Imfsfaffpapers

IMFsfaffpapers

Robert Flood Editor and Committee Chair Ratna Sahay Co-Editor Marina Primorac Assistant Editor Rebecca Coke Research Assistant La Quita Pinkston Administrative Coordinator

Editorial Committee F. Charles Adams Liam Ebrill William E. Alexander Atish R. Ghosh Brian Aitken Nadeem Haque Tamim Bayoumi Donald J. Mathieson Sharmini Coorey Eswar S. Prasad

Among the responsibilities of the International Monetary Fund, as set forth in its Articles of Agreement, is the obligation to "act as a centre for the collection and exchange of information on monetary and financial problems." IMF Staff Papers makes available to a wider audience papers prepared by IMF staff members. The views presented in the papers are those of the authors and do not necessarily reflect the position of the Executive Board or of the IMF To facilitate electronic storage and retrieval of bibliographic data, IMF Staff Papers has adopted the subject classification scheme developed by the Journal of Economic Literature.

Subscription: $US54.00 a volume or the approximate equivalent in the currencies of most countries. Four issues constitute a volume. Single copies may be purchased at $18. The individual academic rate to full-time professors and students of universities and colleges is $27 a volume. Subscriptions and orders should be sent to:

International Monetary Fund Publication Services 700 19th Street, N.W. Washington, D.C. 20431, U.S.A.

Telephone: (202) 623-7430 Fax: (202) 623-7201 E-mail: [email protected] Internet: http://www.imf.org

©International Monetary Fund. Not for Redistribution International Monetary Fund

Volume 46 Number 1 IMFstaffpapers March 1999

©International Monetary Fund. Not for Redistribution EDITOR'S NOTE

The Editor invites from contributors outside the IMF brief comments (not more than 1,000 words) on published articles in IMF Staff Papers. These comments should be addressed to the Editor, who will forward them to the author of the original article for reply. Both the comments and the reply will be considered for publication. The data underlying articles published in IMF Staff Papers are available on the journal©s website (http://www.imf.org/staffpapers). Readers are invited to use these data to expand on the material in the articles, and the journal will consider publishing such work.

The U.S. Library of Congress has cataloged this serial publication as follows: International Monetary Fund Staff papers Ð International Monetary Fund. v. 1- Feb. 1950- [Washington] International Monetary Fund. v. tables, diagrs. 23 cm. Three no. a year, 1950-1977; four no. a year. 1978- Indexes: Vols. 1-27, 1950-80, 1 v. ISSN 1020-7635 = Staff papers Ð International Monetary Fund. 1. Foreign exchangeÐPeriodicals. 2. CommerceÐPeriodicals. 3. Currency questionÐPeriodicals. HG3810.15 332.082 53-35483

©International Monetary Fund. Not for Redistribution Letter from the Editor

Dear Readers,

I am the new Editor of the International Monetary Fund©s academic journal, IMF Staff Papers. I have worked in the Research Department at the IMF for over 11 years and have served at the co-editor level for the Journal of International Economics and the American Economic Review.

Along with my editorship, the journal has instituted several changes. First, the name of Staff Papers has changed to IMF Staff Papers, and the journal has a new look. Second, the refereeing process has been changed to the extent that papers sub- mitted to IMF Staff Papers are now subject to expert review outside the IMF. Third, all papers and underlying data published by IMF Staff Papers are now available freely at our web site (http://www.imf.org/staffpapers). And finally, from time to time the journal will publish issues devoted fully or partly to special topics suggested by staff, subscribers, and other readers.

It is my intention to continue the journal©s long tradition of publishing papers by IMF staff and invited guests on a variety of topics of interest to the IMF, our member countries, and our general readers. In addition, the journal will continue to publish outside comments, criticisms, and interesting replications relating to our published work.

Sincerely,

Robert P. Flood Editor, IMF Staff Papers

iii

©International Monetary Fund. Not for Redistribution This page intentionally left blank

©International Monetary Fund. Not for Redistribution International Monetary Fund

Volume 46 Number 1 Contents March 1999

How Do the Skilled and the Unskilled Respond to Regional Shocks? The Case of Spain Paolo Mauro and Antonio Spilimbergo • 1

Growth, Trade, and Deindustrialization Robert Rowthorn and Ramana Ramaswamy • 18

Regional Trade Agreements or Broad Liberalization: Which Path Leads to Faster Growth? Athanasios Vamvakidis • 42

Exchange and Capital Controls as Barriers to Trade NataliaT. Tamiris a • 69

Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions Tamim Bayoumi and Ronald MacDonald • 89

©International Monetary Fund. Not for Redistribution This page intentionally left blank

How Do the Skilled and the Unskilled Respond to Regional Shocks?

The Case of Spain

PAOLO MAURO and ANTONIO SPILIMBERGO*

Are there any differences in how workers of different skill levels respond to regional shocks? This paper addresses that question using the methodology of Blanchard and Katz (1992) and a unique data set on working-age population, labor force, and employment for five educational groups (ranging from the illiterate to the college-educated) over 196492 for the 50 Spanish provinces. The paper finds that the highly skilled migrate very promptly in response to a decline in regional labor demand, while low- skilled workers drop out of the labor force or stay unemployed. [JEL E24, J61]

When workers in a given region lose their jobs, do they remain unemployed, drop out of the labor force, or migrate? In other words, what are the mechanisms of adjustment to local labor demand shocks? Existing studies, beginning with the seminal paper by Blanchard and Katz (1992) on the 50 U.S. states, and including those by Decressin and Fatas (1995) on the regions of Europe and by Obstfeld and Peri (1998) on the regions of a wide range of industrial countries, have addressed that question with respect to the labor force as a whole. However, owing in part to data limitations, none of

*Paolo Mauro is an Economist in the IMF©s European I Department and Antonio Spilimbergo is an Economist in the Research Department. Paolo Mauro received his Ph.D. from Harvard University. Antonio Spilimbergo received his Ph.D. from M.I.T. The authors gratefully acknowledge helpful comments and suggestions by Jacques Artus, Maria Carkovic, Jorg Decressin, Enrica Detragiache, Robert Flood, Martin Hardy, Alexander Hoffmaister, Michael Mussa, Danny Quah, Peter Wickham, Charles Wyplosz, and seminar participants at the International Monetary Fund and the European Economic Association meetings, as well as high-quality research assistance by Madhuri Edwards, and indebted to Instituto Valenciano de Investigaciones Economicas for providing data.

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimbergo those studies has examined whether the relative speed and strength of these adjustment mechanisms depend on workers© educational levels. To filltha t gap, this paper analyzes the dynamic responses to regional labor demand shocks in Spain, considering separately various educational groups. There are good reasons to expect that workers with different educational levels will respond in different ways to regional labor demand shocks. In fact, the opportunity cost of not working is typically higher for the highly skilled.1 Therefore, in response to a job loss motivated by a collapse in local labor demand, the highly skilled are more likely than low-skilled workers to migrate rather than remaining unemployed or dropping out of the labor force. It is also important to recognize that the adjustment mechanisms to labor demand shocks by workers of different educational levels depend on existing labor market institutions and policies. This can be seen by considering two extreme hypothetical cases: in the presence of very generous unemployment compensation, migration might be an unattractive option for low-skilled workers, though perhaps still not for the highly skilled; by contrast, in the presence of low unemployment compensation, both low-skilled and highly skilled workers might have similarly strong incentives to migrate. This paper focuses on the case of Spain. It uses a data set on employment, labor force, and working-age population by educational level for the 50 provinces of Spain over 1964-92. That data set, published by the Instituto Valenciano de Investigaciones Economicas, is almost unique in that similar data are not easily available for any other countries. Beyond the advantage of data availability, however, the case of Spain is extremely interesting in itself. Not only does Spain have the highest unemployment rate (19 percent in mid-1998) among industrial countries, but it also displays large and persistent unemployment rate differences among its regions. By analyzing how workers with different educational levels respond to regional labor demand shocks, this paper considers those issues from a new angle. It also forms part of a broader research agenda on the regional dimension of unemployment in Europe. Related studies, which focus on the persistence of regional unemployment differences and provide further institutional detail, are presented in Mauro, Prasad, and Spilimbergo (1999). By analyzing the response of workers with different skill levels to regional shocks against the background of Spain©s institutions and policies, this paper may lead to useful policy lessons for Spain, but hopefully some of these lessons may be applicable to other countries as well. At the same time, the paper implies that caution must be exercised in drawing conclusions from studies (such as those by Decressin and Fatas, 1995, and Obstfeld and Peri, 1998) that attribute cross-country differences in the dynamics of adjustment to differences in policies and institutions. In fact, by showing that the adjustment to local labor demand shocks depends on workers© educational levels, this paper suggests that future comparative work should also take into account cross-country differences in the educational composition of the labor force.

1Throughout the paper, the terms "skill" and "education" are used interchangeably, although the estimation is based upon data on educational levels.

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

I. Persistence of Geographic Differences in Unemployment Rates, by Skill Level

Although this paper©s main contribution is to estimate how the skilled and the unskilled respond to regional shocksÐan exercise that has not been conducted before for any other countryÐa brief description of the geographic distribution of unemployment in Spain may be useful, particularly because the Spanish case is interesting in itself. This section shows that Spain is characterized by large and persistent geographic differences in unemployment rates, and that the degree of persistence is higher for low-skilled than for highly skilled workers. Therefore, in the Spanish setting, efforts to reduce geographic unemployment imbalances may need to focus on low-skilled workers. There is a striking variation of unemployment rates among Spain©s 17 regions, ranging from about 11 percent in the Balearic Islands to 30 percent in Andalucia in mid-1998. Considering a finer level of geographical disaggregation, namely that of the 50 provinces (provinces are subsets of regions), unemployment rates vary even more widely, ranging from 8 percent in Lleida, Cataluna, to 38 percent in Cadiz, Andalucia. Although patterns in the geographic distribution of unemployment rates are not easy to identify, a broad generalization could be that the southern, agricultural regions, such as Andalucia and Extremadura, and some of the northern regions with declining industries, such as Pais Vasco, Cantabria, and Asturias, tend to have higher unemployment. In addition to the large differences among regions, there is also substantial variation in unemployment rates among provinces within regions. Again, it is difficult to identify clear patterns, but provinces dominated by large cities seem to have somewhat higher unemployment rates than provinces with only small urban centers. Even though generalizations may not be easy, it is nevertheless clear that a geographic dimension of the unemployment problem exists. In fact, regional dummies explain individuals© employment status to a significant extent when controlling for personal characteristics such as age, gender, and education (Blanchard and others, 1995). Whatever the determinants of the geographic distribution of unemployment rates, however, there is compelling evidence that the current pattern has persisted for many years. In fact, even the sharp increase in nationwide unemployment since the late 1970s has left the regions© or provinces© unemployment ranking almost unchanged, though absolute differences in unemployment rates have widened considerably. Scatter plots of the survey unemployment rates in 1977 and 1992 for the 50 Spanish provinces reveal a remarkable correlation between the provinces that have higher unemployment rates in the 1990s and those that had higher unemployment rates in the 1970s (Figure 1, top left panel).2 The degree of persistence of geographical differences in unemployment varies considerably depending on the labor force participants© skill levels. Based upon data produced by the Institute) Valenciano de Investigaciones Economicas (see Appendix I), low-skilled workers seem to display greater unemployment persistence than the highly skilled, as shown by scatter plots of the unemployment rate in 1977 and 1992 in the 50 Spanish provinces for five groups of labor force participants: illiterate, primary-

2The data by skill level are available only through 1992. Scatter plots tor all workers are similar when applied to data through the late 1990s. They are also similar for the case of the regions, or using registered unemployment data instead of survey unemployment dataÐsee Mauro, Prasad, and Spilimbergo. 1999.

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimbergo

Figure 1. Spanish Provinces: Unemployment Rates by Skill Level, 1977 and 1992

1992 All Workers Illiterate 1992

1992 Primary School Middle School 1992

1992 High School College 1992

Source: Institute) Valenciano de Investigaciones Economicas.

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

Table 1. Unemployment Persistence by Educational Group, 1977-92

Skill Level Coefficient of Correlation Share in the Total Labor Force 1977 1992 All groups 0.83 100.0 100.0 Illiterate 0.50 4.1 1.4 Primary school 0.88 74.8 43.8 Middle school 0.56 15.1 42.8 High school 0.35 3.4 5.9 College 0.24 2.7 6.1 Sources:Institut o Valenciano de Investigaciones Economicas; and estimates by authors. school-educated, middle-school-educated, high-school-educated, and college- educated. The relationship between unemployment in the past and unemployment today tends to be closer among low-skilled workers, and looser among highly skilled workers. Table 1 reports, for each educational group, the coefficient of correlation between unemployment in 1977 and unemployment in 1992, as well as each group©s share of the total labor force in 1977 and 1992. The coefficient of correlation between unemployment in 1977 and unemployment in 1992 tends to be higher, the less educated the labor force participants of a given group, with the exception of the illiterate, a small group for which the quality of the data seems to be not as reliable as for the others.3 In other words, the persistence of geographical unemployment differences, a clear indication of sluggish adjustment to past shocks, is highest among the low-skilled group. The next section analyzes how workers with different skill levels adjust to shocks.

II. How Do Workers with Different Skill Levels Adjust to Shocks? When there is a negative shock to local labor demand, workers who lose their jobs can react in three ways. First, they can keep looking for a job in the area, thus remaining unemployed; second, they can stop looking for a job, thereby exiting the labor force (and becoming "discouraged workers"); or third, they can migrate to another area. The decision among these three options is affected by a number of factors, not only macroeconomic (including the nationwide unemployment rate),4 but also individual,

3The results need to be interpreted bearing in mind that, other things being equal, larger groups (in the present case, primary-school-educated and middle-school-educated) will tend to show a better fit simply because they are subject to fewer idiosyncratic changes. 4For instance, migration flows, both toward other countries and within Spain, were very large in the 1960s, but they dropped sharply beginning in the late 1970s. The main reason for this decline is likely to be that absolute unemployment rates rose in the whole country as well as in the rest of Europe. In fact, it is well known that workers tend not to migrate, regardless of how bad prospects are in their current location, if the chances of finding a job once they reach their destination are low. This phenomenon of falling migration at a time of rising absolute unemployment has been well documented not only in the case of Spain (Bentolila, 1997), but also in other countries, including Germany (Decressin, 1994), Italy (Attanasio and Padoa-Schioppa, 1991), and the United Kingdom (Pissarides and McMaster, 1984).

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimberg o

includin g the level of education . In particular , the decision of moving, like any other investmen t decision, has costs, in this case social, reallocation , and searchin g costs, and benefits deriving from the higher stream of wages in the new location. 5 Both costs and benefits depen d on the level of education . As the opportunit y costs of not working are higher for the skilled than for the unskilled, the former are more likely to migrate than the latter. Similarly, the skilled are less likely to drop out of the labor force than are the unskilled. The relative speed and strength of the adjustmen t mechanism s described above is estimate d using a panel vector autoregressio n (VAR) system of employmen t growth, the employmen t rate, and labor force participation , for the 50 Spanish province s over 1964-92. The framework adopte d is identica l to that developed by Blanchar d and Katz (1992), who first applied it to the Unite d States, and similar to that applied by Decressi n and Fata s (1995) to Europe , and by Bentolila and Jimen o (1995) to the 17 Spanish regions on quarterl y data for 1976-94. As a consequence , the results obtaine d can be compare d to those of the foregoing studies. The system is the following:

Employmen t growth:

∆e i t = αi 1 + β 1 (L) ∆e i t-1 + γ 1 (L) lei t - 1 + δ 1 (L) lpi t - 1 + ε iet Employmen t rate:

lei t = αi 2 + β 2 (L) ∆e i t + γ 2 (L) lei t - 1 + δ 2 (L) lpi t - 1 + ε iut Labor force participatio n rate:

lpi t = αi 3 + β 3 (L) ∆e i t + γ 3 (L) lei t - 1 + δ 3 (L) lpi t - 1 + ε ipt, where all variables are differences between province i and the nationa l average, in order to focus on development s at the provincia l level that are not due to nationwid e developments . ∆e i t is the first difference of the logarithm of employment ; lei t is the logarithm of the ratio of employmen t to the labor force; and lpi t is the logarithm of the ratio of the labor force to the working-age population . There are two lags for each right-han d side variable, to allow for feedback effects from labor force participatio n and the employmen t rate to employmen t growth. (Fo r example, a decrease in labor force participatio n could lower wages, thereb y facilitatin g an increase in employmen t growth.) The system is estimate d by poolin g all observations, though allowing for different province-specifi c constan t terms in each equation , since some province s may have higher average employmen t growth, employmen t rates, and labor force participa - tion rates than others, for reasons that are not capture d by the explanator y variables.6 The size of a typical local shock to labor deman d is large, both in an absolute sense and by compariso n to a typical nationa l shock. The standar d deviation of the residual in the employmen t growth equatio n for all workers amount s to 1.54 percentag e points. 7 When the system above is estimate d as a VAR at the nationa l level (instea d of as a

5Sjaastad (1962) present s the first mode l in which persona l costs and benefits of migration are formally considered . 6Furthe r technica l issues are addressed in Appendix II. 7This is an approximation , since the variables are defined in logarithms .

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

Table 2. Size of Unemploymen t Shocks (Standard Deviations)

Educationa l Level Nationa l Local All groups 0.77 1.54 Illiterat e 1.93 9.57 Primar y School 0.75 3.41 Middle School 3.36 4.56 High School 1.79 5.52 College 2.85 6.67 Source : Authors© calculations .

panel VAR on the deviation s of the provinces© data from nationa l averages), the standard deviation of the residual in the employmen t growth equatio n for all workers amount s only to 0.77 percentag e points. Moreover , the local shocks are larger than the nationa l shocks for each educationa l group (Table 2). The size of shocks seems to be somewhat larger, the higher the workers© skill level, with the notabl e exception of the illiterateÐa group that, as noted , is relatively small. The effects of a fall in employmen t can be traced through time by analyzing the impulse response graphs based upon the estimate d parameter s of the system above. Those effects can be interprete d as resulting from a decline in labor demand , under the reasonabl e assumptio n that most of the year-to-yea r changes in employmen t reflect change s in labor demand , rathe r than labor supply.8 The immediat e response to a decline in labor deman d in a given Spanish province does not differ much from that observed in other countries , though the effects on labor participatio n are higher in some cases. In response to a 1 percentag e point negative shock to employmen t growth, the unemploymen t rate immediatel y increases by 0.31 percentag e point , while the participatio n rate decreases by 0.65 percentag e point (Figur e 2). The remainin g adjustmen t to the fall in employmen t is accounte d for by migration . The simultaneou s impact on the unemploymen t rate is similar to that estimate d by existing studies for both the Unite d States and Europe . The immediat e response of the participatio n rate is similar to that observed in Europe , but much higher in Spain than in the Unite d States, suggesting that the phenomeno n of the "discouraged worker" plays a larger role in the former than in the latter. Ther e are more importan t differences between Spain and other countrie s in the extent and compositio n of adjustmen t to a negative employmen t shock after several years. In the case of Spain, migration is not sufficient to bring the unemploymen t rate back to its preshoc k level even after more than a decade . The participatio n rate rises back toward its preshoc k level, which it reache s after 10 years. These results contras t sharply with those obtaine d by other studies for both the Unite d States and the rest of Europe , where unemploymen t rates return to their preshoc k levels after about five

8 Formally , the identifying assumptio n is that ε iet can be interprete d as an innovatio n in local labor demand . Correspondingly , curren t innovation s in local employmen t growth are allowed to affect local employmen t rates and local participatio n rates, but not vice versa.

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimbergo years. In the United States, adverse employment shocks result in a relatively small decline in the participation rate, a small increase in the unemployment rate, and rapid migration, in the first few years. After about five years, both the participation rate and the unemployment rate are back at their preshock levels, and employment remains per- manently at (or below) the level attained through the initial shock, with migration being entirely responsible for that full adjustment. In the rest of Europe, the overall pattern of the response to an adverse employment shock is fairly similar to that observed in the United States, although the effects on the participation rate and the unemployment rate are much larger in Europe than in the United States during the first few years, as migration is more sluggish in the former than in the latter. The analysis conducted above is also applied to each of the five educational groups for which data are available, showing how workers with different skill levels respond differently to local shocks. Five separate systems are estimated, each of which uses data for only one of the educational groups. Figure 2 presents the impulse response graphs for each of the five educational groups, based upon the estimated parameters of the corresponding systems, for a 1 percentage point fall in the respective group©s employment. There are striking differences in the immediate responses among the various groups, particularly with respect to the participation rate and migration. In response to a 1 percentage point fall in employment, the unemployment rate immediately rises by 0.10-0.30 percentage point for all groups. However, while the participation rate drops by 0.60 percentage point or more in the case of the illiterate and those with a primary- school education, and 0.40 percentage point in the case of those with a middle-school education, it falls only by 0.10 percentage point in the case of the two top educational groups. This result is consistent with the view that the less educated are more likely to become "discouraged workers." Conversely, while some low-skilled workers do migrate in response to an adverse labor demand shock, migration takes place much more rapidly among those who are high-school educated and college educated, for whom the opportunity cost of not being employed is larger, since their salaries tend to be higher.9 Considerable differences can also be observed in the extent and composition of the adjustment to a fall in labor demand, after several years. Rapid migration implies that the unemployment rate returns to its preshock levels after only three years for those educated up to the high-school or college level. By contrast, in the cases of the illiterate, those educated up to primary-school level, and those educated up to middle- school level, about half of the initial increase in the unemployment rate persists after a decade. In all cases, the participation rate tends to return toward its preshock level, but in the case of high-school and college graduates it reaches the preshock level after only three years, perhaps because the initial impact is relatively small, while in the other cases the initial effects are not fully reversed even after 10 years. The impulse response graphs in Figure 2 are clearly different. The fact that the dynamic adjustment differs across educational groups is also confirmed by formal

9This result is consistent with Antolin and Bover©s (1997) finding that, controlling for personal characteristics (including whether unemployed workers are registered or not), higher education implies not only more migration, but also more responsiveness of migration to geographic unemployment differentials.

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

Figure 2. Response to 1 Percent Negative Employment Shock in a Given Province

All Workers Illiterate

Primary School Middle School

High School College

Source: Instituto Valenciano de Investigaciones Economicas; and IMF staff estimates.

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimbergo testing of the null hypothesis that the parameters of the systems for any two educational groups are the same. F-tests reject the null at the 1 percent significance in all of the pairwise tests. The results presented above are robust to changes in the period considered or in the geographical coverage of the sample. For example, when the system is estimated for 1964-77 or 1978-92, or for small groups of provinces (such as those in Andalucia and Extremadura, or in Catalonia), the impulse response graphs for the various educational groups are broadly the same as those obtained using all the 50 provinces. The analysis above has focused on adjustment in terms of "quantities" (changes in the unemployment rate and the participation rate, and migration). Additional information would be gained by considering the role of "price" incentives (in this case, wages) in determining the adjustment of quantities. That information would make it possible to describe the adjustment of quantities into two components, namely, (1) the adjustment of "price" incentives and (2) workers© and firms© responsiveness to such incentives. Unfortunately, wages by skill level are not available for the 50 Spanish provinces over the period considered in this study. Nevertheless, for the labor force as a whole, there is tentative evidence that the adjustment of wages is sluggish, owing to Spain©s de facto centralized bargaining system. Mauro, Prasad, and Spilimbergo (1999) find that wages do not seem to reflect local labor market conditions: in spite of large and persistent geographic differences in unemployment, unit labor costs and real wages do not differ significantly between high-unemployment and low-unemployment areas, so that neither firms nor workers have strong "price" incentives to migrate to correct existing unemployment imbalances. The different speed of adjustment of labor market quantities to local shocks for different skill groups can be partly explained by the following four factors, all of which are largely outside the control of policy makers. · First, as noted above, in a comparison of the costs and benefits of moving, migration is more likely to be an attractive option for highly skilled than low-skilled workers. The benefits of taking up a job in another area of the country compared with remaining unemployed in the area of current residence are much higher for highly skilled than for low-skilled workers, since highly skilled workers have higher wages than low-skilled workers, and unemployment benefits are subject to a ceiling. By contrast, the costs of moving are fairly independent of a person©s skill level. · Second, workers with more years of schooling may be better able to adapt to new jobs, including those that are located in different areas of the country. · Third, information about vacancies in other areas of the country may be much more readily available for jobs that require high skills. · Fourth, given the remarkable improvement in educational achievement over the past decades, low skills are associated with old age. In particular, more than three- fourths of illiterate and primary-school-educated labor force participants were 35 or older in 1993, compared with one-half for the labor force as a whole. As a result, the illiterate and the primary-school-educated groups might move less promptly than other groups partially because they tend to be older. However, there is no clear relationship between age and educational attainment for labor force participants that are educated to the middle-school level and above (Table 3). For example, middle-school-educated labor force participants are, on average, considerably

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

Table 3. Composition of the Labor Force by Age and Educational Attainment, 1993

Illiterate and Middle High Age (in years) Primary School School School College (In percent of total labor force) 16-24 3.21 14.29 0.56 0.28 25-34 6.45 16.84 2.21 2.70 35-44 11.44 8.57 1.81 1.65 45-54 11.90 3.63 1.11 0.85 55 and above 9.87 1.47 0.60 0.56 Total 42.87 44.80 6.29 6.04 (In percent of educational group) 16-24 7.49 31.90 8.90 4.64 25-34 15.05 37.59 35.14 44.70 35-44 26.69 19.13 28.78 27.32 45-54 27.76 8.10 17.65 14.07 55 and above 23.02 3.28 9.54 9.27 Total 100.00 100.00 100.00 100.00 Source: Palafox and others (1995). Notes: The breakdown between the illiterate and primary-school-educated groups is not available by age group. Data by age group are available only for a slightly different breakdown between middle school and high school than that used in the rest of this paper.

younger than high-school- and college-educated ones. Therefore, speedier labor market adjustment for the high-school- and college-educated workers than for those that are middle-school-educated cannot be explained by differences in age. Finally, the result that the highly skilled migrate more promptly than low-skilled workers in response to regional shocks is consistent with evidence from the labor force survey on the unemployed workers© willingness to move to obtain a job, for the various educational categories. Only 24.8 percent of illiterate unemployed workers declare themselves willing to move to obtain a job, compared with 45.7 percent of college- educated workers (Table 4). To find a job, the more highly educated are also less likely to declare themselves willing to accept a different type of job than the one they are looking for. Among college-educated unemployed workers, only 35.6 percent declare themselves willing to accept a job of a different type than the desired one, whereas that proportion amounts to 73.6 percent in the case of illiterate unemployed workers. Therefore, it seems that the highly skilled are more willing to move to find a job of the desired type.

III. Current Labor Market Arrangements and Differences in Adjustment The empirical analysis in the previous sections shows that the unskilled migrate more slowly than the skilled in response to regional shocks. This section attempts to relate that finding to existing labor market arrangements. Although some of these arrangements are

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimbergo

Table 4. Willingness To Move or To Change Type of Job To Become Employed (In percent) Primary Middle High All Illiterate school School School College Workers Willing to move? Yes 24.8 25.8 29.4 39.2 45.7 28.4 No 52.8 47.8 38.1 30.9 26.7 42.3 Did not know or did not answer 22.4 26.4 32.5 29.9 27.6 29.3 Willing to change job? Yes 73.6 72.9 66.3 51.6 35.6 68.9 No 9.2 6.9 9.0 23.3 35.3 8.7 Did not know or did not answer 17.2 20.2 24.7 25.1 29.1 22.4 Source: Palafox and others (1995). Note: Column heads represent the highest level of education achieved. specific to Spain, many are present in other countries. Therefore, several policy lessons may be relevant for other countries as well. In the Spanish context, the key barrier to the reduction of existing geographical unemployment differences and the prompt adjustment to local labor demand shocks seems to be the de facto centralized wage bargaining system. Existing policies and arrangements in other goods and factor markets (e.g., the housing market) also seem to hamper adjustment to shocks. However, these barriers are likely to affect labor force participants of all educational levels in a similar manner.10 This section reviews current policies and arrangements in the labor market that hamper the mobility of low- skilled workers, even though probably not that of other groups. These include programs to help agricultural workers in specific depressed areas (Andalucia and Extremadura), minimum wage legislation, and the unemployment benefit system.11 Programs to help workers in depressed areas reduce incentives for these workers to accept lower wages or to seek jobs elsewhere. An example is the agricultural employment plan (Plan de Empleo Rural), which provides farm workers (who are typically low skilled) in Andalucia and Extremadura with temporary jobs in state- financed infrastructure projects and unemployment assistance for a substantial portion of the remainder of the year.12

10At the same time, it may be argued that illiquid rental markets make it difficult to move, especially for the less affluent groups of workers, that is, typically, the low-skilled group. 11 ·Further institutional detail not only on the labor market, but also other goods and factor markets, is provided in Mauro, Prasad, and Spilimbergo (1999). 12Under that program, which covers about 250,000 workers and accounts for about 5 percent of total expenditure on unemployment benefits in Spain, as few as 40 days© work a year entitle workers to 75 percent of the statutory minimum wage for 40-360 days a year (depending on their age).

