ⎜ μ y ) (m − Z s,i ␥s ⎟ ⎜ μ w ) ⎟⎜ m ( − X f ,i␤f ⎟ ⎜ μ y ) ⎟ ⎜ (m − Z f ,i ␥f ⎟⎠ ⎜⎝ μ w ) (m

* m,i

* m,i

* s,i

* s,i

* f ,i

* f ,i

′ − X m ,i␤m ⎞ ⎟ − Z m ,i ␥m ⎟ ⎟ − X s,i␤s ⎟ , − Z s,i ␥s ⎟ ⎟ − X f ,i␤f ⎟ ⎟ − Z f ,i ␥f ⎟⎠

)

68

E C O N O M I A , Fall 2009

(m) (m) * (m), ⍀k(m), y] k = s, where σ 2(m) = Cov(ym,i , . . . , wf,i 冷␪(m) ji j , ⍀ j , y), μ y * = E[ y k,i 冷␪ (m) (m) (m) f, m, μ w* = E[w *k,i 冷␪ , ⍀ k , y] k = s, f, m. The elements in Qji associated to equations not solved by entities in regime j must be set equal to zero. k,i

k,i

The Gibbs Sampler Gibbs sampling (Casella and George 1992) is necessary to simulate the nonobserved information present in the matrices Qji. The sampler requires the distribution of each y*k,i and w *k,i conditional on the values of the rest of the dependent variables in the corresponding regime. It is well known that these distributions are univariate normal when the unconditional multivariate distribution is normal. Let the means and variances of the conditional 2(m) 2(m) (m) distributions at the m + 1 iteration be μ(m) y* 冷 (−y* ), σ y*冷 (−y *), μ w* 冷 (−w* ), and σ w*冷 (−w*), respectively, where 冷 (−y* ) indicates conditionality on the values of all the other dependent variables (apart from y*k,i) being present in the regime at which entity i belongs. * must be done conditional on its corresponding observed Simulations for yk,i information yk,i. The observed counterpart of y*k,i is dichotomous with y*k,i being positive if yk,i equals one and nonpositive if yk,i equals zero. Accordingly, we simulate y*k,i from a normal distribution with mean μ(m) y* 冷(−y* ) and variance σ y2(m) truncated below at zero if y equals one and truncated above at *冷(−y*) k,i zero if yk,i equals zero. The observed counterparts of variables w*k,i are cate* gorical ordered and deﬁned by equation 3. Correspondingly, we simulate wk,i 2(m) from a normal distribution with mean μ(m) , and variance σ truncated w* 冷 (−w* ) w*冷 (−w *) above at αk,t+1 and truncated below at αk,t when wk,i equals lk,t (k = s, f, m; t = 1, . . . , r). A complete set of starting values y*k,i(0) and w*k,i(0) is required to initiate the Gibbs sampler. We use y*k,i(0) = 0 ∀k, i and w*k,i(0) = wk,i. The simulation is then repeated iteratively until completing sequences y*k,i(1), . . . , y*k,i(K(m)) and w*k,i(1), . . . , w*k,i(K(m)), where K(m) is a number large enough to ensure convergence. Wei and Tanner (1990) recommend starting with a small K(1) and progressively increasing K(m) as m increases. Then eliminate a number kburn of simulations from the beginning of the sequence. The remaining simulations in the sequence are (m) used to estimate the terms σji2(m), μ(m) y* , and μ w* in Qji. k,i

k,i

k

k

k,i

k,i

k

k

k,i

k

k,i

k,i

k

k

k

k,i

k,i

k,i

k,i

M-Step Following Meng and Rubin (1993), it is advisable to replace the M-step by two conditional M-steps. The ﬁrst conditional M-step maximizes E[ᐉc(␪, ⍀冷y)]

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

69

with respect to the elements in ␪ conditional on ␪(m) and ⍀(m). After a little matrix calculus, it is easy to see that the maximizer in this ﬁrst conditional maximization can be written as a generalized least squares estimator −1

⎡ ⎡ −1 ⊗ I j ⎤ X ⎤ X ′ ⎡ ⍀ −1 ⊗ I j ⎤ ␮(m ) , ␪(m +1) = ⎢ X d′ ⎢ ∑ ⍀ ⎥ d⎥ ⎥ y* j j d ⎢∑ ⎣

⎣

j

)⎦

(

⎣

⎦

j

)⎦

(

where I j is a N × N diagonal matrix with Iiij = 1 if entity i belongs to regime j ˜ j−1 contains the elements of ⍀ j−1 in and I iij = 0 otherwise. The 6×6 matrix ⍀ the positions corresponding to the equations solved in regime j, while the remaining elements must be set equal to zero. The block diagonal matrix Xd is deﬁned as ⎡Xm ⎢ ⎢0 Xd = ⎢ ⎢ ⎢ ⎢⎣ 0

(

0

0 ⎤ ⎥ 0 ⎥ m m m m ⎥ ,wheree Z k,i = 0 if yk,i = 0; ␮(y* ) ␮(y* ) ␮(z* ) … ␮(z* ) 0 ⎥ ⎥ Z f ⎥⎦

Zm

0

(

␮ y* ) = ␮ y* ) … ␮ y* ) … ␮ y* ) (m k

(m

k ,1

(m

k ,i

(m

k, N

m

m

f

) , ␮( ) = (␮( ) … ␮( ) … μ( ) ) T

m

m

m

m

z*k

z*k ,1

z*k ,i

z*k , N

T

), T

and ␮ z* ) (m

k ,i

= 0 if yk,i = 0. The second conditional M-step estimates ⍀ (m+1) by maximizing E[ᐉc(␪, ⍀冷y)] with respect to the elements in ⍀ conditional on ␪(m+1) and ⍀ (m). No closed form for ⍀ (m+1) exists; thus numerical optimization techniques must be used at this stage. Thresholds αk,3 < . . . < αk,r are not present in the complete information likelihood function; therefore they cannot be obtained by ﬁrst order condition or by numerical optimization. We proceed the following way to estimate αk,t. First, at every round of the Gibbs sampler at iteration m, keep the minimum value of every sequence obtained when simulating the observations wk,i = lk,t; this produces a set of K(m) − kburn values. Second, keep the maximum value of every sequence obtained when simulating the observations wk,i = lk,t−1. Third, calculate the medians of the two sets obtained in the preceding two steps. Finally, take the average between the two medians, which produces a consistent estimator of αk,t. The E-step and M-step are then repeated until convergence is attained.

70

E C O N O M I A , Fall 2009

Appendix B. The Information Matrix Louis’s identity (Louis 1982) was used in this study to obtain a Monte Carlo estimation of the information matrix ⎡ ⎤ ⎡ ⎤ I ( ␪; y = − Hc ( ␪; x − E ⎢ Sc ( ␪; x Sc ( ␪; x ′ ⎥ + E ⎡⎣ Sc ( ␪; x ⎤⎦ E ⎢ Sc ( ␪; x ′ ⎥ ⎣ ⎦ ⎣ ⎦

)

where Hc (␪; x) =

)

∂2 ᐉc ( ␪; x

)

)

)

)

) and S (␪; x) = ∂ᐉ (␪; x ) are the complete inforc

c

∂␪∂␪′ ∂␪ mation Hessian and score vector, respectively. All of the expectations are estimated at the final MCEM estimators. Monte Carlo estimates of the complete information Hessian and score vectors can be used to estimate the information matrix (Ibrahim, Chen, and Lipsitz 2001). Since thresholds αk,t are not present in the complete information maximum likelihood, their standard errors cannot be obtained from the information matrix presented above. Following Albert and Chib (1993), we consider that estimates of αk,t are uniformly distributed between the two medians calculated in the third step above in appendix A when estimating αk,t . Thus standard errors for our estimates of αk,t were calculated as the square roots of the variances of those distributions.

Comments Eduardo Cavallo: This paper by Hernández-Trillo and Smith-Ramírez makes an important contribution to the credit rating literature. While a lot of attention has been paid recently to developments in the sovereign and corporate ratings markets amid the global ﬁnancial crisis of 2008–09, a lot less is known about the workings of credit ratings for subnational entities (SNGs), particularly in developing countries. Mexico provides an interesting case in point. The Mexican government introduced legislation in 1997 mandating the use of credit ratings for SNG debt as part of a broader overhaul of the regulatory framework for debt management by municipal governments. The main purpose of the legislation, as the authors explain, was to promote market discipline in SNG debt markets. A key element of the process was the commitment on the part of the federal government not to bail out SNGs that enter into solvency problems. Otherwise, in the presence of federal bailout guarantees, credit ratings assessing the likelihood of default by SNGs lack value. In light of the aforementioned, a key question today is whether the commitment by the federal government is credible. In other words, if politically important SNGs face solvency problems, will the federal government really stay on the sidelines, letting these municipalities default on their obligations? Given that there have been no SNG defaults (and therefore bailouts) since the introduction of the legislation, the level of commitment has not yet been tested. This notwithstanding, the authors evaluate whether the commitment appears to be credible to an important group of market actors: the credit rating agencies. The idea, in a nutshell, is to investigate what characteristics predict the probability that an SNG will opt into the rating process, and, conditional on being rated, what the factors are that rating agencies consider in determining the actual rating (that is, the perceived likelihood of default). The authors claim

71

72

E C O N O M I A , Fall 2009

that if the bailout guarantee is still implicit due to a noncredible government promise, then two things should be observed in the data: —that those SNGs that opt into the rating process are those that are more likely to be bailed out, and —that conditional on being rated, a higher rating should be assigned to SNGs that are more politically powerful. A key assumption is that political factors are good proxies for federal bailout likelihood. In particular, the claim is that bigger municipalities (in terms of population size) as well as municipalities that are ruled by the same party that runs the federal government are more likely to be bailed out. While I ﬁnd these to be entirely plausible assumptions, it would be helpful to see some evidence. That would entail a precise deﬁnition of what the authors mean by “bailout” and some historical correlations between bailouts and the aforementioned political factors. The results reported in the paper are consistent with these two hypotheses. In particular, they ﬁnd that politically powerful municipalities (measured by population size) and municipalities ruled by the same party as the national government (the conservative PAN for the time frame used in this study) are more likely to select themselves into the rating process. At the same time, political variables (population size and afﬁnity of political parties) are important determinants of the rating that SNGs obtain conditional on being rated. These results are also novel in the credit rating literature. One idea that is embedded in the analysis is that more politically powerful SNGs (and thus SNGs that are more likely to be bailed out) receive better ratings. On this issue, I have a different interpretation and would thereby propose an alternative dependent variable. In my view, if an SNG is perceived to be “too big” or “too powerful” to fail, then the rating agencies should assign to it the same rating as that of the sovereign, which is the agent that is ultimately assuming the credit risk. Thus, rather than (or in addition to) regressing the actual rating of the SNG on a series of determinants (including political factors) as the authors do, I would instead regress the difference between the sovereign and the SNG ratings against the same determinants. At the very least, this would be a useful robustness check. If this interpretation is right, and, in the presence of bailout guarantees, rating agencies are more likely to assign SNGs the same rating as the federal government, then one characteristic prevalent in these markets would be the presence of a “sovereign ﬂoor”—meaning that no SNG that is likely to be bailed out is granted a rating that is worse than the federal government. Indeed, the evidence reported in the revised version of the paper is consistent with this

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

73

view, as in very few instances in the case of Standard and Poor’s, and in no instances in the case of Moody’s, is an SNG granted a rating below that of the sovereign.1 In my view, this fact should be documented as it stands in sharp contrast to the evidence in favor of a “sovereign ceiling” prevalent for foreign currency debt in corporate debt markets of developing countries (see, for example, Durbin and Ng 2005; Borensztein, Cowan, and Valenzuela 2006; Grandes and Peter 2005; Cavallo and Valenzuela, forthcoming).2 In turn, the connection would work as a bridge between this paper and an important strand of the broader credit ratings literature. Another result reported in the paper is that the three main rating agencies (Standard and Poor’s, Fitch, and Moody’s) appear to give different weights to the factors used in assessing SNGs’ creditworthiness in Mexico. The authors interpret this as evidence of “opaqueness” in the ﬁnancial information reported by Mexican SNGs (see Morgan 2002). The idea is that since the financial information of Mexican SNGs has so little transparency, there is more scope for disagreement among the rating agencies, even if they were looking at the same data. While this interpretation is certainly possible, I would introduce two caveats. On the one hand, the result that rating agencies disagree more than they agree is neither new nor idiosyncratic to this market. For example, Powell and Martínez (2007) and Cavallo, Powell, and Rigobón (2008) show that even in relatively plain vanilla markets such as the sovereign rating market, credit rating agencies disagree more than they agree on actual ratings. In that case, relative “opacity” is less of a problem, and the disagreements most likely reﬂect the use of different information sets by the rating agencies. On the other hand, the authors do not directly control in the regressions for “opacity” of the data at the SNG level, nor do they include an SNG ﬁxed-effect, thereby making the interpretation of this result hard to disentangle. The last point brings me to probably my main critique of the paper, which is the lack of robustness checks and alternative model speciﬁcations. The authors run one set of baseline regressions and do not report results based on alternative speciﬁcations. In the absence of an underlying structural model that determines a single speciﬁcation to be tested empirically, more robustness checks are warranted. For example, how robust are the results to the inclusion or exclusion of certain variables? What about different dependent variables, as mentioned 1. While for some reason the authors do not report the results for Fitch, they should be reported, too, or else an explanation be provided as to why such results have been omitted. 2. The idea behind the “sovereign ceiling” policy is that the risk of a government encountering difﬁculties in servicing its debts can be mitigated by its ability to “transfer” such problems to the domestic private sector via, for example, taxation or capital controls.

74

E C O N O M I A , Fall 2009

before? This critique should not detract from the merits of the paper, but it is intended to encourage the authors to probe deeper into the results. One area for future research along the lines pioneered in this paper would be to evaluate the impact of the bailout guarantees on the perceived riskiness (that is, either credit ratings or actual bond spreads for tradable debt instruments) of the bailout agent. In the case of Mexico, this appears to be particularly relevant, as the country has been granted investment grade status by the rating agencies. The question that arises is whether the bailout guarantees that the authors claim to be implicit in the Mexican case are not so onerous to the federal government or, instead, whether the ﬁnancial situation of Mexico has improved so much in recent years that it warrants investment grade status despite the onerous guarantees. Yet another possibility that would be in direct contradiction to the authors’ results is that the guarantees are not really there. In any event, pursuing research along these lines would foster greater understanding on the workings of the SNG debt markets. Tito Cordella: The interesting paper by Hernández-Trillo and Smith-Ramirez (HTSR hereafter) sheds new light on the Mexican experience of developing a market-friendly regulatory framework for subnational government (SNG) debt. The idea of asking bank regulators to use credit ratings to assign capital risk weights for loans provided to states and municipalities was innovative and created strong incentives for SNGs to be rated. However, the paper, which studies the factors that jointly determine SNGs’ decisions to be rated (as well as the rating score), raises signiﬁcant doubts about the effectiveness of such a model in effecting market discipline. This “negative” result may serve as a lesson for other countries in the region on what one can reasonably expect (and not) from rating agencies in the presence of implicit guarantees. Rating agencies are not very popular these days, to say the least. The common wisdom is that they misbehaved in the buildup to the subprime disaster, and that it is very likely they would misbehave in the very same way in the future. The problem is that rating agencies’ incentives to misrepresent risk follow directly from the way the industry is structured. Indeed, incentive problems are difﬁcult to overcome as long as rating agencies’ fees are paid by the “rated” entities. While HTSR do not address this important issue directly, they provide some evidence that the common wisdom may not be so distant from the truth. The main ﬁndings of this paper are that SNG ratings are opaque because agencies disagree on the rating of states and municipalities, and that, in explaining the ratings, political afﬁnity between subnational and federal government,

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

75

the size of the state (measured by population), the ratio of own-to-total revenue, and investment variables are the most important components. The ﬁrst result of the paper is a very interesting one that, in my view, the authors could have exploited much more. What really caught my attention was the fact that rating agencies disagree more in rating states than municipalities. This is puzzling. If “opaqueness” is driven by the murkiness of information, shouldn’t one expect information to be murkier at the municipal than at the state level? Perhaps that opaqueness might not be the result of different information sets but rather be an equilibrium outcome in which rating agencies implicitly collude by specializing in determined market segments. Assume that rating agencies compete for clients and reputation and that each rating agency decides to be more lenient toward SNGs with particular characteristics; also assume that in so doing it attains a competitive edge in acquiring this type of client. I imagine that one can build up a model in which this kind of “diversiﬁcation” is an equilibrium outcome. Is this what happens in Mexico? While the paper does not address this question directly, the data seem to support such a view. Indeed, S&P’s rating rules assign greater weight to population size (a too-big-to-fail variable) and to the ratio of own-to-total revenues than do the rating rules of the other agencies. Therefore it is no surprise that states with large populations and larger ratio of own revenues to total revenues are more likely to go to S&P than to other rating agencies. This is a very interesting story, and it is somehow unfortunate that many of the variables that enter into the propensity-to-rate equations do not enter into the rating ones. This means that it is difﬁcult to assess whether the specialization story really holds true. This could be an interesting topic for future research in which one could also look at international comparisons between rating agencies’ behavior and see whether the patterns of specialization are global or just local. Regarding the credibility of the regulatory framework, the paper is quite convincing about the fact that the Mexican market-friendly approach has not been able to remove the perception of the existence of an implicit bailout guarantee. Such an implicit guarantee is reflected in the credit rating as (in my opinion) it should be. However, a few additional checks could be useful in further convincing the reader. For instance, I would be interested in knowing how many bailouts occurred at the municipal and state level before and after the reform; whether there have been cases in which a state or municipality defaulted and was not bailed out; which risk weights have been assigned to states or municipalities that have not been rated; and how sensitive risks weights and interest rates are to rating scores. It would also

76

E C O N O M I A , Fall 2009

be interesting to know whether risk weights changed over time or remained pretty stable. Last but not least, the authors show that in a few instances, SNGs are rated above the sovereign ceiling, which is really puzzling in a situation in which subnational ratings mainly reﬂect bailout probabilities. Unfortunately, the puzzle is not solved, and the reader remains curious about which states and municipalities are (at least apparently) doing so well, and why. It would have helped if the authors at least named these states and municipalities and showed whether such “great” performers are the same for S&P and Moody’s. Summing up, this is an interesting paper that, exploiting a nice dataset, sheds new light on the role that rating agencies play in fostering market discipline. As is often the case for interesting papers, it also raises a number of questions that one hopes will be answered by new research, either from the authors or from some of our Economia readers.

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

77

References Albert, J., and S. Chib. 1993. “Bayesian Analysis of Binary and Polychotomous Response Data.” Journal of the American Statistical Association 88 (42): 669–79. Alfonso, A. 2003. “Understanding the Determinants of Sovereign Debt Ratings: Evidence for the Two Leading Agencies.” Journal of Economics and Finance 27 (1): 56–74. ARegional. 2004. Transparencia en estados y municipios. Mexico City. Bevilaqua, Afonso S., and Marcio G. P. Garcia. 2002. “Banks, Domestic Debt, and Crises: The Recent Brazilian Experience.” Brazilian Journal of Political Economy 22 (4): 85–103. Bhatia, Ashok Vir. 2002. “Sovereign Credit Ratings Methodology: An Evaluation.” Working Paper 02/170. Washington: International Monetary Fund. Black, Thomas. 2003. “The State of Mexico Is Virtually Bankrupt” (www.bloom berg.com [August 13]). Borensztein, E., K. Cowan, and P. Valenzuela. 2006. “Sovereign Ceiling ‘Lite’? The Impact of Sovereign Ratings on Corporate Ratings in Emerging Market Economies.” Working Paper 07/75. Washington: International Monetary Fund. Börsch-Supan, A., and V. Hajivassiliou. 1993. “Smooth Unbiased Multivariate Probability Simulators for Maximum Likelihood Estimation of Limited Dependent Variable Models.” Journal of Econometrics 58 (3): 347–68. Cabrero, Enrique. 2004. Experiencias en innovaciones locales en México. Mexico City: Editorial Miguel Angel Porrúa. Cantor, R., and F. Packer. 1995. “The Credit Rating Industry.” Journal of Fixed Income 5 (3): 10–34. ———. 1996. “Multiple Ratings and Credit Standards: Differences of Opinion in the Credit Rating Industry.” Federal Reserve Bank of New York Staff Reports (12): 1–41. Carleton, W. D., and Lerner, E. M. 1969. “Statistical Credit Scoring of Municipal Bonds.” Journal of Money, Credit and Banking 1 (4): 750–64. Casella, G., and George, E. 1992. “Explaining the Gibbs Sampler.” American Statistician 46 (3): 167–74. Cavallo, E., A. Powell, and R. Rigobón. 2008. “Do Credit Ratings Agencies Add Value? Evidence from the Sovereign Rating Business Institutions.” Working Paper 647. Washington: Inter-American Development Bank. Cavallo, E. A., and P. Valenzuela. Forthcoming. “The Determinants of Corporate Spreads in Emerging Markets: An Option-Adjusted Spreads Analysis.” International Journal of Finance and Economics. Cohen, J. 1968. “Weighed Kappa: Nominal Scale Agreement with Provision for Scaled Disagreement or Partial Credit.” Psychological Bulletin 70 (4): 213–20. Dempster, A. P., N. M. Laird, and D. B. Rubin. 1977. “Maximum Likelihood Estimation from Incomplete Observations.” Journal of the Royal Statistical Society: Series B 39 (1): 1–38.

78

E C O N O M I A , Fall 2009

Díaz, A. 2003. “Balance ﬁscal de los estados en el sistema federal.” In De la descentralización al federalismo: Estudios comparados sobre el gobierno local, edited by A. Díaz and J. Martínez, pp. 17–50. Mexico City: Editorial Miguel Angel Porrua. ———. 2006. Federalism, Fiscal Authority and Centralization in Latin America. Cambridge University Press. Durbin, Erik, and David Ng. 2005. “The Sovereign Ceiling and Emerging Market Corporate Bond Spreads.” Journal of International Money and Finance 24 (4): 631–49. Ederington, L. H., J. B. Yawitz, and B. E. Roberts. 1987. “The Information Content of Bond Ratings.” Journal of Financial Research 10 (3): 211–26. Fleck, Robert K. 2001. “Comment: Population, Land, Economic Conditions, and the Allocation of New Deal Spending.” Explorations in Economic History 38 (4): 296–304. Giugalle, M., A. Korobow, and S. Webb. 2001. “A New Model for Market-Based Regulation of Subnational Borrowing: The Mexican Approach.” Mimeo. Washington: World Bank. Grandes, M., and M. Peter. 2005. “How Important Is Sovereign Risk in Determining Corporate Default Premia? The Case of South Africa.” Working Paper 05/217. Washington: International Monetary Fund. Greene, William H. 2000. Econometric Analysis. 4th ed. Upper Saddle River, N.J.: Prentice Hall. Hernández, F. 1998. “Federalismo ﬁscal en Mexico: ¿Como vamos?” Revista Internacional de Finanzas Pu´ blicas, special issue: 97–121. Hernández, F., A. Díaz, and R. Gamboa. 2002. “Bailing Out States in Mexico: Causes and Determinants.” Eastern Economic Journal 28 (3): 365–80. Ibrahim, J. G., M. Chen, and S. R. Lipsitz. 2001. “Missing Responses in Generalized Linear Mixed Models When the Missing Data Mechanism Is Non-ignorable.” Biometrika 88 (2): 551–64. Inter-American Development Bank. 1997. Latin America after a Decade of Reforms. Washington. La Porta, R., and others 1998. “Law and Finance.” Journal of Political Economy 106 (6): 1113–55. Laulajainen, R. 1999. “Subnational Credit Ratings—Penetrating the Cultural Haze.” GeoJournal 47 (4): 501–10. Londero, Elio. 2005. “ ‘Market Discipline,’ Lending Ceilings and Subnational Finance.” Kyklos 58 (4): 575–90. Louis, T. A. 1982. “Finding the Observed Information Matrix When Using the EM Algorithm.” Journal of the Royal Statistical Society: Series B, no. 44: 226–33. Meng, X., and D. Rubin. 1993. “Maximum Likelihood Estimation via the ECM Algorithm: A General Framework.” Biometrika 80 (2): 267–78. Millon, M., and A. Thakor. 1985. “Moral Hazard and Information Sharing: A Model of Financial Information Gathering Agencies.” Journal of Finance 40 (5): 1403–22.