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

Minimum wages may play some role in preventing wages from falling sufficiently to encourage the creation of new jobs at the low end of the pay scale. Their importance in determining labor force participants© willingness to take up jobs is increased by the fact that it affects the level of unemployment assistance and the ceiling and floor for unemployment benefits. Spain©s statutory minimum wage, currently at 32 percent of the average adult wage (after gradually declining from 40 percent in 1985), is not high by international standards.13 Moreover, it is nationwide, with no adjustments for differences in the cost of living in the various areas of the country. While this institutional feature of the labor market is probably of little consequence for the highly skilled, it may have important consequences for low-skilled workers, particularly in areas where productivity and the cost of living are low. Unemployment protection reduces unemployed workers© job search efforts and raises the participation rate, thereby contributing to high unemployment and low labor mobility. Spain©s unemployment benefit system is fairly generous by international standards, though not sufficiently so to explain why Spain©s unemployment rate is higher than in other countries.14 The benefit system is rendered particularly generous by the possibility of cumulating unemployment benefits paid by the state with sizable dismissal benefits paid by the employer.15 Antolin and Bover (1997) find that, after controlling for personal characteristics, the unregistered unemployed, who do not receive unemployment benefits, are more mobile than the employed and the registered unemployed. Such adverse effects are particularly important among the low-skilled workers, because benefits are capped.

IV. Concluding Remarks Using a unique and relatively underexplored data set on the Spanish provinces, this paper has shown that the adjustment to a local labor demand shock varies depending on the educational level of the workers affected by it. Faced with the loss of their job, the highly skilled tend to migrate promptly, whereas low-skilled workers are more likely to drop out of the labor force or to remain unemployed. This is an intuitive result, in light of the fact that the opportunity cost of not working is typically higher for highly skilled than for low-skilled people. The paper also points out that differences in the adjustment process between the highly skilled and the low-skilled groups depend on existing policies and institutions. For example, unemployment protection may reduce the attractiveness of migration for low-skilled workers to a greater extent than it does for the highly skilled. Similarly,

13It is well below the averageÐof more than 50 percent of the average adult wageÐfor European Union countries that have a statutory minimum wage. It is also below the average of the minimum wages set through collective agreements in European Union countries that do not have a statutory minimum wage. l4In particular, a thorough comparison between the unemployment benefit systems and other labor market institutions in Spain and Portugal reveals that the differences between these two countries are rather limited, raising the puzzle of why unemployment is so much worse in the former than in the latter (Blanchard and Jimeno, 1995). Bover, Garcia-Perea, and Portugal (1998) argue that Portugal©s better labor market performance can be attributed to its much lower unemployment compensation prior to 1985. 15Permanent workers hired before May 1997 typically receive their full salary for 45 days per year worked, up to a maximum of 42 weeks, if they are dismissed for "unjustified" economic causes, which tends to be the majority of cases.

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimbergo minimum wages may set an artificial floor on the wages of the low-skilled, although their impact on the highly skilled is probably negligible. Therefore, policy makers should devote particular attention to the impact of these policies and institutions on the low-skilled groups. More generally, this paper©s finding that labor market adjustment is more rapid among the highly skilled provides a new argument in favor of promot- ing the educational level of the workforce. On the specific case of Spain, these considerations may provide clues to why the persistence of geographic unemployment differences is more pronounced among low- skilled workers than among the highly skilled, as well as on why geographic unemployment differences are persistent for the labor force as a whole and, in turn, on why nationwide unemployment is persistently high. There are two avenues for further research. First, the results obtained in this study cannot be entirely divorced from the policies and institutions of Spain. Although it is reasonable to expect that the key results and related policy conclusions may extend beyond the case of Spain, this needs to be confirmed by studies on other countries. Second, owing to data limitations, this study has focused on labor market adjustment to shocks (and geographic unemployment differences) through migration of workers. However, migration of firms seems to be both more desirable and more effective an adjustment mechanism to reduce geographic unemployment imbalances. Finally, this study suggests a caveat to the conclusions drawn by existing studies that attribute cross-country differences in the dynamics of adjustment to differences in policies and institutions. By pointing out that adjustment depends on workers© skill levels, this study suggests that future comparative work should also strive to take into account cross-country differences in the educational composition of the workforce.

APPENDIX I Data Description The data on working-age (16-65) population, labor force, and employment by province and by skill level, for 1964-1992, are drawn from Mas, Perez, Uriel, and Serrano (1995). Nothing comparable to this data set exists for other countries. It provides data on working-age population, active population, and employment for the 50 Spanish provinces, for people belonging to five groups with different skill levels: illiterate, primary-school-educated, middle-school- educated, high-school-educated, and college and above. The data are based on a very comprehensive data collection project conducted by the Instituto Valenciano de Investigaciones Economicas (IVIE). Since 1977, the basic source of information used for that project are the individual replies to the labor force survey by the Instituto Nacional de Estadistica (INE). These individual replies include information on the respondent©s educational attainment, which is typically not reported with any geographical disaggregation by the INE but which constitutes the focus of the IVIE study. Prior to 1977, the information is based on less disaggregated information from the labor force survey, and other sources including censuses and statistics on schooling. All in all, while some judgment may have been applied (particularly for the illiterate in the period prior to 1977) to correct the series obtained from such a wide range of sources, the data seem very reliable. The data on the labor force composition by age and skill level are drawn from Palafox. Mora, and Perez (1995).

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

APPENDIX II Technical Issues This appendi x describes some key features of the individual series of employmen t growth, the unemploymen t rate, and the participatio n rate, and addresses a numbe r of technica l issues related to the estimatio n of the VAR system in Section II.

Characteristic s of Individual Series This section provides summar y statistics on employmen t growth, the unemploymen t rate, and the participatio n rate for all workers and the five educationa l groups, and estimate s the extent to which individua l provincia l series covary with nationa l ones. It also report s the results of unit root tests. The 1964-92 average employmen t growth was lowest for the illiterate and the primary- school-educate d groups, as the working-age populatio n belonging to these groups decreased sharply durin g the past decades. Unemploymen t rates did not differ systematically by educationa l group over the period as a whole, since in the 1960s and early 1970s unemploymen t rates were extremely low for the low-skilled group. This is in sharp contras t to the curren t situation , where high unemploymen t rates are strongly associated with low education . The participatio n rate has always been much higher, the higher the educationa l level, ranging from 19 percen t for the illiterate group to 81 percen t for the college-educate d group over the period (see Table Al). Change s in employmen t growth, the unemploymen t rate, and the participatio n rate at the provincia l level can be decompose d into a nationa l and a provincia l component . To determin e the relative magnitud e of the provincia l components , the following regression is run for each of the 50 Spanish provinces:

Xi t = αi + β i Xt + η i t where I= 1 ... 50,

where Xi t is the provincia l variable (namely , employmen t growth, the unemploymen t rate, and the participatio n rate) at time t, and Xt is the same variable at the nationa l level. Table A1 report s the weighted average (by each province© s share of Spanish population ) of the adjusted R2 for the 50 regressions, for each variable, and for each group of labor force participants . In the case of all workers, the average adjusted R2 amount s to 0.46 for the employmen t growth rate, 0.94 for the unemploymen t rate, and 0.50 for the participatio n rate. In other words, only about half of the changes in provincia l employmen t growth and the participatio n rate are explained by movement s in the correspondin g nationa l variables.16 By contrast , unemploymen t rates at the provincia l level are extremely highly correlate d with nationwid e unemploymen t rates, suggesting that unemploymen t rates seem to vary in near-uniso n throughou t the country , though some provinces have always much higher unemploymen t rates than others. The covariance of provincia l and nationa l variables is similar for all educationa l groups in the case of employmen t growth and the unemploymen t rate, but is much higher for low-skilled than for highly skilled workers in the case of the participatio n rate. Augmente d Dickey-Fulle r unit root tests yield the following results. Employmen t levels are integrate d of order one. Unemploymen t rates typically have a unit root, anothe r sign that they are persistent over time. 17 The evidence on whether participatio n rates have unit roots is rathe r mixed.

16For the 50 U.S. states, Blanchar d and Katz (1992) find an average adjusted R2 of 0.60, suggesting that aggregate shocks explain local development s to a slightly greater extent in the Unite d States than in Spain. 17This result is consisten t with Bentolila and Jimeno (1995), who find high persistence in local Spanish unemploymen t rates. It contrast s sharply with the evidence on the Unite d States, where unem - ploymen t rates are not persistent

©International Monetary Fund. Not for Redistribution Paolo Mauro and Antonio Spilimberg o

Table A1. Individual Series Average Levels and Covariance with the Nationa l Variable s

All Primar y Middl e High Workers Illiterat e Schoo l Schoo l Schoo l College

Employmen t growth Average (in percent ) 2.80 -6.00 -2.00 9.00 4.00 5.00 Average of adjusted R2 0.46 0.37 0.39 0.56 0.32 0.34

Unemploymen t Average (in percent ) 11.00 11.00 9.00 16.00 8.00 8.00 Average of adjusted R2 0.94 0.76 0.93 0.93 0.88 0.88

Participatio n rate Average (in percent ) 50.00 19.00 49.00 49.00 65.00 81.00 Average of adjusted R2 0.50 0.91 0.84 0.95 0.53 0.31

Technica l Issues on the VAR System

The specification of the VAR system estimate d in Section II follows Blanchar d and Katz (1992) exactly, to permit internationa l compariso n of the results. Nevertheless , a numbe r of alternativ e specification s were estimate d to show that the results are robust to specification changes. The results are broadly similar if the system is estimate d by using differences rathe r than levels of the employmen t rate, or differences of employmen t growth and levels of the other two variables. The results are very similar to the ones reporte d in Section II if three or four lags of all the variables are used, instead of two lags. Curren t innovation s in provincia l employmen t growth are allowed to affect provincia l employmen t rates and provincia l participatio n rates but not vice versa, consisten t with the inter- pretatio n of ε iet as an innovatio n in provincia l labor demand . The covarianc e matrix of the residuals confirm s that the contemporaneou s correlatio n between ε iet with the innovation s in the employmen t rate, ε iut, and in the participatio n rate, ε ipt, is very low by compariso n with the variation in ε iet. In other words, the first element s of the second and third row in the covarianc e matri x of the residuals (all educationa l groups) reporte d below are very small compare d with the first elemen t in the first row.

7.4514e-00 4 3.3023e-00 7 1.7100e-00 4 -3.2700e-00 7 -1.6334e-00 4 1.8686e-00 4

The covarianc e matrice s for the other systems for the five educationa l groups are similar to the one reporte d above. Finally , it is well known that using the standar d within-grou p estimato r for dynami c models with fixed individual effects generate s estimate s that are inconsistent , even in the case of a large numbe r of individuals. However, the size of the bias decreases as the length of time period increases, and the panel estimate d in this paper uses about 30 years of data, a relatively long sample period.

©International Monetary Fund. Not for Redistribution HOW DO THE SKILLED AND THE UNSKILLED RESPOND TO REGIONAL SHOCKS?

REFERENCES Antolin, Pablo, and Olympia Bover, 1997, "Regional Migration in Spain: The Effect of Personal Characteristics and of Unemployment, Wage and House Price Differentials Using Pooled Cross-Sections," Oxford Bulletin of Economics and Statistics, Vol. 59 (May), pp. 215-35. Attanasio, Orazio, and Fiorella Padoa-Schioppa, 1991, "Regional Inequalities, Migration and Mismatch in Italy," in Mismatch and Labor Mobility, ed. by Fiorella Padoa-Schioppa (Cambridge: Cambridge University Press), pp. 237-320. Bentolila, Samuel, 1997, "La Inmovilidad del Trabajo en las Regiones Espanolas," Papeles de Economia Espanola, Vol. 72, pp. 168-77. Bentolila, Samuel, and Juan F. Jimeno, 1995, "Regional Unemployment Persistence (Spain, 1976-94)," CEPR Discussion Paper No. 1259 (London: Centre for Economic Policy Research, October). Blanchard, Olivier Jean, and others, 1995, "Spanish Unemployment: Is There a Solution?" (London: Centre for Economic Policy Research). Blanchard, Olivier, and Juan F. Jimeno, 1995, "Structural Unemployment: Spain Versus Portugal," American Economic Review, Papers and Proceedings, Vol. 85 (May), pp. 212-18. Blanchard, Olivier, and Lawrence F. Katz, 1992, "Regional Evolutions," Brookings Papers on Economic Activity: 1, Brookings Institution, pp. 1-61. Bover, Olympia, Pilar Garcia-Perea, and Pedro Portugal, 1998, "A Comparative Study of the Portuguese and Spanish Labour Markets," Bank of Spain Working Paper No. 9807 (Madrid: Bank of Spain). Decressin, Jorg, 1994, "Internal Migration in West Germany and Implications for East-West Salary Convergence," Weltwirtschaftliches Archiv, Vol. 130, No. 2, pp. 231-57. , and Antonio Fatas, 1995, "Regional Labour Market Dynamics in Europe," European Economic Review, Vol. 39 (December), pp. 1627-55. Mas, M., F. Perez, E. Uriel, and L. Serrano, 1995, Capital Humano, Series Historicas, 1964-1992 (Valencia, Spain: Fundacion Bancaja). Mauro, Paolo, Eswar Prasad, and Antonio Spilimbergo, 1999, Perspectives on Regional Unemployment in Europe, IMF Occasional Paper No. 177 (Washington: International Monetary Fund). Obstfeld, Maurice, and Giovanni Peri, 1998, "Asymmetric Shocks," Economic Policy, Vol. 26 (April), pp. 205-47. Palafox, J., J.G. Mora, and F. Perez, 1995, Capital Humano, Educacion y Empleo (Valencia, Spain: Fundacion Bancaja). Pissarides, C.A., and I. McMaster, 1984, "Regional Migration, Wages and Unemployment: Empirical Evidence and Implications for Policy," LSE Centre for Labour Economics Discussion Paper No. 204 (London: London School of Economics). Sjaastad, L., 1962, "The Costs and Returns of Human Migration," Journal of Political Economy, Vol. 70, pp. 80-93.

Growth, Trade, and Deindustrialization ROBERT ROWTHORN and RAMANA RAMASWAMY*

This paper shows that deindustrialization is explained primarily by developments that are internal to the advanced economies. These include the combined effects on manufacturing employment of a relatively faster growth of productivity in manufacturing, the associated relative price changes, and shifts in the structure of demand between manufactures and services. North-south trade explains less than one-fifth of deindustrialization in the advanced economies. Moreover, the contribution of north-south trade to deindustrialization has been mainly through its effects in stimulating labor productivity in northern manufacturing; it has had little enduring effect on the total volume of manufacturing output in the advanced economies. (JEL Ol, O3, Fl, F43)

he share of manufacturing employment has declined continuously for more than Ttwo decades in most advanced economiesÐa phenomenon that is referred to as deindustrialization. For instance, in the group of countries that are classified as "industrial countries" in the IMF©s World Economic Outlook, the share of manufacturing employment declined from about 28 percent in 1970 to about 18 percent in 1994. The main issues of debate regarding deindustrialization are whether the secular decline in the share of manufacturing employment ought to be viewed with concern, and the extent to which this decline is caused by factors that are internal to the advanced economies, as opposed to external factors in the form of expanding economic linkages with the developing countries. The early contributions in this area by Baumol (1967) and Fuchs (1968), which were later extended more systematically by Rowthorn and Wells (1987) and Baumol, Blackman, and Wolff (1989), argued that deindustrialization in advanced economies is

*Robert Rowthorn is a Professor of Economics, Cambridge University. Ramana Ramaswamy is a Senior Economist in the IMF©s Asia and Pacific Department. The authors are grateful to Graham Hacche for raising some of the issues that are discussed in the paper.

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

not necessarily an undesirable phenomenon, but is essentially the natural consequence of the industrial dynamism exhibited by these economies. As the bulk of the workforce in advanced economies is employed in either manufacturing or services, the evolution of employment shares depends mainly on output and productivity trends in these two sectors. In most advanced economies, labor productivity has typically grown much faster in manufacturing than it has in services, while output growth has been about the same in each sector.1 Thus, given the similarity of output trends in the two sectors, lagging productivity in the service sector results in this sector absorbing a rising share of total employment, while rapid productivity growth in manufacturing leads to a shrinking employment share for this sector. This emphasis on differential productivity growth as the main cause of deindustrialization contrasts with Colin Clark©s (1957) influential hypothesis that the evolution of employment structure during economic development is explained by a well-defined sequence of changes in the composition of demand. Clark©s hypothesis essentially consisted of an extrapolation of Engel©s law to the case of manufactures. He argued thatÐjust as, in a poor country, the share of income spent on food declines as per capita income rises, while a growing share is spent on other items such as manufactured goodsÐas the country develops further, demand shifts increasingly toward services and the share of expenditure devoted to manufactures stabilizes and then ulti- mately falls. As a result, the employment share of manufacturing should also stabilize and eventually fall. Thus, according to Clark, deindustrialization in advanced economies would be a natural consequence of the shift in demand away from manufactures toward services. More recent studies seeking to explain the declining share of manufacturing employment, such as for instance those by Sachs and Schatz (1994), Wood (1994 and 1995), and Saeger (1996), broadly concur with the importance assigned to "internal" factors in accounting for deindustrialization. They recognize, however, that "external" factors such as the growth of north-south trade may also have played a significant role in accelerating the decline of manufacturing employment. The role of external factors has been most vigorously stressed by Wood. He argues that manufactured imports from the developing countries are highly labor intensive, and displace many times more workers in the advanced economies than their dollar value would suggest. Thus, even a balanced increase in north-south trade will, under these conditions, reduce manufacturing employment in the north because the number of low-skill jobs lost in the import-competing industries will greatly exceed the new jobs created in the skill-intensive export sector. The main aim of this paper is to assess the relative importance of the forces described by the various hypotheses that have been put forward to explain deindustrialization. The analytical framework used is an extension of the framework provided in Rowthorn and Ramaswamy (1997). The main findings of the current paper are that deindustrialization has been caused primarily by factors that are internal to the advanced economiesÐi.e., by the combined effects of the interactions among shifts in the pattern of demand between manufactures and services, the faster growth of productivity in

1For instance, between 1960 and 1994 output grew at roughly similar rates in manufacturing and ser- vicesÐannual growth rates of 3.6 and 3.8 percent respectively; in contrast, productivity in manufacturing during this period grew at an annual rate of 3.6 percent, while productivity in services grew at only 1.6 percent.

©International Monetary Fund. Not for Redistribution Robert Rowthorn and Ramana Ramaswamy manufacturing as compared to services, and the associated fall in the relative price of manufactures. The regression analysis further indicates that north-south trade has, on average, contributed less than one-fifth to the relative decline of manufacturing employment in the advanced economies. Moreover, the results show that competition from low- wage producers has had little effect on the overall volume of manufacturing output in the advanced economies. The contribution of north-south trade to deindustrialization is shown to have been mainly through its effect in stimulating labor productivity in the manufacturing sector of the advanced economiesÐfirms in the north appear to have responded to the competition from cheaper imports both by utilizing their labor more efficiently and by shifting production increasingly toward higher valued items.

I. Deindustrialization: Some Conceptual Issues Clark©s account of structural change is notable for the weight it assigns to income elasticities of demand in explaining what happens to the output of manufacturing in the course of development. The income elasticity of demand for manufactures is high in poor countries, but low in rich countries, and this explains why the share of manufacturing in output and employment rises at first and falls later on. While there is some empirical basis for this hypothesis (see below), a purely demand-based explanation of deindustrialization is incomplete because it neglects the influence of productivity and prices on the structure of demand, and hence on output and employment. As noted above, labor productivity grows faster in manufacturing than it does in the economy as a whole, and hence the relative price of manufactured goods declines as the economy develops. This in turn encourages the substitution of manufactured goods for other items, especially those services whose relative cost is rising because of the relatively slower growth of productivity in those activities. In the earlier stages of development, the effect of such substitution is to boost the already rapid growth in demand for manufactures, while later on the substitution effect helps to stimulate an otherwise flagging demand for manufactured goods. From a theoretical point of view, the effect of productivity growth on manufacturing employment is ambiguous. As noted above, on the one hand, the faster growth of productivity in this sector makes manufactured goods relatively cheap, thereby stimulating demand for them. On the other hand, less labor is required to manufacture any given volume of output. How these two influences net out in their effect on manufacturing employment is an empirical question that cannot be settled theoretically. As we shall see below, the evidence suggests that the labor-saving impact of faster productivity growth in the manufacturing sector outweighs the demand-creating effect of lower prices, so the net effect is to reduce the share of employment in this sector. Figure 1 provides a schematic illustration of what happens to the manufacturing sector as per capita incomes rise. For convenience, units are chosen so that the shares of manufacturing in real output and in employment are initially the same. The curve labeled "hypothetical" shows how these shares would evolve if productivity growth were uniform across sectors and if relative prices were to remain unchanged through time. Under these conditions, the shares of manufacturing in real output and employment would remain equal, and their evolution would be determined solely by the income elasticity of demand for manufactures. The hypothetical curve is at first

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

Figure 1. Evolution of the Manufacturing Share

upward sloping because the income elasticity of demand for manufactures is greater than unity in the initial stages of economic development, and it later slopes down- ward when this elasticity falls below unity in the more advanced stages of economic development. In practice, neither output nor employment shares follows this hypothetical curve. Faster than average productivity growth in the manufacturing sector causes the relative price of manufactured goods to fall, thereby stimulating demand, raising their share in real output, and causing this share to follow a path indicated by the upper curve in the diagram. It also causes the amount of labor required per unit of manufacturing output to fall rapidly, so that the share of manufacturing in employment follows a much lower trajectory, which normally lies well below the hypothetical curve. The above exposition assumes that the income elasticity of demand for manufactures is less than unity in advanced economies. How does one reconcile this with the findings of studies such as those of Summers (1985) and Falvey and Gemmel (1996) that services as a whole have an income elasticity of demand close to unity? The answer is as follows. In advanced economies, the share of manufactures in output and expenditure is small (and conversely, the share of services is large). Under these conditions, with an income elasticity of demand for services only marginally greater than unity, the income elasticity of demand for manufactures may be well below unity. For example, with an income elasticity of demand for services equal to 1.1, the income

©International Monetary Fund. Not for Redistribution Robert Rowthorn and Ramana Ramaswamy elasticity of demand for manufactures in the typical advanced economy will be around 0.7. This issue is explored in more detail in the appendix. Regarding external factors, foreign trade can affect the internal structure of an economy in various ways. One involves international specialization between manufactures, and other goods and services. A country can become a "workshop" economy, generating a large trade surplus in manufactures that is used to help finance a substantial deficit in nonmanufactured items, such as food, fuels, or services. This is the situation in Germany or Japan, and was also true at one time of the United Kingdom. Alternatively, like Australia, Canada, or the United Kingdom today, a country may have a trade deficit in manufactures that is financed partly through the export of nonmanufactured items such as food, minerals, or services. Such trade patterns have obvi- ous implications for the relative size of the manufacturing sector. Other things being equal, a more positive trade balance in manufactured goods implies a larger share of domestic manufacturing in output and employment. A second avenue through which trade may affect the structure of employment in advanced economies is international specialization within manufacturing production. In recent decades, there has been an evolution in the division of labor whereby advanced economies of the north export skill-intensive manufactured goods in return for labor-intensive manufactures, such as clothing or toys, from developing countries in the south. To manufacture the former goods requires the employment of a modest number of skilled workers, whereas to produce the same value of labor-intensive goods would require the employment of a much greater number of unskilled workers. The effect of such trade should therefore be to reduce manufacturing employment in the north and to alter the skill composition of the manufacturing workforce. Low-wage imports may also reduce employment in the manufacturing sector of advanced economies by increasing competition and forcing firms to utilize their labor more efficiently. Note that aggregate statistics for the manufacturing sector make no distinction between a shift into higher value-added products and greater efficiency in the creation of existing products, and both will show up as an increase in labor productivity. The above discussion describes the evolution of the manufacturing sector under the impact of rising incomes, differential productivity growth, relative price changes, and foreign trade. Superimposed on this evolution is the influence of other factors such as the share of fixed investment in total spending. Investment expenditure is skewed toward manufactured goods, such as machinery and building materials, so that a higher rate of investment will increase the share of manufactured goods in total demand, and thereby raise the share of manufacturing in real output and employment.

II. Econometric Estimations To examine the above relationships empirically, we use annual data from a panel of 18 industrial countries over the period 1963-94, for which a total of 510 observations are available.2 This sample was derived by dropping Ireland, Portugal, and Switzerland

2These countries are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Italy, Japan, the Netherlands, New Zealand, Norway, Spain, Sweden, the United Kingdom, and the United States.

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

from the slightly larger sample of countrie s used in our previous studyÐRowthor n and Ramaswam y (1997). Their exclusion was necessitate d by the absence of sectoral data on prices, output , and productivit y for these countries . The present sample also differs from the one used in our earlier study because the data now includ e observation s for every year.3 The equation s we estimat e are of the following type:4

Productivity:

log RELPROD = α0 + α1 log Y + Σ i>1 αiZ i, (1) where RELPROD is relative labor productivit y in manufacturin g as compare d to labor productivit y in the econom y as a whole, Y is per capita income , and the Zs are addi- tiona l variables reflecting the influenc e of foreign trade and other factors.

Prices:

log RELPRICE = β 0 + β 1 log (RELPROD) + Σ i>1β iZ i, (2) where RELPRICE is the relative price of manufacturin g goods as compare d to the price of nationa l outpu t as a whole.

Output:

2 log OUTSHARE = γ o+ γ 1logY + γ 2(logY) + γ 3log RELPRICE +Σ i>3 γ iZ i, (3) where OUTSHARE is manufacturin g value added as a share of real GDP .

With appropriat e units of measurement , the following equatio n holds identically :

log EMPSHARE = log OUTSHARE - log RELPROD, (4) where EMPSHARE is the share of manufacturin g in total employment .

Eliminatin g the relative price variable from equatio n (3) and using the above identity , we obtain equation s of the following type:

2 log OUTSHARE = δ 0 +δ 1 log Y +δ 2(log Y) +Σ i>2δ iZi; (5)

2 log EMPSHARE = ε 0 + ε 1 log Y + ε 2(log Y) + Σ i>2ε iZ i. (6)

3Dat a on fixed capital formation , per capita income , outpu t and prices are taken from the OECD National Accounts, supplemente d as required by OECD Historical Statistics. Per capita incom e is con- verted to 1986 U.S. dollars by mean s of purchasin g power parities in the IMF World Economi c Outloo k database . Employmen t series were taken from a variety of OEC D sources, includin g Historical Statistics, Labor Statistics, and the Intra-sectoral Data Base. Trade statistics are drawn from the UNCTA D database, and our use of the term "developing country " accord s with curren t Unite d Nation s practice . Thus, Singapor e and Hon g Kong SAR are classified as developing countrie s even though their per capita income s are now amon g the world©s highest and despite the fact that they have been classified as advanced economie s in the World Economic Outlook since May 1997. The term "manufactures " covers SITC sections 5 to 8 excluding division 68 (nonferrou s metals). 4The appendi x contain s a simple mode l that motivate s these equations .