Fausto Hernández-Trillo and Ricardo Smith-Ramírez

79

Moon, C. G., and G. Stotsky. 1993. “Testing the Differences between the Determinants of Moody’s and Standard and Poor’s Ratings: An Application of Smooth Simulated Maximum Likelihood Estimation.” Journal of Applied Econometrics 8 (1): 51–69. Morgan, D. 2002. “Rating Banks: Risk and Uncertainty in an Opaque Industry.” American Economic Review 92 (4): 874–88. Natarajan, R., C. E. McCulloch, and N. Kiefer. 2000. “A Monte Carlo EM Method for Estimating Multinomial Probit Models.” Computational Statistics and Data Analysis 34 (1): 33–50. National Institute of Statistics. 2003. “Sistema de información municipal” (www.inegi. org.mx). Powell, A., and J. Martínez. 2007. “On Emerging Economy Sovereign Spreads and Ratings.” Working Paper 629. Washington: Inter-American Development Bank. Reinhart, C. 2001. “Sovereign Credit Ratings before and after Financial Crises.” Mimeo. Project on the Role of Credit Rating Agencies in the International Economy. New York University–University of Maryland. ———. 2002. “Default, Currency Crises and Sovereign Credit Ratings.” Mimeo. Washington: International Monetary Fund. Remolona E., M. Scatigna, and E. Wu. 2008. “A Ratings-Based Approach to Measuring Sovereign Risk.” International Journal of Finance and Economics 13 (1): 26–39. Sanguinetti, Pablo, and others. 2002. “Decentralization, Fiscal Discipline in Sub-National Governments and the Bailout Problem: The Case of Argentina.” Research Paper R-467. Washington: Inter-American Development Bank. Ter-Minnasian, T. 1999. Fiscal Federalism. Washington: World Bank. Ugalde, L. C. 2003. “Descentralización y corrupción en gobiernos locales: Una correlación relevante?” In Descentralización, federalismo y planeación regional en México, edited by R. Tamayo and F. Hérnandez, pp. 253–58. Mexico City: Editorial Miguel Angel Porrúa. Wall Street Journal. 2004. “Credit Ratings in China Can Be Mere Guesswork.” January 5, p. C1. Wallis, John Joseph. 1998. “The Political Economy of New Deal Spending, Revisited, with and without Nevada.” Explorations in Economic History 35 (2): 140–70. ———. 2001. “The Political Economy of New Deal Spending, Yet Again: A Reply to Fleco.” Explorations in Economic History 38 (2): 305–14. Wei, C., and M. Tanner. 1990. “A Monte Carlo Implementation of the EM Algorithm and the Poor Man’s Data Augmentation Algorithms.” Journal of the American Statistical Association 85 (411): 699–704.

GRACIELA KAMINSKY AMINE MATI NADA CHOUEIRI

Thirty Years of Currency Crises in Argentina: External Shocks or Domestic Fragility? rgentina has had an active presence in international capital markets since its independence in the early nineteenth century. However, its participation has been quite volatile. In the early 1800s, in the midst of the lending boom fueled by the end of the Napoleonic wars, Argentina and many other countries in Latin America were able to issue bonds in London to finance their wars of independence and the civil wars that followed. This lending boom ended in the summer of 1825 when the Bank of England raised the discount rate to stop the drain in its reserves. The tightening of liquidity was followed by stock market crashes, banking problems, and recessions in England and on the Continent. Within months the crisis also spread to Latin America. Argentina defaulted in 1827, in the midst of what is known as the first Latin American debt crisis, only resuming payments in 1857. Similar international capital flow booms to emerging markets occurred in 1867–72, 1880–90, 1893–1913, and 1920–29, fueled by an easing monetary stance in the financial centers of those times and by the financial needs of railway expansion, urbanization, and development of the banking sector of countries in the periphery. While Argentina was heavily involved in all these capital flow bonanzas, its participation was quite volatile, with financial crises often following booms.1

A

Kaminsky is with George Washington University and the National Bureau for Economic Research, and Mati and Choueiri are with the International Monetary Fund. We thank Gastón Gelos, Roberto Rigobon, Federico Sturzenegger, Carlos Winograd, and participants at the Economia meeting for helpful comments. The views expressed herein are those of the authors and should not be attributed to the International Monetary Fund, its executive board, or its management. 1. For example, the boom of the 1880s ended with banking and currency crises as well as a sovereign default, while the end of the capital inflow episode of the 1920s led to Argentina’s abandonment of the gold standard. See Kaminsky (2009) for an analysis of Latin America’s participation in international capital markets from independence to the Great Depression.

81

82

E C O N O M I A , Fall 2009

In the aftermath of the crisis of the 1930s, international capital markets all but disappeared, and Argentina was unable to borrow again until the 1970s. The period from the mid-1970s to 2002 was as tumultuous as that of the earlier eras and was characterized by booms and busts in international capital flows, crises, and failed stabilization programs. During this period, Argentina had eight currency crises, four banking crises, and two sovereign defaults. Many argued that domestic fragilities were at the heart of these crises.2 Others blamed erratic international capital markets by pointing out the lending boom of the late 1970s that ended with defaults across all Latin American countries, or the lending cycle of the 1990s that triggered banking and currency crises in the most active participants in international capital markets, such as Argentina, Brazil, Colombia, Mexico, Peru, and Venezuela.3 This important debate is still unsettled. Now, in the midst of the worst international financial crisis since the Great Depression, untangling the roots of financial distress becomes crucial. This is the question we plan to examine in this paper. We focus on Argentina’s currency crises of the last thirty years and cast our net wide to examine the role of three external shocks and four sources of domestic vulnerability in the development of currency turmoil. Our selection of external shocks centers on the easing and tightening of monetary policy in the world financial centers, financial contagion and overall “international investors’ sentiment” about emerging markets, and real exchange rate misalignments caused by currency depreciations among Argentina’s major trading partners. With respect to domestic vulnerabilities, we focus on the boom-bust cycle of domestic credit and monetary policy, fiscal problems, shocks to economic activity, and increases in households’ risk aversion triggered by spells of hyperinflation, controls on foreign exchange transactions, cycles of controls on prices and wages, and bank deposit confiscations that have plagued Argentina’s recent history. To capture the onset of the crises and track the buildup of fragility during fixed exchange rate regimes, we look at the evolution of foreign exchange reserves of the central bank as a proportion of domestic credit. For short periods of time in the early 1970s and late 1980s, Argentina adopted a dual exchange rate regime, with a fixed exchange rate for commercial transactions and a freely floating exchange rate for capital account transactions. For these episodes, the onset of a crisis is captured by an index of exchange market pressure, which is constructed as a composite index of losses of reserves of the central bank and the dual exchange market premium. 2. See, for example, Mussa (2002) and Perry and Servén (2002). 3. See, for example, Calvo, Leiderman, and Reinhart (1992, 1996).

Graciela Kaminsky, Amine Mati, and Nada Choueiri

83

Structural Vector Autoregression (VAR) techniques are used to identify the effects of domestic and external shocks on the onset of the crises. The next section presents a chronology of Argentina’s currency crises since 1970. This is followed by the presentation of a basic model to underpin the VAR specification. The third section discusses the estimation, presents the data, and examines key empirical results, and the final section summarizes our findings.

Chronology of the Currency Crises4 During most of the post–World War II period, Argentina experienced chronic inflation. Many stabilization programs using the exchange rate as an anchor were launched in the belief that with fixed exchange rates, domestic inflation would converge quickly toward world levels. These programs also included plans for fiscal and monetary austerity (although, in most cases, they were later abandoned). All the programs ended up with currency crises. In addition to failed stabilization attempts, global external factors also contributed to the general instability of the domestic currency. Declining interest rates in the industrialized world fueled capital flows to developing countries in the late 1970s and in the 1990s, and while these capital flow bonanzas are generally considered beneficial to emerging markets, they also trigger real exchange rate appreciations and current account deficits, which often lead to currency crises. Furthermore, these flows are prone to quick reversals whenever monetary policy in the center economies switches to a contractionary stance. Also, fragilities in the domestic financial system as well as forced conversions of deposits were another potential cause of runs against the Argentine peso. Thus our chronology of crises will highlight the evolution of the different stabilization programs implemented in this period as well as the role of world shocks and financial vulnerabilities. To help in our crisis chronology, figure 1 shows the evolution of the central bank’s foreign exchange reserves and the dual market premium from January 1970 to January 2002, the month of the onset of the last crisis. The dates of the currency crises are indicated by the vertical lines. It is clear from figure 1 that 4. This chronology is partly based on Blejer and Liviatan (1987), Cumby and van Wijnbergen (1989), D’Amato, Grubisic, and Powell (1997), De la Torre, Levy Yeyati, and Schmukler (2002), Di Tella and Dornbusch (1989), Dornbusch and de Pablo (1989), Edwards (2002a and 2002b), Giorgio and Sagari (1996), Hausman and Velasco (2002), International Monetary Fund (2004a and 2004b), Kiguel (1989), Montanaro (1990), and Rodriguez (1994).

E C O N O M I A , Fall 2009

84

F I G U R E 1 . Indicators of the Fragility Buildup, 1970–2002a Foreign Exchange Reserves (in Billion Dollars)

24

16

8

Jan-94

Jan-97

Jan-00

Jan-94

Jan-97

Jan-00

Jan-91

Jan-88

Jan-85

Jan-82

Jan-79

Jan-76

Jan-73

Jan-70

0

Dual Exchange Market Premium (in Percent) 400 300 200 100

Jan-91

Jan-88

Jan-85

Jan-82

Jan-79

Jan-76

Jan-73

Jan-70

0

Sources: See appendix A. a. The vertical lines indicate the month of the crises.

almost all crises were preceded by losses of reserves or by sharp increases in the dual market premium when foreign exchange controls were introduced. Table 1 reports crisis dates and the names of the stabilization programs preceding them, as well as the time during which these programs were implemented. To illustrate the severity of each crisis, table 1 shows the loss of foreign exchange reserves of the central bank in the months leading into

Graciela Kaminsky, Amine Mati, and Nada Choueiri

85

T A B L E 1 . Stabilization Plans and Crises, Argentina, 1973–2002

Crisis date

Reserve losses a (percent)

Dual market premium on the month of the crisis (percent)

Devaluation on the month of the crisis (percent)

Cumulative devaluation the first six months after crisis (percent)

March 1975 February 1981 July 1982 September 1987 April 1989 February 1990 March 1995 January 2002

56 45 17 75 62 58 41 50

369 ... ... ... 206 105 ... ...

100 10 148 16 387 220 0 40

628 136 244 133 4025 232 0 265

Stabilization plans Name Gelbard Tablita Alemann Austral Primavera BB Convertibility

Date implemented May 1973 December 1978 December 1981 June 1985 August 1988 July 1989 April 1991

Sources: See appendix A. a. For each episode, reserve losses are computed from the month the stock of reserves held by the central bank peaks until the crisis date.

the crisis, the dual exchange market premium at the onset of the crisis, and the devaluations following the crisis. All speculative attacks ended with a sharp devaluation, with the exception of the one in 1995; in this instance, the central bank managed to avoid a devaluation during the speculative attack despite a 41 percent loss in foreign exchange reserves. The first crisis, associated with the collapse of the stabilization plan implemented by the then minister of finance, José Gelbard, occurred in March 1975 after various speculative attacks that resulted in a 56 percent loss of foreign exchange reserves, even in the presence of many restrictions to free convertibility.5 At that time, the domestic currency in both the commercial and financial markets was devalued by 100 percent and 50 percent, respectively. More than a dozen additional devaluations followed over the course of the year.6 The second crisis and the collapse of the second stabilization plan (the Tablita Plan) occurred in February 1981 when a 10 percent devaluation was announced. Two other devaluations followed: 34 percent in April and 38 percent in June of that year. The Tablita Plan, launched in December 1978, was characterized by a slowly declining, preannounced rate of exchange rate depreciation (the tablita). The program also included fiscal and monetary 5. Part of the exchange rate pressures led to a sharp increase in the financial market premium, which peaked at 369 percent right before the abandonment of the program. 6. During 1975 there were several devaluations, such as the one in June 1975 when Celestino Rodrigo was the finance minister, but these devaluations are not examined separately. Consecutive devaluations less than six months apart from the first devaluation are considered part of the same crisis.

86

E C O N O M I A , Fall 2009

reforms as well as a sweeping financial liberalization plan that led to the complete deregulation of domestic banking activities and a removal of capital account restrictions. This episode coincided with the capital flow bonanza fueled by the savings of OPEC economies after the 1973 oil shock and channeled to emerging markets through the Eurodollar market. By 1980 the boom in capital inflows to Argentina had triggered an explosive growth in domestic credit and overall banking fragilities, which ended with the failure of two of the largest private banks, as well as the liquidation of almost 100 financial institutions.7 The crisis in 1981 also coincided with the reversal in international capital flows triggered by a shift toward anti-inflationary monetary policy in the United States. The third crisis took place in July 1982. Following the February 1981 crisis, a variety of programs to refinance banks and insure holders of foreign currency–denominated debt were implemented, maximum interest rates were reimposed and then abandoned, and dual rates were reintroduced from March to December 1981.8 Naturally, the continuous regulatory changes regarding interest rates and foreign exchange markets contributed to reducing investors’ already jittery confidence in the domestic currency and the banking sector. During this period, inflation continued to accelerate, in part fueled by the bailout of the banking sector. Despite the announcement of a new stabilization plan in December 1981, the so-called Alemann Plan, inflation continued to surge, fueled this time by central bank financing of massive military spending during the Malvinas war. The economy was also hit hard by many adverse external shocks: the decrease in international commodity prices, the increase in foreign interest rates, a worldwide recession, and the beginning of the world debt crisis. After a 17 percent loss in foreign exchange reserves, the crisis culminated in July 1982 with a 148 percent devaluation, the introduction of dual exchange rates with controls on domestic interest rates and the capital account, and an exchange rate that floated for the following three years. During the floating regime, inflation continued to increase, reaching 300 percent during the first half of 1985. In June 1985, a new stabilization plan, the Austral Plan, was launched. A new currency, the austral, was introduced; the dual exchange rate regime was abandoned; the domestic currency was fixed again to the dollar; and interest rate controls were eliminated.9

7. See Baliño (1987) for a detailed analysis of the banking crisis in 1980–1981. 8. See Baliño (1987) and Machinea and Fanelli (1988) for a detailed analysis of the measures adopted during this period. 9. One austral was equivalent to 1,000 pesos.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

87

The introduction of the austral was accompanied by a forced renegotiation and markdown of debt contracts and banking deposits. While inflation was contained, the annual rate of inflation was still at about 100 percent by mid1986, forcing the government to abandon the peg. The government tried to save the program with repeated rounds of enforcement and then relaxation of price controls, and with other restrictions, all with no success. In the first nine months of 1987, reserves of the central bank declined by 1.5 billion dollars (75 percent), leading to a collapse of the Austral Plan in 1987, with the domestic currency being devalued by 16 percent in September and by 33 percent in October. The next two currency crises occurred in the midst of a hyperinflation period.10 The first crisis occurred in April 1989, with a 387 percent devaluation. The second crisis erupted within eight months, with a 175 percent devaluation in December 1989 and a 220 percent devaluation in February 1990. During this period, there were two more stabilization attempts: the Primavera Plan in August 1988 and the BB Plan in July 1989. Both plans included price controls, dual exchange rates, and fiscal and monetary contraction. In both plans, monetary and fiscal restraints were rapidly abandoned, and investors’ confidence immediately deteriorated. In December 1989, the government froze most domestic austral-denominated time deposits and converted them to ten-year, dollar-denominated Bonex bonds. The value of these bonds immediately dropped to less that 30 percent of face value, weakening investors’ faith in the domestic currency. The last two currency crises were in March 1995 and in January 2002. In April 1991, the Convertibility Plan was launched. Its main feature was the creation of a currency board to enforce the one-to-one peg of the peso to the dollar.11 In addition, the plan included a series of privatization and deregulation measures as well as fiscal reforms. Also in the early 1990s, Argentina, along with other emerging markets, witnessed another round of capital inflows triggered by declining interest rates in the United States together with the 1989–1990 Brady Plan agreement for Mexico and other Latin American countries. As in the late 1970s, capital inflows led to a domestic credit explosion and to consumption, real estate, and stock market booms. The real exchange rate appreciated and the current account deteriorated. In 1994 the shift back to

10. From the collapse of the Austral Plan in September 1987 to the implementation of the Convertibilty Plan in 1991, prices in Argentina increased by 4,500. 11. The monetary reform in January 1992 replaced the austral with the peso at a rate of 10,000 australs for 1 peso.

88

E C O N O M I A , Fall 2009

a tight monetary policy in the United States (the Federal Funds interest rate was raised by 250 basis points in 1994 alone) led to worldwide interest rate increases, contributing to banking fragilities and a credit crunch amid a severe recession in Argentina. After the Mexican crisis in December 1994, Argentina’s banking system suffered from large deposit withdrawals. As investors converted pesos into dollars, the central bank’s reserves decreased sharply (41 percent in the first quarter of the year), marking the first currency crisis of the Convertibility Plan. At that time, however, the convertibility program did not end up with a devaluation of the domestic currency, and the reversal of capital flows to Argentina was only transitory. By the end of 1995, capital flows not only had resumed but even surpassed the levels reached before the Mexican crisis. Capital flows to Argentina and Latin America continued to grow even in the midst of the 1997–98 Asian crisis. Eventually, these flows started to diminish as the behavior of international capital markets changed drastically during the Russian crisis and the collapse of Long-Term Capital Management in the fall of 1998. This time around, as in the mid-1980s, the collapse in capital flows was of a more permanent nature. Argentina still fared comparably better than other countries in the region, with capital flows to Argentina still relatively high in the last half of 1999. The relief, however, was only temporary as capital flows to Argentina completely dried up in the last half of 2000 and especially in 2001. By this time, political uncertainty (President Menem’s desire to remain in power for a third term) as well as financial turmoil following Brazil’s crisis in January 1999 had severely affected private investment and consumption in Argentina, with economic activity plummeting through 2001. As the situation continued to deteriorate, the government sought more financing. When the government found it difficult to reschedule its debt, it resorted to compulsory placing of government bonds at banks, with banks becoming increasingly more exposed to government default. By June 2001, a massive bank run had started, sealing the fate of the currency board. In December the government announced a deposit freeze, foreign exchange controls, and a debt moratorium. The currency board was formally abandoned in January 2002 with a 40 percent devaluation of the peso. The convertibility regime was replaced with a dual exchange rate system based on an official exchange rate of 1.4 pesos per dollar for the public sector and most trade-related transactions while all other transactions were conducted at market rates. On February 11, the dual exchange rate was abolished, and the peso depreciated to 1.8 pesos per dollar. By June 2002, the exchange rate had reached 4 pesos per dollar.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

89

A Basic Model The numerous crises in Argentina in the last quarter of a century have stirred a heated debate about the causes behind the periodic collapses of the peg. Throughout the years, several explanations have been offered. Many argue, as it is also evident in our chronology, that at the heart of the crises are large fiscal deficits, leading to rapid growth in money creation and eventually to a depletion of reserves that makes the peg unsustainable. Another view stresses that crises erupt because of real exchange rate misalignments brought about by exchange-rate-based stabilization plans or by devaluations in neighboring countries.12 According to this view, the exchange rate misalignment eventually leads to unsustainable current account deficits and to speculative attacks against the domestic currency. Another version of the “real appreciation” theory of currency crises links the real appreciation with protracted recessions and with governments’ inability to defend the peg in bad times. For example, Drazen and Mason (1994) conclude that in the presence of persistent unemployment, a tough policy (such as one required by the commitment to the currency board in Argentina in the late 1990s) may lower rather than raise the credibility of a no-devaluation pledge and thus trigger a currency attack. The crises of the 1990s brought to the spotlight the fact that crises may be of a contagious nature. While crises could be synchronous across countries because of a common adverse shock (such as a rise in world interest rates), crises may spill over when the infected country is linked to others via trade or finance. For example, Kaminsky, Lyons, and Schmukler (2004) argue that the 1994 Mexican crisis spread to Argentina and Brazil via mutual fund withdrawals as mutual fund managers scrambled for liquidity after investors’ major redemptions from mutual funds specializing in Latin America. Similarly, Kaminsky and Reinhart (2000) conclude that the Mexican default in 1982 propagated to all Latin American countries when U.S. banks, badly damaged by the Mexican default, tried to rebalance the overall risk of their portfolios by calling loans and drying up credit lines not only in Mexico but also in all the Latin American countries where they had exposure. Calvo (1999) provides a different interpretation of the collapse of the peg, which he labels “the sudden stop” syndrome. While this view, like the aformentioned ones, acknowledges the problems of fiscal unsustainability and real

12. See, for example, Reinhart and Végh (2002).

90

E C O N O M I A , Fall 2009

exchange rate misalignments, it places strong emphasis on international financial shocks. At the core of Calvo’s explanation lies an unexpected and persistent stop in international capital flows, such as the one observed after the Russian crisis in August 1998. As explained in Calvo, Izquierdo, and Talvi (2002), the unexpected slowdown in capital flows forces emerging economies, such as Argentina, to drastically adjust their current account deficits to accommodate the shortage of external credit. Naturally, a real exchange rate adjustment becomes the essential ingredient for this adjustment to take place. With sticky prices, this adjustment can only be accomplished with a devaluation. Finally, Kiguel and Neumeyer (1995) and Ericsson and Kamin (1993), among others, have emphasized investors’ jittery confidence in Argentina’s domestic currency and the banking sector due to continuous changes in regulations on interest rates and foreign exchange markets, as well as the forced conversions of bank deposits in 1985, 1989, and 2001, as triggers of runs against the peso. This section will incorporate these features into a small open economy model, which will be estimated afterwards. As discussed earlier, the monetary authority in Argentina alternated between the adoption of fixed and dual exchange rate systems. For example, a fixed exchange rate and full convertibility for both current and capital account transactions were at the core of the Tablita Plan and the Convertibility Plan, while a dual exchange rate system was introduced during the Gelbard Plan. In most cases, when the peg collapsed, the central bank allowed the exchange rate to float for some time. Our model should reflect these changing exchange rate regimes. Thus there are two versions of the model that respectively capture the stylized features of each system.

Fixed Exchange Rate Regime The model is a discrete-time model of an open economy with a fixed and unique exchange rate. The government has a predefined goal for domestic credit, not necessarily consistent with the goal of a fixed exchange rate. Fixing the exchange rate is a secondary goal that can be abandoned if it hinders discretionary monetary policy. This assumption seems to capture quite well monetary and exchange rate policies in Argentina in the post–World War II period. Investors realize that these two goals might conflict and expect the central bank to abandon the peg when it runs out of reserves (Krugman 1979). We follow Blanco and Garber (1986) to model the onset of the crisis, but with a twist. In that paper, the authors only focus on the effect of money supply

Graciela Kaminsky, Amine Mati, and Nada Choueiri

91

shocks. Here, we extend their model to account for foreign shocks as well as other domestic shocks, such as fiscal policy. The money market is the central component of our model. Equilibrium in that market is given by the following equation, (1)

mt − pt = −α it + βct + γ yt + μ td ,

where m and p are, respectively, the logarithms of the money stock and the price level, i is the domestic interest rate, y is the logarithm of output, and µd represents money demand shocks. A negative money demand shock can capture investors’ shift out of pesos into dollars in the midst of the financial instability of the 1980s or the run against deposits due to confiscation risk in 2001. A new feature of the money demand is the component c. As we examine in more detail below, this component will try to capture shifts in international investors’ perception about emerging markets. For example, an increase in c could capture international investors’ renewed interest in emerging markets following the resolution of the debt crisis, whereas a decline in c could typify the sudden stop syndrome, such as the one triggered by the Russian crisis of August 1998, or a contagion effect, such as the reversal in capital flows following the 1994 Mexican crisis. In the open economy, interest rates and prices are determined by (2)

it = i*t + Et et +1 − et + ρt

(3)

pt = et − qt ,

where i* is the world interest rate, e is the logarithm of the nominal exchange rate, ρ is the risk premium, q is the log of the real exchange rate, and E is an expectations operator. Equation 2 allows for deviations from interest parity. Equation 3 allows for deviations from purchasing parity. In equation 3, the log of the foreign price level is normalized to zero. Money supply in the fixed exchange rate system can be written as follows: (4)

mt = d t + rt ,

where r is the ratio of foreign exchange reserves of the central bank to domestic credit in foreign currency and d is the logarithm of the domestic credit

92

E C O N O M I A , Fall 2009

component of money supply. In this simple model, changes in domestic monetary policy or in bank credit to the private sector, changes in the world interest rate, shocks to money demand, and sudden stop or contagion effects will determine the evolution of reserves of the central bank. When reserves are depleted, the central bank will not be able to intervene in the foreign exchange market any longer and will have to let the exchange rate float. Using the money market clearing conditions, we can determine the equilibrium flexible exchange rate e˜t.13 (5)

)

)

d t + α ( i t*+ ρt − βct − γ yt + qt = (1 + α et − αEt et +1 .