©International Monetary Fund. Not for Redistribution Robert Rowthorn and Ramana Ramaswamy

The precedin g discussion implies that coefficient s have the following signs:

α1, γ 1, δ 1, Ε 1 > 0; β 1, γ 2, γ 3, δ 2, ε 2 < 0. (7) As we shall see, the estimate d coefficient s satisfy these inequalities . Most of the estimate d equation s include other variables in additio n to those explicitly identified above. Dumm y variables are used for individual countrie s to correct for internationa l differences in measuremen t practice s and other unexplaine d "fixed" effects. In some cases a time trend or time dummie s are also included . To examin e the influenc e of internationa l trade on economi c structure , we use two main variables, TRADEBAL and LDCIMP. The former is the overall trade balance in manufacture d goods (tota l exports minus total imports) ; the latter is equal to manufacture d import s from developing countries . Both are expressed as a percentag e of GD P measured in U.S. dollars at purchasin g power parity. 5 The role of TRADEBAL is to captur e the effect of overall manufacturin g trade per- formanc e on the structur e of employment . The variable LDCIMP is designed to captur e the effects of competitio n from low-wage countrie s on labor productivit y in norther n economies . These effects include increased efficiency in activities that compet e directly with low-wage producers , togethe r with shifts in the compositio n of norther n manufac - turin g toward higher value-added , skill-intensive , or capital-intensiv e activities. Such effects may occur even when trade is balance d between the nort h and the south, and are in additio n to those north-sout h effects capture d by the variable TRADEBAL. Finally, there is the variable FIXCAP, which is gross domesti c fixed capital for- matio n expressed as a percen t of GD P at constan t prices. As noted earlier, the ratio- nale for using this variable is that capital investmen t is manufacturin g intensive, and a change in the rate of investmen t will therefor e have a greater impact on the deman d for manufacture d goods than on the deman d for the outpu t of other sectors.

Regressio n Results: Productivit y and Prices Table 1 report s the results of pooled regressions in which the dependen t variables are log Y and log RELPROD. Both these variables have a strong upward trend , which account s for the finding in regression equatio n (3) that rising per capita incom e is associated with increasin g relative labor productivit y in manufacturing . The variable LDCIMP is include d to quantify the impact of low-wage import s on manufacturin g productivity . In equatio n (4), the coefficient of this variable is quite large and highly significant. It implies that an increase of 1 percentag e point in the ratio of low-wage import s to GD P will cause manufacturin g productivit y to rise by 8.5 percen t as compare d to productivit y in the economy as a whole. To examine the robustnes s of this finding, we experimente d with two othe r equations : (4AR), which correct s for first-orde r autocorrelation , and (5), which include s a time trend . In each case, the coefficient of LDCIMP turn s out to be smaller than in the original formulation , but is still statistically significant.

5The conventiona l metho d is to normaliz e trade by dividing import s and exports by GD P converte d to U.S. dollars at curren t exchange rates. This metho d is subject to major distortion s caused by large exchange rate fluctuations , which may give the impression that large volume changes have occurre d in import s and exports when this is not the case. The use of purchasin g power parity (PPP ) exchange rates helps to avoid this problem .

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION t l r m s rea fo y fro = s percen * * assume 1 ) e ) ) ) dumm s e 7 ) 7 th g t ye a 0.012* 0.028* 0.7 (4AR tim (2.15 (8.41 (1.34 e -0.0 n e RELPROD . manufacture f o s parity separat Equatio r . a s significanc RELPROD l i import * e (5 powe * = ) e g 2 ) ) lo s g 4 available Ther .02 ye . e 0.029* 0.40* statistica (3.57 (7.84 ar e s Variabl t 1963 LDCIMP purchasin . e t 1963-9 , a denot s k RELPROD sinc 100) d * = * 0 observation lo (4AR ) ) dollar Dependen ) s 0 g . asteris 8 d Prices e ye d elapse 51 199 0.085* 0.30* 0.8 s f an U.S (7.12 x on ) 6 (12.60 o d l an year y an s 198 (inde tota = . n y a i RELPROD h e * ) States asterisk parentheses lo (4 s 7 d whic ) o YEARS n g r econom . incom ye e (i a 0.8 0.41* fo r , Tw Productivit . (22.40 Unite e e parity whol r capit e r th t th Numbe pe n n countries powe i = r RELPROD D g Relativ Y excep parentheses f . * y n i o ) lo (3 e s OEC s ) 5 Equatio g worke 8 r ye ar 1 0.7 0.012* f countr purchasin pe h t (27.26 o d RELPROD logarithm s a l P values eac r adde t- Estimate e e fo GD d f natura consist y e e o t * valu th l lo s ) (2 ) s Y dumm Absolut 3 sampl . rea g y Poole a (1 . ye percen lo 0.024* 0.9 - a denote 1 dat s g (60.37 " e e a ) countr ©International Monetary Fund. Not for Redistribution e reported Th t . "log Tabl x no s i separat prefi m definition e a . manufacturin d n ? ter i t Th respectively . r an , 4 (U.N s exist s level 199 t worke t constan r process e r variables pe y Th countrie d : g excep erro percen r dummie ) 5 y e Y adde g yea Notes e th h YEARS 2 lo LDCIMP AR(1 d n R an a developin Countr Explanator valu eac

25 Robert Rowthorn and Ramana Ramaswamy

Table 2. Pooled Estimates of Relative Prices, 1963-94

(Dependent variable = log RELPRICE) Equation Number (6) (7) Explanatory variables Log RELPROD -0.88** -0.89** (36.91) (28.97) LDCIMP -0.008 (0.85) Country dummies exist? yes yes R2 0.80 0.80 Notes: RELPRICE = implicit price of manufactures - implicit price of GDP (index 1990 = 100). See also the notes to Table 1.

The determinants of relative prices are examined in Table 2. The coefficient of log RELPROD is large, negative, and highly significant, which is consistent with the notion that movements in labor productivity are the major factor influencing the behavior of relative prices. The coefficient of LDCIMP is close to zero and statistically insignificant, suggesting that competition from low-wage imports has had little enduring effect on domestic producer prices once we control for productivity. It may be that competition from such imports does affect producer prices by squeezing profit margins, but this effect must either be small or short lived, and manufacturers are probably able to restore their profit margins by becoming more efficient or shifting into higher value-added products.

Output Table 3 analyzes how the share of manufacturing in real output is determined; three different equations are presented. Equation (8) is based on the standard OLS approach applied to the pooled sample containing all 510 observations. The residuals of this equation exhibit strong autocorrelation, which is largely eliminated in equation (8AR) by assuming an AR(1) error process. Equation (8AVG) is derived by grouping the original data by taking three-year averages centered on the years 1964, 1968, 1972, and so on. While this method sacrifices information, it has the advantage of largely eliminating autocorrelation without having to make explicit assumptions about the nature of the error process.6

6We experimented using country-specific autocorrelation coefficients, but these gave much the same results as in equation (9AR). We also experimented with lagged endogenous variables in the regression analysis, but the coefficients turned out to be very small and statistically insignificant. The t-values shown in Table 3 as well as the other tables are based on the conventional OLS estimates of significance, and are very similar to those obtained using White©s correction for heteroscedasticity.

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

Table 3. Pooled Estimates of the Share of Manufacturing in Real Output, 1963-94

(Dependent variable =log OUTSHARE) Equation Number (8) (8AR) (8AVG) Explanatory variables Log Y 6.32** 5.01** 6.55** (13.63) (3.95) (7.02) (LogY)2 -0.350** -0.269** -0.363** (13.82) (4.00) (7.09) Log RELPRICE -0.589** -0.265** -0.611** (18.37) (6.14) (9.82) TRADEBAL 0.019** 0.004 0.020** (12.96) (3.50) (6.39) LDCIMP 0.000 -0.009 -0.003 (0.01) (1.16) (0.14) FIXCAP 0.016** 0.007** 0.018** (11.71) (5.89) (6.16) Country dummies exist? yes yes yes R2 0.84 0.36 0.84 Turning point $8,390 $10,983 $8,276 Notes: OUTSHARE = manufacturing value added as a percent of GDP at 1990 prices. Equation (8AR) assumes an AR(1) error process; equation (8AVG) is based on three-year averages centered on the years 1964, 1968, 1972, 1976, 1980, 1984, 1988, and 1992. FIXCAP = gross domestic fixed capital formation as a percent of GDP at 1990 prices. The turning point is the value of Y at which the dependent variable starts to fall with increasing Y. See also notes for Tables 1 and 2.

The key points to note in Table 3 are as follows. There is strong evidence of a hump- shaped relationship between log OUTSHARE and log Y. This implies that the income elasticity of demand for manufactures is well above unity when a country is poor, and falls below unity when a country becomes rich. This outcome is found no matter what method of estimation is used. Table 3 also shows the "turning point," which is the level of per capita income at which the income elasticity of demand for manufactures is equal to unity. An interesting feature of these estimations is the coefficient on relative prices. In equations (8) and (8AVG), this coefficient is highly significant and is about 0.6, suggesting that the price elasticity of substitution between manufactures and nonmanufactures is in the region of O.6.7 In the case of equation (8AR), the price coefficient is less

7The coefficient of the regression equation indicates, strictly speaking, how the real expenditure share on manufacturing is affected by changes in relative prices. The appendix demonstrates that this coefficient is approximately equal to the elasticity of substitution between manufactures and nonmanufactures.

©International Monetary Fund. Not for Redistribution Robert Rowthorn and Ramana Ramaswamy

Table 4. Pooled Estimates of the Share of Manufacturing in Employment, 1963-94

(Dependent variable = logEMPSHARE) Equation Number (9) (9AR) (9AVG) (10) Explanatory variables Log Y 11.78** 12.02** 11.65** 9.72** (22.14) (8.70) (10.92) (18.31)

(Log Y)2 -0.649** -0.667** -0.643** -0.510** (22.39) (8.97) (11.01) (17.05) TRADEBAL 0.012** 0.000 0.015** 0.012** (7.11) (0.14) (4.16) (5.88) LDCIMP -0.041** -0.035** -0.041 -0.041** (3.79) (4.36) (1.57) (3.54) FIXCAP 0.013** 0.008** 0.014** 0.002** (8.68) (6.65) (4.26) (0.21) Country dummies exist? yes yes yes yes Time dummies exist? no no no yes R2 0.90 0.70 0.91 0.93 Turning point $8,790 $9,421 $8,673 $13,846 Notes: EMPSHARE = employment in manufacturing as a percent of total employment. For other notes see Tables 1 and 2. than half this value, but remains highly significant. Thus, while there is strong evidence that, as economies develop, the demand for manufactures is stimulated by their falling relative price, there is some uncertainty about the magnitude of this effect. As expected, the real output share of manufactures is boosted by a positive manufacturing trade balance and by a high level of fixed capital formation, but the estimated size of these effects is lower when the AR version of the equation is used. In all output equations, the coefficient of LDCIMP is very small and insignificant, which is to be expected since this variable captures the impact of low-wage imports on productivity rather than output.

Employment Tables 4 and 5 examine how the share of manufacturing employment is determined. The former table uses pooled data, while the latter uses cross-sectional data. There is strong evidence of a hump-shaped relationship, with the employment share of manufacturing rising in the earlier stages of economic development and falling back at high levels of per capita income. The estimated turning point varies somewhat between

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION * ) ) ) 1 4 9 2 ) 6 1 0.4 0.03 0.022* (0.09 (0.05 (3.61 (18 199 -1.0 . 4 o t * 1 ) ) ) 3 s ) 3 0 0 8 ) t 6 1 0.018* 0.5 (1.07 (1.09 (3.80 (17 198 11.6 -0.62 Table e s $11,05 ($2,588 se , * ) ) 0 ) ) 5 4 9 notes 4 ) r Countrie 8 f 1 Employmen 0.017* 0.3 (16 (1.59 (1.58 (2.80 18.5 198 o -0.99 n r othe i $11,64 r ($1,114 g Fo . * Numbe d ) ) ) 4 0 ) ) 0 7 2 8 ) shown 1 an , r (15 198 0.014* 0.4 (1.55 (1.55 (3.20 16.7 -0.89 yea $11,44 Year e ($1,136 , th Manufacturin n * f * o EMPSHARE * ) ) ) o d ) 8 4 e 6 ) 6 log 1 0.014* 0.5 (14 (2.21 (2.21 (3.33 197 22.15 -1.189 ($832 = Shar Parentheses) centere $11,10 n d e (I r * * th * f perio r ) ) ) ) o 6 9 2 ) s 5 variable 1 0.6 0.021* Numbe (13 (2.65 (2.64 (4.03 197 19.51* -1.049* n ($788 $10,91 three-yea a r * * * Equatio Estimate ) ) ) l (Dependent ) ove 4 7 8 ) d 5 1 0.7 0.020* (3.21 (12 (3.30 (3.67 196 14.59* -0.783* $11,09 ($1,147 average a * * * ) ) ) dat ) 8 8 g 4 4 1 0.8 0.019* (5.03 (4.24 (11) (5.18 15.61* 196 ©International Monetary Fund. Not for Redistribution Cross-Sectiona -0.855* usin . ($638 $9,20 d 5 e Tabl estimate e s ar s ) t variable error 2 y d Equation poin : g Y) g Y 2 g Notes TRADEBAL (Standar (Lo R lo Turnin Explanator

29 Robert Rowthorn and Romano Ramaswamy equations, with the cross-sectional equations yielding a somewhat higher turning point than the pooled equations. However, the latter are probably a better guide to the intertemporal process of structural change since they make use of time-series data for individual countries. They suggest a turning point of about $9,000 (1986 purchasing power parity) per capita, which most OECD countries had reached by 1970, and some well before. A number of East Asian economies have also reached or surpassed this point, and the share of manufacturing employment in most of them is now falling; thus the estimated turning point is consistent with the East Asian experience.8 As expected, fixed capital formation exerts a positive influence on manufacturing employment, but its impact varies between equations. There is also evidence that the overall trade balance in manufactures has a major impact on manufacturing employment. In equations (9) and (9AVG) the coefficient of TRADEBAL is quite large and statistically significant; this is also true of the cross-sectional estimates shown in Table 5. However, in the autoregressive formulation (9AR) the coefficient of TRADEBAL is small, which most likely reflects the fact that this variable plays a much greater role in explaining cross-country differences in the structure of employment than it does in explaining intertemporal developments. The coefficient of the other trade variable, LDCIMP, differs somewhat in magnitude and significance between equations. Equations (9), (9AR), and (9AVG) use pooled data, and in all of these equations the coefficient of LDCIMP is negative and of similar size, although it is not always statistically significant. Interestingly, when additional dummies are included to allow for "fixed" time effects, the coefficient of LDCIMP increases in significance, indicating that this variable is not simply acting as a proxy for unidentified time effects. Note that in the cross-sectional regressions, the coefficients of LDCIMP are either small and insignificant or of the wrong sign. This suggests that although imports from the south have influenced the evolution of economic structure through time, they do not account for the large persistent differences in structure between countries. Such contrasts are explained primarily by the pattern of trade surpluses and deficits as reflected in the variable TRADEBAL.

III. Accounting for Deindustrialization In this section we quantify the influence of the various factors responsible for the declining share of manufacturing employment in the industrial countries since 1970. For this purpose we use the regression results shown in Table 4. This table contains a number of equations, all of which yield unbiased estimates, so there is a question as to which is the most appropriate. Equation (9) has autocorrelated residuals, but yields plausible coefficients. Equation (9AR) has the advantage that the residuals are virtually uncorrelated, but the coefficient for TRADEBAL is implausibly low, whereas equation (9AVG) also has uncorrelated residuals but ignores much of the available information. We have chosen the estimates in equation (9) as the basis for the decompositions, but the results would have been virtually the same whichever equation was used.

8See Rowthorn and Ramaswamy (1997) for a more detailed discussion of the East Asian experience regarding deindustrialization.

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

Table 6 decomposes the changes in manufacturing employment into various components using the regression coefficients shown in equation (9). The headings in the table are self-explanatory with the exception of the component labeled "normal growth." This component covers all of the effects that would normally be associated with rising per capita income in a closed economy, and thus takes into account both the income elasticity of demand and the influence of normal productivity and price changes. It is estimated from the coefficients of log Y and (log Y)2 in equation (9). Note that this component excludes the effect of abnormal price and productivity changes, in particular the abnormal productivity growth induced by competition from low-wage imports. The latter is included under the heading "north-south trade." The main conclusion from our decomposition is that the bulk of deindustrialization since 1970 is due to internal factors. In most countries in our sample, between one-half and two-thirds of the relative decline of manufacturing employment is explained by the normal process of economic growthÐvia changing preference patterns, differential productivity growth, and associated price changes. In addition, there has been a substantial fall in the ratio of investment to GDP, which has helped to skew demand away from manufactured goods. Our calculations suggest that this has had quite a large impact on manufacturing employment in some countries, and on average accounts for about one-sixth of deindustrialization. For the average country in the sample, this is similar to the effect of north-south trade. Note that these findings do not depend on the specific equation used for our decomposition; similar results are obtained with equation (9AR) or (9AVG). For most countries in the sample, our decomposition explains with reasonable accuracy what has happened to manufacturing employment since 1970. There is, however, a striking exception. As can be seen from Table 6, the residual for Japan is both large and positive, indicating that Japanese manufacturing employment has declined by much less than predicted. To understand what might be responsible for this anomaly, we examined the behavior of output, productivity, and prices over the period. The data indicate that the share of manufacturing in real output in Japan has risen substantially since 1970, whereas in most other countries this share has fallen. The unusual experience of Japan may be explained by the behavior of relative prices. Since 1970, profit margins have declined sharply in Japanese manufacturing industry, causing the relative price of manufactured goods to fall by far more than is predicted by productivity growth and by far more than in other industrial countries. This has stimulated the demand for manufactured goods in Japan, and helps to explain why the constant price share of manufactured goods in national production has risen since 1970, and hence why the share of manufacturing employment has fallen so little.

North-South Trade We can also estimate the impact of north-south trade on the structure of employment in advanced economies. Suppose that manufactured exports to the south increase by 1 percent of GDP. According to equation (9), this will cause the number of people employed in manufacturing to rise by 1.2 percent. Conversely, if manufactured imports from the south increase by 1 percent of GDP, the result will be a 5.3 percent fall in the number of

©International Monetary Fund. Not for Redistribution Robert Rowrhorn and Romano Ramaswamy . - o m 4 t e s fro g associ l Tabl refer y n " i 6 2 2 1 ) 1. arisin 6. 0. h -0. (9 n Residua normall growth e l growt ar y t equatio l n e tha 4 4 5 5 s "Norma . 0. trad Tota -1. -1. -2. productivit l change regressio e n equation o d d pric r e d 2 3 1 5 abnorma 1. 0. 0. base f an trad -0. e Othe y o s estimate ar e e 4 : th f her h to effect n e o e m e th 0 9 6 6 productivit r Du e e 1970-9 for show trad -1. -2. , -1. -0. ) s " thos North-sout f Chang o particula t n trade. i estimate h , e l effec l (logarithmic e 0 4 4 0 Th r . th s Tota -8. -8. -8. interna -10. changes y "north-sout g nonlinea e include t t Deindustrialization th i g ; o manufacturing t 6 0 9 5 headin f e productivit e (9) o d n -1. -0. -2. -1. e th du r s an Investmen e shar t unde equatio pric Explainin effect d l . n n l h i 6 t 0 0 9 8 e -8. -6. -7. -6. include growt Norma s employmen abnorma Tabl i f e r interactio e o t th coefficien e n th i s latte s e g t effec e n r Th . incom i ©International Monetary Fund. Not for Redistribution th o e e s 4 3 7 5 e include change s th s -9. -8. -3. m -10. Shar Chang Employmen fro countries exclude Manufacturin t d e I residual . e decompose th e g s estimate low-wag t n tabl h incomes s s showin wit effec Thi n n : countrie e Unio l rising n h State d n Notes wit colum incom d e e th ate Industria Europea Th competitio Unite Japa

32 GROWTH, TRADE, AND DEINDUSTRIALIZATION manufacturing workers.9 Thus, one dollar©s worth of imports from the south destroys 4.4 times as many northern manufacturing jobs as are created by one dollar©s worth of exports to the south. These figures indicate the origin of the "balanced trade effect," whereby imports from the south reduce manufacturing employment in the north even when they are matched by an equal value of northern exports. Among the richer countries in our sample, gross imports from the south have eliminated manufacturing jobs equivalent in number to 1.5-4 percent of total employment. For the United States, the figure is 2.2 percent of total employment, and for the average country in our sample it is 1.9 percent. The corresponding estimates for the new manufacturing jobs created by exports to the south are 0.3 percent for the United States and 0.3 percent for the average country. Given that total employment in the countries of our sample is about 350 million, this suggests that about 7 million manufacturing jobs have been lost because of southern competition and about 1 million created by additional exports to the south. The net loss of 6 million jobs is less than one-fifth of manufacturing jobs lost because of deindustrialization since 1970, but the impact on unskilled workers and those with nontransferable skills is greater than this figure suggests. Thus, although deindustrialization is not primarily due to north-south trade, such trade is likely to have had a sizable impact on the demand for certain types of labor. There are two main channels through which competition from low-wage producers can affect employment in northern manufacturing. The first is via its impact on total manufacturing output in the north; the second is through its impact on labor productivity. Our estimates suggest competition from low-wage producers has had little effect on the overall volume of manufacturing output in northern countries. In no country has there been a substantial change in the overall balance of manufacturing trade with the south, and the output regressions reported in Table 3 reveal little connection between aggregate manufacturing output and the volume of manufactured imports from the south. However, as we have seen in Table 1, there is evidence that low-wage competition has contributed to higher labor productivity in northern manufacturing. That is, in the face of low-wage competition, northern countries have responded, not by abandoning manufacturing as Brown and Julius (1994) have claimed, but by increasing labor productivity within the manufacturing sector. This has involved either increasing efficiency to produce more of the same kind of output per unit of labor, or switching to other types of manufactured goods where value- added per worker is higher.10

9Note that these effects are expressed as a percentage of manufacturing employment (not of total employment). They are derived as follows. An increase of 1 percentage point in the ratio of northern manufactured exports to GDP implies a change of + 1 unit in the variable TRADEBAL. According to equation (9), this will cause log EMPSHARE to change by +0.012, which is equivalent to a 1.2 percent increase in the number of manufacturing jobs. Conversely, suppose that the ratio of manufacturing imports from the south to GDP increases by 1 percentage point. This will cause the variables TRADEBAL and LDCIMP to alter by -1 and +1 unit, respectively. From equation (9), it follows that log EMPSHARE will change by (+0.012)*(-l) + (-0.041 )*(+l) = -0.053, which implies a 5.3 percent fall in the number of manufacturing jobs. Note that, in the average OECD country, 5 percent of manufacturing jobs represent about 1 percent of total employment. 10This is a theme that has been stressed by Krugman in his popular writings. See Krugman (1996).

©International Monetary Fund. Not for Redistribution Robert Rowfhorn and Romano Ramaswamy

IV. Conclusion

The main conclusion of this paper is that deindustrialization is explained mainly by factors that are internal to the advanced economiesÐi.e., as a result of the interactions among changing preference patterns between manufactures and services, the faster growth of productivity in manufacturing as compared to services, and the associated relative decline in the price of manufactures. North-south trade has, on average, contributed less than 20 percent to the relative decline in manufacturing employment in the advanced economies. Moreover, the impact of north-south trade on deindustrialization has been mainly through its effect in stimulating labor productivity in the manufacturing sector of the advanced economies; it has had little effect on manufacturing output in the advanced economies. The decline in the ratio of investment to GDP in the advanced economies has also skewed demand away from manufacturing output. The decline in the investment ratio has caused almost one-sixth of total deindustrializa- tionÐwhich is roughly similar to the effect of north-south trade on deindustrialization.

APPENDIX This appendix describes a simple model that motivates the regression equations used in the text. It also derives certain mathematical relationships between price elasticities and between income elasticities.

The Model The model refers to a closed economy that is divided into three sectors: manufacturing, services, and agriculture. Measured at base-year prices, the output of these sectors is Ym, Ys, and Ya, respectively, and total output is given by

(Al) Y = Ya + Ys + Ym

We shall refer to Y as "real" output. The following equation is satisfied identically:

Y = C + I + B, (A2) where C, I, and B are consumption, investment, and the overall foreign trade balance (total exports minus total imports). All are measured at base-year prices. The corresponding identity for manufacturing output is

Y (A3) c + Im + Bm. It follows that

Ym c Cm + Im + Bm Y c Y Y Y Cm B_ I Bm 1 + Im Cm + (A4) c Y I c Y Y ' This equation indicates that an increase in I/Y will raise the share of manufacturing if investment is more manufacturing intensive than consumption (i.e., Im/I > Cm/C).

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

Assuming that total exports and imports are equal, which is approximately true for the countries in our sample, the above equation can be written as follows:

Ym Cm Im Cm I B m (A5) Y C + I C Y + Y Measured at base-year prices, per capita income is given by

Y (A6) v N' where N is population. Real consumption can be decomposed as follows:

c = ca + cs + cm. (A7)

The share of each sector in total consumption is a function of per capita income and relative prices. Thus,

C i = a, s, m, (A8) -i- = fi (y, Pa / Pm, Ps / Pm) where fm(.) = l- fa(.) - fs(.)· We assume that the income elasticity of demand for services is greater than one at all stages of development, while that of agricultural goods is less than one. Thus, if prices are held constant, the share of services in real consumption increases with per capita income, while that of agriculture falls. The decline in agriculture is especially rapid during the early stages of development. Figure Al illustrates what these assumptions imply for the manufacturing sector. As the economy grows, both manufacturing and services initially increase their shares of real consumption at the expense of agriculture. However, this cannot continue indef- initely. As the share of agriculture shrinks, the scope for further shrinkage diminishes, and continued expansion in the service share must eventually be at the expense of manufacturing. At a certain point, the share of manufacturing in real consumption stabilizes and then begins to fall. Thus, for some y*,

dfm -y >0 for y < y* <0 for y>y*. (A9)

We also assume that the share of manufacturing in total investment is a function of relative prices. Thus,

(A 10) I = gm(Pa/Pm, Ps/Pm). Substituting in equation (A5) we obtain

Ym = fm(y, Pa/Pm, Ps/Pm) + hm(y, P/Pm, Pa/Pm)y + Bm (All) where hm(.) = gm(.) -fm(.).

©International Monetary Fund. Not for Redistribution Robert Rowrhorn and Ramana Ramoswamy

Figure A1. Sectoral Composition of Real Output

Productivity Output per worker is assumed to satisfy the following equations: l = ce«

y M /Y m _ c e Mt+QMJY1 A,>1, (A12) K m where c, cm, 9, O, and X are constants, and M\ stands manufacturing imports from the south, measured at base-year prices. The exponential terms allow for the effects of both technical progress and capital accumulation. The model does not explicitly allow for capital accumulation since adequate data on these variables were not available for all of the countries in our sample over the whole period. Labor productivity increases faster in manufacturing than elsewhere in the economy, and manufactured imports from the south stimulate manufacturing productivity in the north by eliminating low-productivity activities and encouraging innovation. Assuming that L N = constant, (A13) it follows that

log [Ym/Lm]yjK = constant + (X-l)log y + ( ?^-. (A14) Y/L

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

Prices The following mark-up equations are satisfied identically:

P Y -W L m m m m m= Wm Lm _ PY - WL (A15) II= WL , where W,(Wm) denotes the wage rate per unit of labor and II,(IIm), the mark-up of price over average cost.

(Pm) [Wm (l + IIm) ] (Ym /Lm ) log (A16) (/Pm) = log Wm (1 + IIm{-Tir)

Regression Analysis

The regression analysis reported in the text is based on the above model. Relative productivity is explained using the following version of equation (A14):

L +a lo logf^f j = oc0 . g-V + I;>iOC/Z., (1) where the Zs are other dummy variables that might influence the profit margin, and the equation numbers (1,2,...) refer to the equation numbering in the main text of this paper. Relative prices are explained using the following version of equation (A16):

pJ Y Lm 2 log(p#) = Po + ^{y/t)Y L +1« Mr ( )

This version is in effect testing how far wage rates and the mark-up in manufacturing differ from the national average. In equation (All), the share of manufactures in real output depends on the relative price of manufactures compared to services and agriculture considered separately. In our regression analysis we use a simpler formulation, in which the demand for manufactures depends on the single relative price Pm/P. The specific functional form is as follows:

-I log(Ym ) = yoYo + Y1logy.iogVy + Y2(log)f + y3log^ j + IM y,Z.. (3)

Provided Yi > 0 and Y2, 73 < 0, the demand for manufactures exhibits the required features. The income elasticity of demand is then greater than unity when Y is small, and less than unity for large Y. Moreover, a reduction in the relative price of manufactures increases the demand for such items.