To obtain the time path of the permanently floating exchange rate e, ˜ we need to specify the stochastic processes that govern domestic credit, risk premium, foreign interest rates, the real exchange rate, output, and the “sentiment” of international investors towards emerging markets. (6)

d t = φgt + d t −1 + μ ts

(7)

ρt = ρ + μ tg

(8)

it* = it*−1 + μ*t

(9)

ct = ct −1 − ωit* + μ ct

(10)

gt = gt −1 + μ tg

(11)

qt = δ qt −1 − χμ st + μ qt

(12)

yt = y + λ qt + μ ty

where g is the fiscal deficit, y– is the logarithm of the full employment level of output, and µs, µg, µ*, µc, µy are shocks to money supply, fiscal policy, world interest rates, foreign investors’ sentiment towards emerging market assets 13. In a pure flexible exchange rate regime, by assumption, the stock of reserves of the central bank drops to zero.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

93

(either exuberance or sudden stop or contagion syndrome), and the shock to aggregate demand, respectively. Finally, µq captures exogenous shocks to the real exchange rate. For example, it may reflect nominal devaluations in trading partner countries, such as the Brazilian depreciation in January 1999. The shocks µ j are normally distributed white noise shocks with zero mean and standard deviation σj, for j ∈ {d, y, s, g, *, c, q}. Equation 6 represents the domestic credit process, where we allow fiscal imbalances to be (partly or totally) financed by money creation. Equation 7 captures a time-varying risk premium, with fiscal deficits triggering a higher premium.14 Equation 8 reflects the process followed by the world interest rate. Equation 9 captures investors’ interest in emerging markets. Naturally, this interest cannot just be explained by shocks to risk aversion triggered by, say, the resolution of the debt crisis in 1989. Fluctuations in interest rates in financial centers can also affect the reallocation of portfolios toward emerging economies. This is why increases in i* in equation 9 affect adversely the reallocation of portfolios toward emerging economies. In equation 10, we model fiscal policy as an exogenous process. Equation 11 models the real exchange rate as a mean reverting process. We allow the real exchange rate to be affected by monetary shocks since expansionary monetary policy in fixed exchange rate regimes will lead to higher inflation and a transitory real appreciation (see Reinhart and Végh 2002). We also allow for other exogenous shocks to the real exchange rate. With these shocks, we would like to capture the effects of a depreciation in a trading partner country, such as the effect of the devaluation of the Brazilian real in January 1999. Finally, output deviates temporarily from the full employment level with fluctuations of the real exchange rate or in response to other aggregate demand shocks. The relationship between the real exchange rate and economic activity in equation 12 is ambiguous since a real depreciation can increase competitiveness and fuel demand for domestic goods (λ > 0), but it also can lead to a contractionary effect because of liability dollarization (λ < 0) (Céspedes, Chang, and Velasco 2004). 14. In models with sovereign debt, a positive risk premium is always associated with the possibility of default. It is argued that as debt increases, it may become unsustainable, or the country may become unwilling to pay it back. These models suggest including foreign debt as an explanatory variable for risk. Unfortunately, for empirical purposes, we cannot relate the risk premium to foreign debt because debt statistics are at best only available at annual frequencies, and our estimations use monthly data. Since government deficits in Argentina have been associated with foreign borrowing, we include the fiscal indicator, for which we have monthly data, as the determinant of the premium.

E C O N O M I A , Fall 2009

94

Using equations 5–12, we obtain the equilibrium flexible exchange rate:

(13)

⎡ 1 − γλ ⎤ et = d t + α (1 + βω i*t + αρ − βct − γ y + ⎢ ⎥ qt ⎢⎣ (1 + α − αδδ ⎥⎦

)

+ αφgt −

)

1 (μ d + γ μ ty − αμ tg 1+ α t

)

The exchange rate depreciates in response to expansionary monetary shocks, fiscal deficits, and positive shocks to world interest rates; it appreciates in response to positive output shocks, increases in investors’ interest in emerging markets, and positive money demand disturbances. Finally, a real exchange rate depreciation has an ambiguous effect on the equilibrium flexible nominal exchange rate. The decline in domestic prices triggering the real depreciation leads to higher real money balances and lower domestic interest rates, which fuel a depreciation of the nominal exchange rate. But the real depreciation may stimulate economic activity and demand for money, which results in an appreciation of the equilibrium exchange rate. The peg will collapse at time t + 1 if e˜t+1 > e. Thus the time t probability of a currency collapse in the next period can be written as follows:

)

1 − F ( kt = Pr ⎡⎣ υ t +1 > kt ⎤⎦ ,

(14) where

)

υi +1 = θ1μ ig+1 + θ2μ*t +1 + (1 − θ3 μ st +1 + θ3μ qt +1 − βμ ct +1

(

− θ4 γ μ ty+1 + μ dt +1

)

)

kt = e − d t − φ (1 + α gt − θ2i*t − θ3 qt − αρ + βct + γ y and

)

)

)

F ′ ( kt > 0, θ1 = φ (1 + α + α (1 + α ;

)

θ2 = α (1 + βω + βω , θ3 =

1− γ λ 1 , θ4 = . + 1 + α (1 − δ 1 ( α

)

)

Graciela Kaminsky, Amine Mati, and Nada Choueiri

95

Knowing the distribution function of the shocks F(kt), agents can form expectations of the future exchange rate, based on the average of the current fixed exchange rate and the rate expected to materialize conditional on a devaluation, both weighed by the respective probabilities of occurrence: (15)

)

) (

)

Et et +1 = F ( kt e + ⎡⎣1 − F ( kt ⎤⎦ Et et +1 υ t +1 > kt .

After linearizing equation 15, we can solve the model in equations 1–4 and obtain the path of reserves in the fixed exchange rate system when there is a chance that there will be an abandonment of the peg: (16)

)

rt = η0 − η1 ( d t − e − η2it* + η3ct + η4 yt − η5 gt + η6 qt + μ dt .

The coefficients ηi are a function of the parameters of the distribution of the shocks and of the structural parameters of the model. Reserves will fall with expansionary monetary and fiscal policies, and with hikes to world interest rates. In contrast, a positive shock to money demand or demand for domestic goods as well as investors’ shift toward emerging markets lead to an increase in foreign exchange reserves. Shocks to the real exchange rate have an ambiguous effect on reserves. The VAR to be estimated is based on equations 6–12 and 16.

Dual Exchange Rate Regime To relieve balance of payment pressures on foreign exchange reserves, albeit temporarily, Argentina implemented dual rates in the early 1970s and in the 1980s, with a fixed exchange rate for trade account transactions and a flexible exchange rate for all other transactions. We now proceed to develop a simple model of the economy under a dual rate regime to examine the behavior of the central bank’s foreign exchange reserves and the dual market premium. The core of our model is still the money market equilibrium condition given by equation 1. Prices continue to be determined by equation 3. The interest parity condition is now written as: (2 ′)

it = it* + Et ft +1 − ft + ρt ,

E C O N O M I A , Fall 2009

96

where ft is the log of the exchange rate for nontrade account transactions. Note that in equation 2 it is assumed that the purchase and sale of assets as well as the interest rate proceeds are channeled through the nontrade account exchange rate market. Using equations 1, 2 , and 3, we can write the equilibrium condition in the money market as: (17)

)

d t + rt − e + α ( it* + ρt − βct − γ yt + qt − μ dt = αft − αEt ft +1 ,

where e is the log of the fixed exchange rate for trade account transactions. Note that reserves at the central bank can still change in response to trade account imbalances because the central bank intervenes to keep the commercial rate fixed. A persistent deficit in the trade account may deplete reserves holdings. When reserves are depleted, the central bank will not be able to intervene again in the foreign exchange market and will have to allow the commercial rate to depreciate. We assume that the foreign exchange rate market is unified after the abandonment of the peg. Naturally, investors will try to forecast as best as they can the time and the size of the devaluation. To examine the likelihood of a devaluation, we need to describe the behavior of the trade account. We assume that the trade account depends on the real exchange rate: Rt − Rt −1 = κ qt .

(18)

The probability of a unique floating exchange rate can be written as:

(19)

)

(

)

Pr ( Rt +1 ≤ 0 = Pr ⎡κ μ qt +1 − χμ st +1 ≤ − Rt − κδ qt ⎤ . ⎣ ⎦

Equation 19 indicates that a devaluation in a trading partner (a negative µq) will worsen the trade account and increase the probability of a currency crisis. Similarly, expansionary domestic monetary policy will trigger higher prices, a real appreciation of the domestic currency, and a deterioration of the trade balance. In the event of a currency crisis, the exchange rate market will be unified, with the exchange rate equal to e. ˜ Note that the expected future value of the financial exchange rate can be written as:

Graciela Kaminsky, Amine Mati, and Nada Choueiri

(20)

) (

)

97

)

Et ft +1 = Pr ( Rt +1 ≤ 0 et +1 + ⎡⎣1 − Pr ( Rt +1 ≤ 0 ⎤⎦ Et ft +1 Rt +1 > 0 .

The expected financial exchange rate is a nonlinear function of monetary and fiscal shocks, investors’ preference for emerging markets, world interest rates, output, and real exchange rate shocks. To aid in the solution, we linearize equation 20. Instead of evaluating separately the path of the financial rate and foreign exchange reserves, we follow the crisis literature and estimate an index of severity of the speculative attack by using a composite indicator tracking foreign exchange reserve losses and the dual market premium.15

(21)

(f

t

)

)

− e − ΔRt = τ 0 + τ1 ( d t − e + τ 2 it* − τ 3ct − τ 4 yt + τ 5 gt − τ 6 qt − μ dt ,

where R is the percent change in foreign exchange reserves of the central bank. In equation 21, the index of exchange market pressure increases with expansionary monetary and fiscal policy and with positive hikes in world interest rates; it decreases with positive shocks to economic activity and money demand as well as with higher investors’ interest in emerging markets. In the following section, the VAR specification that corresponds to the dual markets system is based on equations 6–12 and 21.

Explaining the Nature of Currency Crises In the following discussion, we apply the above described models to identify the nature of the shocks triggering a speculative attack. First, we review the estimation methodology, then we discuss the data, and finally we elaborate on the results.

15. In the crisis literature, the index of exchange market pressure is a composite index that incorporates reserve losses of the central bank, the rate of exchange rate depreciation, and hikes in interest rates. See, for example, Eichengreen, Rose, and Wyplosz (1996) and Kaminsky and Reinhart (2000). Here, we adapt the index to account for the buildup of pressure in the dual exchange rate market.

98

E C O N O M I A , Fall 2009

The VAR Our theoretical model implies estimating the following system: (22)

)

AX t = A ( L X t −1 + C␮ t ,

)

V (␮ t = ⌺,

where X is the vector of variables [i*, c, g, y, (d − e), q, rˆ ], rˆ is the level of foreign exchange reserves as a proportion of domestic credit during the episodes of fixed exchange rates and is a composite index of reserve losses and the dual market premium in episodes with capital account inconvertibility, and µ is the vector of the structural shocks, [µ*, µc, µg, µy, µs, µq, µd]. The theoretical framework of the preceding section provides guidelines for imposing zero restrictions on the elements of A and C. A(L) is a matrix polynomial of order n, where n is the number of lags, and C is a full rank matrix. The covariance matrix of the structural innovations is denoted by ⌺. Under the assumption of zero correlation across innovations, ⌺ is diagonal. The matrices A and C capture the contemporaneous interactions between all the variables in the system. We can now obtain the reduced form VAR representation by multiplying both sides of equation 22 by Aⴚ1: (23)

)

X t = B ( L X t −1 + ␧t .

␧ is the vector of reduced form innovations, [ε*, εc, εg, εy, εs, εq, εd]. The structural and reduced form innovations are related by the following equation: (24)

␧t = A −1C␮ t .

The identification restrictions for both the unified and the dual exchange rate models, as implied by the analysis of the previous section, can be summarized as follows: ε* = μ * ε c = γ 21μ* +μ c

Graciela Kaminsky, Amine Mati, and Nada Choueiri

εg εy εs εq

99

= = = =

μg γ 46μ q + μ y γ 53μ g + μ s γ 65μ s + μ q ε rˆ = γ 71μ* + γ 72μ c + γ 73μ tg + γ 74 μ y + γ 75μ s + γ 76μ q + μ d . Note that the parameters γ are functions of the structural parameters in the system, such as the degree of monetization of the fiscal deficit, and these may be changing over time. For example, with fixed exchange rates and capital mobility, central banks lose their ability to conduct an independent monetary policy. This is not the case with a dual exchange rate regime, making it necessary to estimate the systems for each exchange rate regime separately. Even within a particular exchange rate regime, parameters may vary. For example, the hard peg of 1991, approved by law, certainly introduced more barriers to the conduct of monetary policy than the fixed exchange rate regime implemented in the late 1970s. Again, we need to test parameter stability within a given exchange rate system.

Data Figure 2 shows the evolution of domestic and external indicators from January 1970 to December 2001, the month preceding the last crisis.16 All the indicators are at a monthly frequency, so we can track closely the onset of domestic and external vulnerabilities. The dates of the currency crises are indicated by the vertical lines. The top two panels show the evolution of monetary and fiscal factors. Domestic credit in dollars (including both credit to the public and private sector), shown in the left panel, provides a measure of possible inconsistency between the fixed exchange rate and monetary shocks. The central government deficit (annualized as a proportion of GDP), shown in the right panel, provides a measure of government debt sustainability. While a broader measure of the public sector would have been more appropriate to measure the fiscal stance, long high-frequency time series on local governments and public enterprises are not available.17 The middle panels show the effective real exchange rate (a depreciation is shown as an increase in the real exchange rate index) and the index of manufacturing production. The bottom left panel shows the behavior of the world real interest rate, captured by the U.S. real interest rate. Finally, the bottom 16. See appendix A for data sources and definitions. 17. Information on public sector debt is available, although not at a monthly frequency.

E C O N O M I A , Fall 2009

100

F I G U R E 2 . Domestic and External Indicators, January 1970–2001a Domestic Credit (in Billion Dollars)

Government Deficit (Percent of GDP) 30

80 60

15

40 0

20

Jan-97

Jan-00 Jan-00

Jan-94

Jan-97

Jan-91

Jan-88

Jan-85

Jan-82

Jan-79

Jan-76

Jan-70

Jan-00

Jan-97

Jan-94

Jan-91

Jan-88

Jan-85

Jan-82

Jan-79

Jan-76

Jan-73

Jan-70

Jan-73

-15

0

Industrial Production Index Number

Real Exchange Rate Index Number 140

450 300

100

150

Jan-94

Jan-91

Jan-88

Jan-82

Jan-79

Jan-76

Jan-85

Jan-00

Jan-97

Jan-94

Jan-91

Jan-88

Jan-85

Jan-70

Jan-00

Jan-97

Jan-94

Jan-91

Jan-88

Jan-85

-3

Jan-82

-5

Jan-79

-1

Jan-76

1

0

Jan-73

5

Jan-82

3

Jan-79

5

10

Jan-76

15

Jan-73

Latin American Foreign Exchange Reserves:b Index Number

World Real Interest Rate (Percent per Annum)

Jan-70

Jan-73

Jan-00

Jan-97

Jan-94

Jan-91

Jan-88

Jan-85

Jan-82

Jan-79

Jan-76

Jan-73

Jan-70

Jan-70

60

0

Sources: See appendix A. a. The vertical lines indicate the month of the crises. b. First principal component of the largest Latin American countries (excluding Argentina): Brazil, Chile, Colombia, Mexico, and Venezuela.

right panel shows the first principal component of foreign exchange reserves of the five largest Latin American countries (with the exception of Argentina): Brazil, Chile, Colombia, Mexico, and Venezuela.18 With this index, we try to 18. It would have been preferable to use international capital flow data to emerging markets to proxy “investors’ interest in emerging markets.” However, capital flow data are at best only available at quarterly or even annual frequencies.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

101

provide a measure of investors’ sentiments toward Latin America. Investors’ overall enthusiasm about Latin American markets translates into increases in the first principal component of foreign exchange reserves held by central banks, whereas investors’ worries about Latin America lead to losses of foreign exchange reserves across Latin American countries, again as captured by the first principal component of reserves. Since increases in foreign exchange reserves of central banks can also be affected by changes in interest rates in financial centers, we will separately identify in our estimations the effect of shocks on investors’ preferences (possibly capturing contagion effects) and world interest rate shocks on the fluctuations of the first principal component (as shown in equation 9).

Results As discussed in the chronology, we divide our sample into fixed and dual exchange rate regimes. The fixed exchange rate regimes include the Tablita, Alemann, Austral, and Convertibility Plans; the dual exchange rate regimes include the Gelbard, Primavera, and BB Plans.19 Macropolicies and credibility may vary across and within stabilization plans, affecting the transmission of shocks and making it necessary to test for parameter stability. Since periods with dual exchange rate regimes are very short, we cannot test this hypothesis. Thus we estimate a unique VAR for the Gelbard, Primavera, and BB Plans and for the currency crises that followed the implementation of those plans. Since the fixed exchange rate regime episodes are longer lasting, we can test for parameter stability during these episodes.20 In particular, three possible structural breaks are studied. We examine whether the transmission mechanism during the Tablita-AlemannAustral periods is different from that of the Convertibility Plan, and then also test for two structural breaks during the Convertibility Plan: the crisis in April 1995 (following the Mexican crisis) and the Brazilian crisis in January 1999. Our results indicate that the transmission mechanism of the Tablita-AlemannAustral periods is different from that of the Convertibility period and that the 19. Although there were never controls on foreign exchange transactions during the Tablita and Convertibility Plans, at times during the Alemann and Austral Plans the government allowed different rates for financial and commercial exchange rate transactions. Still, we include these last two episodes in our estimations of the fixed exchange rate episodes because when these plans were launched, a unique exchange rate regime was implemented. 20. To test whether VARs were different, we introduced slope dummies representing various periods into the reserves equation, with a significant slope dummy implying that transmission mechanisms were different across different periods.

102

E C O N O M I A , Fall 2009

dynamics during the Convertibility program change in the aftermath of the Brazilian crisis. Thus our results for the fixed exchange rate episodes will include three VAR systems: the first includes the Tablita, Alemann, and Austral programs; the second one refers to the Convertibility program from its implementation in April 1991 to December 1998; and the third episode, also during the Convertibility program, starts in January 1999 with the Brazilian devaluation and ends with the collapse of the currency board in January 2002. As examined above, our VARs include seven variables: the world real interest rate; the first principal component of foreign exchange reserves of Brazil, Chile, Colombia, Mexico, and Venezuela; government deficit (as a proportion of GDP); industrial production; domestic credit in dollars; the real exchange rate; and foreign reserves (or the index of exchange market pressure for the dual periods). However, in light of the unavailability of data series on industrial production during the Gelbard Plan (1973–75), the VAR for the dual exchange rate regimes only includes six variables. Although some of our variables turned out to be I(1), we estimate an unrestricted VAR in levels in order to allow the data to pick up the underlying long-run cointegrating relationship.21 We allow for two lags in all three systems, which was sufficient to produce serially uncorrelated residuals.22 The results are presented in three complementary ways. First, we examine the impulse responses to assess whether the various shocks have been identified correctly, that is, whether, for example, our “money supply” shock leads to a decline in foreign exchange reserves of the central bank, as predicted by

21. Dickey-Fuller tests failed to reject the unit root hypothesis at the 5 percent significance level for the first principal component of foreign exchange reserves of the five largest Latin American economies, foreign exchange reserves of the central bank of Argentina, and the industrial output and money variables (although there were mixed results on money variables, depending on number of lags ultimately chosen). The hypothesis was rejected for the world real interest rate, the exchange market pressure index, and the deficit. 22. The estimation was done with only one lag for the dual system due to the limited number of observations for the hyperinflation episode. We also estimated the system with one lag for the 1999–2001 unified exchange rate system. Slope dummies were introduced for hyperinflation periods and for periods in which no stabilization plan was being implemented. Based on the model’s assumptions and the significance levels of variables, we formulate the world real interest rate equation as a univariate AR(1). The equation for the first principal component of reserves of the five largest Latin American countries only includes lags of the world real interest rate in addition to lags of the principal component variable itself. Ultimately, we end up estimating a near-VAR using seemingly unrelated regression (SUR) estimation while allowing for a Sims-Bernanke decomposition of the structural innovations.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

103

the first-generation model of currency crises described earlier in this paper. Second, we report the variance decomposition of reserves during the fixed exchange rate regimes and of the index of exchange market pressure during dual exchange rate regimes to assess the importance of domestic and foreign shocks. The variance decompositions provide a measure of the average role of each shock over the whole estimation period, that is, during both tranquil and crisis times. In some cases, such as during dual exchange rate periods, the variance decompositions show the importance of each shock over various stabilization plans and crisis episodes. To untangle the role of domestic and world shocks for each stabilization program and in the unfolding of the currency turbulences for each crisis examined, we then present the historical decompositions of foreign exchange reserves and the index of market pressure for the various stabilization plans, and then estimate the role of each shock from the onset of the fragilities until the crisis month. Figure 3 shows the impulse responses of the ratio of foreign exchange reserves to domestic credit (shown in percent) to domestic and external shocks during episodes of unified exchange markets. The left panels show the impulse responses during the Tablita, Alemann, and Austral Plans; the middle panels show the impulse responses for the first part of the Convertibility Plan, from the implementation of the currency board until the Brazilian crisis; and the right panels show the impulse responses for the second part of the Convertibility Plan, from the Brazilian crisis to the Argentine crisis in January 2002. The top four rows of panels show the responses to the four types of domestic shocks. The effects of shocks to money supply, government deficit, and money demand are all statistically significant and of the expected signs, with positive shocks to money supply and government deficit and negative shocks to money demand triggering losses of reserves, with somewhat more persistent effects during the Tablita-Alemann-Austral Plans. In contrast, shocks to output only show a strong and positive effect on reserves during the second part of the Convertibility Plan. The effect of this shock on reserves is negligible during the other two episodes, as shown in table 2. The bottom three rows of panels show the responses to world shocks. Shocks to the world real interest rate are only statistically significant during the Tablita-Alemann-Austral Plans and the first part of the Convertibility Plan (until the Brazilian crisis), with hikes in world interest rates triggering losses in reserves. Since the 1990s, a continuously increasing amount of research in international finance has emphasized the role of international investors’ sentiments (or “risk appetite”) in creating capital flow bonanzas to particular regions, such as Latin America in the late 1970s, Europe in the early 1990s, or

-1

2 4 6 8 10 12 14 16 18 20 22 24

0 -1

0

2

-0.5

0

0.5

1 0.5 0 -0.5 -1

1

Money Demand

2 4 6 8 10 12 14 16 18 20 22 24

Output

2 4 6 8 10 12 14 16 18 20 22 24

Government Deficit

2 4 6 8 10 12 14 16 18 20 22 24

2 4 6 8 10 12 14 16 18 20 22 24

Money Demand

2 4 6 8 10 12 14 16 18 20 22 24

Output

2 4 6 8 10 12 14 16 18 20 22 24

Government Deficit

Convertibility Plan: April 1991-December 1998 Money Supply 1 0.5 0 -0.5 -1 2 4 6 8 10 12 14 16 18 20 22 24

1

2

2 1 0 -1 -2

1 0.5 0 -0.5 -1

-2

-1

0

1

Tablita, Alemann, and Austral Plans Money Supply

2 1 0 -1

3

3 2 1 0 -1

-2

-1

0

1

-2

-1

0

2 4 6 8 10 12 14 16 18 20 22 24

Money Demand

2 4 6 8 10 12 14 16 18 20 22 24

Output

2 4 6 8 10 12 14 16 18 20 22 24

Government Deficit

2 4 6 8 10 12 14 16 18 20 22 24

Convertibility Plan: January 1999-January 2002 Money Supply 1

F I G U R E 3 . Impulse Responses of Foreign Exchange Reserves to Various Shocks during Fixed Exchange Rate Regimesa

2 4 6 8 10 12 14 16 18 20 22 24

Real Exchange Rate

2 4 6 8 10 12 14 16 18 20 22 24

Investors’ Interest in Emerging Markets

2 4 6 8 10 12 14 16 18 20 22 24

World Real Interest Rate

-0.5

0

0.5

0

0.5

1

-1

-0.5

0

0.5

2 4 6 8 10 12 14 16 18 20 22 24

Real Exchange Rate

2 4 6 8 10 12 14 16 18 20 22 24

Investors’ Interest in Emerging Markets

2 4 6 8 10 12 14 16 18 20 22 24

World Real Interest Rate

1 0.5 0 -0.5 -1

1 0.5 0 -0.5 -1 -1.5

1 0.5 0 -0.5 -1

2 4 6 8 10 12 14 16 18 20 22 24

Real Exchange Rate

2 4 6 8 10 12 14 16 18 20 22 24

Investors’ Interest in Emerging Markets

2 4 6 8 10 12 14 16 18 20 22 24

World Real Interest Rate

Source: Authors’ calculations. a. Dotted lines indicate the one-standard deviation band; solid lines show the impulse response of reserves (as a percentage of domestic credit) to a one percent shock (domestic or external) at time one.