©International Monetary Fund. Not for Redistribution Robert Rowfhorn and Romano Ramaswamy

Elasticity of Substitution This subsection demonstrates that the price elasticity of substitution between manufactures and total output is approximately equal to the price elasticity of substitution between manufactures and nonmanufactures. Measured at base-year prices, real output can be decomposed as follows:

Y=Ym + Yn, (All) where n refers to all nonmanufacturing activities combined. The corresponding equation in current prices is

PY=PJm+PJn- (A18)

In the base year, all prices are equal to unity and this equation coincides with the previous one. Define Y Y pjp A J ) (A 19) om YJY*(pjp) and

PJP *(YJY.) Qmn = (A20) y»./y- *pjp.y

Note that om is equal to -73, the coefficient of \og(Pm/P) in equation (3) of the main text, and amn is the elasticity of substitution between manufactures and nonmanufactures. We shall now consider how omn and cm are related. The following equation can be easily derived:

P Yn 3{PJP) P Y °,,, " *PJP.) (A21) mn AYJY) d{YJY,,)

To convert this equation into a more useful form, let Y P -h, -p--z. (A22) n n

From equations (A21) and (A22), it follows that

y, h Y i (A23) Y l + h' Y \ + h

P z(l + h) P \ + h m = (A24) P ' \ + zh ' P \ + zh

__zLdh_ (A25) °mn ~ h dz'

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIAUZATION

Table Al. Relationship Between aw and amn

h z om amn 0.2 0.5 0.3 0.32 0.4 0.5 0.3 0.33 0.2 2.0 0.3 0.27 0.4 2.0 0.3 0.25 0.2 0.5 0.6 0.62 0.4 0.5 0.6 0.64 0.2 2.0 0.6 0.56 0.4 2.0 0.6 0.54

Moreover,

%PJP) d(z(l + h)\ HPJP) A^*) (A26) _(\ + h) + {\-z)z~ (1+z/o2 and afa/r) _ d ( h \ d(YJYn) dh{\ + h) (A27) 1 (l+hf

Substituting in equation (A21) yields 0 + *K, mn (l + zh) + (\-z)ham (A28) (l-z)h(\-am) 1 + = °m (l + zh) + (l-z)ham

This is the required formula linking the two elasticities of substitution.

In the base year, all prices are equal to unity, so that z = 1 and hence <5mn = am. More generally, cmn is approximately equal to Gm if relative prices are similar to those in the base year, or if the share of manufacturing in real output is small, or if cm is close to unity. Table A1 illustrates the implications of the above equation for values of h and z in the range covered by our sample of countries. It is clear that amn and am are very similar in magnitude. Thus, ~Y3 in equation (3) of the text can be used as an estimator of omn.

©International Monetary Fund. Not for Redistribution Robert Rowrhorn and Romano Ramaswamy

Income Elasticity Formula This subsection examines the relationship between the income elasticities of demand for manufactures and nonmanufactures. Per capita output in base year prices is given by

*=** + *.> (A29) where Ym and Yn are the real per capita output of (expenditure on) of manufactures and nonmanufactures, respectively. Income elasticities of demand are defined as follows:

Y dY = nm Y dY * (A30) Y dYn

\ = Yn dY •

Differentiating equation (A29) with respect to Y, we find that

1 =L K Y \, + Y \ (A31) y Ym = ". T[m + 1- Y Y ^n* which implies that

Y (A32) i-nm=- -1 \ -1 ym .

To illustrate the implications of this equation, suppose that YmIY- 0.20 and that r\n = 1.10. In this case r|m = 0.6. Thus, the income elasticity of demand for nonmanufactures is close to unity, and the corresponding elasticity for manufactures is quite low. This is a direct result of the fact that the share of manufactures in total expenditure is small.

REFERENCES

Baumol, William J., 1967, "Macroeconomics of Unbalanced Growth: The Anatomy of Urban Crisis," American Economic Review, Vol. 57 (June). , Sue Anne Batey Blackman, and Edward N. Wolff, 1989, Productivity and American Leadership: The Long View (Cambridge, Massachusetts: MIT Press). Brown, Richard, and De Anne Julius, 1994, "Is Manufacturing Still Special?" in Finance and the International Economy, Vol. 8, ed. by Richard O©Brien (Oxford: Oxford University Press). Clark, Colin., 1957, The Conditions of Economic Progress (London: Macmillan) Falvey, Rodney E., and Norman Gemmel, 1996, "Are Services Income-Elastic? Some New Evidence," Review of Income and Wealth, Vol. 42 (September), pp. 257-69. Fuchs, Victor R., 1968, The Service Economy (New York: National Bureau of Economic Research, distributed by Columbia University Press). Krugman, Paul, 1996, Pop Internationalism (Cambridge, Massachusetts: MIT Press).

©International Monetary Fund. Not for Redistribution GROWTH, TRADE, AND DEINDUSTRIALIZATION

Lengelle, Maurice, 1966, The Growing Importance of the Service Sector in Member Countries (Paris: OECD). Rowthorn, R.E., and J.R. Wells, 1987, De-industrialization and Foreign Trade (Cambridge: Cambridge University Press). Rowthorn, Robert, and Ramana Ramaswamy, 1997, "Deindustrialization: Causes and Implications," Staff Studies for the World Economic Outlook (Washington: International Monetary Fund, December), pp. 61-77. Sachs, Jeffrey D., and Howard J. Shatz, 1994, "Trade and Jobs in U.S. Manufacturing," Brookings Papers on Economic Activity: 1, Brookings Institution. Saeger, S., 1996, "Globalization and Economic Structure in the OECD" (unpublished Ph.D dis- sertation; Cambridge, Massachusetts: Harvard University). Summers, Robert, 1985, "Services in the International Economy," in Managing the Service Economy, ed. by Robert P. Inman (Cambridge: Cambridge University Press), pp. 27-48. Wood, Adrian, 1994, North-South Trade, Employment, and Inequality: Changing Fortunes in a Skill-Driven World (Oxford: Clarendon Press). , 1995, "How Trade Hurt Unskilled Workers," Journal of Economic Perspectives, Vol. 9 (Summer).

Regional Trade Agreements or Broad Liberalization: Which Path Leads to Faster Growth?

ATHANASIOS VAMVAKIDIS*

Should a closed economy open its trade to all countries or limit itself to participation in regional trade agreements (RTAs)? Based on time-series evidence for a data set for 1950-92, this paper estimates and compares the growth performance of countries that liberalized broadly and that of those that joined an RTA. The comparisons show that economies grew faster after broad liberalization, in both the short and the long run, but slower after participation in an RTA. Economies also had higher investment shares after broad liberalization, but lower ones after joining an RTA. The policy implications support broad liberalization. [JEL F43]

Globalization has become one of the most popular topics of the 1990s. Articles on the formation of the new open world economy abound. Do countries benefit from free trade? What is the impact of a large open economy on welfare, growth, investment, technology, and income equality? What policies can maximize the benefits from globalization while minimizing the costs? These are some of the questions often asked by both international economists and policy makers. Alongside the globalization process, countries have been increasing their regional economic links through regional trade agreements (RTAs). Global versus regional integration has become an important policy dilemma that needs to be addressed by both economists and politicians.

*Athanasios Vamvakidis is an Economist in the IMF©s Asia and Pacific Department, and was in the Research Department when this paper was written. He holds a Ph.D. from Harvard University. He thanks Robert Sharer, Antonio Spilimbergo, Leslie Teo, Peter Wickham, Jonathan Wright, and the participants in an IMF Research Department Seminar at which this paper was presented for very helpful comments.

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

Do open economies prosper more than closed economies? Based on cross- country regressions for the 1970s and 1980s, studies1 have found that economies with low trade barriers grow faster. However, this is not a robust result. Other studies have found that openness variables are not significant in growth regressions that include investment over GDP as an independent variable.2 Nonetheless, openness is significant in regressions that have the investment share as the dependent variable. Do member countries of an RTA prosper more than nonmembers? Even though there is a considerable theoretical literature on trade creation versus trade diversion in RTAs, there are very few empirical growth studies, and no theoretical ones, that address the issue of opening to the world economy versus opening to an RTA. Participation in an RTA does not explain cross-country growth differences.3 Open economies grow faster while closed economies grow slower regardless of their participation in an RTA.4 What are the policy implications of the cross-country evidence? Can we infer that if an economy liberalizes it will grow faster? The answers in the existing literature to these questions are not satisfying and have often been criticized. One reason is that faster growth may be causing more trade and not the other way around. Another reason is that openness variables may be proxies for other country characteristics that have very little to do with trade. For example, most of the developing countries that have reduced trade barriers in recent decades have also implemented a variety of other policy reforms in fiscal and monetary policy, capital flows, financial regulation, and labor markets. Furthermore, most of the economies with high trade barriers are often also characterized by government intervention in internal competition and the financial sector, subsidy and tax programs favoring specific sectors in the economy, inefficient bureaucracy, inconsistent macroeconomic policies, and high inflation. Therefore, policy implications for the impact of openness on growth based on cross- country regressions should be treated with caution. Variables that attempt to measure trade distortions may be capturing other distortions instead. These problems can be tackled to a satisfactory extent by using time-series evidence. With this approach, many of the characteristics that differ across countries and are correlated with trade intervention do not influence the estimates.5 Comparing the growth performance of countries before and after trade liberalization can suggest what will happen when other countries follow similar policies in the future. Estimates of the impact of discriminatory, versus nondiscriminatory, liberalization on growth can lead to similar policy implications. This methodology can resolve the issue of whether a closed economy should liberalize to all countries or opt for a discriminatory approach via RTA. It also allows us to disentangle both the short- and the long-run effects of changes in trade policy.

1For example, see Dollar (1992), Edwards (1992), Barro and Sala-i-Martin (1995), Sachs and Warner (1995), Wacziarg (1998), and Vamvakidis (1997 and 1998). 2See Levine and Renelt (1992) and Baldwin and Seghezza (1996a). 3See Vamvakidis (1997 and 1998). 4Ben-David (1993) has shown that European Union (EU) members have experienced convergence. This is possible even though an EU dummy is not significant in cross-country regressions. 5Given that changes in trade policies may coexist with other reforms, we consider only liberalization episodes that feature trade openness as their main element.

©International Monetary Fund. Not for Redistribution Arhanasios Vamvakidis

Based on a data set for the period 1950-92, this paper estimates the growth performance of countries that liberalized to world trade or joined an RTA. The paper considers 109 cases of participation in an RTA and 51 cases of broad liberalization, between 1960 and 1989. The results suggest that closed economies that want to open their markets to free trade and face the dilemma of global versus regional integration should choose the first path. According to the estimates in the paper, economies have grown faster on average after liberalization, in both the short and the long run, but not after joining an RTA. The impact of RTAs on growth is actually negative and statistically significant in most empirical specifications. The results also suggest that broad liberalization leads to higher investment shares, while RTAs lead to lower ones. International economists have long been saying that openness fosters growth, but evidence in this paper shows that this advice must be qualified. Empirical findings show that only nondiscriminatory openness fosters growth. Thus, a closed economy seeking to open its markets will be better off with world integration, as opposed to regional integration, in terms of benefits for short- and long-run growth. This time-series evidence is robust, in contrast to cross-country regressions that have found mixed evidence on whether the impact of openness on growth is direct or indirect (through investment). To test the robustness of the estimates, this paper uses two measures of opennessÐan openness index constructed by Sachs and Warner (1995) and the trade share (total trade/GDP). Both are positive and statistically significant in fixed-effects regressions, when either growth or the investment share is the dependent variable. Even though RTAs and broad liberalization are not mutually exclusive in theory, evidence in this paper shows that countries rarely follow them simultaneously. In most cases, RTAs and broad liberalization have been two alternative paths of liberalization, and indeed this paper treats them as two alternatives. However, in more recent liberalization episodes, which are not considered in this paper, the two types of liberalization have to some extent been followed simultaneously. This paper assumes that RTAs have regional trade liberalization as one of their main purposes, but this has not always been the case. Indeed, even though most RTAs seem to have led to more trade among their members, there are some cases where this effect is not very large. However, our main conclusions remain valid when we consider each RTA separately. This paper does not address the issue of whether an already open economy should introduce RTAs. Most of the countries in our sample that participated in RTAs had high trade barriers. This is not always the case for more recent RTAs, but it would be useful for future research to measure their impact on growth.

I. Theoretical Aspects of the Impact of Liberalization on Growth The theoretical literature on trade and growth has changed its predictions in the last two decades. As Krueger (1997, p. 1) points out:

Ideas with regard to trade policy and economic development are among those that have changed radically. Then and now, it was recognized that trade policy

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

was central to the overall design of policies for economic development. But in the early days, there was a broad consensus that trade policy for development should be based on "import substitution" . . . The contrast with views today is striking. It is now widely accepted that growth prospects for developing countries are greatly enhanced through an outer-oriented trade regime and fairly uniform incentives . . .

The empirical literature on the openness-growth connection has contributed significantly to this shift. Open economies have grown much faster than economies with high protection during the last three decades. In addition, some of the economies that followed import substitution policies experienced economic crisis and collapsed during the 1980s and 1990s.6 These stylized facts have motivated a rich theoretical literature that attempts to explain them. Most studies have focused on two main channels through which trade fosters economic growth: technology and investment. The first channel has been supported mainly by Grossman and Helpman (1989, 1990, and 1991), Rivera-Batiz and Romer (1991a and 1991b), Romer (1990) and Krugman (1990, ch. 11). The main conclusion from this literature is that countries open to free trade benefit for the following four reasons: (1) a large international market provides technological spillover effects; (2) there are economies of scale in the research and development sector; (3) a large international market provides higher profits to innovators; and (4) there is avoidance of replication of research and development efforts across countries. Coe and Helpman (1995) and Coe, Helpman, and Hoffmaister (1997) have provided empirical evidence for these arguments, showing that trade affects the rate of technological progress. Other theoretical studies, however, have argued that investment is the main link between trade and growth. Baldwin and Seghezza (1996a and 1996b) have presented models whereby trade fosters investment for the following three reasons: (1) the traded sector is more capital intensive than the nontraded sector; (2) the production of investment goods uses imported intermediates; and (3) competition in the international market of machinery and capital equipment lowers the price of capital. Lee (1993 and 1994) presented neoclassical growth models in which domestic production uses imports of capital equipment as primary inputs. His models show that trade liberalization fosters growth through a rise in imports of capital goods. Empirical evidence by Levine and Renelt (1992), Baldwin and Seghezza (1996a), and Wacziarg (1998) supports the argument that trade fosters growth through its positive impact on investment. The first two studies find that this is the only channel, and some economists take this as a suggestion that the impact of openness on growth is not robust. Both of these channels may be important, but it is very difficult to empirically disentangle the effects of investment and technology, since most investment incorporates new technology and most new technology results in more investment. It is surprising that previous theoretical openness-growth literature does not address the issue of multilateral trade versus regional integration through RTAs. However, all of the above models would support opening up to free trade without any discrimination, versus opening only trade with a few neighboring countries, while still

6See Sachs and Warner (1995).

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis intervening to distort trade with the rest of the world. Nonetheless, more research on the theoretical links of regional integration with growth would help considerably in designing trade policy.

II. Data and Methodology The data set includes all countries with available data for the period from 1950 to 1992. This is a quite large sample, allowing us to estimate growth performance before and after liberalization. Data for real GDP per capita, investment share, and population growth are from the Penn World Table data set described by Summers and Heston (1991). Trade shares are from the World Tables (World Bank, 1994), and school enrollment ratios are from Barro and Lee (1994). A variety of measures have been used in the previous literature to measure openness, including trade shares, growth of exports, tariff and nontariff barriers, and black market premiums. The problem with these measures is that they cannot suggest the precise year that a country opened for free trade. However, Sachs and Warner (1995) have constructed a measure of openness that deals successfully with this problem. Based on their definition, an economy is defined as open if all five of the following conditions are met: (1) average tariff rate is less than 40 percent; (2) average nontariff barriers are less than 40 percent; (3) black market premium is less than 20 percent of the official exchange rate; (4) it does not have a communistic government; and (5) there is no state monopoly on major exports. Using this approach, Sachs and Warner review regulations that have changed trade policy in most countries of the world since the 1950s, and they provide the dates of liberalization for countries that are considered open. This paper uses the Sachs and Warner methodology to determine the year a country liberalized to international trade. The authors are very clear on which of the above cri- teria are the main elements of each of the liberalization episodes they consider. Based on this information, we consider only liberalization episodes that have trade openness as their main element (we do not consider ex-communist countries or countries for which the date of independence is defined as their liberalization date). According to this approach, there were 51 cases of broad liberalization between 1958 and 1989.7 Another way to deal with the possibility that the Sachs and Warner measure may be too broad, since policy changes in trade protection often coexist with other policy reforms, is to use the trade share ((exports + imports)/GDP) as an alternative measure of openness. (Other openness measures have only cross-country variations and therefore are not relevant for our analysis). The use of trade shares has been criticized in the past, because these shares are negatively correlated with country size. However, this criticism applies only to cross-country regressions. The present paper is on time- series evidence and therefore the use of trade shares is justified (the results do not change when we control for country size). Sachs and Warner also consider failed liberalization episodes, in which countries opened to free trade for a period of time and then closed again. Trying to explain why trade liberalization was reversed in these cases, the authors look at the growth performance of

7We do not consider cases of liberalization after 1989, which are too recent.

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

the respective countries. What they find is that in all cases countries grew faster during these liberalization episodes than before and after liberalization (this led the authors to conclude that political economy arguments and not growth performance might be driving these policy changes). According to the authors, this contradicts the argument that the suc- cessful liberalization episodes led to faster growth by definition (since the failed episodes also led to faster growth). For this reason, and to make the analysis simpler, this paper does not consider empirical estimates for failed liberalization episodes. However, the results remain robust if such episodes are included in the regressions. The dates of participation in an RTA are available in UNCTAD (1994). The data set includes 18 RTAs with a total of 109 member countries, with all RTAs for the period 1958-89. If a country participates in more than one RTA, the earliest entry is regarded as the date of participation. Most of the member countries have data available for the period in consideration. However, there are not enough observations yet to estimate the effect of more recent RTAs. The appendix provides a list of the RTAs this paper considers. Most RTAs since 1950 have been south-south agreements. With very few exceptions, they include developing countries that are generally small, highly protected, and similar in their economic endowments. Such agreements have often been part of the import substitution policies their members were following and, as a result, they may have diverted trade from more efficient external sources of production. Even though some of these agreements were not fully implemented, it is important to estimate their impact on economic growth, to determine why they have or have not succeeded. Most participating countries thought that these agreements were an alternative path to broad liberalization, and therefore they are relevant for the inquiry of this paper. The calculations use a dummy variable for participation in an RTA and therefore treat all RTAs equally. This may be a problem since, as mentioned above, some RTAs were implemented more fully than others, and also not all RTAs had regional trade as their main purpose (political cooperation was sometimes the main reason for an RTA). However, the conclusions of this paper do not change if we treat each RTA separately, with a dummy variable for each of them. The methodology in this paper is very straightforward. One very simple test compares the average real GDP per capita growth rate 10 years before and 10 years after both discriminatory and nondiscriminatory liberalization. In this test the estimations do not include liberalization cases and RTAs of the 1980s. Another simple approach would have been to estimate the change in the trend of GDP per capita after liberalization, or an RTA, for each country in the sample. However, the GDP series follow a unit root for almost all countries in the sample and therefore use of this methodology is not appropriate.8 A solution is to use the growth rate as the dependent variable, and then estimate its change after the date of opening for free trade or for an RTA. Following this methodology, this paper estimates a fixed-effects growth model.9 The observations are five-year averages for the period 1950-92.10 The model controls

8See Campbell and Perron (1991). 9The fixed-effects model allows for the constant to vary across countries. 10The openness and RTA dummies take the value of one if the change in trade policy occurred in the first two years of the five-year period.

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis for the initial GDP per capita, population growth, the share of investment, secondary school enrollment, growth of world GDP per capita, a dummy for participation in an RTA, the Sachs and Warner (1995) openness dummy, the trade share, and the growth of world GDP per capita. The choice of variables follows Levine and Renelt (1992), with the addition of the growth of world GDP per capita. The inclusion of this variable in the regression controls for the effect of world output fluctuations on domestic output. We may need to control for such an impact because the period after liberalization or participation in an RTA in most cases includes the 1970s, a decade characterized by worldwide productivity slowdown. However, the results are not sensitive to the inclusion of this variable in the regression. Most studies have estimated this model using cross-country regressions. However, a fixed-effects model will be more useful for policy implications. As pointed out in the introduction of this paper, time-series evidence is more relevant than cross-country regressions in determining the impact of trade policy changes on growth performance. The time-series approach also has interesting implications for convergence, as in this case convergence occurs in terms of country-specific long-term values. Finally, the paper estimates an empirical specification of the fixed-effects model with the investment share as the depended variable. It then investigates the impact of RTAs and broad liberalization, measured by the openness dummy or the trade share.

III. Growth Before and After Broad Liberalization and RTAs This section compares growth performance 10 years before and 10 years after discriminatory and nondiscriminatory liberalization. Table 1 includes countries that liberalized during 1960-80, and indicates the year of their liberalization and their growth rates 10 years before and 10 years after this date. Table 2 shows similar information for countries that joined an RTA. Based on the Sachs and Warner definition of openness, most countries that conducted RTAs during the 1960s and 1970s had high protection. The European Union (EU) was the only RTA that included only open economies. Therefore, it seems that most RTAs during this period represented efforts to increase the market size available to closed economies, and may have delayed broader liberalization. Table 2 also lists countries that opened at the same time (three years before or after) that they joined an RTA. For the EU, only three countries (the United Kingdom, Ireland, and Denmark) liberalized before joining (in addition, these three countries had already been members of earlier RTAs). Furthermore, the productivity slowdown during the 1970s may influence the results, since these countries joined the EU in 1973. The tables report the results for the EU, but any conclusions from them may be misleading. According to the evidence in both Tables 1 and 2, economies grew faster 10 years after broad liberalization or participation in an RTA, compared with 10 years before, but this effect seems to be larger, though not significantly, for the first case. (The impact of openness on growth is even higher if we exclude Germany and Japan from the sample, since during the decade before their liberalization they were recovering from World War II.) Most of the countries grew faster after opening their markets to international trade, although they did not experience economic "miracles." This result does not change if

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

Table 1. Growth 10 Years Before and 10 Years After Liberalization

Year of Growth 10 Years Growth 10 Years After Countries Openness Before Liberalization Liberalization Australia 1964 2.09 2.55 Austria 1960 5.68 4.22 Belgium 1960 2.46 4.08 Botswana 1979 9.17 2.02 Chile 1976 -0.46 1.83 Denmark 1960 3.59 3.29 France 1959 3.61 4.78 Finland 1960 4.04 4.36 Germany 1959 6.74 3.92 Greece 1959 4.59 6.62 Indonesia 1970 1.08 6.97 Ireland 1966 3.46 3.97 Italy 1959 4.75 5.51 Japan 1962 7.85 6.77 Jordan 1965 6.92 5.78 Korea 1968 4.77 6.43 Luxembourg 1959 2.29 1.78 Netherlands 1959 2.23 4.62 Norway 1960 2.95 3.72 Portugal 1960 4.24 6.85 Spain 1960 4.81 5.46 Sweden 1960 3.10 2.92 Taiwan 1963 3.58 6.29 United Kingdom 1960 2.26 2.55 Average 3.99 4.47 we exclude countries that joined an RTA at the same time they liberalized (three years before or after joining an RTA). However, these comparisons can be misleading, since it is not clear what is driving the results. To understand the full impact of these two different forms of liberalization on growth, we need to control for changes in other variables, as the following sections show.

IV. Estimation of a Fixed Effects Growth Model Estimation of a fixed-effects model shows that economies grow faster after broad liberalization. Table 3 presents these results. Regressions (l)-(5) include the Sachs and Warner openness dummy, while regressions (6)-(10) include the trade share. Both variables have positive and statistically significant coefficients, at least at the 5 percent level.11 These results are not sensitive to the inclusion of other independent variables.

11Levine and Renelt (1992) have found based on cross-country regressions that trade fosters growth only indirectly, through higher investment. We find that this is not true for a fixed-effects model.

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis

Table 2. Growth 10 Years Before and 10 Years After Participation in an RTA

Year in Growth 10 Years Growth 10 Years RTA RTA Open Before RTA After RTA MRU 1.41 0.50 Guinea 1973 no -0.49 2.84 Liberia 1973 no 2.72 -1.06 Sierra Leone 1973 no 2.00 -0.29

UDEAC 1.29 3.62 Cameroon 1966 no 1.04 2.85 Central Africa 1966 no -1.15 1.14 Chad 1966 no -0.46 -1.20 Congo 1966 no -0.67 5.03 Gabon 1966 no 7.71 10.27

CEPGL 3.50 0.50 Burundi 1976 no 2.01 2.15 Rwanda 1976 no 6.80 1.80 Zaïr e 1976 no 1.70 -2.45

ASEAN 2.98 5.78 Indonesia 1967 "yes" -0.63 5.29 Malaysia 1967 no 2.83 6.02 Philippines 1967 no 1.85 2.79 Singapore 1967 no 3.34 10.53 Thailand 1967 yes 5.58 4.27

ACM 3.00 2.33 Egypt 1964 no 2.77 1.49 Iraq 1964 no 3.28 2.49 Jordan 1964 "yes" 7.55 3.83 Mauritania 1964 no -1.61 1.52

CACM 2.08 2.78 El Salvador 1960 no 1.84 2.48 Guatemala 1960 no 0.83 2.03 Honduras 1960 no 0.83 1.78 Nicaragua 1960 no 3.04 3.99 Costa Rica 1960 no 3.88 3.64

CARICOM 4.51 1.033 Barbados 1973 yes 5.42 1.49 Jamaica 1973 no 5.61 -1.65 Trinidad & Tobago 1973 no 2.50 3.26

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

Table 2. (concluded)

Year in Growth 10 Years Growth 10 Years RTA RTA Open Before RTA After RTA LAIA/LAFTA 1.70 2.58 Mexico 1961 no 2.64 3.95 Argentina 1961 no 1.16 2.10 Bolivia 1961 no -0.82 4.01 Brazil 1961 no 3.54 3.64 Chile 1961 no 1.81 2.59 Colombia 1961 no 1.19 2.62 Ecuador 1961 no 2.07 2.56 Paraguay 1961 no -0.55 1.84 Peru 1961 no 3.08 2.78 Uruguay 1961 no 1.67 0.54 Venezuela 1961 no 2.95 1.78

EU 3.81 3.00 Belgium 1958 "yes" 2.37 3.80 France 1958 "yes" 3.91 4.37 Germany 1958 "yes" 7.35 3.88 Italy 1958 "yes" 4.92 5.51 Luxembourg 1958 "yes" 2.58 0.64 Netherlands 1958 "yes" 3.21 4.49 United Kingdom 1973 yes 2.55 1.04 Ireland 1973 yes 4.10 2.16 Denmark 1973 yes 3.29 1.13

Average 2.56 2.74 Notes: Acronyms are spelled out in the appendix. "yes" indicates countries that liberalized within three years before or after joining an RTA.