1 0.5 0 -0.5 -1

-1

0

1

2

1 0 -1 -2 -3

106

E C O N O M I A , Fall 2009

East Asia in the mid-1990s. This same literature also has singled out the role of international investors’ sentiments in capital flow reversals. As the impulse responses in figure 3 show, those effects, captured by shocks to the first principal component of foreign exchange reserves, are only important during the early 1990s up to the Brazilian crisis in January 1999. Finally, shocks to the real exchange rate (attempting to capture exogenous shocks, such as fluctuations in trade partners’ real exchange rates due to crises or the adoption of a stabilization plan) are never statistically significant. Figure 4 shows the dynamic responses during dual exchange rate regimes. Domestic shocks are also important during these episodes, with positive shocks to money supply and fiscal deficits and negative shocks to money demand all leading to currency turbulences, as captured by increases in the index of exchange market pressure. While the dual market system implemented in Argentina implied the use of controls on capital flows to insulate the domestic economy from world shocks, our results indicate that fluctuations in world real interest rates and shocks to investors’ interest in emerging markets have statistically significant effects on the index of exchange market pressure, with hikes in world interest rates and negative shocks to investors’ sentiment leading to currency turmoil in Argentina. As with fixed exchange rate regimes, real exchange rate shocks are never statistically significant. Tables 2 and 3 report the variance decompositions for the fixed exchange rate and the dual exchange rate episodes, respectively. As Table 2 shows, the shocks that move the currency market vary across all these episodes. Only money demand shocks are important across the whole sample (explaining between 18 and 80 percent of the conditional variance of foreign exchange reserves—as a proportion of domestic credit in dollars—at all horizons), suggesting that changes in rules as well as improvement or abandonment of property rights dramatically affect households’ behavior and are at the core of all bonanzas and crises in Argentina. Interestingly, the period starting in April 1991, with the adoption of the currency board, and ending with the Brazilian crisis in January 1999 looks different from other episodes. During the earlier part of the currency board episode, world shocks—as captured by shocks to world real interest rates and investors’ interest in emerging markets—account for about 20 to 70 percent of the conditional variance of foreign exchange reserves (as a proportion of domestic credit in dollars). In contrast, vulnerabilities in domestic indicators—fiscal deficits and shocks to economic activity—are the main drivers of reserve fluctuations during the last part of the Convertibility Plan, accounting for about 60 percent of the variance in foreign exchange reserve forecasting errors. Finally, currency booms and busts during the

Graciela Kaminsky, Amine Mati, and Nada Choueiri

107

F I G U R E 4 . Impulse Response of the Index of Market Pressure to Various Shocks during Dual Exchange Regimesa Percent

Money Supply

32 12 -8 -28 1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Government Deficit

20 10 0 -10 -20

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Money Demand

10 -10 -30 -50 1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 World Real Interest Rate

25 15 5 -5 -15 1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 (continued)

E C O N O M I A , Fall 2009

108

F I G U R E 4 . Impulse Response of the Index of Market Pressure to Various Shocks during Dual Exchange Regimesa (Continued ) Investors’ Interest in Emerging Markets

Percent 12 6 0 -6 -12 1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 Real Exchange Rate

15 5 -5 -15 1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Source: Authors’ calculations. a. Dotted lines indicate the one-standard deviation band; solid lines show the impulse response of the index of market pressure to a one percent shock (domestic or external) at time one.

Tablita, Alemann, and Austral Plans are mostly explained by world real interest rate shocks and by money supply and demand shocks (in line with the crisis chronology). Table 3 reports the variance decomposition for the index of exchange rate pressure during dual exchange rate episodes. Again, as during the fixed exchange rate episodes, money demand shocks explain a substantial part of currency market ups and downs (between 20 and 50 percent of the forecasting variance of the index of exchange market pressure for all horizons). The dual exchange rate episodes look very similar to the Tablita-Alemann-Austral Plans, with money supply shocks and world interest rates explaining about 50 percent of the forecasting variance of the index of market pressure. To track in real time the effects of the identified domestic and world shocks on currency bonanzas and crises, we present the historical decompositions of the foreign exchange reserves (as a percentage of domestic credit in dollars)

Fraction of variance due to shocks to Convertibility Plan: April 1991–December 1998

Convertibility Plan: January 1999–January 2002

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

1 2 3 3 3 3 3 3 3 3 3 3 3 3 3 2 2 2 2 2 2 2 2 2

0 0 2 5 7 9 10 10 11 11 11 11 10 10 10 10 9 9 9 9 9 8 8 8

80 80 76 70 65 59 55 50 46 43 40 38 36 34 32 31 30 29 28 27 27 26 26 26

0 1 2 5 8 11 15 18 21 24 27 29 31 33 34 35 36 37 38 38 38 39 39 39

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

6 4 4 4 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5

8 7 6 5 5 4 4 4 3 3 3 3 3 3 3 2 2 2 2 2 2 2 2 2

0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

81 77 69 61 55 50 46 43 40 38 36 34 33 31 30 29 28 27 27 26 25 25 24 24

0 3 7 11 14 17 19 21 23 24 25 25 26 26 27 27 27 27 28 28 28 28 28 28

Source: Authors’ calculations. a. This table shows the variance decomposition of reserves as a percentage of domestic credit in dollars. b. Captures responses to exogenous shocks to the real exchange rate, such as nominal devaluations in trading partner countries.

19 17 16 16 17 17 17 18 18 18 19 19 20 20 21 21 21 22 22 23 23 24 24 24

4 9 13 17 20 22 25 26 28 29 31 32 33 34 35 36 36 37 38 39 39 40 40 41

0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

0 2 3 4 4 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5

10 17 19 20 20 20 21 21 21 21 21 21 20 20 20 20 20 20 20 20 20 19 19 19

15 33 40 43 45 45 46 46 46 46 46 45 45 45 45 45 44 44 44 44 43 43 43 42

71 44 33 28 25 23 22 21 21 20 20 20 20 19 19 19 19 19 19 19 18 18 18 18

3 1 1 1 1 1 1 1 1 1 2 2 2 3 3 4 4 5 6 6 7 7 8 9

0 1 2 2 3 3 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4 4

0 1 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3

Investors’ Investors’ Investors’ World interest World interest World interest Horizon real in Real real in Real real in Real in Money Govt. Money interest emerging exchange Money Govt. Money interest emerging exchange Money Govt. Money interest emerging exchange months supply deficit Output demand rate mkts. rate b supply deficit Output demand rate mkts. rate supply deficit Output demand rate mkts. rate

Tablita, Alemann, and Austral Plans

T A B L E 2 . Variance Decomposition for Foreign Exchange Reserves during Fixed Exchange Rate Regimes a Percent

110

E C O N O M I A , Fall 2009

T A B L E 3 . Variance Decomposition for the Index of Exchange Market Pressure during Dual Market Regimes Percent Fraction of variance due to shocks to

Horizon in months 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Money supply

Government deficit

Money demand

World real interest rate

Investors’ interest in emerging markets

51 48 44 40 37 35 33 32 31 30 29 29 28 28 27 27 26 26 25 25 25 24 24 24

0 3 7 10 12 13 13 13 12 12 11 11 11 11 11 11 11 11 11 11 11 12 12 12

47 47 45 42 40 37 35 34 32 31 30 29 29 28 27 27 26 26 25 25 25 24 24 24

1 1 2 4 7 9 11 14 16 18 19 21 22 24 25 26 27 27 28 29 29 30 30 31

0 1 2 2 3 3 3 4 4 4 4 5 5 5 5 5 5 5 5 5 6 6 6 6

Real exchange rate a 0 0 1 2 3 4 4 5 5 5 5 5 5 5 5 5 5 5 5 4 4 4 4 4

Source: Authors’ calculations. a. Captures responses to exogenous shocks to the real exchange rate, such as nominal devaluations in trading partner countries.

and the index of exchange market pressure. Figures 5 and 6 show, respectively, the historical decomposition for the fixed exchange rate and the dual exchange rate episodes from the implementation of each stabilization plan until the crisis. In these figures, the solid line shows the difference between the actual value of reserves (as a percentage of domestic credit in dollars) or index of market pressure and the corresponding forecasted value based on information at the start of the stabilization plan, while the dotted lines show the part explained by either domestic or international shocks. Since the Alemann Stabilization Plan only lasted a few months, and this plan also helped maintain the fixed exchange rate regime launched with the Tablita Plan, we report the historical decomposition jointly for both plans.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

111

F I G U R E 5 . Sources of the Fluctuations in Foreign Exchange Reserves during Fixed Exchange Rate Regimesa Tablita and Alemann Plans Domestic

20

External

20

10

10

0

0

-10

-10

-20 Dec-78 Jul-79 Feb-80 Sep-80 Apr-81 Nov-81 Jun-82

-20 Dec-78 Jul-79 Feb-80 Sep-80 Apr-81 Nov-81 Jun-82

Austral Plan Domestic

20 10

10

0

0

-10

-10

-20 Jun-85

Dec-85

Jun-86

External

20

Dec-86

Jun-87

-20 Jun-85

Dec-85

Jun-86

Dec-86

Jun-87

Convertibility Plan: April 1991-December 1998 Domestic

20 10

10

0

0

-10

-10

-20 Apr-91

Jan-92

Oct-92

Jul-93

External

20

Apr-94

Jan-95

-20 Apr-91

Jan-92

Oct-92

Jul-93

Apr-94

Jan-95

Convertibility Plan: January 1999-January 2002 Domestic

20 10

10

0

0

-10

-10

-20 Jan-99

External

20

Jul-99

Jan-00

Jul-00

Jan-01

Jul-01

-20 Jan-99

Jul-99

Jan-00

Jul-00

Jan-01

Jul-01

Source: Authors’ calculations. a. In each panel, the solid line shows the difference between the level of reserves (as a share of domestic credit in dollars, in percent) and the level that would have been forecasted based upon the history of the system up through the implementation of the stabilization plan. Thus it reflects the cumulative impact of both domestic and foreign shocks. The dotted line shows the actual path of reserves that would have prevailed if either domestic or foreign shocks had hit the system.

112

E C O N O M I A , Fall 2009

F I G U R E 6 . Sources of Fluctuations in the Index of Market Pressure during Dual Exchange Rate Regimesa Gelbard Plan 300

External

Domestic 300

200

200

100

100

0

0

-100 Feb-74 May-74 Aug-74 Nov-74 Feb-75

-100 Feb-74 May-74 Aug-74 Nov-74 Feb-75

Primavera and BB Plans Domestic

External

300

300

150

150

0

0

-150 Aug-88 Dec-88

Apr-89 Aug-89 Dec-89

-150 Aug-88 Dec-88

Apr-89 Aug-89 Dec-89

Source: Authors’ calculations. a. In each panel, the solid line shows the difference between the level of the index of exchange market pressure (in percent) and the level that would have been forecasted based upon the history of the system up through the implementation of the stabilization plan. Thus it reflects the cumulative impact of both domestic and foreign shocks. The dotted line shows the actual path of the index of exchange market pressure that would have prevailed if either domestic or foreign shocks had hit the system.

The results in Figure 5 indicate that the capital flow bonanza, as captured by the increases in reserves, in the year following the implementation of the stabilization plans was mostly fueled by better domestic fundamentals across all episodes. This finding agrees with the conventional wisdom in both academic and policy circles that the launching of the stabilization plans coincided, at least transitorily, with fiscal and monetary reforms as well as deregulation of the financial sector.23 Our results for the Tablita Plan and first part of the Convertibility Plan also show that the capital flow bonanzas in the year following the implementation of the plans were triggered by favorable external conditions. We feel confident about our identification since the implementa23. See, for example, International Monetary Fund (2004a, 2004b), Blejer and Liviatan (1987), Kiguel (1989), and Machinea and Fanelli (1988).

Graciela Kaminsky, Amine Mati, and Nada Choueiri

113

tion of the Tablita Plan in the late 1970s and Convertibility Plan in the beginning of the 1990s coincided with episodes of low world real interest rates and with a surge in investors’ interest in emerging markets following the Brady debt relief program of 1989–90. The historical decompositions in figure 5 indicate that capital flow reversals and the onset of the crises in 1981–82 and 2001 were caused in part (totally for the crisis in 1987) by deteriorating domestic fundamentals. In contrast, the reversal in the path of reserves starting in 1994 was mostly due to unfavorable world conditions. Only during the four months preceding the 1995 crisis were fragilities observed on the home front, mostly driven by bank deposit runs. Figure 6 shows the historical decompositions for the stabilization plans during the dual exchange rate regimes. In this figure, we jointly examine the two stabilization plans during the hyperinflation episode. In contrast to the stabilization plans during the fixed exchange rate regimes, the implementation of the Gelbard, Primavera, and BB Plans does not trigger a reduction (even transitorily) in exchange market pressures. In the Gelbard Plan, a large part of the initial vulnerabilities is triggered by adverse external conditions, driven by hikes in world interest rates in 1974. In the stabilization plans in the late 1980s, the exchange market pressure is mostly explained by rapidly deteriorating monetary conditions. Remember that these plans take place at the height of the hyperinflation period, which only ends with the implementation of the Convertibility Plan. Figures 5 and 6 only assess the combined effect of all domestic shocks or that of external shocks. Also, the historical decompositions in these figures cover times of both bonanza and crisis. Table 4 provides a higher resolution picture of crisis times. First, it untangles the various sources of domestic fragility into money supply, government deficit, output, and money demand shocks. Second, it sorts out the origins of external vulnerability into world real interest, investors’ interest in emerging markets (or contagion), and real exchange rate shocks. Third, it concentrates on the onset of the currency turmoil until the crisis, that is, it shows the historical decompositions from the times reserves start to fall or the index of market pressure starts to increase. Each cell in this table shows the share of the fluctuations in foreign exchange reserves as a percentage of domestic credit in dollars, or, alternatively, in the index of exchange market pressure, explained by each single shock. As shown in table 4, all currency crises are preceded by domestic vulnerabilities, with the exception of the 1995 crisis. Monetary shocks are at the core of the domestic fragilities for all crises, with the exception of the 2002 crisis, when dramatic adverse shocks to economic activity seal the fate of the currency board.

114

E C O N O M I A , Fall 2009

T A B L E 4 . Role of Domestic and External Shocks in the Onset of Crises, 1973–2009a Percentage share Losses of Reserves or Increases in the Index of Exchange Market Pressure: Percentage Share Explained by External shocks

Crisis March 1975 February 1981 and July 1982 September 1987 April 1989 and February 1990 March 1995 January 2002

Total

World real interest rate

Investors’ interest in emerging markets

41

34

−3

42 0

53 ⴚ14

41 93 11

17 69 22

Domestic shocks Real exchange rate

Total

Money supply

9

59

36

−9 6

−2 8

58 100

12 24 −20

12 −1 8

59 7 91

Government deficit

Output

Money demand

7

n.a.

16

18 23

2 ⴚ14

−3 14

40 77

72 ⴚ1 13

2 ⴚ3 ⴚ7

n.a. −4 63

ⴚ16 15 22

Source: Authors’ calculations. a. This table focuses on explaining the onset of crises. For the crises during fixed exchange rate regimes, the historical decomposition starts in the month when foreign exchange reserves (as a share of domestic credit in dollars) start to fall and continues up to the month of the crisis. For the crises during dual exchange rate regimes, the historical decomposition starts in the month the index of exchange market pressure starts to increase and continues up to the month of the crisis. Numbers in bold signify that impulse responses for the individual shocks are significant for most horizons.

Notably, monetary shocks do not just reflect money supply shocks. In particular, money demand shocks are very important during the Tablita, Alemann, and Austral crises. As described in our chronology, these are episodes plagued by numerous regulatory changes regarding interest rates and foreign exchange markets—as during the period from February 1981 to June 1982—or by the stop-and-go cycles of controls on prices, wages, and public utilities during the Austral Plan.24 Naturally, these continuous changes in rules on financial and price contracts fuel uncertainty, reducing households’ faith in the financial system and overall ability of the authorities to maintain the peg. Finally, our results do not uncover an important fiscal effect at the onset of the crises. These results may be due to our “fiscal deficit” indicator that includes the central government but not the local governments and public enterprises, which ran particularly large deficits during the Tablita Plan and the latter part of the Convertibility Plan. Since our fiscal indicator captures only partially the fiscal 24. The management of prices was a central part of the Austral Plan. Prices and wage controls were introduced at the start of the program, but the first adjustment in prices was implemented in April 1986. In July 1986, the government introduced ceilings for monthly increases in prices as well as limits on wage increases. By the last months of 1986, prices were again fluctuating freely. In February 1987, a price freeze was again announced, only to be relaxed in May 1987.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

115

deficit of the consolidated public sector, fiscal shocks may, in fact, be captured by the shocks to money supply in equation 6.25 External shocks are also important in explaining currency turbulences. For example, hikes in world interest rates have a major impact on currency vulnerabilities during the Tablita and Alemann Plans, when world real interest rates increased from −1 percent in July 1980 to about 10 percent on average in 1981–1982, and during the Convertibility Plan in 1994, when the Federal Reserve increased its policy rate by 250 basis points. Indeed, our results indicate that 53 percent of the total decline in the historical forecast error of Argentina’s reserves between October 1979 and June 1982 is explained by world interest rate shocks, with the world interest rate effect increasing to 69 percent in the 1995 crisis. In the 1990s, external shocks are not limited to those fueled by changes in monetary conditions in industrialized countries. Spillover effects from other Latin American countries (as captured by a shock to investors’ interest in emerging markets) magnify reserve losses triggered by monetary tightening in financial centers. Our empirical estimations suggest that about one-fourth of the fall in reserves from December 1993 to February 1995 can be explained by contagion factors. However, contrary to theories advocating sudden stops as an explanation of the 2002 Argentine crisis (Calvo, Izquierdo, and Talvi 2002), we find that adverse shocks to investors’ interest in emerging markets played no role in explaining the collapse of the currency board since capital inflows already had dried up following the Russian crisis in late 1998. By 2001 investors had already started regarding Argentina as a country with problems of its own. We also examined the costs of crises fueled by domestic fragilities and those triggered by adverse external shocks, even in the presence of immaculate domestic fundamentals. Table 5 shows various costs for these two types of crises. First, we looked at the severity of the crises as captured by reserve losses in the six months before the crises and the real exchange rate depreciation in 25. Since the results in table 2 show that even our partial measure of fiscal shocks can explain 20 percent of the variance decomposition for foreign exchange reserves during the last part of the Convertibility Plan (January 1999–December 2001), we examine the possibility of a time-varying effect of the fiscal shock. The historical decomposition in table 4 is further decomposed into two episodes. The first episode starts in January 2001, from the onset of currency turmoil, and lasts until July 2001. The second episode starts in August 2001 and ends in December 2001, with the collapse of the Convertibility Plan. During the first episode, increases in government deficit explain 18 percent of the losses of reserves. But on 29 July 2001, the Argentine Congress passes the Zero Deficit Law, requiring a balanced budget by the fourth quarter of 2001. In August 2001, the deficit starts to decline while reserves losses continue to increase, explaining the almost zero cumulative net effect of fiscal shocks on reserves from January to December 2001, as shown in table 4.

116

E C O N O M I A , Fall 2009

T A B L E 5 . Costs of Crises, 1975–2002 Percent

Crisis

Reserve losses Real exchange in the rate depreciation 6 months in the 6 months before crisis after crisis

Output changes in the year of crisis and following year

Export changes in year after crisis

Import changes in year after crisis

March 1975 February 1981 July 1982 September 1987 April 1989 February 1990 March 1995 January 2002

40 56 10 36 47 20 16 25

181 53 30 27 66 −20 5 174

−5 −5 −8 1 −8 9 3 −15

20 30 19 34 35 6 35 4

−46 −50 −6 −4 −39 4 10 −16

Average costs of currency crises Crises with domestic vulnerabilities Crises with external adverse shocks

33 16

73 5

−5 3

21 35

−23 10

Source: Authors’ calculations.

the six months after the crises. On average, losses of reserves for crises fueled by domestic vulnerabilities reach 33 percent but only reach 16 percent for crises triggered by external shocks. Similarly, real depreciations are far larger (73 percent) for crises triggered by fragile domestic fundamentals than for crises with only adverse external shocks (5 percent). Second, we examined the impact of the crisis on the economy. Output losses in the year of the crisis and the following year average 5 percent for crises with domestic fragilities while the economy grows 3 percent during the crisis triggered by adverse external shocks. Finally, we examined the external adjustment following the crises. Access to international capital markets can be severely impaired in the aftermath of crises, with countries having to run sizable current account surpluses to repay their debt. We examined the size and type of the adjustment across these two types of crises. In the case of crises with domestic vulnerabilities, most of the adjustment occurs on the import side, with imports falling approximately 23 percent in the year following the crisis and exports only growing 21 percent, despite large depreciations during this type of crisis. This evidence suggests that Argentina might have been unable to attract trade credits to finance exports when its economy was quite fragile. In contrast, in the aftermath of the 1995 crisis, booming exports were at the heart of the recovery of the current account (35 percent increase) and even imports continued to increase (10 percent).

Graciela Kaminsky, Amine Mati, and Nada Choueiri

117

Conclusions Economists have puzzled at length over the causes and severity of currency crises. As a result, research in this area has surged, especially since the Exchange Rate Mechanism crises in 1992–93. Most of the empirical research has focused on predicting crises by using reduced-form estimations and failed to uncover the effect of policy and structural shocks on the changing severity of currency turmoil. This paper uses an old methodology to study this new problem, implementing VAR techniques to quantify the role of different shocks in the severity of currency crises. Our case study is Argentina, a country that not only has been at center stage in every single episode of international financial turmoil (such as the 1982 debt crisis, the 1994 tequila crisis, and the 1999 Brazilian crisis) but also has had many currency collapses of its own. Thus, while our analysis is confined to one country, it does provide a glimpse into the nature of worldwide currency turbulences. Our results confirm previous findings in the literature but also suggest new results. The major conclusions that emerge from our analysis are as follows. First, our estimations confirm the results obtained by Calvo, Leiderman, and Reinhart (1992) regarding the role of monetary tightening in industrial countries during the episodes of capital flow reversals of the early 1980s and mid 1990s. Both the collapse of the Tablita-Alemann Plans and the speculative attack against the peso in 1994–95 in the midst of the Convertibility Plan were in large part precipitated by the shift to a contractionary monetary stance in the United States. Second, as expected, inconsistent monetary and exchange rate policies did trigger many of the main speculative attacks against the peso. But as our event chronology and historical decompositions suggest, loose monetary policy was not the only culprit. The erratic nature of capital account restrictions and interest rate and credit controls, as well as the stop-and-go cycles on price and wage controls in the mid-1980s, also played a key role—with the uncertainty triggered by forced conversion of contracts leading to capital flight and downward pressure on money demand. Third, the mid-1990s look somewhat different. Spillovers from Mexico and other Latin American countries seem to have been a source of financial distress for Argentina, explaining about 25 percent of the severity of the speculative attack in 1995. This is not surprising given that the extent of Latin American integration into international capital markets increased sharply in 1990s. It was also in the 1990s that mutual funds became important players in Latin America. Naturally, this provided a new channel for spillovers, as

118

E C O N O M I A , Fall 2009

was the case when mutual funds retreated from several countries in Latin America after the losses they suffered from the Mexican devaluation. Fourth, the origin of the 2002 crisis lies in the sharp depression that started in the last half of 1998 and continued and deepened throughout the precrisis period. As the economy slid into recession, the currency board became a liability as the government was constrained to carry out a contractionary monetary policy in the midst of a profound recession. Financial contagion from Brazil or other Latin American countries was found to play no role in explaining the collapse of the currency board in 2001. Finally, our results show that the participation in international capital markets can be risky and that crises may occur even in the presence of immaculate domestic fundamentals. Still, the costs of crises triggered just by adverse external shocks are far smaller than those of crises fueled by fragile domestic fundamentals.

Appendix A. Data Sources and Definitions The data used in the VAR estimation are at monthly frequency and cover the periods 1970:1–2001:12.

Data Sources All data are from the International Monetary Fund’s International Financial Statistics, except as noted below.

Definitions and Units of the Variables r: Ratio of Argentina’s foreign exchange reserves to domestic credit in U.S. dollars (in percent). (f − e): Percentage difference between the black–dual exchange market exchange rate and the commercial exchange rate. d − e: Total domestic credit of the banking sector, measured in billions of dollars at the commercial exchange rate. q: Real effective exchange rate with respect to Argentina’s main trading partners.

Graciela Kaminsky, Amine Mati, and Nada Choueiri

119

i*: U.S. real interest rate: nominal interest rate on one-year U.S. Treasury bills adjusted for consumer price index inflation (in percent). c: First principal component of foreign exchange reserves of the following countries: Brazil, Chile, Colombia, Mexico, and Venezuela. The principal component is constructed as a linear combination of the five series, where the weights correspond to the eigenvector associated with the largest eigenvalue of the covariance matrix of the individual series (See Drhymes [1974] for an explanation of principal components analysis.) g: Annualized central goverment deficit measured as a proportion of GDP, in percent. Obtained from the Ministry of Finance and from the International Monetary Fund’s Government Finance Statistics database (www.imf.org/ external/pubs/ft/gfs/manual/gfs.htm) and staff reports. y: Monthly index of industrial production. Obtained from the Fundación de Investigaciones Económicas Latinoamericanas database.

Comment Carlos Winograd: The paper by Kaminsky, Mati, and Choueiri makes an interesting contribution toward disentangling the role of both domestic and foreign factors in shaping currency crises in Argentina since 1970. Nevertheless, I suggest that the following issues be addressed. The link between the motivation and the sample period used in the econometric estimations is not clear cut. On the one hand, the paper starts with a motivation concerned with the long view of Argentina and international financial markets since the beginning of the nineteenth century, but the sample only starts in 1970. On the other hand, the authors mention that “in the midst of the worst current international financial crisis since the Great Depression, untangling the roots of financial distress becomes crucial.” While the question seems relevant, the authors only extend their focus through 2001. I suggest that the authors make clearer the link between the motivation and the sample period, explaining carefully the choices made. The authors should make an effort to clarify why they think the technique they use is the most appropriate for the problem at hand. This issue was already mentioned at the Thirteenth Annual Meeting of the LACEA in Rio in 2008. In particular, while (structural) VAR models seem suitable to evaluate the impact of external and domestic variables over continuous variables, they are less appropriate for modeling dichotomic variables. Since currency crises are not continuous but dichotomic, maybe a dynamic probit-logit approach would be more appropriate. The authors seem to recognize this point when they compute the probability of a currency collapse in equation 14. Since the structured VAR results heavily depend on the restrictions suggested by the theoretical model presented in the second section, more care should be taken in developing and justifying the theoretical model. For example, what is the relevance of foreign variables when the country does not have access to international markets? Why doesn’t the fiscal deficit depend on output? Why doesn’t output depend on credit and vice versa? Why does the 120

Graciela Kaminsky, Amine Mati, and Nada Choueiri

121

fiscal deficit follow a random walk? These are just a few questions that should be addressed in the theoretical section. The authors also should carefully explain why they prefer to split the sample into fixed and dual exchange rate regimes instead of estimating, for example, a Markov switching regime model with two regimes. This would have the advantage of leading to more efficient estimates than just splitting the sample. It is also unclear why the authors did not handle the presence of breaks within the VAR framework. There is no table including the output of the structural VAR estimations, together with residual diagnostic tests. A presentation and discussion of such results (significance of variables, multivariate normality, multivariate absence of autocorrelation, and so forth.) should be included. The section on results can be made much more reader friendly by streamlining it to emphasize those results that are more relevant to the main goal of the paper. Finally, the conclusion could provide clear policy prescriptions. Hence, based on the estimation results, which policies would the authors suggest to help Argentina avoid or mitigate the consequences of a currency crisis?