Based on regression (5), economies grew on average 1.5 percent faster after broad liberalization. Based on regression (10), an increase in the trade share of a country by 10 percent leads to faster growth by 0.56 percent on average. The estimated coefficients for the other variables confirm previous findings in the literature. Keeping everything else constant, a low GDP per capita, a high investment share, and a low population growth rate lead to faster growth. The impact of secondary school enrollment is not robust, which also agrees with previous findings. Finally, the growth rate of world GDP per capita has a positive and statistically significant coefficient, justifying its inclusion in the regression. Fluctuations in world output do have an effect on domestic output. The estimations in this and the next section also confirm these results. In contrast to broad liberalization, participation in an RTA does not foster growth. Table 4 actually shows the impact of RTAs, if any, to be negative. The coefficient of the RTA dummy is always negative, and is statistically significant in

©International Monetary Fund. Not for Redistribution Afhanasios Vamvakidis ) ) ) ) ) ) 0 6 7 7 0 5 ) 6 5 3 10 49 (10 0.3 0.63 0.05 0.03 0.20 (3.394 (1.817 (5.060 (5.610 -1.09 -0.06 (-4.001 (-9.062 ) ) ) ) ) 5 4 6 6 7 4 ) 5 3 (9 10 49 0.3 0.21 0.02 0.05 (6.045 (1.226 (4.791 -1.13 -0.07 (-4.117 (-11.731 ) ) ) ) 4 3 4 1 1 5 ) 3 (8 10 49 0.3 0.05 0.20 0.03 (5.511 (1.668 (4.597 -0.07 (-11.292 ) ) ) 0 9 6 8 ) 7 5 (7 13 60 0.2 0.12 0.06 (5.403 (3.871 -0.07 h (-12.668 ) . 8 ) 7 6 Growt 7 5 ) d (6 13 0.06 60 0.2 (6.171 -0.07 an (-12.68 , parentheses n ) ) ) ) ) ) i 0 e 2 9 9 5 6 7 5 ) 9 Share ar 8 (5 s 52 e 0.2 0.13 0.01 0.23 (2.084 (6.523 (2.137 -1.04 -0.05 -0.64 (-0.320 (-4.618 (-9.265 Trad , ) ) ) ) ) r-statistic . 8 2 6 3 6 6 ) 5 9 8 (4 52 0.2 0.01 0.23 (6.772 (2.291 -0.00 -1.08 -0.05 (-0.273 (-4.756 (-8.996 numer n Openness . ) ) ) ) 6 7 9 2 3 2 ) 5 9 e 8 (3 regressio 52 0.2 0.00 0.01 0.22 e (2.552 (6.310 (0.299 -0.05 (-8.606 th Tabl t ) ) ) 3 5 1 0 ) 9 5 represen (2 10 76 s 0.2 0.02 0.19 (7.124 (4.527 -0.04 ©International Monetary Fund. Not for Redistribution (-11.572 ) ) 5 1 4 9 ) parenthese 5 n (1 76 i 0.1 0.02 s (5.214 -0.04 10 (-10.258 ; s head n h l y e a d P s t t ) countrie shar growt e schoo Colum worl t capit f f : f n dumm r y GD s s o o l o r r pe h shar P e capita Notes r 2 observation GD enrollmen pe Numbe Numbe Secondar Populatio R Growt Investmen Opennes Independen ln(initia Trad Variable

52 REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

Table 4. RTAs and Growth

Independent Variables (1) (2) (3) (4) (5) (6) (7) ln(initial GDP -0.028 -0.032 -0.042 -0.045 -0.048 -0.064 -0.055 per capita) (-7.776) (-9.202) (-7.847) (-8.411) (-8.854) (-8.991) (-8.856)

RTA dummy -0.007 -0.006 -0.007 -0.006 -0.006 -0.003 -0.003 (-2.251) (-1.909) (-1.761) (-1.415) (-0.494) (-0.659) (-0.678)

Trade share 0.057 (5.087)

Openness dummy 0.016 (2.234)

Investment share 0.167 0.209 0.219 0.209 0.198 0.227 (6.722) (6.377) (6.790) (6.490) (5.495) (6.388)

Secondary school 0.015 0.005 0.003 0.037 -0.006 enrollment (0.751) (0.265) (0.167) (1.823) (-0.300)

Population growth -0.927 -0.893 -1.077 -1.033 (-4.320) (-4.180) (-3.937) (-4.519)

Growth of world 0.158 0.603 0.133 GDP per capita (2.634) (3.081) (2.093) Number of countries 147 147 109 109 109 105 89 Number of observations 943 943 621 621 621 493 525 R2 0.11 0.16 0.20 0.23 0.24 0.36 0.27 Notes: Column heads in parentheses represent the regression number. t-statistics are in parentheses. regressions (l)-(3), at least at the 10 percent level. The t-statistic decreases as we add more independent variables in the regression. The estimates, however, are not small. For example, based on regression (5), growth decreases after participation in an RTA by 0.6 percent. The results do not change significantly if we use a different dummy variable for each RTA (and therefore we do not report these results). As was mentioned above, all RTAs differ, so that using a different dummy for each of them may be the appropriate methodology. However, the estimates for the RTAs are still negative for most of them, but not always statistically significant, as is also the case with the results we report. Broad liberalization has a significant effect on growth even after controlling for participation in an RTA. As regressions (6) and (7) of Table 4 show, both the openness dummy and the trade share have positive and statistically significant coefficients.

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis

To summarize, even controlling for changes in other economic variables, RTAs do not appear to have a robust impact on growth (if there is any impact, it seems to be negative). In contrast, countries have grown faster on average after nondiscriminatory liberalization. This result is true not only for cross-country regressions, but also for time-series regressions, as this paper has shown, and it is robust to different measures of openness and specifications of the empirical model.

V. Short-Run Versus Long-Run and Lagged Effects of Broad Liberalization and RTAs

Did countries experience the growth effects of changes in their trade policies in the short run, the long run, or both? If trade policy has an impact on growth with a lag, the results in the previous two sections may not be accurate. In any case, this is an interesting question. Policy makers and academics often disagree on whether countries experience the effects of changes in trade policy soon enough to make them worth- while. To address this important issue, the empirical model includes lags for the openness and the RTA dummies. One lag measures the impact of changes in trade policy on growth after five years, while two lags measure this impact after a decade. If the estimate of the lagged liberalization (broad or RTA) is insignificant, this means that the short-run effect of the policy continues in the longer run. If it is positive (negative) and significant, it means that the impact of liberalization is greater (smaller) in the long run. RTAs do not seem to have a statistically significant impact, positive or negative, in the long run. Results in Table 5 show that RTAs lead to slower growth in the short run (coefficients of the RTA dummy), a result statistically significant at least at the 10 percent level in regressions (l)-(3) and (6)-(8). In contrast, the long-run effect of RTAs (coefficients of the one and two lags of the RTA dummy) does not always have the same sign and is not significant in any of the regressions. Short- and long-term effects are both positive and statistically significant only in the case of broad liberalization. Countries grow faster, in both the short and the long run, after they open their market without discrimination to international trade. The results in Table 6 show that the openness dummy has a significant coefficient, at least at the 10 percent level, implying that countries grew faster within the first 5 or 10 years after broad liberalization. The coefficient for the one lag of the openness dummy is significant in most specifications, implying that the growth effect of openness not only continues in the period following the first five years after opening for free trade, but is also stronger. The coefficient of the two lags of the openness dummy is always significant, implying that the short-run impact of liberalization on growth becomes stronger in the period 10 years after liberalization. These results suggest that closed economies intending to open their markets to international trade will grow faster if they follow the path of broad as opposed to regional liberalization. All time-series evidence in this section show that RTAs, if any- thing, lead to slower growth, especially in the short run. In contrast, broad liberalization fosters growth in both the short and the long run. These results hold even after keeping constant other independent variables.

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION ) ) ) ) ) ) ) 6 6 3 9 2 7 7 ) 4 9 1 10 62 (10 0.2 0.21 0.16 0.00 (0.351 (2.705 (6.566 -0.00 -0.90 -0.04 -0.00 (-1.579 (-4.259 (-8.205 (-1.375 ) ) ) ) ) ) 3 3 9 6 1 5 ) 3 9 1 (9 10 62 0.00 0.00 0.22 0.2 (1.453 (6.863 (0.436 -0.94 -0.00 -0.04 (-4.395 (-7.733 (-1.306 ) ) ) ) ) 8 0 1 5 7 ) 0 9 1 (8 10 62 0.00 0.21 0.01 0.2 (1.204 (6.430 (0.895 -0.04 -0.00 (-1.672 (-7.229 ) ) ) ) 0 6 7 2 n 6 ) 7 3 (7 14 94 0.16 0.00 0.1 Ru (0.041 (6.718 -0.00 -0.03 g (-8.843 (-1.840 . Lon ) 3 ) ) 8 8 d ) 1 7 3 (6 14 0.000 94 an 0.1 (0.081 t -0.02 -0.00 (-7.471 (-2.179 parentheses n i Shor ) ) ) e ) ) ) ) e 9 6 6 1 3 1 8 4 ) 9 1 ar s th (5 10 62 0.2 0.16 0.20 0.00 n (2.593 (6.474 (0.179 -0.00 -0.04 -0.00 -0.89 i (-8.799 (-1.230 (-4.179 (-0.184 h r-statistic ) ) , ) ) ) ) 9 8 5 2 5 7 ) 3 9 1 Growt (4 10 62 0.2 0.00 0.21 0.00 d (0.230 (0.482 (6.787 -0.00 -0.91 -0.04 (-1.476 (-4.263 (-8.381 number n an s ) ) ) ) ) 3 4 4 9 9 ) 0 9 1 RTA (3 . regressio 10 62 0.2 0.20 0.00 0.01 e (0.681 (0.820 (6.383 5 -0.04 -0.00 (-7.875 (-1.941 e th t ) ) Tabl ) ) 8 3 9 5 ) 6 7 3 represen (2 14 s 94 0.1 0.00 0.16 (6.763 (1.211 -0.03 -0.00 (-2.255 (-9.274 ©International Monetary Fund. Not for Redistribution ) ) ) 4 9 8 ) parenthese 1 7 3 n (1 i 14 94 0.00 0.1 s (0.946 -0.02 -0.00 (-7.822 (-2.372 s head n h l e a d e e P y y s th y t t ) f countrie shar growt th schoo Colum worl t f capit f f : f n r y o GD s o o s o l o r r g h pe dumm dumm P dumm capita lag Notes A la A r o e A 2 observation GD enrollmen pe RT RT Numbe Secondar ln(initia Numbe R Populatio Growt Investmen RT On Tw Independen Variable

55 Afhanasios Vamvakidis ) ) ) ) ) ) ) 4 1 2 7 4 7 5 ) 8 5 9 8 52 (10 0.2 0.14 0.01 0.23 0.01 (2.233 (6.687 (2.865 (2.025 -1.03 -0.01 -0.06 (-4.578 (-0.839 (-9.626 ) ) ) ) ) ) 2 6 6 6 7 1 ) 7 5 9 (9 8 52 0.01 0.24 0.01 0.2 (2.751 (6.942 (2.193 -0.01 -1.06 -0.06 (-0.768 (-4.726 (-9.322 ) ) ) ) ) 0 1 8 5 7 ) 3 5 9 (8 8 52 0.2 0.01 0.23 0.01 (2.450 (2.795 (6.484 -0.00 -0.06 (-0.220 (-8.992 n ) ) ) ) 9 2 0 8 Ru ) 1 9 5 g (7 10 76 0.01 0.19 0.2 0.01 (3.687 (2.550 (7.356 -0.05 (-11.527 Lon d ) ) ) 3 9 5 an t ) 4 . 9 5 (6 0.00 0.02 10 76 0.1 (4.567 (1.814 -0.04 (-10.021 Shor e ) ) ) ) ) ) ) th parentheses 8 7 1 1 8 3 9 n 7 ) n 5 i 9 i e 8 (5 52 0.2 0.12 0.22 0.01 0.00 h (1.898 (6.425 ar (1.675 (1.254 -1.01 -0.00 -0.05 s (-4.413 (-0.415 (-9.121 ) ) ) ) ) ) Growt 3 3 8 0 3 7 d ) r-statistic 6 5 , 9 (4 8 52 0.2 0.23 0.01 0.01 (1.518 (6.631 (1.730 an -0.00 -1.03 -0.05 (-4.500 (-0.392 (-8.897 s number n ) ) ) ) ) 2 4 1 4 7 ) 3 5 9 (3 8 52 0.2 0.22 0.00 0.01 0.01 (0.096 (1.783 (2.131 (6.161 regressio -0.05 Opennes (-8.791 e . 6 th t e ) ) ) ) 3 2 9 5 ) 1 9 5 Tabl (2 10 76 0.2 0.19 0.01 0.01 represen (7.132 (2.683 (2.180 s -0.04 ©International Monetary Fund. Not for Redistribution (-11.588 ) ) ) 9 2 5 ) 4 9 5 parenthese (1 n 10 76 0.01 0.01 0.1 i (3.271 (2.139 s -0.04 (-10.340 s head y y h l n y e a d e e P s i th t t ) f countrie shar growt th schoo worl dumm dumm t Colum capit f f f dumm : f n s s r y o s GD s o o s o l o r r g pe h P capita lag Notes la r o e 2 GD observation opennes enrollmen opennes pe Growt Secondar Numbe Populatio Numbe R Investmen On Independen Variable ln(initia Opennes Tw

56 REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

Another interesting question is whether the positive impact of trade on growth is increasing or decreasing as the trade share rises. The estimates show the second case to be true. Table 7 shows that the estimated coefficient of the trade share is positive and significant, but that of the trade share squared is negative and significant. The total effect of trade on growth, however, is still positive. In addition, the estimated coefficient of the trade squared, although negative, is very small. The estimates of regression (5), for example, imply that an increase of the trade share by 10 percent results in 0.8 percent faster growth. It is interesting that the inclusion of the trade share squared in the regression increases the coefficient of the trade share (compare Table 7 with Table 3).

VI. Impact of Broad Liberalization and RTAs on Investment As mentioned earlier, one part of the literature has found evidence, using cross-country regressions, that the impact of trade on growth is only through higher investment. In

Table 7. Openness and Growth in the Short and Long Run

Independent Variable (1) (2) (3) (4) (5) ln(initial GDP -0.076 -0.075 -0.074 -0.075 -0.063 per capita) (-12.665) (-12.654) (-11.227) (-11.644) (-8.805)

Trade share 0.103 0.092 0.088 0.083 0.092 (5.760) (5.102) (4.745) (4.516) (5.050)

Trade share squared -0.013 -0.012 -0.012 -0.010 -0.012 (-2.489) (-2.249) (-2.426) (-2.002) (-2.476)

Investment share 0.123 0.194 0.210 0.191 (3.715) (5.341) (5.862) (5.350)

Secondary school 0.028 0.020 0.033 enrollment (1.374) (1.008) (1.612)

Population growth -1.071 -1.007 (-3.872) (-3.692)

Growth of world 0.696 GDP per capita (3.696) Number of countries 137 137 105 105 105 Number of observations 605 605 493 493 493 R2 0.27 0.29 0.32 0.34 0.37 Notes: Column heads in parentheses represent the regression number. t-statistics are in parentheses.

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis

contrast, the results of the previous sections based on time-series variation suggest that there is a direct effect of openness on growth. This section investigates whether there is also an indirect effect, through higher investment. If economies invest more after broad liberalization, then they will grow faster, all else being constant. This section also estimates the impact of RTAs on investment. For reasons of comparison and simplicity, the estimates in this section use the same models as in the previous sections, but with investment over GDP as the dependent variable. The results of this inquiry are presented in Tables 8 to 12. The main conclusion is that broad liberalization leads to higher investment shares, while RTAs, if they have any impact, lead to lower shares. Therefore, there is also an indirect effect of liberalization on growthÐpositive for broad liberalization, and negative, if any, for RTAs. Both the openness dummy and the trade share are always positive and significant at least at the 5 percent level (Tables 8 and 9). The estimates of regression (4) in Table 8 imply that the investment share of an economy increases by 2.7 percent after opening its markets to free trade. The estimates of regression (9) in the same table imply that an increase of the trade share of an economy by 10 percent increases its investment share by 0.56 percent. In contrast, the RTA dummy is always negative and significant at least at the 10 percent level in all specifications (Table 9). The estimates of regression (6) in Table 9 imply that the investment share of an economy decreases by 1.6 percent after joining an RTA. The negative impact of RTAs on investment is for both the short and the long run, but not always significant (Table 10). The estimates in Table 10 show that the RTA dummy is always negative, but not always significant. When it is significant, the level of significance is only 10 percent. The "one lag of the RTA dummy" variable is always negative, but the "two lags of the RTA dummy" variable is always positive. However, none of them is significant, which implies that the short-run effect of RTAs continues in the long run. The positive impact of openness is significant in the short run, but its long-run impact seems to be slightly higher or slightly lower, depending on the specification (Table 11). The estimates in Table 11 show that the openness dummy is always positive and significant. The "one lag of the openness dummy" variable is always positive, but significant only in regression (3), and at the 10 percent level. In contrast, the "two lags of the openness dummy" variable is always negative, but significant only in regression (5). Since these lagged variables do not have robust coefficients, we could argue that the impact of broad liberalization on investment is the same in the short and the long run. Finally, even though the trade share has a positive coefficient, the trade share squared has a negative coefficient but is not always significant (Table 12). This suggests that the positive impact of trade on investment decreases as the trade share rises. The coefficient of the trade share is once again greater in regressions that include the trade share squared. Regarding the estimates for the other variables, the only puzzling result is the coefficient for the secondary school enrollment, which is negative and significant. Even though the inclusion of this variable in the regression does not change the other results, it is very difficult to explain its negative coefficient. Perhaps better measures of human capital are necessary to obtain a more reasonable estimate.

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION ) ) ) ) ) 9 6 6 4 5 ) 9 5 3 (8 10 49 0.0 0.88 0.71 0.05 0.01 (5.060 (2.285 (2.682 (1.567 -0.08 (-3.095 ) ) ) ) 9 4 8 4 8 ) 5 3 (7 10 49 0.0 0.84 0.00 0.04 (2.178 (0.405 (3.065 -0.10 (-3.679 ) ) ) 3 0 2 6 ) 5 3 (6 10 49 0.0 0.05 (0.290 (3.145 -0.00 -0.11 (-3.994 t ) ) 4 9 4 7 ) 5 (5 . 13 60 0.0 0.06 Investmen (4.227 -0.00 d (-1.038 an , parentheses n ) ) ) ) ) i 2 9 2 1 7 e 9 ) 5 9 Share ar 8 (4 e 52 s 0.0 0.20 0.02 0.02 0.48 (2.644 (2.786 (1.556 (2.334 -0.09 (-3.640 Trad , r-statistic , ) ) ) ) 8 9 9 7 7 ) 5 9 8 (3 number 52 0.0 0.02 0.02 0.43 n (3.380 (2.996 (1.408 -0.09 (-3.615 Openness . 8 e regressio e ) ) ) th t 8 4 7 Tabl 7 ) 5 9 8 (2 52 0.0 0.02 0.02 (3.337 (2.916 -0.10 (-3.837 represen s ©International Monetary Fund. Not for Redistribution ) ) 1 1 ) 7 9 5 parenthese (1 n 10 76 0.0 0.02 0.02 i (3.794 (3.050 s s head h l n y a d P s t t ) countrie growt e schoo worl Colum f capit f dumm : n f r y s GD s o o l o r r pe h shar P e capita Notes r 2 GD observation enrollmen pe Numbe R Secondar Populatio Growt Opennes Trad Variable ln(initia Numbe Independen

59 Athanasios Vamvakidis

Table 9. RTAs and Investment

Independent Variables (1) (2) (3) (4) (5) (6) ln(initial GDP 0.029 0.034 0.035 0.030 0.017 0.026 per capita) (5.926) (4.762) (4.911) (3.998) (1.699) (3.147)

RTA dummy -0.007 -0.009 -0.010 -0.011 -0.013 -0.016 (-1.679) (-1.699) (-1.831) (-1.886) (-2.228) (-2.549)

Trade share 0.052 (3.308)

Openness dummy 0.016 (2.234)

Secondary school -0.108 -0.102 -0.104 -0.087 0.032 enrollment (-4.032) (-3.790) (-3.859) (-3.044) (3.281)

Population growth 0.494 0.533 0.937 0.557 (1.681) (1.819) (2.430) (1.801)

Growth of world 0.209 0.537 0.202 GDP per capita (2.549) (1.939) (2.353) Number of countries 147 109 109 109 105 89 Number of observations 943 621 621 621 493 525 R2 0.04 0.05 0.06 0.07 0.11 0.10 Notes: Column heads in parentheses represent the regression number. t-statistics are in parentheses.

VII. RTAs in the Past, Present, and Future: How Different Are They?

As mentioned earlier, some of the RTAs this paper considers were not fully implemented. However, they did result in very significant increases in trade among their members. Table 13 presents intragroup trade as a percentage of total trade for each RTA from 1960 to 1991. Even though the share of intragroup trade over total exports is very small for most RTAs, it did increase significantly during this period, in some cases by more than 100 percent. There may be a lot of driving forces for this increase, but it is reasonable to argue that the implementation of RTAs was one of them. The fact that the share of intragroup trade was so small in most RTAs suggests that their members should have expected that their effort to create large regional markets would fail and valuable time for broader liberalization would be lost. As was empha- sized above, this has been the case for the majority of RTAs, with few exceptions. As Table 13 shows, at least for the RTAs with available data, EU is the only one with high average intragroup trade shares.

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION ) ) ) ) ) ) 5 6 7 8 7 1 ) 7 9 1 (8 10 0.54 0.20 0.00 0.02 62 0.0 (1.865 (2.494 (1.043 (3.596 -0.10 -0.01 (-3.950 (-1.956 ) ) ) ) ) 3 1 6 7 1 6 9 ) 1 (7 10 62 0.0 0.51 0.00 0.03 (1.736 (1.162 (4.390 -0.10 -0.01 (-3.897 (-1.910 ) ) ) ) 6 1 2 0 5 9 ) 1 n (6 10 62 0.0 0.00 0.03 (1.077 (4.257 -0.11 -0.01 (-4.132 (-1.768 Ru g Lon . ) ) d ) 1 9 8 4 7 3 ) an (5 14 94 t 0.0 0.00 0.02 (5.625 (0.176 -0.00 (-1.659 parentheses Shor n e i e th ) ) ) ) ) ) ar n 6 0 2 7 5 8 s 7 ) i 9 1 t (4 10 62 0.0 0.51 0.22 0.03 (2.660 (1.758 (4.045 -0.10 -0.00 -0.00 (-1.294 (-3.792 (-0.774 f-statistic , ) ) Investmen ) ) ) 2 0 2 1 5 d number 6 9 ) 1 n (3 10 62 0.0 0.49 0.03 an (1.665 (4.866 -0.10 -0.00 -0.01 (-0.113 (-3.768 (-1.567 s regressio RTA e . th t ) ) 10 ) ) 8 9 4 1 e 5 ) 9 1 (2 10 62 0.0 0.03 (4.736 -0.10 -0.00 -0.00 Tabl (-3.993 (-1.391 (-0.250 represen s ©International Monetary Fund. Not for Redistribution ) ) ) 6 0 4 ) 4 parenthese 7 3 n (1 14 i 94 0.03 0.0 s (6.008 -0.00 -0.00 (-0.995 (-0.758 s head n h l a d e e y P y s th y t t ) f countrie growt th Colum schoo worl f capit f : f f n r y o GD s o o s o l o r r g pe h dumm dumm P dumm lag capita Notes A la A r o e A 2 observation GD enrollmen RT RT pe Growt Numbe Secondar R Numbe Tw Populatio RT On Independen Variable ln(initia

61 Athanasios Vamvakidis ) ) ) ) ) ) 5 5 4 7 7 7 9 5 ) 9 8 (8 52 0.0 0.47 0.19 0.02 0.02 (1.530 (2.287 (2.733 (2.816 -0.00 -0.09 (-3.413 (-0.829 ) ) ) ) ) 0 4 1 9 8 8 ) 5 9 8 (7 52 0.0 0.43 0.03 0.02 (3.439 (1.382 (3.027 -0.00 -0.09 (-3.368 (-0.944 ) ) ) ) n 0 9 9 8 7 ) 5 9 8 (6 Ru 52 0.0 0.03 0.02 (3.417 (2.950 g -0.09 -0.00 (-3.570 (-0.980 Lon d . ) ) ) an 6 8 5 t 8 9 ) 5 (5 10 76 0.0 0.02 0.02 (4.449 (3.566 -0.01 (-2.339 Shor e parentheses n th i e n i ar ) ) ) ) ) ) t 0 3 3 3 2 7 9 5 ) 9 8 (4 52 0.0 0.01 0.54 0.18 0.02 0.01 (2.108 (1.730 (1.952 (2.220 (1.439 -0.10 (-3.730 statistics t- , Investmen d ) ) ) ) ) 2 6 2 4 1 number 8 ) 5 an 9 n 8 s (3 52 0.0 0.51 0.02 0.02 0.01 (1.750 (1.636 (2.405 (2.303 -0.10 (-3.730 regressio e th Opennes t ) ) ) . ) 3 8 3 1 ) 8 5 9 11 8 (2 52 e 0.0 0.01 0.02 0.02 (2.448 (1.539 (2.279 -0.10 (-3.958 represen s Tabl ©International Monetary Fund. Not for Redistribution ) ) ) 0 1 1 ) parenthese 7 9 5 n (1 10 i 76 0.02 0.02 0.00 0.0 s (3.463 (0.140 (2.500 s head y y n h l y a d e e P s th t t ) f countrie growt th schoo Colum dumm dumm worl capit f f f : dumm f n s s r y o s GD s o o s o l o r r g pe h P lag capita Notes la r o e 2 observation GD enrollmen opennes opennes pe Secondar On Independen Variable ln(initia Numbe Growt Numbe Opennes R Tw Populatio

62 REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

Table 12. Openness and Investment In the Short and Long Run

Independent Variables (1) (2) (3) (4) ln(initial GDP -0.008 0.003 0.005 0.018 per capita) (-0.990) (0.366) (0.504) (1.773)

Trade share 0.094 0.081 0.085 0.093 (3.817) (3.133) (3.285) (3.623)

Trade share squared -0.011 -0.011 -0.013 -0.015 (-1.556) (-1.528) (-1.780) (-2.105)

Secondary school -0.117 -0.109 -0.093 enrollment (-4.142) (-3.837) (-3.247)

Population growth 0.923 0.975 (2.361) (2.515)

Growth of world 0.776 GDP per capita (2.908) Number of countries 137 105 105 105 Number of observations 605 493 493 493 R2 0.04 0.07 0.08 0.10 Notes: Column heads in parentheses represent the regression number. t-statistics are in parentheses.

When a small developing economy joins an RTA, the agreement by definition (since it is regional) will mainly include small developing economies. Very few developing economies are in the same region with developed economies. In addition, very few closed economies are neighbors of open economies. As a result, intratrade shares will also be small by definition (since intra-industry trade is very high among developed economies but not among developing economies). Most RTAs have been among small developing economies with relatively high protection, and unless trade agreements stop being regional, this trend will probably persist. The common argument that recent RTAs are different appears invalid, except in a few cases. All closed economies today are either developing or ex-socialist economies in transition. Given that most countries in the same region with them are economically homogenous, any RTA they can conduct, as an alternative path to broad liberalization, will be similar to most RTAs cited in this paper. Therefore, the empirical results of this paper not only describe the past but may also provide useful policy implications for the present and future. Possible reasons for most RTAs not resulting in faster growth may be that intragroup trade, even if it increased after the agreement, remained small; trade with more

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis : a 6 3 8 9 3 7 3 1 5 0 17.7 59.0 71.4 77.4 23.8 199 15 162. 118.6 1960 4083 o t Change : a 0 3 3 2 3 9 5 2 5 22.7 75.3 51.7 21.0 42.2 198 1960 156.2 193.3 22 o t Change . 1 2 3 4 4 5 2 3 1 3. 5. 6. 4. s 1970 199 13. 19. 16. 61. g a s beginnin Change d a e s 0 9 5 6 7 6 7 3 6 5 3. 5. 5. 4. 0. 199 10. perio 18. 15. 60. e RTA n th r i e fo n Percentag d take 0 Trad s 5 8 7 9 2 4 1 p i 3. 2. 0. 6. 2 4. e an 198 13. 17. , 2 chang e th Exports o l Intra-Grou s . , 5 8 9 5 6 1 7 7 9 1 3. 6. 0. 3. 0. 13 197 12. 13. 52. 21. Tota e 1960 f r . o t fo e Tabl 0 appendix Percen 2 1 9 8 7 1 4 3 2 e availabl t 1. 7. 2. 3. 0. 197 10. th 14. 26. 48. no n e i t ©International Monetary Fund. Not for Redistribution ou wer d a dat , . 0 spelle 6 2 6 2 7 5 5 e 1. 1. 2. 7. 7. 4. 196 34. ar s (1994) CEPGL D d an N acronym e A UNCTA : Th M : S ASEA r C N L a M Fo Source Note U M O A U E UDEA EC ECOWA CAC LAIA/LAFT AC CARICO MR CEPG ASEA RT

64 REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION efficient developed countries was diverted; technological spillover effects, if any, were very small; and high trade barriers still kept these economies closed to most of the rest of the world. In the meantime, the members of most RTAs lost the opportunity to benefit from broader liberalization. There are good theoretical reasons to believe that these results do not apply to some of the recent RTAs, which include large and developed economies. However, the results do suggest that broad liberalization will be more beneficial for growth than would most RTAs. The main question of this paper is how a closed economy should liberalizeÐ through an RTA or through nondiscriminatory opennessÐand the answer clearly supports the second path. Given the strong empirical evidence on the positive impact of nondiscriminatory liberalization on growth, it is still a puzzle why countries avoid full liberalization and target regional integration instead. To summarize, when a closed (which implies developing) economy faces the dilemma of nondiscriminatory openness versus participation in an RTA, the actual dilemma is usually between nondiscriminatory openness and an RTA with other also small, developing, and closed economies. Note that the two strategies are not necessarily conflicting. As we have seen, some countries liberalized and joined RTAs at the same time. However, this happened mainly in EU countries. Recent history has shown that countries often joined RTAs instead of following broader liberalization. Finally, it would be interesting to estimate trade creation and trade diversion for RTAs and then test if the growth impact of each RTA depends on these estimates. The negative estimated coefficient found in this paper for the impact of RTAs on growth may be explained by trade diversion effects. If more trade leads to faster growth, RTAs that result in more trade diversion than trade creation should lead to slower growth. Future research on this issue would be interesting.