122

E C O N O M I A , Fall 2009

References Baliño, Tomás. 1987. “The Argentine Banking Crisis of 1980.” Working Paper WP/87/77. Washington: International Monetary Fund. Blanco, H., and P. M. Garber. 1986. “Recurrent Devaluations and Speculative Attacks on the Mexican Peso.” Journal of Political Economy 94 (1): 148–66. Blejer, M., and N. Liviatan. 1987. “Hyperinflation: Stabilization Strategies in Argentina and Israel, 1985–1986.” International Monetary Fund Staff Papers 34 (3): 409–38. Calvo, G. 1999. “The Russian Virus.” Mimeo. University of Maryland. Calvo, G., A. Izquierdo, and E. Talvi. 2002. “Sudden Stops, the Real Exchange Rate, and Fiscal Sustainability: Argentina’s Lessons.” Mimeo. Washington: Inter-American Development Bank. Calvo, G., L. Leiderman, and C. Reinhart. 1992. “Capital Inflows to Latin America: The 1980s and the 1990s.” Working Paper 85. Washington: International Monetary Fund. ———. 1996. “Inflows of Capital to Developing Countries in the 1990s.” Journal of Economic Perspectives 10 (2): 123–39. Céspedes, L., R. Chang, and A. Velasco. 2004. “Balance Sheets and Exchange Rate Policy.” American Economic Review 94 (4): 1183–93. Cumby, R. E., and S. van Wijnbergen. 1989. “Financial Policy and Speculative Runs with a Crawling Peg: Argentina 1979–1981.” Journal of International Economics 27 (1–2): 111–27. D’Amato, L., E. Grubisic, and A. Powell. 1997. “Contagion, Bank Fundamentals or Macroeconomic Shocks? An Empirical Analysis of the Argentine 1995 Banking Problems.” Paper prepared for the 1997 Latin American and Caribbean Economic Association meeting. Bogota, Colombia, October 2–4. De la Torre, A., E. Levy Yeyati, and S. L. Schmukler. 2002. “Argentina’s Financial Crisis: Floating Money, Sinking Banking.” Washington: World Bank. Di Tella, G., and R. Dornbusch, eds. 1989. The Political Economy of Argentina, 1946–1983. University of Pittsburgh Press. Dornbusch, R., and J. C. de Pablo. 1989. “Debt and Macroeconomic Instability in Argentina.” In Developing Country Debt and the World Economy, edited by J. Sachs, pp. 37–56. University of Chicago Press. Drazen, A., and P. Mason. 1994. “Credibility of Policies versus Credibility of Policymakers.” Quarterly Journal of Economics 109 (3): 735–54. Drhymes, P. 1974. Econometrics: Statistical Foundations and Applications. New York: Springer Verlag. Edwards, S. 2002a. “The Argentine Debt crisis of 2001–2002: A Chronology and Some Key Policy Issues.” Mimeo. University of California at Los Angeles. ———. 2002b. “The Great Exchange Rate Debate after Argentina.” Working Paper 9257. Cambridge, Mass.: National Bureau of Economic Research. Eichengreen, B., A. Rose, and C. Wyplosz. 1996. “Speculative Attacks on Pegged Exchange Rates: An Empirical Exploration with Special Reference to the European

Graciela Kaminsky, Amine Mati, and Nada Choueiri

123

Monetary System.” In The New Transatlantic Economy, edited by M. Canzoneri, P. Masson, and V. Grilli, pp. 191–235. Cambridge University Press. Ericsson, N., and S. Kamin. 1993. “Dollarization in Argentina.” International Finance Discussion Paper 460. Federal Reserve, Board of Governors. Giorgio, L., and S. Sagari. 1996. “Argentina’s Financial Crises and Restructuring in the 1980s.” In Bank Restructuring: Lessons from the 1980s, edited by A. Sherg. Washington: World Bank. Hausman, R., and A. Velasco. 2002. “The Argentine Collapse: Hard Money’s Soft Underbelly.” Mimeo. Harvard University (April). International Monetary Fund. 2004a. “Lessons from the Crisis in Argentina.” Occasional Paper 236. Washington. ———. Independent Evaluation Office. 2004b. “The IMF and Argentina, 1991–2001.” Washington. Kaminsky, G. 2009. “Two Hundred Years of Financial Integration: Latin America since Independence” (http://home.gwu.edu/∼graciela/HOME-PAGE/RESEARCHWORK/WORKING-PAPERS/LA-200-years.pdf). Kaminsky, G., R. Lyons, and S. Schmukler. 2004. “Managers, Investors, and Crises: Investment Strategies of Mutual Funds.” Journal of International Economics 64 (1): 113–34. Kaminsky, G., and C. Reinhart. 2000. “On Crises, Contagion, and Confusion.” Journal of International Economics 51 (1): 145–68. Kiguel, M. 1989. “Inflation in Argentina: Stop and Go Since the Austral Plan.” Policy, Planning, and Research Working Paper 162. Washington: World Bank (March). Kiguel, M., and P. Neumeyer. 1995. “Seigniorage and Inflation: The Case of Argentina.” Journal of Money, Credit, and Banking 27 (3): 672–82. Krugman, P. 1979. “A Model of Balance of Payments Crises.” Journal of Money, Credit and Banking 11 (3): 311–25. Machinea, J., and J. Fanelli. 1988. “Stopping Hyperinflation: The Case of the Austral Plan in Argentina, 1985–87.” In Inflation Stabilization: The Experience of Israel, Argentina, Brazil, Bolivia, and Mexico, edited by M. Bruno and others, pp. 111–52. MIT Press. Montanaro, E. 1990. “The Banking and Financial System in Argentina: The History of a Crisis.” In The Future of Financial Systems and Services, edited by E. Gardener. New York: St. Martin’s Press. Mussa, M. 2002. Argentina and the Fund: From Triumph to Tragedy. Washington: Peterson Institute for International Economics. Perry, G., and L. Servén. 2002. “The Anatomy of a Multiple Crisis: Why Was Argentina Special and What Can We Learn From It?” Mimeo. Washington: World Bank (May). Reinhart, C., and C. Végh. 2002. “Do Exchange Rate-Based Stabilizations Carry the Seeds of Their Own Destruction?” Mimeo. University of Maryland. Rodriguez, C. 1994. “Argentina: Fiscal Disequilibria Leading to Hyperinflation.” In Public Sector Deficits and Macroeconomic Performance, edited by W. Easterly, C. A. Rodriguez, and K. Schmidt-Hebbel, pp. 101–66. Oxford University Press.

SEBASTIÁN NIETO-PARRA

Who Saw Sovereign Debt Crises Coming? his paper studies sovereign debt crises through the prism of the primary sovereign bond market and describes the behavior and interactions among the principal actors in the sovereign bond market before and after a sovereign debt crisis. The study ﬁnds that investment banks price sovereign default risk well before crises occur and before investors detect default risk. As early as three years before a crisis, countries that will eventually enter into a debt crisis pay underwriting fees that are almost twice as high as the underwriting fees paid by the average emerging market country. In contrast, sovereign bond spreads do not seem to be good leading indicators of debt crises. Between three years and one year before a crisis, there is almost no difference between the bond spreads paid by countries that will eventually enter into a crisis and the spread paid by the average emerging market country. The paper also shows that investment banks’ behavior differs depending on the type of sovereign debt crisis. While Investment banks charge higher underwriting fees to countries that will later enter into crisis because of high risk of sovereign default, they do not appear to charge higher fees to countries that will eventually suffer a liquidity crisis driven by external factors or banking problems. Given that my results suggest that investment banks price default risk well before investors do, they raise the puzzle of why underwriting fees, which contain valuable publicly

T

Sebastián Nieto-Parra is with the Development Centre of the Organization for Economic Cooperation and Development. I am especially grateful to Marc Flandreau and Ugo Panizza for illuminating and valuable comments on previous versions of this paper. I thank Thomas Dickinson for excellent research assistance provided during interviews in New York in 2006–07 with market participants in the primary sovereign bond market. I am also grateful to Rolando Avendaño, Jeff Dayton-Johnson, Enestor Dos Santos, Alicia Garcia Herrero, Helmut Reisen, Javier Santiso, and Camilo Tovar for valuable feedback, and to seminar participants at the Fifteenth World Congress of the International Economic Association and at the Economía Panel for their comments and suggestions. The views expressed in this article are those of the author and should not be attributed to the Organization for Economic Cooperation and Development.

125

126

E C O N O M I A , Fall 2009

available information, are not used in pricing bonds issued by emerging market countries. While there are many papers that study how emerging market countries access the international bond market (Grigorian 2003; Gelos, Sahay, and Sandleris 2004; Fostel and Kaminsky 2007), studies of the formation of prices in the emerging sovereign bond market are rare and tend to focus on the incidence of pricing of certain covenants, such as collective action clauses (for different perspectives, see Eichengreen and Mody 2004; Gugiatti and Richards 2003; Becker, Richards, and Thaicharoen 2003).1 This stands in contrast to the massive literature on the determinants of underwriting fees in the primary corporate market. This literature shows that the main determinants of underwriting fees are the characteristics of the issue (such as maturity, amount, regulation, and currency denomination), the characteristics of the issuer (for example, credit risk as measured by credit rating, size of the firm, profitability indicators, and activity sector group), and a set of market variables that include secondary market conditions and volatility of prices (West 1967; Higgins and Moore 1980; Rogowski and Sorensen 1985; Lee and others 1996; Livingston and Miller 2000; Kollo and Sharpe 2006; Melnik and Nissim 2003; Hua Fang 2005). These studies tend to find an inverse relationship between the quality of the issuer and the level of underwriting fees. This is usually interpreted as a consequence of the greater effort required from intermediaries when they act as underwriters of lower-quality issues (Altinkihc and Hansen 2000). In this paper, I ask how important credit risk is for the underwriting fees charged by investment banks in the sovereign bond market. While most of my results are based on formal econometric analysis, I also show that my results are corroborated by the responses to a survey that covered the main institutional investors in and originators of sovereign bonds issued by emerging market countries.2

1. Additionally, there is almost no research on the conﬂict of interest between the research departments and origination departments. Calomiris (2003), referring to emerging market crises, notes the possible “cooperation” between these departments. Concerning the Latin American sovereign bond market, Nieto-Parra and Santiso (2007) ﬁnd that when an investment bank is acting as lead manager, 90 percent of its recommendations are positive. 2. Interviews were undertaken with the following institutional investors and investment banks on Wall Street between October 2006 and March 2007: Alliance Bernstein, Alliance Capital, Fidelity, GE Asset Management, GMO, Goldman, Invesco, and Western Asset. The investment banks interviewed were Bear Stearns, Citibank, Deutsche Bank, Goldman Sachs, J. P. Morgan, Lehman Brothers, Merrill Lynch, and Morgan Stanley.

Sebastián Nieto-Parra

127

How the Primary Sovereign Bond Market Works The analysis of the primary market is a key element to understanding the behavior of investment banks and investors, and the formation of fees and primary sovereign bond spreads. This section describes the structure of the primary sovereign bond market and the main risks faced by investment banks when they act as lead managers in this market (for more details on the structure of the primary market and the critical steps of the issue process, see Flandreau and others 2010b). A simple version of the structure of the sovereign bond market is summarized in Figure 1, which illustrates the interactions among actors in the sovereign bond market throughout the execution of a ﬁnancial transaction. It is investment banks that act as lead managers in the emerging sovereign bond market.3 Investors’ participation in the primary market or in the secondary market (by buying or selling securities) is all done via investment banks.4 Most of investment banks’ income is derived from these transactions. In particular, the fee is deducted from the price offered to investors in the primary market, and it is agreed between investment banks and governments before the determination of the price of the bond in the primary market.5 Aside from the role of intermediary between issuers and investors, one of the most important responsibilities of underwriters concerns their role in promoting the bond. A preliminary prospectus (called a red herring) is made available with all the information about the issue (with the exception of the offer price and the effective date of issuance, which are not known at the time of preparation of the red herring). With the red herring in hand, the underwriter and the issuer promote the bond through presentations, conference calls, publications, and, occasionally, “road shows.” The investment banks’ research departments also circulate regular publications covering various 3. I use the term lead manager to refer to agents who place bonds in the market. I do not differentiate between underwriters, lead managers, or book runners, and I assume that these three types of agents have the same responsibilities with respect to the issue during a sovereign bond issue. 4. I take investors as a single group (for example, no account is made for differences that might stem from foreign versus domestic, individual versus institutional, or international versus local). Studies of differences in investor behavior include Calvo (2002), Borensztein and Gelos (2000), and Santiso (2003). 5. In this paper, I use the terms “fee” and “underwriting fee” to refer to the remuneration paid by issuers to underwriting banks in the primary market. Other terms are used in the literature and in the ﬁnancial jargon (for example, underwriting spread, gross spread, underwriting discount).

128

E C O N O M I A , Fall 2009

F I G U R E 1 . Structure of the Prices in the Sovereign Bond Marketa P t − fee

Governments

Investment Banks PT

P T − −t +

PT

+ C

T

−

Pt

−t +

Investors

a. T, maturity date; t, issue date; C, commission paid by investors to investment banks; P, price of the sovereign bond; fee, underwriting fee paid by governments to investment banks.

issuers. These publications contain advice for investors regarding the kind of government bonds they should purchase. These activities of dissemination and promotion are a source of effort and risk for investment banks acting as underwriters. The ﬁrst and foremost risk concerns the potential loss of reputation in the event of government default. In a historical analysis of the sovereign bond market, Flandreau and Flores (2009) show the role of underwriters’ reputations in guiding investors’ portfolio allocations. Concerning the corporate market, Hua Fang (2005) ﬁnds that reputable banks charge higher fees, which can be interpreted as economic rents on reputation.6 6. This research challenges that of Livingston and Miller (2000) and James (1992), in the sense that it takes into account that reputable banks may have chosen (self-selected) to underwrite higher-quality issues precisely due to reputation concerns. Thus Hua Fang (2005) argues that “failing to control for this type of self-selection could lead to incorrect conclusions.” From a theoretical and empirical analysis of the “underpricing” in the corporate market, Carter and Manaster (1990) show that prestigious underwriters charge higher underwriter spreads and are associated with lower-risk offerings.

Sebastián Nieto-Parra

129

Besides the reputation risk, investment banks face other risks related to the transaction itself. Key aspects of each bond issue are formalized by lead managers and the government before the bond goes to the market. One central aspect of the agreement is the distribution system. Even if most bonds are now placed with a “best efforts” contract—in which investment banks pledge to help ﬁnd customers but are not forced to acquire any bonds in the absence of buyers—investment banks can incur some risks. First, they need to buy the issue before selling it on to the investors and thus face a “settlement risk” during the “book-building” process. In this process, which lasts for a couple of hours, the lead managers (also called book runners) build up a list of “orders” at a speciﬁed price. Investors are contacted by telephone and Bloomberg messages, and are asked for “expressions of interest” aimed at helping the lead manager and the government set the price of the issue. The risk for investment banks occurs when the expression of interest is not conﬁrmed and the bonds cannot be allocated. Second, investment banks have the responsibility to place the bonds in the market and make an effort to stabilize the price of the bonds in the secondary market for an unspecified time. Bond prospectuses usually indicate that the underwriting bank is not forced to make a secondary market for the bonds but that it plans to make one. According to interviews in origination departments of investment banks, “market making” activities on the secondary market can extend until the maturity of the bond.7 These interviews also suggested that the quality of this service is related to the fee paid to the underwriter, and to the desire that the underwriter has to acquire a reputation as a good supporter, thus increasing the likelihood of secure future contracts.8 Because of the efforts and risks described above, investment banks are likely to charge fees that are positively correlated with the probability of a debt crisis. Therefore, the fee may contain important information about investment banks’ perception of sovereign risk.9 But why do investment

7. The question asked to origination departments of investment banks was, “How long does a lead manager make a market in the secondary market?” For more than one half of the managers interviewed, market making activities may continue for the duration of the bond. 8. An example is provided by Citigroup: “As lead manager of a bond, one of the things you are compensated for is to maintain markets for the bond.” 9. A crucial difference between the underwriting market of today versus the past is that, in the past, a signal of credit risk was the prestige of the underwriting bank charged to place the bond (Flandreau and others 2010a). Today, it would appear that the fee contains more information about credit risk than a bank’s reputation.

130

E C O N O M I A , Fall 2009

banks have more information than investors? The answer has to do with the fact that underwriting involves a close, regular, and often privileged relationship with important actors in government. Such a close relationship with the treasury and ministry of finance of a country confers a unique vantage point for observing both the economic aspects as well as the behavioral patterns of the issuing government. As pointed out by one of the investment banks interviewed, “The information you get from underwriting is very important—not insider information, but a lot of knowledge on what a sovereign tends to do.” The direct and strict link investment banks have with issuers can define a different type of behavior between investment banks and investors regarding sovereign risk. This difference in behavior can be particularly important in the case of countries in which the risk of default of the public bonds is high. Although public information is available (at least ex post) to distinguish these countries from other emerging countries (see appendix A), investment banks could be more concerned about these aspects than investors are. In fact, research shows that investors tend to herd and may have weak incentives to learn about individual countries (Calvo 1998). One way to show that herding is more important for investors than for investment banks is to regress the perception of risk of investors (that is, the primary bond spread) and that of investment banks (the fee) against a variable that measures external market conditions, and then compare the fit (as measured by the R squared) of the two regressions. The finding that the R squared of the investors’ regression is higher than the R squared of the investment banks’ regressions would be consistent with the idea of more herding by investors (who seem to be more concerned than investment banks with external conditions in pricing sovereign risks). This is exactly what I find when I run simple regressions of fees and bond spreads over external risks as measured by the implied volatility of S&P 500 index options (VIX index).10

10. More precisely, I test the basic following OLS model: risk perception agent kit = α 1 + α 2 i VIX it + ε it , and the results are Underwriting fee = 0.32 + 0.015 * VIX and R2 = 0.03 (4.16) (4.07) Primary bond spread = 1.55 + 0.10 * VIX and R2 = 0.11 (4.92) (6.92), with underwriting fee and bond spread in percent, and t statistics in parentheses.

Sebastián Nieto-Parra

131

Hypothesis and Empirical Strategy I analyze the behavior of market participants by comparing prices before sovereign debt crises with prices in tranquil periods. In particular, I test the following two hypotheses: Hypothesis 1: Before sovereign debt crises, underwriting fees include information on the quality of the bonds that is not incorporated in sovereign bond spreads. Hypothesis 2: The difference between the behavior of investment banks and investors is particularly important for countries that will enter into a crisis because of bad fundamentals (that is, sovereign default risk countries) and less important for countries that suffer liquidity crises. These hypotheses imply that there is valuable information in underwriting fees that is not captured by sovereign bond spreads. The hypotheses are validated when, before sovereign bond crises, the fees of “crisis” countries cannot be fully explained by the behavior of sovereign bond spreads (hypothesis 1) and when this effect is stronger in countries that will enter into crisis because of their bad fundamentals (hypothesis 2). Under the alternative hypotheses, investment banks do not have any information advantage and sovereign bond spreads of “crisis” countries can explain the behavior of fees before the onset of crises. Figure 2 shows the average annual fee and primary sovereign bond spreads. Squares indicate the fee and bond spreads between three (T − 3) and one (T − 1) year before the onset of a crisis. Fees are substantially higher (given the bond spread) for countries that will eventually suffer a crisis, relative to other emerging countries. On average, sovereign default risk countries had to pay 1.10 percent of the amount issued to investment banks between one and three years before the onset of crisis, almost twice the emerging countries’ average during the sample period (0.56 percent). By contrast, when I compare the level of primary sovereign bond spreads between one and three years before crisis with respect to the total for emerging countries, I ﬁnd that the former is on average only slightly higher than the latter (385 basis points versus 319 basis points) and much lower than the primary sovereign spread at the onset of this crisis (603 basis points). Moreover, as crisis countries approach the onset of the crisis, the retention coefficient (fee over sovereign bond spread) decreases, showing that information on underwriting fees is less relevant with respect to bond spreads (Figure 3). As one moves away from the onset of a sovereign debt crisis, the information obtained from the underwriting fee regarding the sovereign debt

E C O N O M I A , Fall 2009

132

F I G U R E 2 . Fees and Primary Sovereign Bond Spreads, Annual Basis, 1993–2006 Fee (as percentage of amount issued) 1.8

Brazil (T-1)

1.6

Turkey (T-3) Argentina (T-3) Argentina (T-2)

1.4

Argentina (T-1)

1.2

Russia (T-1) Russia (T-2)

1

Turkey (T-1) Turkey (T-2)

Mexico (T-2)

0.8 0.6

Average fee

Brazil (T-2)

0.4 0.2 0 0

100

200

300

400

500

600

700

800

900

Sovereign bond spread (basis points) Source: Author’s calculations based on Dealogic database. The fee and sovereign bond spreads for countries between three and one year before the onset of a twin crisis (date T) are indicated with a square. Twin crises refer to the combination of sovereign default risk crisis and currency crisis. The Argentinean crisis occurred in 2001, the Brazilian in 1998, the Mexican in 1995, the Russian in 1998, and the Turkish in 2000. The average of the fees and sovereign bond spreads for the rest of sovereign bond issues of emerging countries are indicated with a diamond.

crisis is more relevant than that contained in bond spreads. This is consistent with ﬁgure 2 showing that underwriting fees during precrisis periods are abnormally high (controlling for bond spreads) more than twelve months before the onset of the crisis. I test hypotheses 1 and 2 using a bond-level panel that covers twenty-nine emerging market countries for the period 1993–2006. I start with a simple model in which I regress underwriting fees over bond spreads, time ﬁxed effects, country ﬁxed effect, and a set of dummies that track the evolution of the crisis. In particular, I test the following model: (1)

FEEit = α1 + α 2 i SBSit +

∑ β K (T + K )it 5

K = −5

+ τ t + υi + ε it ,

where FEE is the underwriting fee (i and t are index countries and time, respectively), SBS is the primary sovereign bond spread, T + K is a dummy variable that takes the value of 1 for countries placed at the year K with

Sebastián Nieto-Parra

133

F I G U R E 3 . Fees over Primary Sovereign Bond Spreads before a Crisis and for Crisis Countries, Monthly Basis, 1993–2006a Fee over sovereign bond spread (percent) 1 0.8 0.6 0.4 0.2 0 -24

-20

-16

-12

-8

-4

0

Months before sovereign debt crises Source: Author’s calculations based on Dealogic database. a. Each point represents a sovereign bond issue in the primary bond market. Symbols: diamond, Argentina; square, Russia; triangle, Turkey; asterisk, Brazil.

respect to the onset of crisis (T ) and the value of 0 in all other periods, τ is a year ﬁxed effect, υ is a country ﬁxed effect, and ⑀ is the error term. With this setup, α2 measures the elasticity of underwriting fees with respect to sovereign bond spreads in tranquil periods, and βK measures the difference between fees in tranquil periods and around crisis periods (from [T − 5] to [T + 5] years before and after the onset of crisis) after controlling for sovereign bond spreads.11 Next, I check whether the relationship between bond spreads and fees changes in the run-up to a crisis by testing the following model: (2)

FEEit = α1 + α 2 i SBSit + α 3 i CRISISit + τ t + υi + ε it ,

where CRISIS is a dummy variable that takes the value of 1 in the X years before crisis (that is, T − X, T − X + 1, . . . , T − 1) and the value of 0 in 11. I also experiment with the lag structure, using dummy variables for different precrisis periods, starting with the set of periods from T − 5 to T − 1 and ﬁnishing with T − 1, the year before crisis. Results are provided upon request.

134

E C O N O M I A , Fall 2009

all other periods (X years is determined using the significance of the βK coefficients in equation 1). Within this setup, α2 measures the elasticity of underwriting fees with respect to sovereign bond spreads in tranquil periods, and α3 measures the difference between underwriting fees in tranquil and precrisis periods. The ﬁrst hypothesis is validated when α3 is positive and statistically signiﬁcant. The alternative hypothesis is that α2 is positive and signiﬁcant and α3 is not signiﬁcant. In that case, investment banks observe crisis countries at the same time as investors do. The same procedure is used to test hypothesis 2. I only differentiate among sovereign debt crises to test that the difference between investment banks and investors in pricing a crisis is especially noticeable in the case of sovereign risk default countries.

Data and Typology of Sovereign Debt Crises This section analyzes two inputs for the study of the behavior of investment banks and investors around sovereign debt crises. The ﬁrst is the dataset used; the second, the variety of sovereign debt crises included in the sample.