VIII. Conclusions and Policy Implications This paper has presented time-series evidence that economies have grown faster on average, in both the short and the long run, after broad liberalization, but not after joining an RTA. The estimates for the RTAs are actually negative, and statistically significant in some specifications. These results are true for a variety of empirical specifications of the estimated model. This paper has also found that time-series variation shows a robust impact of openness on growth, in contrast to some of the cross country evidence. Openness has a direct effect on growth, but also an indirect effect, through higher investment. In contrast, RTAs are found to have a negative, though not always statistically significant, impact on both growth and investment. There are good reasons to believe that the effect of broad liberalization on growth will be higher in the decades to come than the results in this paper suggest has been the case in the past. Vamvakidis (1997) has shown that the benefits of trade liberalization depend positively on the openness of the world economy. Since the world economy is moving toward more open trade regimesÐformer socialist economies have joined the world market and many developing countries seem to have abandoned import-substitution policies in favor of more outward-oriented policiesÐone would

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis expect the benefits from free trade to have greater potential in the future. While this is what existing theory predicts, only future developments can confirm these predictions. Future research should focus on establishing theoretical foundations to support this paper©s empirical results. Many of the arguments in this paper need to be formal- ized by better theoretical models that address directly the impact of RTAs on growth, versus that of broad liberalization. Future empirical research should attempt to explain the determinants of trade policy. Better liberalization policies could be designed if it were known why countries protect trade in the first place and why they resist cutting trade barriers even though the growth benefits have been so well documented.

APPENDIX Regional Trade Agreements and Their Members

Arab Common Market (ACM): Egypt, Iraq, Jordan, Libya, Mauritania, Yemen. Andean Common Market (ANCON): Bolivia, Colombia, Ecuador, Peru, Venezuela. Association of South East Asian Nations (ASEAN): Indonesia, Malaysia, Philippines, Singapore, Thailand. Central American Common Market (CACM): El Salvador, Guatemala, Honduras, Nicaragua, Costa Rica. Caribbean Community and Common Market (CARICOM): Antigua and Barbuda, Barbados, Jamaica, St. Kitts and Nevis, Trinidad and Tobago, Belize, Dominica, Grenada, Montserrat, St. Lucia, St. Vincent and the Grenadines. Economic Community of the Great Lakes Countries (CEPGL): Burundi, Rwanda, Zaïre . European Union (EU): Belgium, France, Germany, Italy, Luxembourg, Netherlands, United Kingdom, Ireland, Denmark. Latin American Integration Association / Latin American Free Trade Association (LAIA/LAFTA): Mexico, Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Paraguay, Peru, Uruguay, Venezuela. Mano River Union (MRU): Guinea, Liberia, Sierra Leone. Union Douaniere des Etats de l©Afrique Centrale (UDEAC): Cameroon, Central Africa, Chad, Congo, Gabon. European Free Trade Association (EFTA): Norway, Switzerland, Iceland. BENELUX: Belgium, Netherlands, Luxembourg. Indian Ocean Commission (IOC): Comoros, Madagascar, Mauritius, Seychelles. South Africa Development Community (SADC): Angola, Botswana, Lesotho, Malawi, Mozambique, Namibia, Swaziland, Tanzania, Zambia, Zimbabwe. Economic Community of West African States (ECOWAS): Cape Verde, Gambia, Ghana, Guinea-Bissau, Nigeria, Togo. South Pacific Regional Trade and Economic Cooperation Agreement (SPARTECA): Australia, Cook Islands, Fiji, Kiribati, Nauru, New Zealand, Papua New Guinea, Solomon Islands, Tonga, Western Samoa. Economic Cooperation Organization (Eco): Iran, Pakistan, Turkey. Gulf Cooperation Council (GCC): Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, United Arab Emirates.

©International Monetary Fund. Not for Redistribution REGIONAL TRADE AGREEMENTS OR BROAD LIBERALIZATION

REFERENCES Baldwin, Richard E., and Elena Seghezza, 1996a, Trade-Induced Investment-Led Growth, NBER Working Paper No. 5582 (Cambridge, Massachusetts: National Bureau of Economic Research, May). , 1996b, Testing for Trade-Induced Investment-Led Growth, NBER Working Paper No. 5416 (Cambridge, Massachusetts: National Bureau of Economic Research, January). Barro, Robert J., and Jong-Wha Lee, 1994, Data set for a panel of 138 countries, Harvard University, Cambridge, Massachusetts. , and Xavier Sala-i-Martin, 1995, Economic Growth (New York: McGraw-Hill). Ben-David, Dan, 1993, "Equalizing Exchange: Trade Liberalization and Income Convergence," Quarterly Journal of Economics, Vol. 108 (August), pp. 653-79. , and David H. Pappel, 1995, "The Great Wars, the Great Crash, and Steady State Growth: Some New Evidence About an Old Stylized Fact," Journal of Monetary Economics, Vol. 36 (December), pp. 453-75. Campbell, John Y., and Pierre Perron, 1991, "Pitfalls and Opportunities: What Macroeconomists Should Know About Unit Roots," NBER Macroeconomics Annual, 1991 (Cambridge, Massachusetts: MIT Press). Coe, David T., and Elhanan Helpman, 1995, "International R&D Spillovers," European Economic Review, Vol. 39 (May), pp. 859-87. , and Alexander W. Hoffmaister, 1997, "North-South R&D Spillovers," Economic Journal, Vol. 107 (January), pp. 134-39. Dickey, David A., and Wayne A. Fuller, 1981, "Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root," Econometrica, Vol. 49 (June), pp. 1057-72. Dollar, David, 1992, "Outward-Oriented Developing Economies Really Do Grow More Rapidly: Evidence from 95 LDCs, 1976-1985," Economic Development and Cultural Change, Vol. 40 (April), pp. 523-44. Edwards, Sebastian, 1992, "Trade Orientation, Distortions and Growth in Developing Countries," Journal of Development Economics, Vol. 39 (July) pp. 31-57. Grossman, Gene M., and Elhanan Helpman, 1989, "Product Development and International Trade," Journal of Political Economy, Vol. 97 (December), pp. 1261-283 . , 1990, "Comparative Advantage and Long Run Growth," American Economic Review, Vol. 80 (September), pp. 796-815. 1991, Innovation and Growth in the Global Economy (Cambridge, Massachusetts: MIT Press). Krueger, Anne O., 1997, "Trade Policy and Economic Development: How We Learn," American Economic Review, Vol. 87 (March), pp. 1-22. Krugman, Paul R., 1990, Rethinking International Trade (Cambridge, Massachusetts: MIT Press). Lee, Jong-Wha, 1993, "International Trade, Distortions, and Long-Run Economic Growth," Staff Papers, International Monetary Fund, Vol. 40 (June), pp. 299-328. , 1994, "Capital Goods Imports and Long-Run Growth," NBER Working Paper No. 4725 (Cambridge, Massachusetts: National Bureau of Economic Research). Levine, Ross, and David Renelt, 1992, "A Sensitivity Analysis of Cross-Country Growth Regressions," American Economic Review, Vol. 82 (September), pp. 942-63. Rivera-Batiz, Luis A. and Paul M. Romer, 1991a, "International Trade with Endogenous Technological Change," European Economic Review, Vol. 35 (May), pp. 971-1004.

©International Monetary Fund. Not for Redistribution Athanasios Vamvakidis

, 1991b, "Economic Integration and Endogenous Growth," Quarterly Journal of Economics, Vol. 106 (May), pp. 531-55. Romer, Paul M., 1990, "Endogenous Technological Change," Journal of Political Economy, Vol. 98 (October), pp. S71-102. Sachs, Jeffrey D., and Andrew Warner, 1995, "Economic Reform and the Process of Global Integration," Brookings Papers on Economic Activity: 1, Brookings Institution, pp. 1-118. Summers, Robert, and Alan Heston, 1991, "The Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950-1988," Quarterly Journal of Economics, Vol. 106 (May), pp. 327-68. United Nations Conference on Trade and Development, 1994, Handbook of International Trade and Development Statistics (New York: United Nations). Vamvakidis, Athanasios, 1997, International Integration and Economic Growth (Ph.D. thesis; Cambridge, Massachusetts: Harvard University, Economics Department). , 1998, "Regional Integration and Economic Growth," World Bank Economic Review, Vol. 12 (May), pp. 251-70. Wacziarg, Romain, 1998, Measuring the Dynamic Gains from Trade, Policy Research Working Paper No. 2001 (Washington: World Bank). World Bank, 1994, World Tables (Baltimore, Maryland: published for the World Bank, Johns Hopkins University Press).

Exchange and Capital Controls as Barriers to Trade

NATALIA T.TAMIRISA*

This paper considers the effect of exchange and capital controls on trade in the gravity- equation framework, in which bilateral exports depend on the distance between countries, the countries' size and wealth, tariff barriers, and exchange and capital controls. The extent of exchange and capital controls is measured by unique indices. In view of the degree to which countries have liberalized their exchange systems, controls on current payments and transfers are found to be a minor impediment to trade, while capital controls significantly reduce exports into developing and transition economies. Thus, further capital account liberalization could significantly foster trade. [JEL F13, F31]

n 1944, the Bretton Woods conference recognized the fundamental link between Iexchange and capital controls1 and international trade. One of the purposes of the International Monetary Fund, which was created at the conference, was to assist in "the elimination of foreign exchange restrictions which hamper the growth of world trade."2 However, the maintenance of capital controls was not viewed as inconsistent with this objective, partly because capital controls were considered necessary for sup- porting the system of fixed exchange rates and thus fostering trade. More than 50 years

*Natalia Tamirisa is an Economist in the IMF©s Policy Development and Review Department. She was an Economist in the IMF©s Monetary and Exchange Affairs Department when this paper was written. The author is grateful to Barry Johnston for encouragement and insights. She appreciates useful discus- sions with Giovanni Dell©Ariccia, Brad McDonald, Mark Swinburne, and Athanasios Vamvakidis and valuable comments from Gian Maria Milesi-Ferretti and two anonymous referees. The author also thanks Virgilio Sandoval for assistance with data collection and Natalie Baumer for helpful editorial suggestions. 1Hereinafter, the term "controls on current payments and transfers" refers to exchange controls over current international transactions, while "capital controls" encompasses controls pertaining to capital account transactions. The term "exchange and capital controls" covers both of the above types of controls. 2Article I of the IMF©s Articles of Agreement (IMF, 1993).

©International Monetary Fund. Not for Redistribution Natalia T. Tamirisa

later, the question about the economic effects of exchange and capital controls is again at the forefront of economic policy debates. Most countries have liberalized controls on current payments and transfers, and the focus of economic policy is increasingly shifting toward liberalizing capital account transactions. The effect of exchange and capital controls on international trade depends on the structure and effectiveness of controls and their interaction with other distortions in the economy. Exchange controls act as a tax on the foreign currency required for purchasing foreign goods and services and, by raising the domestic price of imports, they tend to reduce trade. Besides this basic effect, exchange and capital controls can influence trade through other channels, for example, transaction costs, exchange rates, foreign exchange risk hedging, and trade financing. Capital controls, in particular, can affect trade in goods by reducing intertemporal trade and portfolio diversification, which may substitute or complement intratemporal trade. Given the importance and ambiguity of the link between exchange and capital controls and trade, the systematic empirical evidence on the matter is critical, but it remains limited. This paper examines the effect of exchange and capital controls on trade for 1996 in the empirical gravity-equation framework, in which bilateral exports depend on the distance separating the countries, the countries© size and wealth, tariff barriers, and exchange and capital controls. The extent of exchange and capital controls is measured by unique indices, which aggregate information on 142 individual types of control based on the IMF©s Annual Report on Exchange Arrangements and Exchange Restrictions. Overall, the paper finds that exchange and capital controls have a significant negative impact on bilateral exports. However, this result varies depending on the level of development in the country and the type of exchange and capital control. Controls on current payments and transfers are a minor barrier to trade. In contrast, capital controls significantly reduce exports into developing and transition economies and not into industrial countries. These results may reflect the extent to which restrictions on current payments and transfers have been liberalized generally, while the liberalization of controls on capital flows have so far been focused largely on industrial countries.

I. Theoretical Evidence Theoretically, the impact of exchange and capital controls on trade is ambiguous. With respect to net flows, the effects of controls on trade and capital flows are closely related in the context of a standard balance of payments accounting.3 Since measurement problems are more severe for capital flows than trade flows, the analysis of the relationship between exchange and capital controls and net trade flows may help enhance understanding of the effect on net capital flows. This is not, however, an

3In the balance of payments, the sum of the current account, the capital account, and the change in reserves is by definition equal to zero. If exchange and capital controls are effective and have a statistically significant impact on the capital account, the balance of payments identity implies that they must also affect the current account and/or reserves in a regime of managed or fixed exchange rates, and the current account under the floating exchange rate regime. The author is grateful to an anonymous referee for underscoring this point. For a detailed review of the literature on capital controls, see Dooley (1996).

©International Monetary Fund. Not for Redistribution EXCHANGE AND CAPITAL CONTROLS AS BARRIERS TO TRADE objective of this paper. The paper mainly focuses on the impact of exchange and capital controls on gross trade flows.4 Exchange and capital controls can affect trade through a multitude of (inter- related) channels, including the domestic price of imports, transaction costs, the volatility of exchange rates, intertemporal trade, and portfolio diversification. The overall effect of exchange and capital controls on trade through these channels depends critically on the structure and effectiveness of exchange and capital controls and their interaction with other distortions in the economy. The main effects of exchange and capital controls on trade are discussed in more detail below. The basic economics of exchange controls is similar to that of quantitative restrictions on imports of various goods and services. By taxing foreign money required to purchase foreign goods and services, exchange controls5 cut the quantity imported and/or raise the domestic relative price of imports.6 Moreover, if the government allo- cates foreign exchange according to noncompetitive rules, low-valued uses often get approved instead of higher-valued ones, decreasing trade further. Exchange and capital controls often raise transaction and other trade-related costs, thus reducing trade. Costs and uncertainty associated with international transactions increase, because exchange controls tend to stifle the development of liquid and efficient foreign exchange markets and modern payment instruments. Additionally, exchange and capital controls often encourage evasion and rent-seeking, which impose unproductive costs on firms. Furthermore, exchange and capital controls can reduce trade by limiting the transfer of technology, managerial expertise, and skills through foreign direct investment. Controls on repatriation of profits and dividends, surrender requirements, and direct controls on foreign investment in certain sectors are likely to discourage direct foreign investment and thus limit the dissemination of technological and managerial knowl- edge and learning by doing. The empirical evidence indicates that foreign direct investment tends to increase host countries© exports and imports (although the impact on imports is relatively weak).7 In the presence of tariff barriers, however, controls on foreign direct investment may encourage trade. Foreign direct investment and exports are alternative strategies in this case, and, if foreign direct investment is allowed, a multinational company may prefer to avoid paying tariffs by supplying the host country©s market through a subsidiary company. Capital controls often limit business opportunities for hedging foreign exchange risks and financing trade, thus inhibiting trade. In the presence of capital controls, financial intermediation is less efficient, and savings are not allocated to the most

4For a discussion of the relationship between the intensity of capital controls and net capital flows, see Johnston and others (forthcoming). The study developed a methodology for constructing simple, yet comprehensive, indices of exchange and capital controls and found that the intensity of the controls is negatively correlated with net direct, portfolio, and other capital flows. The present paper uses these indices to examine the impact of exchange and capital controls on trade flows. 5It can be shown that dual exchange rates are equivalent to capital controls, while exchange controls are similar to trade restrictions, according to Adams and Greenwood (1985) and Greenwood and Kimbrough (1987), respectively. 6See, for example, Greenwood and Kimbrough (1987) and Stockman and Hernandez (1988). 7For the review of the literature on foreign direct investment, see, for example, World Trade Organization (1996).

©International Monetary Fund. Not for Redistribution Natalia T. Tamirisa efficient uses. The intermediation margin is often high, and local financial institutions enjoy substantial market power. The range of available financial products and services tends to be narrow. As a consequence, opportunities for hedging foreign exchange risks and financing trade are either unavailable or costly, and trade is likely to fall. Notwithstanding the above, however, capital controls may foster trade indirectly by serving prudential objectives and helping to protect weak financial systems. Fundamentally, capital controls affect trade by decreasing intertemporal trade and portfolio diversification. The impact on trade in goods depends on whether this intratemporal trade substitutes for or complements intertemporal trade and portfolio diversification. If trade in goods and trade in factors are substitutes (for example, as found in the basic Heckscher-Ohlin model), the volume of trade in goods is likely to fall. The terms of trade effect is unclear and depends on changes in the patterns of consumption and production in the recipient and the source countries (also known as the transfer problem) for clarification. If trade in goods and trade in factors are complements (as, for example, in some models with increasing returns to scale), the volume of trade in goods increases. In addition, a number of macroeconomic channels through which capital controls can potentially help foster trade have been suggested in theory.8 The specific effect of capital controls on trade through these macroeconomic channels depends critically on the interaction of capital controls with other distortions and on specific characteristics of the economy. In principle, capital controls may help limit short-term speculative capital flows and hence exchange rate volatility. With a stable exchange rate, trade is likely to increase (particularly if domestic financial markets are not well developed and do not offer adequate opportunities for hedging foreign exchange risk). Exchange and capital controls, on the other hand, are often associated with an overvalued exchange rate, which can inhibit trade. Moreover, if capital controls can help retain domestic savings, and higher savings lead to higher investment in export sectors, trade may increase. When the taxation of foreign source income is nonenforceable, capital controls could help expand the domestic tax base. The adequate tax revenues raised by domestic taxes may induce the government to lower tariff rates, stimulating trade. These effects, however, are likely to be inconsequential in practice, because they tend to be offset by capital flight and the decrease in capital inflow owing to capital controls. Not surprisingly, these arguments have received only limited empirical support so far. Likewise, the empirical evidence on the effects of exchange and capital controls on trade remains scarce. Most of the earlier studies (see, for example, Lee, 1993) used the black market premium to measure the extent of exchange and capital controls and found that exchange and capital controls tend to reduce trade. Although the black market premium often indicates the circumvention of restrictive regulations, it is an imperfect measure of the extent of exchange and capital controls. It may capture the effects of other nontariff barriers to trade, for example, import quotas. Also, information on

8See Dooley (1996) for a review of the literature on capital controls. The empirical literature suggests that capital controls may affect yield differentials, but their role in improving the balance of payments is limited (see, for example, Johnston and Ryan, 1994).

©International Monetary Fund. Not for Redistribution EXCHANGE AND CAPITAL CONTROLS AS BARRIERS TO TRADE

the size of the black market premiu m is not always reliable. Moreover , the black market premiu m canno t isolate the effects of control s on curren t payment s and transfers and capital controls . This paper, in contrast , uses uniqu e indices of the extent of exchange and capital control s to examine their effect on internationa l trade in the empirica l gravity-equatio n framework.

II. An Empirical Model of Trade with Exchange and Capital Controls

The gravity mode l has been used extensively in empirica l studies of internationa l economic s since the 1960s. Accordin g to this static general equilibrium model, bilateral trade is determine d by the wealth and size of countries , the distanc e between them , and other factors that distort trade. The theoretica l foundation s of the gravity mode l are based on the theor y of trade under imperfec t competitio n and have recentl y been integrate d with the factor-proportion s and demand-base d theorie s of internationa l trade. 9 The basic gravity equatio n is given by

α1 α2 α3 α4 α5 α6 Xkj = αo(Qk/N k) (Nk) (Qj/N j) (Nj) (Dkj) (Akj) ekj, (1)

where Xkj is exports from countr y k to countr y j, (Qk/N k) and (Qj/N j) are the per capita income s of countrie s k and j; N k and N j are the population s of countrie s k and j; Dkj is the geographica l distanc e between countrie s k and j; Akj denote s factors distortin g trade ; and ekj is a log normall y distribute d error term. For the empirica l analysis, the above equatio n is modified by taking natura l logs and defining tariffs and exchange and capital control s as trade distortions , as follows:

1nXkj=α0 + α11n(Qk/N k)+α21nN k + α31n(Qj/N j) (2)

+ α4lnN j + α5lnDkj + α6ln(l + T jk)+ α7Ej +ε kj,

where Tjk is the impor t duty imposed by countr y j on import s from countr y k, and Ej is an aggregate measure of exchange and capital control s in countr y j. The intercep t account s for the effect of unmeasure d trade distortion s on exports. The mode l can be estimate d by the ordinary-least-square s method .

III. Data The estimatio n of the mode l requires cross-sectiona l data on bilateral exports of goods and services, population , gross domesti c produc t (GDP ) per capita, and measure s of tariff barriers and exchange and capital control s by countr y for a given year. The mode l is estimate d for a sample of 40 industrial , developing, and transitio n countries . The data described below refer to 1996, unless specified otherwise.

9For more details on the general-equilibriu m foundation s of the gravity model, see Anderson (1979), Helpma n and Krugman (1985), Helpma n (1987), and Bergstrand (1985, 1989, and 1990).

Data on exports of goods and services (denoted by "EX") are from the IMF©s Direction of Trade Statistics Yearbook. GDP per capita (denoted by "GDPIM" and "GDPEX" for importing and exporting countries, respectively) are adjusted according to the purchasing power parity and come from the World Bank©s World Tables. Population data (denoted by "POPIM" and "POPEX" for importing and exporting countries, respectively) are for 1996 or the latest available year, as published in the IMF©s International Financial Statistics. The geographic distance (denoted by "DIST") is measured as the direct-line distance between the capital cities of countries.10 Trade restrictions are represented by mean tariff rates (denoted by "TAR") by country. The tariff data for 1995 or the latest available year come from the World Bank©s World Development Indicators Database. Tariff rates are adjusted to take into account free trade agreements, as reported in the World Trade Organization©s Annual Report. This measure of trade restrictions is imperfect because it does not reflect the extent of nontariff barriersÐfor example, import quotas and voluntary export restraintsÐwhich tend to cover a substantial share of imports, particularly in developing countries. The measurement of the intensity of nontariff barriers is challenging, and the available measures are inadequate. Therefore, in this study, the effect of nontariff barriers (other than exchange and capital controls) is not measured separately but is accounted for in the intercept. The extent of national exchange and capital controls is captured in three aggregate measures: the indices of controls on current payments and transfers (CCI), capital controls (KCI), and exchange and capital controls in their entirety (ECI). The indices summarize information on 142 individual types of national exchange and capital control from the IMF©s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER).11 Table 1 depicts individual types of exchange and capital control and their aggregation into categories and indices. The appendix describes the methodology used to construct the indices.12 Each index ranges from zero (the lowest extent) to one (the highest extent). CCI measures the extent of controls on current payments and transfers, and KCI reflects the pervasiveness of controls on capital movements. ECI comprises capital controls as well as controls on payments and transfers for current international transactions and hence reflects the overall extent of exchange and capital controls. It can be also interpreted as a broad measure of capital controls that takes into account the scope for the circumvention of capital controls through current international transactions. Table 2 shows the indices of exchange and capital controls for the countries in the sample. Despite their limitations, the indices have some advantages over alternative measures of the extent of exchange and capital controls, for example, the black market premium and dummy variables. Unlike the black market premium, the indices do not

10Fitzpatrick and Modlin (1986). 11In 1997, the information in the AREAER was presented for the first time in a new tabular format, which classified and standardized the information on members© exchange systems and expanded the coverage of capital controls. The classification of the AREAER information with this new tabular format has made it possible to develop and apply more comprehensive indices of the extent of exchange and capital controls for 1996. 12For more details on the indices of exchange and capital controls, see Johnston and others (forthcoming).

Table 1. Structure of Indices of Exchange and Capital Controls

A. Index of Controls on Current Payments and Transfers

Exchange arrangement Exchange rate structure (dual, multiple) Exchange tax Exchange subsidy Forward exchange market (prohibited, official cover of forward operations required)

Arrangements for payments and receipts Prescription of currency requirements Bilateral payments arrangements (operative, inoperative) Other payments arrangements (regional, clearing, barter, and open accounts) International security restrictions (in accordance with IMF Executive Board Decision No. 144-(52/51), in accordance with UN sanctions, other) Payments arrears (official, private) Controls on trade in gold (coins and/or bullion) (on domestic ownership and/or trade, on external trade) Controls on exports and imports of banknotes (on exports and imports of domestic and foreign currency)

Resident accounts Foreign exchange accounts held domestically (prohibited, approval required) Foreign exchange accounts held abroad (prohibited, approval required)

Nonresident accounts Foreign exchange accounts (prohibited, approval required) Domestic currency accounts (prohibited, approval required) Blocked accounts

Imports and import payments Foreign exchange budget Financing requirements for imports (minimum financing, advance payments, advance import deposit) Documentation requirements for release of foreign exchange for imports (domiciliation requirements, preshipment inspection, letters of credit, import licenses used as exchange licenses, other) Import taxes collected through the exchange system

Exports and export proceeds Documentation requirements (letters of credit, guarantees, domiciliation, preshipment inspection, other) Export taxes collected through the exchange system

Table 1. (continued)

Payments for invisible transactions and current transfers Freight and insurance (prior approval, quantitative limits, indicative limits/bona fide test) Unloading and storage costs (prior approval, quantitative limits, indicative limits/bona fide test) Administrative expenses (prior approval, quantitative limits, indicative limits/bona fide test) Commissions (prior approval, quantitative limits, indicative limits/bona fide test) Interest payments (prior approval, quantitative limits, indicative limits/bona fide test) Profit and dividends (prior approval, quantitative limits, indicative limits/bona fide test) Payments for travel (prior approval, quantitative limits, indicative limits/bona fide test) Medical costs (prior approval, quantitative limits, indicative limits/bona fide test) Study abroad costs (prior approval, quantitative limits, indicative limits/bona fide test) Subscriptions and membership fees (prior approval, quantitative limits, indicative limits/bona fide test) Consulting and legal fees (prior approval, quantitative limits, indicative limits/bona fide test) Foreign workers© wages (prior approval, quantitative limits, indicative limits/bona fide test) Pensions (prior approval, quantitative limits, indicative limits/bona fide test) Gambling and prize earnings (prior approval, quantitative limits, indicative limits/ bona fide test) Family maintenance and alimony (prior approval, quantitative limits, indicative limits/bona fide test) Credit card use abroad (prior approval, quantitative limits, indicative limits/bona fide test)

B. Index of Capital Controls

Proceeds from exports, invisibles, and current transfers Repatriation requirements for export proceeds Surrender requirements for export proceeds Repatriation requirements for proceeds from invisibles and current transfers Surrender requirements for proceeds from invisibles and current transfers Restrictions on use of funds

Controls on capital and money market instruments On capital market securities (purchase in the country by nonresidents, sale or issue locally by nonresidents, purchase abroad by residents, sale or issue abroad by residents) On money market instruments (purchase in the country by nonresidents, sale or issue locally by nonresidents, purchase abroad by residents, sale or issue abroad by residents) On collective investment securities (purchase in the country by nonresidents, sale or issue locally by nonresidents, purchase abroad by residents, sale or issue abroad by residents)

Table 1. (concluded)

Controls on derivatives and other instruments Purchase in the country by nonresidents Sale or issue locally by nonresidents Purchase abroad by residents Sale or issue abroad by residents

Controls on credit operations Commercial credits (by residents to nonresidents, to residents from nonresidents) Financial credits (by residents to nonresidents, to residents from nonresidents) Guarantees, sureties, and financial backup facilities (by residents to nonresidents, to residents from nonresidents)

Controls on direct foreign investment Outward direct investment Inward direct investment

Controls on liquidation of direct investment

Controls on real estate transactions Purchase abroad by residents Purchase locally by nonresidents Sale locally by nonresidents

Provisions specific to commercial banks and other credit institutions Borrowing abroad Maintenance of accounts abroad Lending to nonresidents (financial or commercial credits) Lending locally in foreign exchange Purchase of locally issued securities denominated in foreign exchange Differential treatment of nonresident deposit accounts and/or deposit accounts in foreign exchange (reserve requirements, liquid asset requirements, interest rate controls, investment regulations, credit controls, open foreign exchange position limits)

Provisions specific to institutional investors Limits (max.) on portfolio invested abroad Limits (min.) on portfolio invested locally Currency-matching regulations on assets/liabilities composition

Table 2. Indice s of Exchange and Capita l Controls, 199 6

Countr y CCI1 α KCI2 b ECI3 C Argentina 0.03 0.19 0.11 Australia 0.04 0.20 0.12 Brazil 0.31 0.60 0.46 Canad a 0.09 0.06 0.07 Chile 0.22 0.89 0.56 Chin a 0.33 0.73 0.53 Czech Republic 0.04 0.33 0.19 Denmar k 0.02 0.07 0.05 Egypt 0.12 0.30 0.21 Franc e 0.04 0.16 0.10 German y 0.04 0.07 0.05 Greec e 0.06 0.06 0.06 Hungar y 0.10 0.57 0.33 Indi a 0.22 0.87 0.55 Indonesi a 0.18 0.50 0.34 Israel 0.16 0.54 0.35 Italy 0.10 0.06 0.08 Japan 0.09 0.16 0.12 Kazakhsta n 0.30 0.95 0.62 Kenya 0.05 0.17 0.11 Korea , Republic of 0.10 0.70 0.40 Latvia 0.10 0.10 0.10 Mexico 0.05 0.36 0.21 Morocc o 0.27 0.72 0.49 Netherland s 0.05 0.01 0.03 New Zealan d 0.02 0.09 0.05 Norwa y 0.01 0.05 0.03 Pakistan 0.31 0.66 0.48 Philippine s 0.16 0.47 0.32 Polan d 0.12 0.69 0.40 Russia 0.27 0.91 0.59 Saudi Arabia 0.03 0.21 0.12 South Africa 0.29 0.56 0.43 Spain 0.04 0.11 0.08 Thailan d 0.17 0.63 0.40 Tunisia 0.21 0.81 0.51 Turkey 0.16 0.36 0.26 Unite d Kingdom 0.03 0.07 0.05 Unite d States 0.05 0.13 0.09 Urugua y 0.09 0.13 0.11 αInde x of control s on curren t payment s and transfers. bInde x of capital controls . cInde x of exchange and capital controls .

©International Monetary Fund. Not for Redistribution EXCHANGE AND CAPITAL CONTROLS AS BARRIERS TO TRADE reflect the effects of other nontariff barriers, such as import quotas, and focus exclu- sively on exchange and capital controls. Unlike dummy variables, the indices summarize information about a broad array of controls, and thus can capture a variety of changes in the regulatory regime. The indices, however, do not explicitly take into account the supervision and enforcement of exchange and capital controls and hence reflect legal (de jure) rather than actual (de facto) incidence of controls.13 The study analyzes a cross-section of 40 industrial, developing, and transition economies for which the indices of exchange and capital controls are available. The countries represent various geographical regions and levels of economic development. All but two of these countries (Brazil and Egypt) have accepted the obligations of Article VIII of the IMF©s Articles of Agreement.14 The sample includes 15 industrial countries (Australia, Canada, Denmark, France, Germany, Greece, Israel, Italy, Japan, the Netherlands, New Zealand, Norway, Spain, the United Kingdom, and United States), 19 developing countries (Argentina, Brazil, Chile, China, Egypt, India, Indonesia, Kenya, Republic of Korea, Mexico, Morocco, Pakistan, Philippines, Saudi Arabia, South Africa, Thailand, Tunisia, Turkey, and Uruguay), and 6 transition economies (Czech Republic, Hungary, Kazakhstan, Latvia, Poland, and Russia).15 Summary statistics and correlations are presented in Tables 3 and 4, respectively. The exchange system in industrial countries is highly liberal, while developing and transition economies have more extensive exchange and capital controls. For instance, the mean ECI for industrial and for developing and transition economies is 0.09 and 0.35, CCI is 0.05 and 0.17, and KCI is 0.12 and 0.54, respectively. Controls on current payments and transfers (as measured by CCI) are less pervasive than capital controls (KCI) in industrial and developing and transition economies. Another interesting observation is that controls on current payments and transfers and capital controls are highly correlated with each other (correlation coefficient is above 0.8), and, of course, with the overall measure of exchange and capital controls, ECI (correlation coefficients are above 0.9).16

IV. Empirical Evidence We estimate equation (2) with three alternative measures of exchange and capital con- trolsÐCCI, KCI, and ECI17Ðdenoting the respective equations as (2a), (2b), and (2c). The results suggest that exchange and capital controls are a notable barrier to trade in

13Although the intensity of exchange and capital controls is not taken into account explicitly, the indices are found to be robust to weighing by subjective intensity measures. 14Under Article VIII of the IMF©s Articles of Agreement, members undertake obligations to avoid imposing restrictions on the making of payments and transfers for current international transactions, without the approval of the IMF. 15The study uses the IMF©s classification of industrial, developing, and transition countries. 16For the analysis of correlation between the indices and measures of economic development, the efficiency of the financial system, foreign direct and portfolio investment, exchange rate volatility, and trade policy, see Johnston and others (forthcoming). 17Including both CCI and KCI in the model intensifies multicollinearity, since the indices are highly correlated with each other (correlation coefficients of 0.8-0.9). Testing for redundant coefficients shows that CCI is redundant. Testing for the stability of coefficients suggests that they are unstable at the 5 percent level of significance.

©International Monetary Fund. Not for Redistribution Natalia T. Tamirisa 8 1 0 2 3 2 3 5 9 9 7 5 5 7 9 58 93 0.3 0.1 0.1 0.6 0.1 0.0 0.6 0.0 0.3 0.0 0.0 0.2 ECI 1,51 8 1 0 0 4 6 5 4 2 2 8 1 5 1 9 58 93 0.2 0.5 0.1 0.9 0.0 0.1 0.1 0.0 0.3 0.9 0.5 0.3 KCI 1,51 8 1 3 0 3 3 0 4 7 6 3 1 1 5 9 93 58 0.1 0.1 0.0 0.3 0.1 0.0 0.3 0.0 0.0 0.0 0.1 0.1 CCI 1,51 0 0 6 8 0 3 0 0 0 9 1 3 7 5 9 93 58 3.5 14.5 1,51 13.4 +TAR 119.3 156.3 100.0 100.0 156.3 100.0 110.5 113.9 105.3 1 0 0 4 8 8 1 0 2 0 0 8 0 0 7 9 93 58 1,51 s 1,380.0 1,380.0 2,598.7 5,319.7 7,490.8 3,488.1 GDPIM 18,940.0 11,710.0 10,730.4 19,465.6 26,980.0 26,980.0 0 0 0 8 8 8 1 0 6 7 8 0 0 5 9 Statistic 93 58 y 1,51 1,380.0 1,380.0 1,380.0 GDPEX 7,531.8 7,485.2 7,408.3 10,836.2 10,412.2 10,674.0 26,980.0 26,980.0 26,980.0 Summar 0 0 8 4 1 9 1 1 6 6 9 7 1 7 9 . 58 93 3 2.4 3.5 2.4 1,51 52.4 66.6 e 141.4 107.4 289.9 266.5 235.4 POPIM 1,221.5 1,221.5 Tabl 0 0 0 1 8 9 3 3 5 9 1 5 9 9 9 93 58 2.4 2.4 2.4 1,51 107.2 107.3 107.3 235.1 235.4 236.1 POPEX 1,221.5 1,221.5 1,221.5 3 0 6 0 0 9 6 5 8 8 1 0 0 0 9 58 93 1,51 137.0 187.0 137.0 DIST 3,449.1 3,873.0 4,928.9 4,475.5 4,879.7 4,800.2 ©International Monetary Fund. Not for Redistribution 62,333.0 79,635.0 79,635.0 3 0 1 8 6 0 4 4 7 0 1 1 1 1 9 58 93 0.0 0.0 0.0 1,51 EX 719.2 1,897.7 2,758.4 8,268.5 3,800.3 12,679.6 56,760.8 .64,761.4 s n n n s d 1 164,761.4 s an deviatio deviatio m m m deviatio countrie m m m g d d d countrie n l t t t n n n countrie Standar Mea Minimu Maximu Coun Mea Maximu Coun Minimu Maximu Coun Standar Minimu Mea Standar l transitio Developin Industria Al

80 EXCHANGE AND CAPITAL CONTROLS AS BARRIERS TO TRADE 0 ECI 1.00 0 0 KCI 1.00 0.99 9 1 0 CCI 1.00 0.82 0.90 3 5 1 0 1,00 +TAR 0.61 0.58 0.61 1 8 1 5 1 0 1.00 GDPIM -0.63 -0.66 -0.67 -0.60 s 3 2 2 1 7 0 1.00 0.02 0.02 0.02 GDPEX -0.03 -0.05 Correlation . 4 e 8 0 4 4 9 1 0 1.00 0.55 0.42 0.34 0.37 0.00 Tabl POPIM -0.23 8 6 8 8 5 5 1 0 1.00 0.00 0.02 POPEX -0.00 -0.00 -0.02 -0.23 -0.00 9 9 7 6 7 9 2 4 0 1.00 DIST 0.00 0.01 0.09 0.00 -0.00 -0.00 -0.00 -0.00 ©International Monetary Fund. Not for Redistribution 0 6 3 7 4 5 3 6 1 0 EX 1.00 0.23 0.03 0.03 0.23 -0.12 -0.10 -0.12 -0.13 -0.11 TAR + 1 CCI GDPEX GDPIM KCI ECI POPIM EX DIST POPEX

81 Natalia T. Tamirisa

developing and transition economies but not in industrial economies. Controls on current payments and transfers reduce bilateral trade flows insignificantly. Estimation results are summarized in Table 5. The adjusted R2s are above 0.70, and F-statistics are significant at the 99 percent level.18 Tests of the stability of coefficients and the recursive analysis of coefficients indicate that coefficients are stable at the 95 percent significance level. The estimated intercept is negative, implying that unmeasured trade distortions tend to reduce exports. Distance has a significant negative effect on bilateral exports, in part because trade costs (e.g., transportation and communication) are likely to increase with distance. Tariff barriers in the importing countries also tend to have a negative, albeit insignificant, effect on exports into these countries. Per capita GDP and population, on the other hand, have significant positive effects on bilateral exports. Overall, exchange and capital controls (as measured by ECI) represent a notable nontariff barrier. The negative parameter on ECI is significant at the 95 percent level for the full sample, suggesting that exchange and capital controls in their entirety significantly reduce bilateral exports. Another interpretation of this result is that capital controls in a broad sense (i.e., including capital controls and controls on current payments and transfers that are used to prevent the circumvention of capital controls) are a significant barrier to trade. The effect of exchange and capital controls on trade, however, varies depending on the type of control. Capital controls (as measured by KCI) are found to be a significant barrier to trade for the full sample. In contrast, controls on current payments and transfers (as measured by CCI) do not reduce exports significantly. Most countries in the sample have already liberalized exchange controls on current payments and transfers, and the remaining exchange controls, including those on current invisible payments such as tourism, do not affect trade noticeably. Very few countries presently maintain significant exchange controls on trade-related transactions or factor services. In contrast, capital controls remain more widespread, particularly in developing and transition economies. The variation in the extent of the liberalization of exchange and capital controls across industrial and developing and transition countries is reflected in the estimation results for the respective subsamples. Exchange and capital controls are a barrier to exports into developing and transition economies, but not to exports into industrial countries. This finding can be attributed to capital controls, which noticeably reduce bilateral exports into developing and transition economies, and have only a minor negative impact on bilateral exports into industrial countries. The reason is that industrial economies have relatively liberal regimes for international capital movements, while many developing and transition economies continue to maintain various capital controls. Controls on current payments and transfers represent only a minor barrier to bilateral exports into all countries, since these controls have been substantially liberalized worldwide. The results appear to be robust with respect to the type of the exchange rate regime and individual country effects. To check whether the effect of exchange and capital controls on export flows varies depending on the exchange rate regime, we add

18Since heteroscedasticity may be a problem due to differences in the country size, standard errors and covariances are calculated on the basis of the White heteroscedasticity-consistent matrix.

©International Monetary Fund. Not for Redistribution EXCHANGE AND CAPITAL CONTROLS AS BARRIERS TO TRADE - * * * * * * * * 8 2 7 ) coef 93 1.06 1.52 1.99 0.7 0.96 a -0.54* -1.07 -0.0 e 334.63 -38.99 s d denot s an * * * * * * * (2c g 8 2 2 5 ) Countrie 93 1.41 1.98 1.06 0.95 0.7 n -1.04 -0.6 -0.6 asterisk 332.21 -38.08 o tw ; Developin s * * * * * * * * level (2b 2 8 t 8 ) 93 1.49 1.99 1.06 0.7 0.96 -0.76* -1.06 -0.1 Result 334.10 -38.76 c percen 9 9 e Basi : th * * * * * * * * t (2a 1 0 1 ) a t 58 1.77 0.8 0.93 0.99 0.99 -7.37 -0.7 -0.58 319.21 -33.74 Transitio s Controls l significan s i * * * * * * * * t (2c 1 0 8 ) Countrie 58 1.77 l 0.8 0.95 0.99 0.93 tha t -6.77 -2.1 -0.60 Capita 319.07 -33.29 d an e coefficien * * * * * * * * (2b 0 0 1 a ) s 58 1.77 0.8 0.94 0.99 0.97 -0.58 -7.21 -1.2 319.42 -33.57 denote k Exchang h * * * * * * * * (2a 6 3 9 ) wit asteris l e 1.03 1.37 1.90 0.7 0.94 1,51 -0.42 -0.91 -0.7 697.26 -37.11 On . Mode Industria s y * * * * * * . * number (2c 6 3 9 9 n ) 1.39 1.03 1.90 0.7 0.94 1,51 level -0.8 -0.91 -0.8 Gravit t . Countrie 694.76 -37.34 l 5 equatio Al ©International Monetary Fund. Not for Redistribution e e percen 5 * * * * * * * * (2b Tabl 9 6 3 9 ) e indicat s 1.03 1.37 1.90 0.7 0.94 th (2a t 1,51 -0.7 -0.66 -0.91 697.05 -37.13 a t heading n s significan s Colum i f : t c o r TAR) tha t + Notes 1 2 GDPIM POPIM POPEX GDPEX DIST ( observation n n n n n n CCI c I Numbe R F-statisti ficien I I I ECI KCI I I

83 Natalia T. Tamirisa

a measure of the exchange rate regimeÐa dummy variable indicating a fixed exchange rate system (EXRD) or the average monthly volatility of the U.S. dollar exchange rate (EXRV)Ðto the set of regressors.19 Coefficients for both measures of the exchange regime are statistically insignificant (Table 6), while the results con- cerning the effect of exchange and capital controls on trade are found to be consistent, independent of the type of the exchange regime. Next we check robustness of the results to alternative experimental designs. We allow the intercepts and slopes, and then only intercepts, to vary across countries and use an F-test to check the validity of these alternative specifications. The null hypothesis is rejected for all models, implying that the differentiated country effects are statistically insignificant. The results should be interpreted with caution, in view of the potential endogeneity and measurement problems. The endogeneity problem may emerge because exchange and capital account regulations depend on the level of economic development and trade flows in a given year. The simultaneous equation bias, however, is likely to be limited in the gravity model of bilateral trade flows, because exchange and capital controls are likely to depend (if at all) on aggregate, rather than bilateral, trade flows. In turn, the measurement problem can be traced to the fact that the indices of exchange and capital controls do not account for the enforcement of controls. Controlling for this measurement error requires using the instrumental variable approach and is left for a future study. The measure of trade barriers (mean tariff rate) does not account for differences in actual tariff rates across export partners other than those due to free trade agreements. To control for this measurement problem, we use several alternative measures of trade barriers: import duties as a share of imports (calculated on the basis of the IMF©s Government Finance Statistics Yearbook), both adjusted and unadjusted for free trade agreements; mean tariff rates unadjusted for free trade agreements; and simple average tariff rates from the trade policy database compiled by the IMF.20 The results are found to be robust to the alternative measures of trade barriers.

V. Conclusion After analyzing the foregoing results, we have determined on an overall basis for 1996 that exchange and capital controls represent a significant barrier to trade. This finding, of course, depends on the level of development in each country and the type of exchange and capital controls in place. Controls on current payments and transfers are a negligible impediment to trade. Capital controls, in contrast, reduce bilateral trade for developing and transition economies, but not for industrial countries. These results reflect the variation in the extent of liberalization across countries and types of control: controls on current payments and transfers have been largely abolished world-

19The former variable is constructed on the basis of the AREAER, and figures for the latter come from the International Financial Statistics. 20The trade policy database is compiled by the IMF©s Trade Policy Division of the Policy Development and Review Department, on the basis of various sources (among others, the International Monetary Fund, the World Trade Organization, and the United Nations Conference on Trade and Development). The author thanks Robert Sharer and the staff of the Trade Policy Division for providing the data.

©International Monetary Fund. Not for Redistribution EXCHANGE AND CAPITAL CONTROLS AS BARRIERS TO TRADE a * e * * * * * * * 8 2 1 4 93 1.06 1.99 1.52 0.7 0.0 0.96 -1.06 -0.0 -0.55* ) 292.52 -39.02 denot * s * * * * * * * 2 8 8 8 1.56 1.99 93 1.06 0.7 0.96 -0.54* -1.07 -0.0 -0.0 s 292.79 -39.17 asterisk * * * d * * * * o 8 1 2 2 8 an 93 1.06 1.41 1.98 tw 0.7 0.95 0.0 (2c ; g -0.5 -1.04 -0.7 ) Countrie * * * * * n 2 1 2 5 8 level t 1.43 1.98 93 1.06 0.7 0.95 -0.0 -1.04 -0.6 -0.6 290.43*290.44 * * Developin * * * * * 2 2 4 8 percen 1.99 1.06 1.49 93 0.7 0.96 0.0 9 : -0.1 -1.06 (2b ) 9 292.09 e * * * * * * 6 2 7 8 th t 93 1.06 1.51 1.99 0.7 0.96 -0.1 -0.0 -0.75**-0.78* -1.06 a 292.12 -38.89*-38.81*-38.17*-38.12 t * * * * * * * * Controls l 0 0 1 1 58 1.77 0.8 0.93 0.99 0.0 0.99 e (2a -0.7 -0.58 -7.36 ) * * * * * * significan 0 7 1 1 s i Capita 58 1.02 1.77 0.8 0.0 t 0.94 0.99 -0.7 -0.58 -6.94 d Transitio Regim 279.00*278.82 s -34.19*-33.74 e * * * * * * * * tha t 0 1 2 1 an 58 e 1.77 0.8 0.0 0.95 0.99 0.93 Rat (2c -2.2 -0.60 ) 278.71 -33.27 e * * * * * * Countrie * l 0 1 9 2 coefficien 58 1.77 0.8 0.0 0.98 0.96 0.97 -0.60 -6.12**-6.71 -2.3 a 279.07 -33.89 s Exchang * * * * * * * * 1 0 1 1 h Exchang 58 1.77 0.8 0.0 0.94 0.97 0.99 r (2b -0.58 -7.18 -1.2 ) denote wit 279.01 -33.56 l k fo * * * * * * * * 1 0 1 7 g 58 1.01 1.77 0.8 0.94 0.0 0.98 -6.73 -1.2 -0.58 asteris -34.07 279.23 e Mode * * * * * * * * 9 6 9 1 y On 1.03 1.37 1.90 . 0.7 0.93 0.0 1,51 (2a -0.43 -0.91 -0.6 ) Controllin -37.16 * * * * * * * . 9 6 0 2 75*609.76 Gravit . . 1.03 1.37 1.90 0.0 0.7 0.94 number 1,51 -0.43 -0.91 -0.7 6 n 609 level t e * * * * Industria * * 9 s 6 8 4 1 1.03 1.90 1.39 0.7 0.0 0.94 (2c 1,51 Tabl -0.9 -0.7 -0.91 ) equatio percen e 5 ©International Monetary Fund. Not for Redistribution * * * * * 9 6 0 2 1 9 e Countrie 1.90 1.03 1.39 0.7 0.0 l 0.94 1,51 -0.9 -0.91 -0.8 th 607.52*607.58 t -37.43*-37.46*-37.18 Al indicat s a * * * * * * * * t 9 6 2 7 1.90 1.03 1.37 0.7 0.0 0.94 (2b 1,51 -0.69 -0.6 -0.91 ) 609.61 -37.19 * * * * * * * * heading (2a 9 3 6 9 n significan s 1.37 1.90 1.03 0.0 0.7 0.94 -0.6 i -0.68 -0.91 t 609.57 1,51 -37.21 s Colum f : tha t c o r EX TAR) + Notes 1 GDPEX POPIM POP GDPIM ( observation n n n n n CCI KCI F-statisti I I I i EXRD ECI c \nDIST Numbe EXRV I coefficien

85 Natalia T. Tamirisa wide, while controls on capital flows continue to prevail in many developing and transition economies, but not in industrial countries. An implication of this study is that further liberalization of exchange and capital controls can discernibly foster trade.21

APPENDIX

Indices of Exchange and Capital Controls

The tabular presentation of the IMF©s Annual Report on Exchange Arrangements and Exchange Restrictions identifies 142 individual types of exchange and capital control. These are aggregated hierarchically into 16 categories; these categories are aggregated into indices, which measure the extent of exchange and capital controls (Table 1). The index of controls on current payments and transfers includes exchange controls pertaining to the exchange arrangement, arrangements for payments and receipts, resident and nonresident accounts, import payments, and export proceeds. The index of capital controls encompasses controls on capital and money market securities, derivatives, credit operations, foreign direct investment, real estate transactions; provisions specific to commercial banks, other credit institutions and institutional investors; and surrender and repatriation requirements. The index of exchange and capital controls covers controls on current payments and transfers and capital movements. The presence of control i in country j is reflected in a dummy variable Dy, which is assigned a value of 1 when the individual type of control is in place and 0 otherwise, according to the conventions described below. The index of controls in category k (denoted by Chj) is defined as the actual number of controls normalized by the total feasible number of controls in the category (A^), as follows:

n yNi n (Al)

The indices of controls on current payments and transfers and capital controls (CCIj and KCIj, respectively) are the averages of the indices for the respective categories:

cci - - y Nca ci (A2) cc Ll C J ~ NCCI V

Kcr=^-^'Cikr (A3) iy KCl where NCci and NKCI denote the number of categories in CCI and KCI, respectively. The overall index of exchange and capital controls (ECIj) is the average of CCIj and KCI/.

21The findings concern the relationship between exchange and capital controls and bilateral exports and thus cannot be interpreted to judge the effect of controls on net trade, net capital flows, or the balance of payments; the latter issues are a topic for future research.

ECIJ=^(CCIJ + KCIJ). (A4)

Conventions for assigning values of the dummy variables Dy are as follows. The value of 1 corresponds to prohibitions, quantitative limits, approval and registration requirements,22 restrictions on investors© opportunity set (for example, the type and maturity of securities), as well as the transactions infeasible due to the absence of the respective markets. The value of 0 is assigned for measures for statistical purposes, administrative verification,23 optional official cover of forward operations, liberal granting of licenses, and the lack of access to the formal market for foreign exchange transactions.24

REFERENCES Adams, Charles, and Jeremy Greenwood, 1985, "Dual Exchange Rate Systems and Capital Controls: An Investigation," Journal of International Economics, Vol. 18 (February), pp. 43-63. Anderson, James E., 1979, "A Theoretical Foundation of the Gravity Equation," American Economic Review, Vol. 69 (March), pp. 106-16. Bergstrand, Jeffrey H., 1985, "The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical Evidence," Review of Economics and Statistics, Vol. 67 (August), pp. 474-81. Ð, 1989, "The Generalized Gravity Equation, Monopolistic Competition, and the Factor- Proportions Theory in International Trade," Review of Economics and Statistics, Vol. 71 (February), pp. 143-53. , 1990, "The Heckscher-Ohlin-Samuels Model, the Linder Hypothesis, and the Determinants of Bilateral Intra-Industry Trade," Economic Journal, Vol. 100 (December), pp. 1216-29. Dooley, Michael P., 1996, "A Survey of Literature on Controls over International Capital Transactions," Staff Papers, International Monetary Fund, Vol. 43 (December), pp. 639-87. Fitzpatrick, Gary L., and Marilyn J. Modlin, 1986, Direct-Line Distances: International Edition (Methuchen, New Jersey: Scarecrow Press). Greenwood, Jeremy, and Kent P. Kimbrough, 1987, "An Investigation in the Theory of Exchange Controls," Canadian Journal of Economics, Vol. 20 (May), pp. 271-88. Helpman, Elhanen, 1987, "Imperfect Competition and International Trade: Evidence from Fourteen Industrial Countries," in International Competitiveness, ed. by Michael A. Spence and Heather A. Hazard (Cambridge, Massachusetts: Ballinger), pp. 197-220.

22Likewise, registration requirements are treated as restrictions in World Bank (1997). 23Under the IMF©s jurisdiction, registration or licensing used to monitor rather than restrict payments and verification requirements, such as a requirement to submit documented evidence that a payment is bona fide, does not constitute an exchange restriction, unless the process results in undue delays. With indicative limits, authorities approve all requests for foreign exchange for bona fide current international transactions in excess of limits or for transactions for which there is no basic allocation of foreign exchange. If the public is made aware of such a policy, indicative limits do not constitute a restriction. 24On average, 99 percent of the AREAER data on exchange and capital controls are available for the countries in the sample. Nonetheless, the baseline indices are defined as the averages of the indices calculated under two alternative assumptions about missing data: controls and no controls.

, and Paul R. Krugman, 1985, Market Structure and Foreign Trade: Increasing Returns, Imperfect Competition and the International Economy (Cambridge, Massachusetts: MIT Press). International Monetary Fund, 1993, Articles of Agreement of the International Monetary Fund (Washington: IMF). , Annual Report on Exchange Arrangements and Exchange Restrictions (Washington: IMF, various issues). , Direction of Trade Statistics Yearbook (Washington: IMF, various issues). , International Financial Statistics (Washington: IMF, various issues). , Government Finance Statistics Yearbook (Washington: IMF, various issues). Johnston, R. Barry, and others, forthcoming, Exchange Rate Arrangements and Currency Convertibility: Developments and Issues, Chapter III (Washington: International Monetary Fund). Johnston, R. Barry, and Chris Ryan, 1994, The Impact of Controls on Capital Movements on the Private Capital Accounts of Countries' Balance of Payments: Empirical Estimates and Policy Implications, IMF Working Paper 94/78 (Washington: International Monetary Fund). Lee, Jong-Wha, 1993, International Trade, Distortions, and Long-Run Economic Growth, Staff Papers, International Monetary Fund, Vol. 40 (June), pp. 299-328. Stockman, Alan C, and Alejandro Hernandez D., 1988, Exchange Controls, Capital Controls, and International Financial Markets, American Economic Review, Vol. 78 (June), pp. 362-74. World Bank, 1997, World Development Indicators database (Washington: World Bank). World Trade Organization, 1996, Annual Report (Geneva: World Trade Organization).