Data I focus on the period 1993 to 2006 and cover twenty-nine emerging economies included in the Emerging Markets Bond Index (EMBI) Global for which I could obtain information on fees. My main source of information on the structure of the primary sovereign bond market is the DCM Analytics database created by Dealogic. In building my sample, I use the following four criteria: —I only take into account sovereign bond issues for which I have information on the ISIN (International Securities Identifying Number) reference of the issues and for which I have data on the fee and the sovereign bond spread. —I exclude issues with ﬂoating coupon rates, which alter the true value of the bond spread.12 12. For these kinds of issues, the primary sovereign bond spread reported corresponds to basis points added to the benchmark rate used to determine the coupon rate. For instance, for the Brazilian Global Bond 21/06/04 (ISIN number US105756BC32), the coupon rate is 3 months Libor + 575 basis points, and the primary sovereign bond spread reported is 575 basis points.

Sebastián Nieto-Parra

135

—I only use issues denominated in euros, yen, or U.S. dollars, which are the most commonly used currencies in the sovereign bond market. —I exclude issues partially or totally guaranteed by international organizations, such as the World Bank, or the regional development banks. This set of restrictions yields a sample of 436 bond issues.13 Table 1 presents a description by country of the sovereign bonds used in the sample. The total amount of sovereign bonds used in the sample exceeds U.S.$300 billion, and the average amount issued by country and per issue is close to U.S.$700 million. The total income received by underwriting banks is more than U.S.$1.5 billion (on average, close to U.S.$3.5 million per issue). The averages of the fee and of the sovereign bond spread in the sample are 0.54 percent of the amount issued and 329 basis points, respectively. The number of lead managers in the emerging sovereign bond market is small. Like the U.S. corporate bond market (Livingston and Miller 2000; Hua Fang 2005), approximately 90 percent of the issues were realized by the top ten book runners and more than 75 percent by the seven most important book runners. Table 2 shows the investment bank market share for the top ten lead managers of the emerging sovereign bonds used in the sample.14 I measure secondary sovereign bond spreads with the EMBI Global spread, calculated by J. P. Morgan. My set of controls includes the following variables: —the characteristics of the bond issues: the collective action clauses, lead managers variables, number of bonds issued by country, rating at launch, years to maturity, and value of proceeds, available from the Dealogic database; —solvency ratios: average maturity of the external debt, short-term debt over total debt, interests of the public external debt over exports, interests of short-term debt over GDP, external debt services over reserves, the total debt over reserves, total external debt over GDP, and public debt over GDP, available from the World Bank’s Global Development Finance Online database; —macroeconomic data: GDP growth, exchange rate depreciation, inﬂation rate, and current account variables, available from International Finance Statistics, obtained online from the International Monetary Fund (IMF); 13. The number of bonds issued by year from 1993 to 2006 is (the ﬁrst is 1993 and last is 2006): 14, 7, 10, 19, 30, 34, 56, 55, 48, 27, 35, 42, 37 and 22. Regarding currency denomination, 67 percent of these issues are denominated in U.S. dollars, 27 percent are denominated in euros, and the rest are denominated in yen. 14. There is a vast research literature that uses market share as proxy for reputation (Megginson and Weiss 1991; Livingston and Miller 2000; Hua Fang 2005). For the case of capital markets, see Bloomberg (2006).

136

E C O N O M I A , Fall 2009

T A B L E 1 . Description of Sovereign Bonds Used in Sample, 1993–2006a Units as indicated

Country Argentina Brazil Bulgaria Chile China Colombia Dominican Republic Ecuador Egypt El Salvador Hungary Indonesia Lebanon Malaysia Mexico Morocco Pakistan Panama Peru Philippines Poland Russia South Africa Thailand Turkey Ukraine Uruguay Venezuela Vietnam Totalb

Number of bonds

Maturity average ( years)

Total amount issued (U.S.$ millions)

Amount issued (average U.S.$ millions)

Total fee (U.S.$ millions)

Underwriting fee (average percent of amount issued)

Primary bond spread (average basis points)

53 44 3 5 18 32 1 1 2 4 13 4 27 5 30 2 2 14 8 37 20 6 11 4 55 4 18 12 1 436

9.2 12.4 9.7 7.7 14.7 13.8 5.0 5.0 7.5 19.0 7.7 15.1 6.0 8.6 11.3 5.0 5.0 18.1 17.4 12.1 11.4 9.4 9.0 5.7 8.8 6.4 11.3 10.6 10.2 10.1

36,233.6 36,205.3 220.8 3,964.2 10,634.9 17,733.6 500.0 497.9 2,993.9 1,703.9 11,582.6 4,359.3 15,722.1 7,135.3 34,453.9 611.6 649.5 6,529.3 3,922.1 29,169.2 20,825.3 4,627.8 6,956.1 1,597.5 40,971.3 1,803.3 4,321.1 5,182.5 736.7 311,844.7

710.5 842.0 220.8 792.8 590.8 554.2 500.0 497.9 1,497.0 426.0 891.0 1,089.8 582.3 1,427.1 1,188.1 305.8 324.8 502.3 490.3 767.6 1,041.3 1,156.9 632.4 399.4 744.9 450.8 240.1 471.1 736.7 692.2

344.2 217.3 8.1 23.9 34.7 81.4 2.5 2.5 7.0 7.7 28.3 6.5 48.5 17.1 170.1 2.2 2.6 33.0 10.9 61.4 34.2 103.7 32.5 5.6 195.8 9.5 22.2 30.2 4.9 1,548.3

1.21 0.63 0.55 0.30 0.48 0.66 0.50 0.70 0.45 0.51 0.36 0.28 0.50 0.44 0.57 0.50 0.50 0.55 0.28 0.38 0.30 0.96 0.52 0.45 0.61 0.65 0.59 0.70 0.65 0.54

449 459 340 159 104 446 569 470 305 339 59 278 387 220 263 142 378 348 432 397 76 597 231 51 482 528 263 525 256 329

Source: Author’s calculations based on Dealogic database. a. The amount issued corresponds to the deal value of the proceeds. b. For the total sample, the maturity, amount issued, underwriting fee, and bond spreads are calculated as the simple average (average of countries in the sample). The total fee (that is, income received by underwriting banks) is calculated as the product of the deal value of the issue and the underwriting fee.

Sebastián Nieto-Parra

137

T A B L E 2 . Market Share for Top Ten Investment Banks, 1993–2006a Percent Investment bank

Africa

Asia

Europe

Latin America

Middle East

Total

JP Morgan Citigroup Morgan Stanley Deutsche Bank Merrill Lynch Credit Suisse Goldman Sachs UBS BNP Paribas Dresdner K. W.

10.0 (2) 9.4 (1) 8.3 (2) 5.0 (2) 19.6 (5) 3.5 (1) 10.0 (2) 7.0 (2) 3.0 (1) 9.4 (1)

19.6 (29) 13.9 (16) 9.1 (14) 12.1 (21) 6.8 (10) 9.9 (14) 4.6 (5) 12.2 (16) 1.1 (2) 0 (0)

18.9 (34) 15.6 (25) 15.7 (22) 9.1 (18) 2.5 (7) 5.6 (11) 2.3 (3) 6.3 (11) 4.9 (7) 6.4 (7)

22.4 (65) 12.4 (43) 11.9 (36) 8.9 (36) 9.8 (32) 4.8 (26) 10.8 (28) 5.2 (19) 2.7 (11) 2.1 (8)

2.0 (2) 2.4 (1) 7.3 (6) 6.9 (3) 13.9 (8) 24.0 (11) 0 (0) 1.3 (1) 30.7 (7) 0 (0)

19.5 (132) 12.9 (86) 12.1 (80) 9.3 (80) 7.8 (62) 6.9 (63) 6.9 (38) 6.5 (49) 4.5 (28) 2.9 (16)

Source: Author’s calculations based on Dealogic database. a. The market share is calculated from the deal value of the proceeds. In the case of multiple book runners for an issue, the deal value of the proceeds is divided by the number of book runners in the operation. The number of issues underwritten is in parentheses.

—political variables: index of freedom status and years of presidential elections, available online from Freedom House and the World Bank’s Database of Political Institutions, respectively; and —external variables: U.S. Treasury bill rate and VIX index, obtained from Thomson Datastream.15 In order to analyze the relevance of the information received by investors from investment banks concerning the primary bond market, I also examined for the period July 1997 through December 2007 the publications of thirteen investment banks that are active in trading and issuing emerging countries’ sovereign debt.16 In these publications, investment banks state their views for each emerging country, providing input for their clients—namely, the “buy side” of the market (such as portfolio asset managers, mutual funds, hedge funds, and pension funds).

15. See Freedom House, “Freedom in the World” (www.freedomhouse.org/template.cfm? page=15). 16. The name of the publications used are Emerging Markets Fortnightly (ABN AMRO), LatAm Drivers Fortnightly (Barclays Capital), Global Emerging Markets Monthly (Bear Stearns), Economics/Strategy (Citigroup), Debt Trading Monthly (Credit Suisse), Emerging Markets Monthly (Deutsche Bank), EM Strategist (Dresdner Kleinwort Wasserstein), Global Interest Rate Strategy (Goldman Sachs), Emerging Markets Outlook and Strategy (J. P. Morgan), Emerging Markets Compass (Lehman Brothers), Emerging Markets Debt Monthly (Merrill Lynch), EMD Perspectives Quarterly (Morgan Stanley), and Emerging Markets Debt Strategy Perspectives (UBS).

138

E C O N O M I A , Fall 2009

Typology As there is no consensus on the deﬁnition of a sovereign debt crisis (see Pescatori and Sy 2007; Panizza, Sturzenegger, and Zettelmeyer 2009), I apply the deﬁnition that is commonly used in the literature on the “early warning models” (see Manasse, Roubini, and Schimmelpfennig 2003; Manasse and Roubini 2005; Ciarlone and Trebeschi 2005; Fioramanti 2006). According to this deﬁnition, a country is in a debt crisis if it is classiﬁed as being in default by Standard & Poor’s (S&P) or it receives a large nonconcessional IMF loan deﬁned in excess of 100 percent of quota. Concerning the ﬁrst part of the deﬁnition, there is heterogeneity in the types of default included in the S&P deﬁnition (see appendix B). Debt restructurings may have either followed a sovereign default or been undertaken preemptively in an effort to avoid default.17 The latter can be associated with cases in which liquidity problems, possibly driven by external shocks, triggered the debt restructuring. These cases differ from the “post default” cases in which solvency problems were more likely to be the main driver of the crisis. On the basis of these considerations, I divide countries classiﬁed as in default by S&P into two groups, depending on the restructuring case: preemptive and post default.18 The second part of the definition considers countries that would have defaulted without the intervention of the IMF, and these countries also are divided into two groups, depending on the level of their external public debt before the crisis. (I use a debt risk index that depends on four external debt indicators; see appendix C). Countries with high levels of external public debt are referred to here as countries with “public bonds vulnerabilities.” Figure 4 shows the different types of sovereign debt crises studied in this paper and also highlights countries that suffered twin (currency and debt) crises.19 17. For the case of Moody’s, Argentina (2001) and Russia (1998) were the only default countries in the contemporaneous era. 18. Duration and intensity of default also vary considerably among countries. For instance, the Argentinean default lasted four years (from 2001 until 2005) while the Dominican Republic (2005) and Uruguayan (2003) defaults lasted only one year. Additionally, the recovery rates of these defaults are also different. Concerning the reduction of the principal of the debt restructured and according to IMF (2006), Argentina obtained a reduction of 56 percent, in contrast to the Dominican Republic (0.0 percent), Ukraine (0.0 percent), or Uruguay (1 percent). 19. In order to determine which of the sovereign debt crises are combined with a currency crisis, I construct an index of currency market turbulence, in the spirit of Eichengreen, Rose, and Wyplosz (1996). See Nieto-Parra (2008) for a detailed description of the construction of this index and the connection between currency crises and sovereign debt crises.

Sebastián Nieto-Parra

139

F I G U R E 4 . Typology of Sovereign Debt Crisesa Sovereign Debt Crises

Default (S&P’s definition)

Preemptive

Post default

Ukraine (Sept. 1998)* Uruguay (May 2003)* Pakistan (Jan. 1999) Dominican Rep. (Feb. 2005)*

Russia (August 1998)* Argentina (Nov. 2001)* Ecuador (Sept. 1999)

IMF large package

Public bonds vulnerabilities (PBV) Mexico (Feb. 1995)* Turkey (Dec. 2000)* Brazil (Dec. 1998)* Brazil (August 2001)

No PBV Indonesia (Nov. 1997)* Thailand (Aug.1997)*

Source: Author’s calculations based on S&P (2006, 2007), the World Bank’s Global Development Finance, International Financial Statistics (IMF). The Economist, Organization for Economic Cooperation and Development economic surveys, and CRS Reports for Congress. a. Asterisk denotes countries that also experienced a currency crisis in the twelve months before and after the sovereign debt crisis. See Nieto-Parra (2008) for the definition of currency crises and the combination of both crises (currency and sovereign debt crises).

The sovereign debt crises shown in Figure 4 can be reclassified into two groups, depending on the fragility of the public sector or the capacity of governments to repay public debt. The first group, which includes postdefault restructuring countries and countries with public bonds vulnerabilities, consists of high-risk countries, which are labeled sovereign default risk (SDR) countries. I assume that in the second group, sovereign debt crises are triggered by liquidity or banking problems; this group is referred to as no-SDR countries. It includes preemptive default countries and countries that received large IMF packages but have moderate levels of external public debt. SDR countries exhibited higher default risk than the average emerging market country (see appendix A). From the sample of 436 bond issues studied for this paper, 184 bonds are issued during the eleven-year window around sovereign debt crises (from year 5 before to year 5 after the onset of the crisis). Table 3 shows the distribution of bond issues by crisis countries around their crises. Most of the bonds are issued between three years before the crisis (T − 3) and the onset of the crisis (T). Postcrisis (from T + 1 to T + 3), the number of issues drops, before rising again for some default countries (from T + 4 to T + 5). The

140

E C O N O M I A , Fall 2009

T A B L E 3 . Number of Bond Issues around Crisesa Country Argentina Brazil Dominican Rep. Ecuador Indonesia Mexico Pakistan Russia Thailand Turkey Ukraine Uruguay Total

T−5

T−4

T−3

T−2

T−1

T

T+1

T+2

T+3

T+4

T+5

3

4

10 4 1

15 2

13 13

1 6

3

1

2

8

4

4

2

1

3 1

6

5

6

6

3 3 3

13

8

3 11

15

17

1 1 1 1 2 1

1

3

5

7

4 1 9

2 6

3 9

18

3 26

3 40

21

Total 46 43 1 1 1 11 2 6 4 52 3 14 184

Source: Author’s calculations based on Dealogic database. a. T is the onset of the crisis, and it is constructed on an annual basis. T − 1, one year before onset of crisis; T + 1, one year after onset of crisis.

table also shows considerable differences regarding the number of bond issues in crisis countries. Robustness checks are performed to deal with the heterogeneity of the number of bonds issued by crisis countries in the sample.

Results and Robustness Checks This section provides the econometric results and the robustness checks based on the models described in equations 1 and 2.

Results Table 4 summarizes the ordinary least squares (OLS) results (excluding country and time ﬁxed effects) for some variants of equation 1. Column 1 shows a positive and statistically signiﬁcant relationship between fees and primary sovereign bond spreads (SBS), with a point estimate of approximately 0.05 percentage points. With fees averaging 0.6 percent, this is equivalent to 8 percent of the average fee. Assuming that the ﬁxed component of the fee is 0.4 percent (the constant coefﬁcient in the regression), the variable component of the fee is 0.2 percent, and the impact of bond spreads on the variable component of the fee is 25 percent (0.05/0.2). Column 2 shows that the introduction of the dummy variables that track sovereign debt crises does not change the signiﬁcance of the primary bond spread. The crisis dummies

Sebastián Nieto-Parra

141

T A B L E 4 . Fee, Primary Sovereign Bond Spread, and Debt Crises: OLS Regression Modela Explanatory variable SBS

SBS

Total crisis SC

Twin crisis SC and CC

SDR crisis

No-SDR crisis

0.047*** (3.92)

0.027** (2.55) 0.345** (2.18) 0.341*** (2.61) 0.961*** (9.13) 0.531*** (6.46) 0.614*** (8.38) 0.272*** (2.64) −0.083 (0.55) 0.004 (0.03) −0.239* (1.80) −0.208 (1.42) −0.104 (0.70) 0.399*** (9.61)

0.029*** (2.74) 0.345** (2.20) 0.345** (2.53) 1.038*** (9.62) 0.543*** (6.55) 0.612*** (8.44) 0.269*** (2.63) −0.088 (0.59) −0.001 (0.00) −0.243* (1.85) −0.209 (1.43) −0.056 (0.38) 0.394*** (9.56)

0.020* (1.90) 0.328* (1.73) 0.413** (2.43) 1.037*** (9.73) 0.596*** (6.92) 0.699*** (9.13) 0.283*** (2.70) −0.068 (0.46) 0.016 (0.11) −0.232 (1.46) −0.202 (1.30) −0.102 (0.66) 0.427*** (10.45)

0.048*** (3.97) 0.302 (0.91) 0.130 (0.48) −0.218 (0.47) 0.036 (0.14) −0.041 (0.18) 0.162 (0.35) —

T−5 T−4 T−3 T−2 T−1 T T+1 T+2 T+3 T+4 T+5 Constant Summary statistic No. observations R squared

0.449*** (9.23) 419 0.04

419 0.35

419 0.36

419 0.38

— −0.335 (1.25) −0.381 (0.82) −0.059 (0.11) 0.445*** (8.88) 419 0.04

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. Dependent variable is the underwriting fee as a percentage of the amount issued. Absolute value of t statistics shown in parentheses. Abbreviations: SBS, primary sovereign bond spread; SC, sovereign debt crises; CC, currency crises; SDR crisis, countries with sovereign default risk.

are positive and statistically signiﬁcant between ﬁve years before the crisis (T − 5) and the onset of the crisis (T). Similar results are obtained in column 3 for the twin crises (sovereign debt crises and currency crises). Columns 4 and 5 differentiate between sovereign default risk (SDR) countries and no-SDR countries. For SDR countries, there is a positive and signiﬁcant difference between fees in tranquil periods and fees in crisis periods (from T − 4 to T) with the sovereign bond spread only weakly signiﬁcant. For no-SDR countries, only sovereign bond spreads are statistically signiﬁcant, implying that this kind of crisis does not impact fees.

142

E C O N O M I A , Fall 2009

The regressions of table 4 do not fully control for common factors that may have affected the evolution of the sovereign debt market over the period 1993–2006. High fees before the onset of crises could be explained by the reemergence of this market in the 1990s. Similarly, low levels of fees in tranquil periods could be explained by the fact that the market is becoming more mature. To deal with this problem, table 5 reports results of an OLS estimation that includes year ﬁxed effects (equation 1 without country ﬁxed effect). As in table 4, the elasticity of fees with respect to bond spreads is positive and signiﬁcant (column 1), and the elasticity of fees with respect to bond spreads

T A B L E 5 . OLS Time Fixed-Effect Regression Modela Explanatory variable SBS

SBS

Total crisis SC

Twin crisis SC and CC

SDR crisis

No-SDR crisis

0.036*** (3.47)

0.029*** (2.88) 0.078 (0.53) 0.056 (0.46) 0.739*** (7.09) 0.306*** (3.90) 0.403*** (5.87) 0.082 (0.85) −0.079 (0.57) −0.01 (0.07) −0.036 (0.29) −0.023 (0.17) −0.112 (0.84) 0.646*** (4.89)

0.031*** (3.05) 0.084 (0.57) 0.065 (0.52) 0.816*** (7.54) 0.337*** (4.25) 0.406*** (5.97) 0.087 (0.91) −0.097 (0.71) −0.012 (0.09) −0.039 (0.32) −0.019 (0.14) −0.101 (0.77) 0.649*** (4.99)

0.025** (2.37) 0.06 (0.33) 0.098 (0.61) 0.826*** (7.75) 0.378*** (4.50) 0.475*** (6.50) 0.114 (1.16) −0.108 (0.80) −0.013 (0.09) −0.039 (0.26) −0.016 (0.11) −0.099 (0.71) 0.649*** (5.08)

0.035*** (3.23) −0.123 (0.46) −0.190 (0.86) −0.053 (0.14) −0.042 (0.19) 0.028 (0.15) −0.335 (0.88) —

T−5 T−4 T−3 T−2 T−1 T T+1 T+2 T+3 T+4 T+5 Constant Summary statistic No. observations R squared

0.622*** (4.39) 419 0.40

419 0.51

419 0.52

419 0.53

— −0.003 (0.01) −0.035 (0.09) 0.121 (0.26) 0.635*** (4.14) 419 0.40

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a Dependent variable is the underwriting fee as a percentage of the amount issued. Absolute value of t statistics shown in parentheses. For abbreviations, see table 4.

Sebastián Nieto-Parra

143

in tranquil periods is positive and signiﬁcant (column 2). The crisis dummies are positive and statistically signiﬁcant between three (T − 3) and one (T − 1) year before the onset of a debt crisis, indicating that there is a positive difference between fees in tranquil periods and precrisis periods, after controlling for spreads (column 2 of table 5). This result contrasts with those reported in table 4, where fees were signiﬁcant between ﬁve years before the crisis (T − 5) and the onset (T) of the crisis. The same holds for the comparison of twin crises (column 3) and SDR countries (column 4) in tables 4 and 5. High fees in the precrisis periods with respect to fees in tranquil periods are signiﬁcant between three (T − 3) and one (T − 1) year before the onset of a crisis. Column 5 focuses on no-SDR countries, and for this group, there still is no difference between underwriting fees in tranquil and precrisis periods. Figure 5 plots results obtained in table 5, column 2, and the average primary sovereign bond spread around crises. Between three and one year before a crisis, the component of the fee not explained by the spread is high and statistically F I G U R E 5 . Fee and Primary Sovereign Bond Spread around Sovereign Debt Crises, Annual Basisa Fee (percent)

Primary bond spread (percent)

1.2

6

1

5.5

0.8

5

0.6

4.5

0.4 4 0.2 3.5

0 -0.2

3

-0.4

2.5

-0.6

2 T-3

T-2

T-1

T

T+1

T+2

T+3

Variable component of fee not explained by spread 95 percent CI Primary bond spread Source: Authors’ calculations. a. CI refers to confidence interval. T is the onset of a crisis, and it is constructed on an annual basis. T – 1, one year before onset of crisis; T + 1, one year afer onset of crisis.

144

E C O N O M I A , Fall 2009

signiﬁcant, meaning that crisis governments pay investment banks more before a crisis (for example, 0.7 percent of the amount issued three years before the crisis). At the onset of the crisis (time T) and thereafter, the variable component of the fee is instead fully driven by other factors. This result conﬁrms that as crisis countries approach the onset of a crisis, the “information value” of the fees with respect to sovereign bond spreads decreases considerably and then completely disappears when the crisis actually takes place. Next, I include country ﬁxed effects to check whether my results are robust to controlling for a host of other time-invariant country-speciﬁc factors that can affect underwriting fees (such as country credit risk or specific factors of markets in which investors and investment banks are trading). Table 6 includes time and country ﬁxed effects (equation 1) and shows that sovereign bond spreads are not statistically signiﬁcant (column 1), meaning that other factors can explain fees better than sovereign bond spreads. The introduction of the crisis dummies confirms the results of table 5. Between three (T − 3) and one (T − 1) year before the onset of a crisis, the dummy variables for sovereign debt crisis, twin crises, and SDR crisis are positive and signiﬁcantly correlated with underwriting fees (from column 2 to column 4). The results of table 5 for crisis dummy variables of no-SDR countries (column 5) are robust to controlling for country ﬁxed effects (crisis dummies are not signiﬁcant for this type of crisis). The results show that underwriting fees between three and one year before the crisis are signiﬁcantly higher than in tranquil periods, after controlling for bond spreads, time effects, and country effects (equation 1). I also check the robustness of this result by estimating several variants of my baseline model and by experimenting with the lag structure. I ﬁnd the best ﬁt when I use as dummy variable the set of periods between three years and one year before a crisis (an R squared of 0.53 and a t statistic for the crisis dummy variables close to 4.0).20 Table 7 summarizes the results of the benchmark model (equation 2), determined with the best lag structure found in previous regressions. The results of table 6 are robust to this new speciﬁcation. The coefﬁcient of the 20. For each X ∈[1,5], a regression is estimated according to the following model:

(

)it + τ

FEEit = α 1 + α 2 i SBSit + α 3 i CRISIS T − X , . . . , T − 1

t

+ υ i + ε it ,

where CRISIS(T − X, . . . , T − 1) is a dummy variable that takes the value of 1 between the X years before crisis and 1 year before the crisis, and 0 in all other periods. Results are provided upon request.

Sebastián Nieto-Parra

145

Table 6. OLS Time and Country Fixed-Effects Regression Modela Explanatory variable SBS

SBS

Total crisis SC

Twin crisis SC and CC

SDR crisis

No-SDR crisis

−0.013 (0.86)

−0.002 (0.14) −0.001 (0.01) 0.006 (0.04) 0.651*** (4.63) 0.251** (2.32) 0.365*** (3.97) 0.186 (1.44) 0.013 (0.08) 0.062 (0.40) 0.087 (0.59) 0.008 (0.05) −0.113 (0.79) 0.095 (0.24)

−0.002 (0.14) −0.003 (0.02) 0.001 (0.00) 0.649*** (4.53) 0.250** (2.25) 0.364*** (3.88) 0.185 (1.42) 0.012 (0.08) 0.061 (0.39) 0.085 (0.57) 0.007 (0.04) −0.114 (0.78) 0.586 (1.56)

−0.004 (0.26) −0.039 (0.19) 0.003 (0.01) 0.713*** (4.93) 0.339*** (2.83) 0.449*** (4.45) 0.251* (1.84) 0.034 (0.22) 0.083 (0.52) 0.121 (0.68) 0.033 (0.20) −0.079 (0.52) 0.246 (0.65)

−0.015 (0.92) −0.243 (0.77) −0.256 (0.91) 0.248 (0.50) −0.197 (0.70) −0.093 (0.39) −0.217 (0.43) —

T−5 T−4 T−3 T−2 T−1 T T+1 T+2 T+3 T+4 T+5 Constant Summary statistic No. observations R squared

0.489 (1.27) 419 0.51

419 0.56

419 0.56

419 0.57

— −0.078 (0.28) −0.102 (0.25) 0.275 (0.46) 0.465 (1.19) 419 0.51

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. Dependent variable is the underwriting fee as a percentage of the amount issued. Absolute value of t statistics in parentheses. For abbreviations, see table 4.

crisis dummy for three, two, and one year before a crisis is high and signiﬁcant at 1 percent (column 1).21 The elasticity of fees with respect to sovereign bond spreads in tranquil periods is negative but always statistically insignificant, conﬁrming hypothesis 1. Before the onset of sovereign debt crises, investment banks charge an additional fee to risky countries in comparison to nonrisky countries (0.23 percent of the amount issued), whereas investors do not price this risk before the onset of crises. 21. When I cluster the standard errors by year in the benchmark model, t statistic of the crisis is signiﬁcant at 5 percent.

419 0.53

419 0.53

0.287 (0.75)

0.002 (0.13) 0.460*** (3.16) −0.539* (1.75)

−0.010 (0.68) 0.228*** (3.74)

0.299 (0.79)

(2)

(1)

419 0.55

0.536 (1.44)

0.334*** (5.32)

−0.004 (0.29)

(3)

419 0.55

0.521 (1.40)

0.462*** (3.10) −0.024 (0.95)

0.001 (0.09)

(4)

419 0.53

0.267 (0.70)

0.274*** (4.15)

−0.013 (0.86)

(5) 0.004 (0.25)

(6)

419 0.54

0.239 (0.63)

0.672*** (4.05) −0.868*** (2.61)

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. Dependent variable is the underwriting fee as a percentage of the amount issued. Absolute value of t statistics in parentheses. For abbreviations, see table 4.

Summary statistic No. observations R squared

Constant

Interactive no-SDR crisis

No-SDR crisis

Interactive SDR crisis

SDR crisis

Interactive twin crisis SC-CC

Twin crisis SC-CC

Interactive total SC

Total crisis SC

SBS

Explanatory variable

T A B L E 7 . OLS Time and Country Fixed-Effects Regression Model with Interactive Dummies Variablesa

419 0.51

0.490 (1.27)

−0.024 (0.14)

−0.014 (0.87)

(7)

419 0.51

−0.064 (0.20) 0.016 (0.14) 0.491 (1.27)

−0.014 (0.88)

(8)

Sebastián Nieto-Parra

147

Column 2 tests the robustness of the results of column 1 by interacting the crisis dummy with sovereign bond spreads. The interaction term is negative but only weakly signiﬁcant, indicating there is almost no difference in elasticity between crisis and noncrisis periods. Moreover, the elasticity of the fee before crisis with respect to tranquil periods is high and signiﬁcant (0.46 percent). The same holds for columns 3 and 4 that analyze the case of twin crises. Therefore crisis countries pay an extra fee between three and one year before crises in comparison to tranquil periods. Again, I split sovereign debt crises into two groups (SDR and no-SDR countries) and test whether the predictive power of fees is higher in SDR countries (hypothesis 2). Column 5 focuses on SDR countries and shows results similar to those of table 6 (column 4) and thus conﬁrms hypothesis 2. The fee charged to SDR crisis countries is signiﬁcantly higher than in tranquil periods and well above that charged for total debt crisis (0.27 versus 0.23 percent of the amount issued, reported in columns 5 and 1, respectively). Column 6 introduces the interaction term, and the robustness test of column 5 is conﬁrmed, validating hypothesis 2. Investment banks charge a high and signiﬁcant fee with respect to tranquil periods, and well above that charged for the total of debt crisis (0.67 versus 0.46 percent of the amount issued, reported in columns 6 and 2, respectively). Finally, columns 7 and 8 summarize the results for no-SDR countries. As before, they show that no-SDR countries do not pay an additional fee before a crisis with respect to tranquil periods. To sum up, investment banks’ behavior differs depending on the type of crisis. They charge high fees to countries with bad fundamentals. In contrast, countries that enter into a crisis because of liquidity or banking crises pay a fee comparable to that of tranquil periods.

Robustness Checks Table 8 tests the robustness of my results by including additional control variables. I consider the following characteristics of the issue: amount issued (value proceeds), the maturity of the bond (maturity bond), a dummy variable indicating whether the top underwriter placed the bond (top 1 banks dummy), the median of the market share of investment banks (market share banks), a dummy variable indicating whether the bond issued is investment grade (rating), the number of times that the lead investment bank acted as underwriter with a given issuer (stability bank issuer). I also employ the following standard macroeconomic variables: ratio of short-term debt over total external debt (short-term debt/total debt), ratio of external public debt over GDP (public

Market share banks (median)

Top 1 banks dummy

Maturity bond

Value proceeds

Interactive no-SDR crisis

No-SDR crisis

Interactive SDR crisis

SDR crisis

Interactive twin crisis SC-CC

Twin crisis SC-CC

Interactive SC

Total SC

SBS

Explanatory variable

−7.61E-11 (1.62) 0.006** (2.31) −0.119** (2.18) −0.006 (0.67)

9.53E-03 (0.44) 0.579*** (3.50) −0.069** (2.00)

−7.22E-03 (0.36) 0.291*** (3.54)

−7.75E-11 (1.64) 0.005** (2.14) −0.117** (2.15) −0.005 (0.62)

(2)

(1)

−6.70E-11 (1.44) 0.006** (2.33) −0.129** (2.41) −0.006 (0.72)

0.410*** (5.14)

−1.70E-03 (0.09)

(3)

−6.69E-11 (1.44) 0.006** (2.38) −0.128** (2.39) −0.006 (0.73)

0.557*** (3.35) −0.037 (1.01)

5.08E-03 (0.24)

(4)

−7.55E-11 (1.60) 0.005** (2.18) −0.122** (2.23) −0.004 (0.50)

0.347*** (3.86)

−1.38E-02 (0.69)

(5)

−7.37E-11 (1.58) 0.006** (2.47) −0.126** (2.34) −0.005 (0.58)

0.842*** (4.33) −0.108*** (2.86)

7.79E-03 (0.37)

(6)

(7)

−9.38E-11* (1.95) 0.006** (2.40) −0.113** (2.02) −0.006 (0.72)

0.01 (0.05)

−1.25E-02 (0.60)

T A B L E 8 . OLS Time and Country Fixed-Effects Regression Model with Interactive Dummies Variables and Control Variablesa

0.031 (0.09) −0.009 (0.07) −9.36E-11* (1.94) 0.006** (2.40) −0.113** (2.02) −0.006 (0.71)

−1.25E-02 (0.60)

(8)

365 0.58

0.05 (0.43) 0.010* (1.93) 0.006 (1.04) 0.008 (1.32) 0.001 (0.46) 0.006 (0.65) −0.020** (1.97) 6.66E-12 (1.20) 0.012 (1.49) −0.002 (0.94) −0.217 (0.40) 365 0.58

0.056 (0.48) 0.012** (2.15) 0.006 (1.03) 0.007 (1.08) 0.001 (0.66) 0.006 (0.66) −0.018* (1.73) 6.68E-12 (1.21) 0.013 (1.55) −0.002 (1.03) −0.184 (0.34) 365 0.60

0.051 (0.45) 0.011** (2.07) 0.004 (0.68) 0.007 (1.15) 0.001 (0.28) 0.01 (1.12) −0.013 (1.42) 7.88E-12 (1.44) 0.013* (1.65) −0.004 (1.59) 0.097 (0.18) 365 0.60

0.054 (0.47) 0.011** (2.18) 0.004 (0.72) 0.006 (1.07) 0.001 (0.38) 0.009 (1.03) −0.013 (1.43) 7.82E-12 (1.43) 0.013 (1.60) −0.004 (1.53) 0.106 (0.20) 365 0.58

0.07 (0.60) 0.011** (2.01) 0.006 (1.08) 0.01 (1.60) 0.001 (0.12) 0.003 (0.33) −0.021** (2.09) 6.05E-12 (1.09) 0.01 (1.26) −0.002 (0.95) −0.289 (0.53) 365 0.59

0.075 (0.65) 0.013** (2.48) 0.006 (1.10) 0.008 (1.25) 0.001 (0.19) 0.002 (0.27) −0.018* (1.78) 5.76E-12 (1.05) 0.01 (1.28) −0.002 (1.03) −0.227 (0.42)

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. Dependent variable is the underwriting fee as a percentage of the amount issued. Absolute value of t statistics in parentheses. For abbreviations, see table 4.

Summary statistic No. observations R squared

Constant

Number of bonds

Average maturity

Ext. public debt outstanding

Interests/exports

GDP growth

Exchange rate

Public debt/GDP

Short-term debt/total debt

Stability bank issuer

Rating at launch

365 0.56

0.045 (0.37) 0.011* (1.96) 0.006 (1.05) 0.005 (0.85) 0.001 (0.46) −0.002 (0.19) −0.006 (0.58) 5.48E-12 (0.96) 0.009 (1.11) −0.003 (1.12) 0.064 (0.12) 365 0.56

0.046 (0.38) 0.011* (1.95) 0.006 (1.05) 0.005 (0.85) 0.001 (0.46) −0.002 (0.20) −0.006 (0.59) 5.47E-12 (0.96) 0.009 (1.10) −0.003 (1.12) 0.059 (0.11)

150

E C O N O M I A , Fall 2009

debt/GDP), annual exchange rate depreciation (exchange rate), GDP growth rate (GDP growth), ratio of interests of external public debt over exports (interests/exports), the external public debt outstanding (ext. public debt outstanding), the average maturity of the external public debt (average maturity), and the number of bonds placed in the past in international markets (number of bonds).22 I ﬁnd that the crisis dummy is statistically signiﬁcant at 1 percent (column 1), confirming that fees are higher before a crisis than they are during tranquil periods and after controlling for the augmented set of control variables. The additional fee paid is close to 0.3 percent, equivalent to 50 percent of the average fee in the sovereign bond market for the period 1993–2006. Only three control variables are signiﬁcant and positive: the maturity of the bond, the ratio of interests of external public debt over exports, and the dummy variable of the top investment bank (that is, 1 if J. P. Morgan and 0 otherwise). This last variable was coded using both market share—see table 2—and interviews with institutional investors on Wall Street.23 Column 2 conﬁrms the results of table 7. The “precrisis” dummy variable is signiﬁcant, and in addition to the signiﬁcant variables of column 1, the stability bank issuer variable and the interaction term are also significant. In columns 3 and 4, I focus on twin crises and again find that the results are similar to those of table 7 (columns 3 and 4). There is a positive and signiﬁcant difference between the fee before a crisis versus one in tranquil periods, and the interaction term is not signiﬁcant. The years to maturity of the bond, the top 1 bank dummy variable, and the stability bank issuer are signiﬁcant. Columns 5 and 6 show the results for the SDR crisis and conﬁrm the main result of table 7. The underwriting fee before the SDR crisis is higher than during tranquil periods, after controlling for a set of variables. Results of columns 7 and 8 of table 7 (no-SDR crisis) are also robust to this new speciﬁcation, indicating that fees charged to no-SDR countries are not statistically different with respect to the fees in tranquil periods. Table 8 therefore conﬁrms both hypotheses. First, investment banks charge a high and signiﬁcant fee before a crisis. Second, this result is driven by SDR countries. The extra fee is 0.84 percent of the amount issued (column 6),

22. I checked for the presence of multicollinearity of the control variables introduced by computing the variance inﬂation factors. The tolerance for all control variables included in the model was close to 1, conﬁrming the absence of collinearity between regressors. 23. The question asked was, “Which investment banks have the best reputation as underwriters?”

Sebastián Nieto-Parra

151

higher than that paid during the total of sovereign crisis (0.58 percent according to column 2).24 A potential problem of the bond-level regressions presented above is that results might be driven by countries that issued a large number of bonds before a crisis (such as Argentina, Brazil, and Turkey). As a consequence, I reestimated my benchmark model by using countries as a unit of observation. Column 1 of table 9, panel A, focuses on simple OLS (that is, equation 2 without ﬁxed effects) and shows that the elasticity of underwriting fees with respect to sovereign bond spreads is positive and signiﬁcant. A Hausman test shows that the ﬁxed-effects model is preferable to a random-effect model.25 Column 2 of panel A reports the ﬁxed-effects model. The results of table 7 (column 1) are robust to this new speciﬁcation. There is a statistically significant difference between the fees charged in precrisis periods versus those in tranquil periods. The elasticity of underwriting fees with respect to bond spreads is negative but not signiﬁcant in the ﬁxed-effects model. Column 3 of panel A reports ﬁxed effects with year dummy variables. Again, I ﬁnd a positive and signiﬁcant difference between the fee during precrisis periods and that during tranquil periods, after controlling for sovereign bond spreads (conﬁrming hypothesis 1). The same holds for twin crises—sovereign debt crisis and currency crisis—from columns 4 to 6 of panel A. Crisis governments pay to investment banks an additional fee before crisis in comparison to other countries and tranquil periods. Next, I differentiate between types of crises (panel B, from column 1 to column 3 for SDR countries, and from column 4 to column 6 for no-SDR countries) to test whether the predictive power of fees is higher for SDR countries (hypothesis 2). All speciﬁcations for SDR countries conﬁrm hypothesis 2. In particular, a ﬁxed-effects model with year dummies (column 3 of panel B) shows that the additional fee that SDR countries have to pay before a crisis is signiﬁcant and higher than for total sovereign debt crises (0.26 percent versus 0.18 percent in column 3 of panel A). Again, the results of table 7 are conﬁrmed regarding no-SDR countries. The additional fee paid before a crisis by these countries is not signiﬁcant in all speciﬁcations (see columns 4 to 6 of panel B). 24. When I include a dummy variable for issues incorporating collective action clauses (CAC dummy), hypotheses 1 and 2 are also confirmed. Results are provided upon request. The CAC dummy variable is measured from 1993 to 2002 by a dummy variable that takes the value of 1 for issues underwritten under the U.K. governing law and 0 otherwise (Drage and Hovaguimian 2004). 25. Results are provided upon request.

152

E C O N O M I A , Fall 2009

T A B L E 9 . Panel Data for Fee, Primary Sovereign Bond Spread, and Debt Crisesa Total SC Explanatory variable Panel A SBS Total SC

Twin crisis SC-CC

OLS

FE

FE with time effect

0.024** (1.99) 0.374*** (5.44)

−0.023 (1.29) 0.338*** (4.31)

−0.006 (0.42) 0.181*** (3.23)

Twin SC-CC Constant Summary statistic No. observations R squared

0.443*** (10.08) 169 0.18

0.596*** (9.77)

0.884*** (9.63)

169 0.13

169 0.64

OLS 0.025** (2.06)

0.386*** (5.49) 0.441*** (10.06) 169 0.18

SDR countries

Panel B SBS SDR crisis

OLS

FE

FE with time effect

0.020* (1.69) 0.463*** (6.15)

−0.021 (1.20) 0.448*** (5.06)

−0.007 (0.47) 0.260*** (4.10)

Summary statistic No. observations R squared

0.456*** (10.62) 169 0.22

FE with time effect

−0.023 (1.29)

−0.006 (0.42)

0.338*** (4.31) 0.598*** (9.82)

0.181*** (3.23) 0.885*** (9.65)

169 0.13

169 0.64

No-SDR countries

No-SDR crisis Constant

FE

OLS

FE

FE with time effect

0.327** (2.52)

−0.026 (1.40)

−0.013 (0.83)

0.007 (0.04) 0.645*** (10.10)

−0.048 (0.42) 0.725*** (8.73)

0.588*** (9.84)

0.700*** (8.95)

0.015 (0.09) 0.456*** (9.53)

169 0.17

169 0.66

169 0.04

169 0.01

169 0.62

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. Dependent variable is the underwriting fee as a percentage of the amount issued. OLS, FE (fixed effects), and FE with time-effect regression models. Absolute value of t statistics in parentheses. For abbreviations, see table 4.

Finally, I test the out-of-sample properties of the model (table 10). First, I include underwriting fees in a model that estimates the probability of sovereign debt crises. If fees before onset of sovereign debt crises are high with respect to tranquil periods, the fee should be an early warning indicator of these crises. I estimate a logit model (1 if sovereign debt crisis and 0 otherwise) by including the lagged value of the underwriting fee and controlling for the lagged values of the macroeconomic and political variables employed

Sebastián Nieto-Parra

153

T A B L E 1 0 . Determinants of Sovereign Debt Crisis and High Increases of the EMBI Spreada Probability (crisis) Explanatory variable

1

Underwriting fee Total external debt/GDP Total debt over reserves Internal short-term debt/GDP External debt services/reserves Current account/GDP Openness/GDP US Treasury bill rate GDP growth rate Inflation volatility Dummy high inflation (> 50 percent) Presidential election Index of freedom status Constant

Summary statistic No. observations Pseudo–R squared

0.002 (0.13) 0.005* (1.87) 0.965 (0.72) 0.008** (2.21) −0.160 (1.26) −0.014 (1.51) 0.216 (0.76) −0.100 (1.55) 0.214** (2.25) 0.620 (0.60) −0.057 (0.06) −0.256 (0.61) −5.219*** (3.84)

339 0.25

Probability (high increases of EMBI) 2

2.786** (1.96) 0.083** (2.04) 0.015* (1.65) −3.345 (1.21) −0.001 (0.07) −0.601** (2.18) −0.011 (0.68) 0.876 (1.47) −0.369** (2.33) 0.490*** (2.92) 0.243 (0.23) — −2.843** (2.20) −12.588*** (3.43)

136 0.48

3

4

−0.045 (1.31) 0.003 (0.66) 4.879 (1.39) −0.001 (0.14) 0.051 (0.63) −0.061** (2.47) −0.196 (0.61) −0.013 (0.14) 0.217* (1.68) 2.700* (1.84) 1.949*** (2.82) 0.643 (0.64) −0.472 (0.22)

1.417** (2.02) 0.199* (1.81) −0.344*** (2.76) 47.384*** (2.60) 0.03 (1.60) 0.378 (1.21) −0.488** (2.53) −0.562 (0.80) 0.578*** (3.02) −0.196 (0.90) 3.270* (1.94) 4.645*** (3.08) 0.137 (0.29) −1.407 (0.98)

138 0.40

107 0.72

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. Dependent variables are sovereign debt crisis (regressions 1 and 2) and high increases of the EMBI spread (regressions 3 and 4). Logit model with lagged regressors of one year. Robust z statistics in parentheses.

154

E C O N O M I A , Fall 2009

in previous work aimed at predicting debt crises (Manasse, Roubini, and Schimmelpfennig 2003).26 I use a robust variance estimator (Huber-White sandwich estimator) with country-speciﬁc variances. Column 1 of table 10 reports the results of the estimation for the set of macroeconomic and political variables. Excluding the dummy of presidential elections, all lagged regressors have the expected sign. In particular, there is a positive and signiﬁcant impact of the external debt service over reserves and the volatility of the inﬂation rate on the probability of sovereign debt crises. The same holds for the ratio of total debt over reserves, but it is only weakly signiﬁcant. Column 2 of table 10 includes the lag of the underwriting fee in the estimation shown in column 1. An increase of the underwriting fee one year before the sovereign debt crises helps to predict these crises. The lag of this variable is positive and statistically signiﬁcant, and its inclusion improves the ﬁt of the model (the pseudo–R squared is close to 0.48 versus 0.25 in the regression of column 1). Moreover, while sovereign debt crises are endogenous to the standard variables presented above (that is, the increases and decreases of these variables around crises are the causes—and consequences— of crises), the underwriting fee is likely to be an exogenous variable. Finally, in this new speciﬁcation, there is a positive impact on the lag of the total external debt over GDP and the lag of the volatility of inﬂation on the probability of crises. In contrast, there is a negative and signiﬁcant impact of the lag of the current account over GDP, the lag of the GDP growth rate, and the lag of the freedom status on the probability of a crisis. Is the fee a good predictor of high increases in the sovereign bond spread? To test this hypothesis, I deﬁne high increases of the EMBI Global spread when this index is above the eightieth percentile of the sample (that is, above 792 basis points) and build a dummy variable that takes a value of 1 when the EMBI spread crosses this threshold and 0 otherwise.27 Column 3 reports the results of the logit model using the same explanatory variables used by Manasse, Roubini, and Schimmelpfennig (2003). A reduction of openness, a high level and volatility of inﬂation, and presidential elections are signiﬁcant 26. These variables are total external debt over GDP, short-term debt over international reserves, external debt service over reserves, current account balance over GDP, openness, U.S. Treasury bill rate, real GDP growth, inflation volatility, the dummy for inflation higher than 50 percent, the dummy for presidential election, and an index of freedom status (for a description of these variables, see Manasse, Roubini, and Schimmelpfennig 2003). 27. If the EMBI spread is above the eightieth percentile in the four years following the initial increase of the spread, the dummy variable is 0.

Sebastián Nieto-Parra

155

variables to explain high increases of the sovereign bond spreads in the subsequent year.28 Column 4 introduces the lag of the underwriting fee in the estimation and shows that its coefficient is positive and statistically significant. Summing up, the results presented in table 10 show the out-of-sample properties of the model. An increase in the fee one year before a crisis is a good early warning indicator of sovereign debt crises.

Underwriting Fees and Financial Markets Actors Information on fees can easily be found in Bloomberg’s databases or in Dealogic’s DCM Analytics database.29 There is, however, a lag of one to seven days between an issue’s announcement date and the day that ﬁnancial databases release information on the fee.30 Because market actors have access to information on the underwriting fee after the issue date, it is valuable to study secondary market prices to determine if investors track this information to price sovereign bonds issued by potential “crisis” countries. To that end, I replicate the analysis presented in table 8 for the benchmark model (equation 2), and I use the EMBI Global spread for one day (EMBI1) after the issue date, as well as the ten-day EMBI Global spread average (EMBIAV10) after the issue dates of the sovereign bonds used in the sample. Results are reported in tables 11 and 12, respectively. Column 1 of table 11 uses the secondary sovereign bond spread one day after the issue (EMBI1). It confirms my main results as it shows that the

28. However, the high level and volatility of the inflation are only weakly significant at 10 percent. 29. Although investment banks have no obligation to submit deal information and consequently fees for issues, they have an incentive to provide this information because such databases compile rankings of primary bond market leaders (“league tables”) through the deals investment banks make. This information is an important benchmark for market makers, issuers, analysts, and ﬁnancial media where investment banks’ reputation is measured in market share (see Bloomberg 2006). 30. According to a Dealogic employee in the United Kingdom at the end of 2007, “For about 80 percent of large deals (more than USD 200 million equivalent), we should have the fee within one day. For the remaining 20 percent, it would be, on average, within one week. For smaller deals [for example, small medium-term notes], it may take two to three weeks until we receive the pricing supplement.” Regarding Bloomberg, for instance, for the Colombian Global Bond 09/08/06 (ISIN number XS0213272122), the information concerning the fee was obtained one week after the issue date. Moreover, this piece of information was neither disclosed by the Colombian government nor the investment banks in their external documents or websites on the day of the issue.

379 0.56

379 0.57

0.855** (1.97)

−0.018 (1.09) 0.667*** (3.64) −0.077** (2.54)

−0.034** (2.22) 0.238*** (3.32)

0.764* (1.76)

(2)

(1)

379 0.58

0.926** (2.20)

0.364*** (4.98)

−0.028* (1.82)

(3)

379 0.58

0.892** (2.11)

0.605*** (3.21) −0.045 (1.39)

−0.021 (1.34)

(4)

379 0.56

0.774* (1.79)

0.260*** (3.52)

−0.036** (2.35)

(5)

379 0.57

0.898** (2.08)

0.762*** (4.00) −0.088*** (2.85)

−0.020 (1.20)

(6)

379 0.54

0.992** (2.26)

−0.064 (0.24)

−0.036** (2.26)

(7)

379 0.54

−0.190 (0.24) 0.035 (0.17) 0.995** (2.27)

−0.036** (2.26)

(8)

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. With interactive dummies variables. Dependent variable is the fee as a percentage of the amount issued. Absolute value of t statistics in parentheses. EMBI1 refers to the one-day EMBI Global spread one day after the issue date. For abbreviations, see table 4.

Summary statistic No. observations R squared

Constant

Interactive no-SDR crisis

No-SDR crisis

Interactive SDR crisis

SDR crisis

Interactive twin crisis SC-CC

Twin crisis SC-CC

Interactive total SC

Total crisis SC

EMBI1

Explanatory variable

T A B L E 1 1 . Fee, Secondary Sovereign Bond Spread (EMBI1), and Debt Crises: OLS Time and Country Fixed-Effects Regression Modela

379 0.56

379 0.57

0.842** (1.98)

−0.018 (1.10) 0.690*** (3.87) −0.081*** (2.78)

−0.035** (2.34) 0.236*** (3.30)

0.771* (1.79)

(2)

(1)

379 0.58

0.930** (2.23)

0.362*** (4.94)

−0.028* (1.92)

(3)

379 0.58

0.896** (2.15)

0.637*** (3.48) −0.052 (1.64)

−0.021 (1.36)

(4)

379 0.56

0.778* (1.82)

0.258*** (3.50)

−0.037** (2.47)

(5)

379 0.57

0.935** (2.42)

0.785*** (4.25) −0.092*** (3.10)

−0.019 (1.21)

(6)

379 0.55

1.003** (2.32)

−0.068 (0.25)

−0.037** (2.41)

(7)

379 0.55

−0.205 (0.28) 0.037 (0.20) 1.007** (2.32)

−0.037** (2.42)

(8)

*Statistically significant at the 10 percent level; ** statistically significant at the 5 percent level; *** statistically significant at the 1 percent level. a. With interactive dummies variables. Dependent variable is the fee as a percentage of the amount issued. Absolute value of t statistics in parentheses. EMBIAV10 refers to the ten-day EMBI Global spread average after the issue date. For abbreviations, see table 4.

Summary statistic No. observations R squared

Constant

Interactive no-SDR crisis

No-SDR crisis

Interactive SDR crisis

SDR crisis

Interactive twin crisis SC-CC

Twin crisis SC-CC

Interactive total SC

Total crisis SC

EMBIAV10

Explanatory variable

T A B L E 1 2 . Fee, Secondary Sovereign Bond Spread (EMBIAV10), and Debt Crises: OLS Time and Country Fixed-Effects Regression Modela

158

E C O N O M I A , Fall 2009

crisis dummy (which takes value 1 between T − 3 and T − 1 before the onset of crisis T) is positive and statistically signiﬁcant. When I include the interactive dummy variable, the interaction term is negative and signiﬁcant, and the crisis dummy is again positive and statistically signiﬁcant. The ﬁxed cost crisis countries paid to investment banks remains high before the onset of a crisis (higher than 0.60 percent of the amount issued). The main result holds for twin crises (columns 3–5). When I split sovereign debt crises into two groups, I ﬁnd that the fee is high and signiﬁcant at 1 percent for SDR countries only (column 5). By including the interactive dummy variable (column 6), the t statistic of the slope is negative and signiﬁcant at 1 percent. By contrast, for no-SDR countries, there is no signiﬁcant difference between the fee paid before a crisis and in tranquil periods. My results are robust to using the average of the EMBI spread ten days after the issue date (table 12). These results also hold—for both EMBI1 and EMBIAV10—when I control for the set of economic and ﬁnancial variables of table 8.31 To sum up, my results suggest that investors do not make use of information on fees in pricing bonds in the secondary market. This ﬁnding was conﬁrmed by a series of interviews with investment ﬁrms in which institutional investors were asked questions about their perceptions of the structure of the primary sovereign bond market.32 Seven investors out of the eight interviewed said that underwriting fees play no role in their investment decisions.33 Investors argued that because fees are formed by the connection between investment banks and governments, they are of no interest to them. As investors appear to give some weight to the opinion expressed in investment banks’ publications on emerging sovereign bond markets, I surveyed 600 publications issued by thirteen investment banks over the period 1997–2007.34 These publications present detailed information related to the 31. Results available upon request. 32. The questions asked regarding the relevance of the fee for investors were: “Are underwriting fees of any relevance to an investor?” and “Is underwriting fee a good indicator of credit risk?” 33. The one institutional investor interested in fees is not directly linked to the impact of fees on credit risk analysis. According to that investor, fees can serve to indicate the effort the underwriters will put into the performance of the issue. (“Fees can tell me how the banker is biased about issuance performance.”) 34. The investment banks are ABN AMRO, Barclays Capital, Bear Stearns, Citigroup (the former Salomon Smith Barney), Credit Suisse (the former Credit Suisse First Boston), Deutsche Bank, Dresdner Kleinwort Wasserstein, Goldman Sachs, J. P. Morgan, Lehman Brothers, Merrill Lynch, Morgan Stanley, and UBS.

Sebastián Nieto-Parra

159

primary bond market. In particular, a vast description of the structure of bonds issued (for example, outstanding amount, coupon rate, maturity, primary bond spread, and currency denomination) is presented, as well as forecasts concerning future public bond issues, depending on the public financing needs of each emerging country. However, no publication reports information on underwriting fees. The survey and empirical results show that underwriting fees are not used as a tool in determining portfolio allocations. It is puzzling that useful, publicly available information is not tracked by investors. This is directly connected with the study of the efﬁciency of the sovereign bond market. Why do investors not pay attention to the evolution of fees? It is possible that the sort of market inefﬁciency documented in this paper is driven by the phenomenon of “herding behavior.” Individual investors may not pay attention to useful public information because they are only concerned with the variables, which are considered “leading indicators” by the rest of the market.35 This is just another example of Keynes’s “beauty contest.”36

Conclusions This paper analyzes the behavior of investment banks and investors around sovereign debt crises, by studying the structure of the primary sovereign bond market over the period 1993–2006. It ﬁnds that one cannot reject the hypothesis that investment banks price sovereign default risk well before crises emerge, well before investors do. Investment banks charge a much higher underwriting fee between three years and one year before a crisis than they do during tranquil periods. This result is statistically signiﬁcant after controlling for sovereign bond spreads and other variables. My results suggest that the additional fee paid by governments to investment banks before crisis (0.31 percent of the amount issued) is equivalent to 50 percent of the average fee in the sovereign bond market. 35. Moreover, this paper has used standard bond issues—ﬁxed-coupon, denominated in the most important currencies (U.S. dollar, euro, and yen), and no international guarantee—to analyze underwriting fees. The determinants of fees in emerging markets could be more complex, making it more difﬁcult to extract from them information that could usefully serve as an early warning indicator. At a ﬁrst glance, individual investors may therefore have no incentive in incurring the ﬁxed cost of analyzing fees. 36. In that context, see Lamont and Thaler (2003) for the case of mispricing in the tech stock market, even after the arbitrage opportunities in that market were made public (through the press).

160

E C O N O M I A , Fall 2009

Moreover, I ﬁnd that investment banks’ behavior differs depending on the type of debt crisis. When I split debt crises into two groups, based on the sovereign default risk (SDR) before crisis, I ﬁnd that the additional fee paid by SDR crisis countries (0.39 percent of the amount issued) is signiﬁcantly higher than they pay during tranquil periods and above that charged for total debt crisis. Finally, I show that underwriting fees can be used as early warning indicators of debt crises, after controlling for standard economic variables, and that increases of the secondary sovereign bond spread could be explained by the underwriting fee as well. These results show that underwriting fees provide valuable information. It is puzzling that investors do not use this potentially useful public information in order to allocate efﬁciently their portfolios of emerging market ﬁxedincome assets. Should policymakers promote the dissemination of underwriting fees? If so, how can policymakers increase the use of this type of information by market participants?37 These questions do not have straightforward answers since they involve several trade-offs. Advertising the importance of underwriting fees and promoting their use may improve market efﬁciency, but it may also lead investment banks to adopt less transparent pricing schemes that would hide the information currently incorporated in the fees. Along similar lines, sovereign issuers with poor fundamentals would also have incentives to alter the fees in order not to be charged higher bond spreads.

37. Market inefﬁciency in the primary market has several consequences for the three main actors of the sovereign bond market: investors, investment banks, and governments. Investors’ losses incurred upon default are higher than in an efﬁcient market. In turn, due to market inefﬁciency, investment banks are likely to obtain higher proﬁts (that is, higher fees in the primary market and higher commissions in the secondary market before the onset of a debt crisis). Moreover, it is uncertain whether market inefﬁciency is ultimately beneﬁcial or harmful to risky issuers in this context. On the one hand, it is probable that, in exchange for a higher fee, issuers can sell a bond for a higher price than would be accepted in a perfectly competitive market. Second, given that in a perfectly competitive market, self-fulﬁlling effects can trigger crises, investors’ lack of information may serve to sustain demand for the issues of a country with bad fundamentals, maintaining its access to ﬁnancing through a risky period and perhaps helping to avert a crisis. Nevertheless, market inefﬁciency may also induce governments to increase outstanding debt beyond ﬁnancial capacity. Thus a debt crisis may only be postponed, leaving a country with a higher debt burden and worse fundamentals than an efﬁcient market would have allowed. In this case, market inefﬁciency may serve to aggravate a looming debt crisis.

Sebastián Nieto-Parra

161

Appendix A. Debt Risk Index before the Onset of a Crisis, Annual Basisa Total debt Type of debt crisis and country

Public debt

Public bonds

Time

Outstanding index

Over MI countries

Index

Over MI countries

Index

Over MI countries

T−3 T−2 T−1

5 4 4

0.8 0.6 0.6

1 1 1

0.7 0.5 0.4

5 4 4

1.1 0.8 0.8

Pakistan 1999

T−3 T−2 T−1

17 15 16

2.5 2.2 2.4

15 13 14

3.2 2.8 3.0

0 0 0

0.1 0.2 0.2

Uruguay 2003

T−3 T−2 T−1

9 10 16

1.3 1.4 2.4

4 5 8

2.5 3.1 4.9

8 9 15

1.7 1.9 3.1

Dominican Rep 2005

T−3 T−2 T−1

7 12 7

1.0 1.9 1.0

2 4 2

1.1 2.6 1.4

6 12 7

1.3 2.6 1.4

T−3 T−2 T−1

6 7 7

0.9 1.0 1.0

0 0 0

0.1 0.1 0.2

6 6 7

1.3 1.4 1.4

Ecuador 1999

T−3 T−2 T−1

12 13 15

1.9 1.9 2.3

6 6 8

3.6 3.5 4.6

12 13 15

2.6 2.7 3.2

Argentina 2001

T−3 T−2 T−1

15 18 18

2.2 2.7 2.7

9 11 12

5.4 6.4 7.2

11 13 14

2.3 2.8 3.0

9 9 13

1.4 1.3 1.9

8 6 10

1.7 1.4 2.2

4 3 6

2.6 2.0 3.6

Default preemptive Ukraine 1998

Post–sovereign default Russia 1998

IMF package, public bonds vulnerabilities Mexico 1995 T−3 T−2 T−1 Brazil 1998

T−3 T−2 T−1

10 11 13

1.5 1.6 1.9

5 4 4

2.9 2.4 2.4

8 7 6

1.6 1.4 1.4

Turkey 2000

T−3 T−2 T−1

8 9 10

1.2 1.3 1.5

2 2 3

1.3 1.4 1.5

6 6 6

1.3 1.3 1.4

Brazil 2001

T−3 T−2 T−1

17 22 19

2.6 3.3 2.8

4 5 6

2.2 3.2 3.4

7 9 9

1.5 2.0 1.9 (continued)

162

E C O N O M I A , Fall 2009

(Continued ) Total debt Type of debt crisis and country

Time

Outstanding index

IMF package, no public bonds vulnerabilities Indonesia 1997 T−3 12 T−2 12 T−1 11 Thailand 1997

Total MI countries

Public debt

Public bonds

Over MI countries

Index

Over MI countries

Index

Over MI countries

1.8 1.8 1.7

9 8 7

1.9 1.8 1.5

0 0 0

0.0 0.0 0.1

T−3 T−2 T−1

5 5 6

0.7 0.8 0.8

2 2 2

0.4 0.4 0.3

0 0 0

0.1 0.1 0.1

1995–2002

7

1.0

2

1.0

5

1.0

Source: Author’s calculation based on the World Bank’s Global Development Finance database. a. GDF only calculates these indicators for the total debt. I have adapted these definitions to the public debt and public bonds. The indicators used to calculate this index are: debt service over exports of goods and services, interest payments over exports of goods and services, debt over exports of goods and services, international reserves over debt, debt over GNP, and interest payments over GNP. In order to construct this index, I give the same weight to each indicator according to its value for middle income (MI) countries during the period 1995–2002 (the period that encloses the entire crises sample). MI countries are defined according to the World Bank; see “Data and Statistics” (www.worldbank.org). All the countries studied in this paper are included inside this category.

2001 2005 1999 1999 1998 1998 2003

Country

Argentina Dominican Republic Ecuador Pakistan Russia Ukraine Uruguay

2005 2005 1999–2000 1999 1998–2000 1998–2000 2003

Debt restructuring yearb 33 92 44 65 18 69 66

Average trading price (percent of par)c 30 95 60 65 50 60 85

PV ratio of cash flows (percent)d 56.0 0.0 37.3 −1.0 17.2 0.0 1.0

Nominal principal reduction (percent debt restructured)e 64–82 n.a. 19–47 29–32 50–75 22–35 5–20

Haircut (percent) f

13.65 0.39 1.32 n.a. 6.32 n.a. 0.12

Weight in the secondary market (percent total emerging sovereign debt) g

Post default Preemptive Post default Preemptive Post default Preemptive Preemptive

Restructuring case b

Sources: Bedford, Penalver and Salmon (2005), IMF (2006), Moody’s (2006), Standard and Poor’s (2006), and Sturzenegger and Zettelmeyer (2005). a. According to Standard & Poor’s (2006) b. According to IMF (2006). c. Thirty-day postdefault price or predistressed exchange trading price (Moody’s 2006). d. Ratio of the present value (PV) of cash flows received as a result of the distressed exchange versus those initially promised, discounted using yield to maturity immediately before default (Bedford, Penalver and Salmon 2005). e. Negative numbers indicate an increase in principal (IMF 2006). f. According to Sturzenegger and Zettelmeyer (2005). g. Corresponds to the weight of each country one month before the onset of the crisis (according to the weight of the EMBI Global index calculated by J. P. Morgan).

Default year a

Units as indicated

Appendix B. Debt Restructuring Cases and Recovery Rates

164

E C O N O M I A , Fall 2009

Appendix C. Debt Risk Index at Onset of the Crisisa Mexico 1995

Indonesia 1997

Thailand 1997

Brazil 1998

Turkey 2000

Brazil 2001

Middle income 1995–2002

10.6 1.6 1.6

10.9 1.6 1.7

7.2 1.1 1.1

17.0 2.6 2.4

10.2 1.5 1.5

18.1 2.7 2.7

6.7 1.0

Public debt index 1995–2002b Crisis datec

8.6 1.9 1.6

6.2 1.3 1.3

2.0 0.4 0.4

7.0 1.5 1.5

6.7 1.4 1.5

8.4 1.8 2.0

4.7 1.0

Public bonds index 1995–2002b Crisis datec

4.6 2.7 3.0

0.2 0.1 0.1

0.3 0.2 0.2

3.7 2.2 2.2

3.0 1.8 1.6

5.3 3.2 3.1

1.7 1.0

Debt risk indexes Total debt outstanding index 1995–2002b Crisis datec

Source: Author’s calculation based on the World Bank’s Global Development Finance database. a. Sample: nonconcessional IMF loans/quota > 100 percent. For details regarding the construction of the debt indexes, see appendix A. b. Debt indicator of the crisis country divided by the average of the period 1995–2002 of the debt indicator for middle income countries. c. Debt indicator of the crisis country divided by the debt indicator for middle income countries at the crisis year.

Comment Alicia Garcia Herrero and Enestor Dos Santos: Nieto-Parra’s paper clearly touches upon an interesting topic, namely, the microstructure of emerging countries’ sovereign bond markets. The idea that fees contain important economic information seems fruitful. Although the fact that investment banks price sovereign default risk well before crises occur and before investors detect default risk is, per se, interesting, it seems to us that the main contribution of the paper is to suggest that this information can be used to foresee a sovereign debt crisis. Some evidence regarding this utilization of data on fees is presented by the author, but we think that this issue could be more carefully studied in future works. In doing so, it would be interesting to look deeper into the fees charged across the whole sample and not only in the crisis years, as well as to see how underwriting fees evolve during noncrisis times. The comparison of noncrisis with crisis years seems natural and pertinent. The regressions presented in the paper are based on bond level. This partially addresses the problem of deﬁning a biased frequency (yearly, quarterly, or monthly). Throughout the paper, however, the author notes that investment banks charge higher fees three years before a crisis. Although we can ﬁnd in ﬁgure 3 information on how the retention coefﬁcient (fee over sovereign bond spread) performs months before the onset of the crisis, it would have been interesting to consider using quarterly or even monthly data throughout the study. Yearly aggregation can hide some important features. Moreover, it seems natural to think that some crises could be anticipated months before they occur rather than years. Although we have not observed purely sovereign debt crises for emerging economies in recent years, another notable way to extend the current work (and to check the assumptions tested) would be the use of recent data (up to 2009) to see how fees and spreads behaved before the current crisis for different groups of emerging countries. 165

166

E C O N O M I A , Fall 2009

There are some other stories that could be told to explain why fees and spread evolve in different ways before, during, and after crises—for example, different cost structure or different risk aversion between investors and investment banks or differences in the markets in which they are traded. It would be interesting to test the importance of these other stories and how robust the results are to the inclusion of variables capturing them. As emphasized by the author, it is puzzling that underwriting provides useful information that investors do not employ when making their portfolio decisions. This puzzle has diverse implications for public policy, and its revelation by the author is certainly another signiﬁcant contribution of the paper.

Sebastián Nieto-Parra

167

References Altinkihc, O., and R. Hansen. 2000. “Are There Economies of Scale in Underwriting Fees? Evidence of Rising External Financing Costs.” Review of Financial Studies 13 (1): 191–218. Bedford, P., A. Penalver, and C. Salmon. 2005. “Resolving Sovereign Debt Crises: The Market-Based Approach and the Role of the IMF.” Financial Stability Review (June): 99–108. Becker T., A. Richards, and Y. Thaicharoen. 2003. “Bond Restructuring and Moral Hazard: Are Collective Action Clauses Costly?” Journal of International Economics 61 (1): 127–61. Bloomberg. 2006. “The Bloomberg 20. The World’s Best-Paid Investment Banks.” Bloomberg Markets (April). Borensztein, E., and R. Gelos. 2000. “A Panic-Prone Pack? The Behavior of Emerging Market Mutual Funds.” Working Paper WP/00/198. Washington: IMF (December). Calomiris, C. W. 2003. “Lessons from Argentina and Brazil.” Cato Journal 23 (1): 33–45. Calvo, G. A. 1998. “Varieties of Capital Market Crises.” In The Debt Burden and Its Consequences for Monetary Policy, edited by G. Calvo and M. King, pp. 181–202. Basingstoke: Palgrave Macmillan. ———. 2002. “Contagion in Emerging Markets: When Wall Street Is a Carrier.” In Proceedings from the International Economic Association Congress, vol. 3. Buenos Aires. Carter, R., and S. Manaster. 1990. “Initial Public Offering and Underwriter Reputation.” Journal of Finance 45 (4): 1045–67. Ciarlone, A., and G. Trebeschi. 2005. “Designing an Early Warning System for Debt Crises.” Emerging Market Review 6 (4): 376–95. Drage, J., and C. Hovaguimian. 2004. “Collective Action Clauses (CACS): An Analysis of Provisions Included in Recent Sovereign Bond Issues.” Bank of England International Finance Division (November). Eichengreen, B., and A. Mody. 2004. “Do Collective Action Clauses Raise Borrowing Costs?” Economic Journal 114 (495): 247–64. Eichengreen, B., C. Rose, and C. Wyplosz. 1996. “Contagious Currency Crises.” Working Paper 5681. Cambridge, Mass.: National Bureau of Economic Research. Fioramanti, M. 2006. “Predicting Sovereign Debt Crises Using Artificial Neural Networks: A Comparative Approach.” Working Paper 72. Rome: Istituto di Studi e Analisi Economica (October). Flandreau, M., and J. H. Flores. 2009. “Bonds and Brands: Foundations of Sovereign Debt Markets, 1820–1830.” Journal of Economic History 69 (3): 646–84. Flandreau, M., J. H. Flores, N. Gaillard, and S. Nieto-Parra. 2010a (forthcoming). “The End of Gatekeeping: Underwriters and the Quality of Sovereign Bond Markets, 1815–2007.” In NBER International Seminar on Macroeconomics 2009, edited by L. Reichlin and K. West. University of Chicago Press.

168

E C O N O M I A , Fall 2009

———. 2010b (forthcoming). “Two Centuries of Government Bond Underwriting.” In Institutions, Markets and Capital Flows from the 1880s to the Present: Why Are Financial Centres Attractive? edited by Y. Cassis and L. Quennouëlle-Corre. Oxford University Press. Fostel, A., and Kaminsky, G. L. 2007. “Latin America’s Access to International Capital Markets: Good Behavior or Global Liquidity?” Working Paper 13194. Cambridge, Mass.: National Bureau of Economic Research. Gelos, G., R. Sahay, and G. Sandleris. 2004. “Sovereign Borrowing by Developing Countries: What Determines Market Access?” Working Paper 04/221. Washington: IMF (November). Grigorian, D. 2003. “On the Determinants of First-Time Sovereign Bond Issues.” Working Paper 03/184. Washington: IMF (September). Gugiatti M., and A. Richards. 2003. “Do Collective Action Clauses Inﬂuence Bond Yields? New Evidence from Emerging Markets.” Research discussion paper. Reserve Bank of Australia (March). Higgins, W. W., and B. J. Moore. 1980. “Market Structure versus Information Costs as Determinants of Underwriters’ Spreads on Municipal Bonds.” Journal of Financial and Quantitative Analysis 15 (1): 85–97. Hua Fang, L. 2005. “Investment Bank Reputation and the Price and Quality of Underwriting Services.” Journal of Finance 60 (6): 2729–61. IMF (International Monetary Fund). Policy Development and Review Department. 2006. “Cross-Country Experience with Restructuring of Sovereign Debt and Restoring Debt Sustainability.” Washington (August). James, C. 1992. “Relationship-Speciﬁc Assets and the Pricing of Underwriter Services.” Journal of Finance 47 (5): 1865–85. Kollo, M. G., and I. G. Sharpe. 2006. “Relationships and Underwriter Spreads in the Eurobond Floating Rate Note Market.” Journal of Financial Research 29 (2): 163–80. Lamont O. A., and R. H. Thaler. 2003. “Can the Market Add and Subtract? Mispricing in Tech Stock Carve-outs.” Journal of Political Economy 111 (2): 227–68. Lee, I., S. Lochhead, J. Ritter, and Q. Zhao. 1996. “The Costs of Raising Capital.” Journal of Financial Research 19 (1): 59–74. Livingston, M., and R. E. Miller. 2000. “Investment Bank Reputation and the Underwriting of Nonconvertible Debt.” Financial Management 29 (2): 21–34. Manasse, P., and N. Roubini. 2005. “Rules of Thumb for Sovereign Debt Crises.” Working Paper 0542. Washington: IMF. Manasse, P., N. Roubini, and A. Schimmelpfennig. 2003. “Predicting Sovereign Debt Crises.” Working Paper 221. Washington: IMF. Melnik, A., and D. Nissim. 2003. “Debt Issue Costs and Issue Characteristics in the Market for U.S. Dollar Denominated International Bonds.” European Finance Review 7 (2): 277–96.

Sebastián Nieto-Parra

169

Megginson, W. L., and K. Weiss. 1991. “Venture Capitalist Certiﬁcation in Initial Public Offerings.” Journal of Finance 46 (3): 879–901. Moody’s. Global Credit Research. 2006. “Default and Recovery Rates of Sovereign Bond Issuers, 1983–2005.” Special Comment (April). Nieto-Parra, S. 2008. “Who Saw Sovereign Debt Crises Coming?” Working Paper 274. Paris: Organization for Economic Cooperation and Development, Development Centre. Nieto-Parra, S., and J. Santiso. 2007. “The Usual Suspects: A Primer on Investment Banks’ Recommendations and Emerging Markets.” Working Paper 258. Paris: Organization for Economic Cooperation and Development, Development Centre. Panizza U., F. Sturzenegger, and J. Zettelmeyer. 2009. “The Economics and Law of Sovereign Debt and Default.” Journal of Economic Literature 47 (3): 651–98. Pescatori, A., and A. N. R. Sy. 2007. “Are Debt Crises Adequately Deﬁned?” IMF Staff Papers 54 (2): 306–37. Rogowski, R. J., and E. H. Sorensen. 1985. “Deregulation in Investment Banking: Shelf Registrations, Structure and Performance.” Financial Management 14 (1): 5–15. Santiso, J. 2003. The Political Economy of Emerging Markets: Actors, Institutions and Financial Crises in Latin America. New York: Palgrave Macmillan. Standard and Poor’s. 2006. “Default Study: Sovereign Defaults at 26-Year Low, to Show Little Change in 2007.” New York (September 18). ———. 2007. “Sovereign Defaults and Rating Transition Data: 2006 Update.” New York (February 1). Sturzenegger, F., and J. Zettelmeyer. 2005. “Haircuts: Estimating Investor Losses in Sovereign Debt Restructurings, 1998–2005.” Working Paper 05137. Washington: IMF. West, R. R. 1967. “Determinants of Underwriters’ Spreads on Tax Exempt Bond Issues.” Journal of Financial and Quantitative Analysis 2 (3): 241–63.