Deviations of Exchange Rates from Purchasing Power Parity: A Story Featuring Two Monetary Unions

TAMIM BAYOUMI and RONALD MACDONALD*

We examine the mean-reverting properties of real exchange rates, by comparing the unit root properties of a group of international real exchange rates with two groups of intranational real exchange rates. Strikingly, we find that while the international real rates taken as a group appear mean reverting, the intranational rates are not. This is consistent with the view that while nominal shocks may be mean reverting over the medium term, underlying real factors do generate long-term trends in real exchange rates. [JEL C12, C23, F31]

he proposition that exchange rates are volatile when allowed to float freely has Tbecome something of a stylized fact in the international finance literature (see, for example, Frenkel and Mussa, 1980; MacDonald and Taylor, 1992; and Frankel and Rose, 1995). Indeed, the volatility of exchange rates during the recent floating experience has led economists to advocate moving from an international monetary regime based on flexible exchange rates toward one based on greater exchange rate fixity (McKinnon, 1988; Mundell, 1992; and Williamson, 1987) and is also one of the central arguments made by proponents of greater monetary integration in Europe. The volatility of nominal exchange rates has also had implications for the behavior of real exchange rates. In particular, because prices in goods markets are generally regarded as being sticky (certainly in the short run), volatility in nominal exchange rates is transferred into comparable real exchange rates. This violation of purchasing power parity (PPP) may be viewed as a second stylized fact in international finance.

*Tamim Bayoumi is an economist in the IMF©s Asia and Pacific Department, and Ronald MacDonald is a professor of economics at the University of Strathclyde.

The failure of PPP to hold continuously is well documented empirically (see the summaries in Froot and Rogoff, 1995, and MacDonald, 1995). However, there is now growing evidence to suggest that although PPP does not hold on a month-to-month or quarter-to-quarter basis, it does hold as a long-run phenomenon (see, for example, Edison, 1987; Frankel, 1988; and Diebold, Husted, and Rush, 1991). The main explanation for this follows on directly from our discussion in the previous paragraph and the perceived source of the deviations from PPP. If the predominant force upsetting the PPP relationship is nominal, this will have only a transitory effect on deviations from PPP (this is essentially the story in the seminal Dornbusch, 1976, model). If, however, the source of PPP disturbances are truly "real" in nature (as suggested by Stockman, 1987), we argue this will have a permanent, or more permanent, effect on the real exchange rate.1 In this paper, we propose a way of gaining a perspective on the importance of nominal shocks in generating deviations from PPP. We do this by comparing the behavior of relative prices across countries with those within countries.2 While exchange rates across countries include both real and nominal disturbances, it appears reasonable to assume that relative prices movements within countries are dominated by real factors, with little or no nominal influence.3 In this context, it is interesting to know whether relative prices within countries are dominated by long- run trends, and hence are nonstationary or not. Our method involves constructing a panel data set for the real exchange rates of 20 countries and comparing the time-series properties of these data with comparable data sets within two monetary unions (namely, the United States and Canada). We confirm the findings of others that relative price variability within countries is considerably lower than across countries,4 and that real exchange rates appear stationary, or mean reverting, across countries.5 However, we also find that relative prices within countries are nonstationary. The implication is that underlying real factors can create long-run trends in relative prices even within a fairly homogeneous economic environment. The implication for real exchange rates is that, although they may appear stationary in the longer run compared with their short-term behavior, these results in all probability mask long-run trends caused by real behavior.

lThe issue of whether the mean reversion observed in long runs of dataÐa half-life of four years is the standard finding (see MacDonald, 1995)Ðis in fact consistent with purely nominal shocks is not uncontroversial. Rogoff (1996), for example, argues it is not. One way in which such reversion could be consistent with purely nominal shocks is if the initial real exchange rate deviation is not immediately offset because of the pricing-to-market policies pursued by multinational companies and the inability of agents to arbitrage away potentially profitable misalignments. That is the interpretation we offer. 2Previous work comparing the behavior of real exchange rates in inter- and intranational data sets includes Engel and Rogers (1996) and Engel, Hendrickson, and Rogers (1998). 3Although price connections might be different within and between countries, we believe that the fundamental distinguishing feature of an intra- and intercountry comparison is the absence of differential nominal disturbances within a monetary union. 4See, for example, Vaubel (1978), Eichengreen (1992), Bayoumi and Thomas (1995), Engel (1993), and Parsley (1996). 5See, for example, Frankel and Rose (1996) and MacDonald (1996a).

I. Panel Unit Root Method s

We use panel unit root method s to compar e the time-serie s propertie s of real exchange rates within and across countries . Such tests have a clear statistical advantage over uni- variate tests, such as the Dickey-Fulle r class of statistics, because they have greater power to reject the null of a unit root when it is in fact false. Pane l unit root tests may be motivate d by considerin g the following regression equation: 6

∆q it = αi + δq t-1 + Σγ iDi + Σγ tDt + Σβ iti+ υ it, (1) i t i

where q denote s a real exchange rate, i denote s a currency , Di and Dt denote , respectively, country-specifi c and time-specifi c fixed effects dumm y variables, ti denote s a country-specifi c time trend , αi, δ , γ i, yt, and β i are estimate d coefficients, and υ it is an error term. 7 Equatio n (1) is essentially the panel analogue to the standar d Dickey- Fuller autoregression . Of particula r interest is the magnitud e of 8, which indicate s the speed of mean reversion and its significance as judged by the estimate d t-ratio. As Levin and Lin (1992 and 1993) have demonstrated , the critical values for the t-rati o are affected by the particula r deterministi c specification used. In circumstance s where all of the deterministi c element s in equatio n (1) are excluded apart from the single constan t term, a, Levin and Lin (1992) demonstrat e that the t-statisti c on δ converges to a standar d norma l distribution . Includin g individual specific effectsÐeither (Σ i yi D i ) or (Σ i β i ti) or bothÐbut excluding time-specifi c intercepts , Levin and Lin (1992) demonstrat e that the t-rati o converges to a noncentra l norma l distribution , with substantia l impact on the size of the unit root test (and they tabulat e critical values). However, Levin and Lin (1993) argue that unless there are very strong ground s for exclusion, time-specifi c intercept s should always be include d in these kinds of panel tests. The reason for this is that the inclusion of such dummie s is equivalent to subtractin g the cross-sectiona l average in each period. This subtractio n may be dispensed with in cases where the units in the panel are independen t of each other ; however, in cases where this is not the case such a subtractio n is vital to ensure independenc e across units. In additio n to facilitatin g the removal of time means, the panel method s of Levin and Lin (1993) have a numbe r of other advantages, such as allowing the residual term to be heterogeneousl y distribute d across individuals (in terms of both nonconstan t variance and autocorrelation) , rathe r than a white noise process. The testing metho d has the null hypothesi s that each individual time series in the panel has a unit root, against the alternativ e that all individual units taken as a panel are stationary . The pro- cedur e consists of four steps, which we now briefly note (these steps do not corre- spond exactly to the steps in Levin and Lin).

6A similar equatio n forms the basis of a cross-countr y panel study by Franke l and Rose (1996). 7Henc e our tests are robust to the criticism made of earlier studies by Papell (1997). Additionally, our subtractio n of cross-sectiona l mean s for each time period addresses the point made by O©Connel l (1997) that the qi series within a given panel are not independent . Huste d and MacDonal d (1997) have demon - strated that, having controlle d for cross-sectiona l means, the use of a S.U.R.E.-typ e estimato r makes little difference to the adjusted t-ratio s reporte d in this paper.

The first step involves subtracting the cross-sectional mean from the observed exchange rate series. Thus we have qih where / runs from 1 to N and N denotes the total number of real exchange rates in the panel. We construct qt = (l/N)^=\qit. In the following steps, the term qit is interpreted as having been adjusted by qt. Step 2 involves performing regression (2) and (3) for the de-meaned data:

p, d d (2) lit iLA%-L + mi mt + eit, H, =L=]

^8iVi+e/r (3) The ^-ratio calculated on the basis of 5, is the panel equivalent to an augmented Dickey-Fuller (ADF) statistic. To control for heterogeneity across individuals, both eit and vit_\ are deflated by the regression standard error from equation (3); these adjusted errors are labeled eit and vit_\. Under the null hypothesis, these normalized innovations should be independent of each other; this can be tested by running the following regression as step 3:

^ = Svi+4- (4) Under the null hypothesis that 8/ = 0 for all / = 1,.. JV, the asymptotic theory in Section 4 of Levin and Lin (1993) indicates that the regression ^-statistic, t$, has a standard normal distribution in a specification with no deterministic terms, but diverges to negative infinity in models with deterministic terms. However, Levin and Lin (1993) demonstrate that the following adjusted ^-ratio has an ^(0,1) distribution and the critical values of the standard normal distribution can be used to test the null hypothesis that Si = 0 for all i = !,...,#:

L-NfS^RSEi^ii* (5) £=

The terms in equation (5), other than t& are calculated in step 4. Specifically, S = (\INyil\qi, where qt = 6qi/6ei,6ei is the residual standard error from equation (4) and 6qi is an estimate of the long-run standard deviation of qt, RSE(8) is an estimate of the reported standard error of the least-squares estimate of 8, a£ is the estimated standard error of regression (4), T= (T-p - 1) is the average number of observations per individual in the panel, and

N p = (i/yv)X Pi (6) 1=1

* * is the average lag order for the individual ADF statistics. Σ MT and mT represen t the mean and standar d deviation adjustments , respectively, and are tabulate d in Table 1 of Levin and Lin (1993) for different deterministi c specifications .

II. Data Sources and Data Descriptio n In line with our earlier discussion we have constructe d three annua l data sets: an internationa l data set and two intranationa l data sets. The internationa l data set consists of two real bilateral exchange rates defined for 20 countrie s relative to the Unite d States (the countrie s are listed in Table 1), constructe d using relative wholesale and consume r prices (which are the most widely studied real exchange rates in the literature) . The internationa l data run from 1973 through 1993, and the wholesale and consume r prices and the exchange rates are taken from the IMF© s International Financial Statistics.8 The two monetar y union s we focus on are Canad a and the Unite d States. For the former country , real exchange rates are defined using consumptio n indices, while for the Unite d States productio n indices are used. Although these series were chosen because of their availability, there is a debate in the literatur e regarding the most appropriat e price series to use in defining a real exchange rate (see, for example, Frenkel , 1978). Since our chosen indices may be interprete d as representin g two extreme forms of price series, they should help to determin e if a particula r intracoun - try result is driven by the choice of price index or is independen t of the index used. Mor e specifically, for Canad a we have collected data on provincia l nondurabl e consumption , and the real exchange rate is measure d as (the log of) the relative price of a particula r province with respect to Ontario . The Canadia n sample period is 1972-94. The U.S. data consist of gross state produc t data for 48 states (we exclude Alaska and Hawaii) and the real exchange rates are constructe d relative to New Jersey (again in logs).9 The total U.S. sample period runs from 1963 through 1992. We have used this full sample but, to be consisten t with the internationa l data sample, we also constructe d a subsample correspondin g to the recent floating period. 10 We believe it is importan t to run our panel tests for a variety of samples since it is well known that the panel estimator s we use are most efficient when the dimension s of the panel are approximatel y square; that is, when the cross-sectiona l dimension s are approximatel y equal to the time-serie s dimensions . For the internationa l data set, this will be true for the recent floating period and it will also be approximatel y true for the U.S. data over the full sample period. Before conductin g formal tests, it is useful to examine some of the propertie s of the data graphically. Given our interest in the mean-revertin g propertie s of the data, we compar e the relationshi p between curren t and future movement s in real exchange

8The wholesale price series is line 62, the consume r price line series is line 63, and the exchange rate is line ae. As our interest is in the low frequenc y characteristic s of the data, annua l data are sufficient. 9As the regressions use dumm y variables for each year, the choice of numerair e has no impact on the results. New Jersey was selected as it has been used in the numerair e in some other studies using intrastat e data from the Unite d States. 10We also experimente d with a subsample of 20 large states, to see if the results were sensitive to the numbe r of regions being considered . As the results were very similar to those with the full sample, they are not reported .

©International Monetary Fund. Not for Redistribution Tamim Bayoumi and Ronald MacDonald rates. Accordingly, the data in the three panels were divided into successive five-year periods, and the change in the logarithm of the real exchange rate in the first year was compare d with the average change over the next four years within each five-year period for each country/state/province . The results from this exercise are shown in Figure 1. Two differences are immediately apparen t between the internationa l data set and those for the Unite d States and Canada . The first is the much larger degree of real exchange rate movement s in the internationa l data, consisten t with the notio n that internationa l relative prices are driven by larger underlyin g shocks than their intranationa l counterparts . The variability for the Canadia n data set is also considerabl y smaller than that for the Unite d States, presumabl y because of the much stronger forces toward equalizatio n of the consume r prices used in the Canadia n panel compare d to the produce r prices used in the U.S. panel. By contrast , in the internationa l data set differences in behavior across alternativ e types of prices indices are minima l (not reporte d for the sake of brevity). The second difference is in the predictabilit y of future movement s in relative prices. The internationa l data show almost no correlatio n between curren t relative price movement s and movement s over the next four years, which is consisten t with the weak mean reversion found in most studies of the internationa l data. By contrast , curren t increases in relative prices between U.S. states are clearly positively correlated with further increases in relative prices in the future, implying that changes in relative prices have considerabl e momentu m over time (the behavior across Canadia n province s is difficult to assess because of the much lower level of variability). In short, the intranationa l data show much less evidence of mean reversion. The next section examines this issue more formally using the panel unit root tests discussed earlier.

III. Univariat e and Panel Unit Root Results Before implementin g the panel unit root tests we examine the univariat e unit proper - ties of each of the real exchange rates using standar d ADF statistics. These results for our range of real exchange rates are reporte d in Tables 1 through 3. With very few exception s the internationa l data set, reporte d in Table 1, confirm s the now standar d result that on a univariat e basis, and for the recent float, real exchange rates are non- stationar y variables. Tables 2 and 3 confirm that this internationa l result also holds for real exchange rates within our two monetar y unions . What happens , though , when we take these groupings as panels? Two aspects of our panel results should be empha - sized: first, the speed of adjustment , as represente d by 8, and second, our estimate d t- ratios. Table 4 report s panel unit root tests for our two internationa l data sets (using CPI s and WPIs), the U.S. panel over the full sample period and the subperiod since 1973, and the Canadia n panel. The estimate d adjustmen t speeds for our different regressions are reporte d in the rows labeled "δ " in Table 4 (the layout of the table is explained in the footnotes) . It is noteworth y that the adjustmen t speeds in the internationa l and intranationa l panels are negative and are therefor e all indicative of mean reversion. However, adjustmen t is much more rapid in the internationa l data sets relative to the nationa l ones. For example, the average value across the latter regressions is -0.12, while the average

©International Monetary Fund. Not for Redistribution DEVIATIONS OF EXCHANGE RATES FROM PURCHASING POWER PARITY a Canad n Withi s Rate e . Exchang l U.S.A Rea f n o r Withi Behavio . 1 e Figur ©International Monetary Fund. Not for Redistribution . s details e Countrie mor s r fo t Acros tex e Se : Sources

95 Tamim Bayoumi and Ronald MacDonald

Table 1. Augmented Dickey-Fuller Unit Root Tests: Levels, International Results

CPI WPI

Country tu tx t^ tx Australia -1.98 -2.08 -1.21 -1.99 Austria -2.28 -2.26 -2.05 -2.13 Belgium -1.12 -0.79 -1.69 -2.51 Canada -2.27 -1.46 -1.71 -1.71 Denmark -1.54 -1.36 -1.58 -1.51 Finland -4.17* -4.42* -2.91 -3.58 France -3.58* -5.68* -1.64 -1.64 Germany -1.84 -1.76 -1.95 -1.99 Greece -1.84 -2.09 -1.50 -1.57 Ireland -2.44 -3.87* -0;89 -1.62 Italy -0.46 -1.45 -0.91 -3.27 Japan -1.77 -2.56 -1.44 -2.26 Netherlands -1.87 -1.53 -1.22 -2.28 New Zealand -2.05 -1.89 -2.14* -3.05 Norway -3.92* -3.77* -4.29* -4.34* South Korea -1.80 -2.14 -2.28 -2.29 Spain -1.46 -1.90 -0.83 -1.53 Sweden -2.47 -2.44 -1.66 -0.88 Switzerland -3.12* -2.99 -2.89 -3.16 United Kingdom -1.10 -1.61 -1.41 -1.45

Notes: The numbers in the columns labeled t^ and tx are ADF f-ratios from an autoregression with, respectively, a constant and a constant plus a time trend included. An asterisk denotes significance at the 5 percent level.

for the international data sets is two and a half times greater at -0.29. These figures translate into half-lives of two and six years, respectively, for the international and intranational data sets. The average half-life from our international data sets is shorter than the estimates reported by Frankel and Rose (1996) (they report average half- lives of four years), but nevertheless reinforces the importance of using panel data when defining PPP deviations. The average half-life for the intranational data,

Table 2. Augmented Dickey-Fuller Unit Root Tests: U.S. Results

State 'n h New England Connecticut -1.98 -1.76 Maine -1.81 -1.68 Massachusetts -1.53 -1.53 New Hampshire -0.75 -0.99 Rhode Island -2.08 -2.00 Vermont -1.42 -0.91 Mid-East Delaware -2.22 -1.14 Maryland -1.19 -1.75 New York -1.69 -1.60 Pennsylvania -1.73 -2.61 Great Lakes Illinois -1.53 -1.56 Indiana -0.37 -0.86 Michigan 0.07 -0.91 Ohio -2.62 -3.09 Wisconsin -0.47 -1.13 Plains Iowa -0.18 -1.18 Kansas -0.73 -1.48 Minnesota -1.65 -0.95 Missouri -0.85 -1.53 Nebraska -0.22 -1.10 North Dakota -1.18 -1.32 South Dakota -1.97 -1.61 Southeast Alabama -1.96 -2.11 Arkansas -1.34 -0.92 Florida -0.95 -0.79 Georgia -1.92 -0.93 Kentucky -0.87 -1.59 Louisiana -0.95 -1.03 Mississippi -1.63 -1.71 North Carolina -1.59 -1.07 South Carolina -3.04* -3.02 Tennessee -0.39 -1.41 Virginia -0.27 -1.56 West Virginia -1.28 -1.90 Southwest Arizona -1.31 -1.05 New Mexico -0.95 -0.79 Oklahoma -1.74 -1.25 Texas -1.71 -1.39

Table 2. (concluded)

State h t% Rocky Mountains Colorado -1.73 -1.50 Idaho -2.02 -1.14 Montana -1.76 -1.66 Utah -1.62 -1.25 Wyoming -1.73 -0.63 Far West California -1.61 -1.26 Nevada -1.97 -1.11 Oregon -2.19 -0.72 Washington -2.04 -1.83 Notes: See Table 1.

Table 3. Augmented Dickey-Fuller Unit Root Tests: Canadian Results

Province h tx Alberta -1.41 -1.97 British Columbia -1.68 -2.51 Manitoba -0.75 -0.99 New Brunswick -2.08 -2.00 Newfoundland -1.42 -0.91 Nova Scotia -2.22 -1.14 Prince Edward Island -1.19 -1.75 Quebec -1.69 -1.60 Saskatchewan -1.73 -2.61 Notes: See Table 1.

although much shorter than the international value, is still longer than the average value culled from single-country estimates for the recent float (which would imply a very long half-life of about 20 to 30 yearsÐsee MacDonald, 1995). However, a cru- cial issue is whether the mean reversion exhibited in our panel data sets is statistically significant; that is, are the negative adjustment speeds significantly different from zero or not? The estimated unadjusted r-ratio, that is, t$, is in all cases larger in absolute value than -4.0 and, in terms of the original Levin and Lin (1992) critical values, these t- ratios would be statistically significant. However, as we have noted, the unadjusted t-

Table 4. Panel Unit Root Tests

International Panel INT/CPI INT/WPI

t5 -8.23 -8.72 tt -2.28 -2.58 2-tail (0.02) (0.00) 1-tail (0.01) (0.00)

6 -0.276 -0.308

U.S. Panel US/47/full US/47/sub

r6 -10.11 -10.21 tt -1.04 0.49 2-tail (0.29) (0.62) 1-tail (0.14) (0.31)

8 -0.079 -0.146

Canadian Panel Province/Ontario ?5 -4.47 tt -0.39 2-tail (0.69) 1-tail (0.34)

5 -0.126 Notes: The numbers in the rows labeled fg and /| are, respectively, the unadjusted and adjusted panel unit-root ^-ratios, defined in the text. The latter statistic has a standard normal distribution; numbers in parentheses are marginal significance levels. The numbers in the rows labeled 5 are the adjustment speeds defined in the text. The columns labeled INT/CPI and INT/WPI denote the international panels using, respectively, consumer and wholesale prices to define the real exchange rate. The columns labeled US/47/full and US/47/sub denote the U.S. panel samples over the full and subsample period (see text for further details). The column headed Province/Ontario denotes the Canadian real exchange rates defined for each province with respect to Ontario.

ratios are biased to minus infinity and it is not appropriate to draw inferences on the basis of these test statistics. Interestingly, the estimated adjusted r-ratiosÐthe t% valuesÐgive a dramatically different picture. For all of the currency union samples the estimated value of t% is insignificantly different from zero, but for the international data set both real exchange rate data sets produce statistically significant adjusted r-ratios. Given that we use two very different price series for the monetary unions, we do not believe our results are a result of the particular series used. We offer an interpretation in the following concluding section.

IV. Conclusion

Recent empirical work on the behavior of exchange rates has gone through a number of distinct phases. The first phase involved testing the hypothesis that rates were a random walk, and hence unpredictable in the long run. More recent work indicates that while the random-walk model is a reasonably good approximation to short-run dynamics, real exchange rates show mean-reverting tendencies over the medium to long term. The evidence in this paper can be seen as adding a further layer of complexity to this story. To abstract from the nominal factors, which are often thought to generate much of the short-term dynamics, we studied the behavior of relative prices across regions within a country. The results indicate that these relative prices have significant long-run trends. This implies that underlying real factors can create long-run trends in relative prices even in a fairly homogeneous environment. The implication we draw is that, while nominal shocks may be mean reverting over the medium term, generating the observed mean reversion in real exchange rates, this medium-term effect obscures the fact that underlying real factors generate long-term trends in real exchange rates. The next task for empirical researchers is to identify and quantify these effects.11

REFERENCES Bayoumi, Tamim, and Alun Thomas, 1995, "Relative Prices and Economic Adjustment in the United States and the European Union: A Real Story About European Monetary Union," Staff Papers, International Monetary Fund, Vol. 42 (March), pp. 108-133. Diebold, Francis X., Steven Husted, and Mark Rush, 1991, "Real Exchange Rates Under the Gold Standard," Journal of Political Economy, Vol. 99 (December), pp. 1252-71. Dornbusch, Rudiger, 1976, "Expectations and Exchange Rate Dynamics," Journal of Political Economy, Vol. 84 (December), pp. 1161-76. Edison, Hali J., 1987, "Purchasing Power Parity in the Long Run: A Test of the Dollar/Pound Exchange Rate (1890-1978)," Journal of Money, Credit and Banking, Vol. 19 (August), pp. 376-87. Eichengreen, Barry, 1992, "Is Europe an Optimum Currency Area?" in The European Community After 1992: Perspectives from the Outside, ed. by Silvio Borner and Herbert Gruebel (Houndmills, Basingstoke, England: Macmillan), pp. 138-61. Engel, Charles, 1993, "Real Exchange Rates and Relative Prices: An Empirical Investigation," Journal of Monetary Economics, Vol. 32 (August), pp. 35-50. , and John H. Rogers, 1996, "How Wide Is the Border?" American Economic Review, Vol. 86 (December), pp. 1112-25. Engel, Charles, Michael K. Hendrickson, and John H. Rogers, 1998, "Intra-national, Intra-continental, and Inter-Planetary PPP," Journal of Japanese and International Economies (December), pp. 480-501. Faruqee, Hamid, 1995, "Long-Run Determinants of the Real Exchange Rate: A Stock-Flow Perspective," Staff Papers, International Monetary Fund, Vol. 42 (March), pp. 80-107.

11Some evidence on this can be found in Faruqee (1995), Gagnon and Rose (1996), and MacDonald (1996b).

100

Frankel, Jeffrey A., 1988, "International Capital Flows and Domestic Economic Policies," in The United States in the World Economy, NBER Conference Report, ed. by Martin Feldstein (Chicago: Chicago University Press), pp. 559-627. , and Andrew K. Rose, 1995, "A Survey of Empirical Research on Nominal Exchange Rates," Chapter 33 in The Handbook of International Economics, ed. by Gene M. Grossman and Kenneth Rogoff, Vol. 3 (Amsterdam: North-Holland), pp. 1689-1729. , 1996, "A Panel Project on Purchasing Power Parity: Mean Reversion Within and Between Countries," Journal of International Economics, Vol. 40 (February), pp. 209-24. Frenkel, Jacob, 1978, "Purchasing Power Parity: Doctrinal Perspective and Evidence from the 1920s," Journal of International Economics, Vol. 8, pp. 169-91. , and Michael Mussa, 1980, "The Efficiency of Foreign Exchange Markets and Measures of Turbulence," American Economic Review, Vol. 70 (May), pp. 374-81. Froot, Kenneth, and Kenneth Rogoff, 1995, "Perspectives on PPP and Long-Run Real Exchange Rates," in The Handbook of International Economics, ed. by Gene M. Grossman and Kenneth Rogoff (Amsterdam: North-Holland). Gagnon, Joseph E. and Andrew K. Rose, 1996, "Panel Evidence on the Determinants of Real Exchange Rates," paper presented at the American Economics Association meeting, San Francisco, January 5-7. Husted, Steven, and Ronald MacDonald, 1998, "Monetary-Based Models of the Exchange Rates," Journal of International Financial Markets, Institutions, and Money, Vol. 8 (January), pp. 1-19. International Monetary Fund, 1998, International Financial Statistics (CD-ROM; Washington: IMF). Levin, Andrew, and Chien-Fu Lin, 1992, "Unit Roots in Panel Data: Asymptotic and Finite Sample Properties" (unpublished; San Diego: University of California at San Diego). , 1993, "Unit Roots in Panel Data: Asymptotic and Finite Sample Properties," Discussion Paper 93-56 (San Diego: University of California at San Diego). MacDonald, Ronald, 1995, "Long-Run Exchange Rate Modeling: A Survey of the Recent Evidence," Staff Papers, International Monetary Fund, Vol. 42 (September), pp. 437-89. , 1996a, "Panel Unit Root Tests and Real Exchange Rates," Economics Letters, Vol. 50 (January), pp. 7-11. , 1996b, "What Determines Real Exchange Rates? The Long and Short of It," Journal of International Financial Markets, Institutions, and Money, Vol. 8 (June), pp. 117-53. , and Mark P. Taylor, 1992, "Exchange Rate Economics: A Survey," Staff Papers, International Monetary Fund, Vol. 40 (March), pp. 1-53. McKinnon, Ronald, 1988, "Monetary and Exchange-Rate Policies for International Financial Stability: A Proposal," Journal of Economic Perspectives, Vol. 2 (Winter), pp. 83-103. Mundell, Robert A., 1992, "The Global Adjustment System," in Global Disequilibrium in the World Economy, ed. by Mario Baldassarri, John McAllum, and Robert A. Mundell (New York: St. Martin©s), pp. 351-464. O©Connell, Paul G.J., 1997, "The Overvaluation of Purchasing Power Parity," Journal of International Economics, Vol. 44 (February), pp. 1-19. Papell, David H., 1997, "Searching for Stationarity: Purchasing Power Parity Under the Current Float," Journal of International Economics, Vol. 43 (November), pp. 313-32. Parsley, David C, 1996, "Convergence to the Law of One Price Without Trade Barriers or Currency Fluctuations," Quarterly Journal of Economics, Vol. 111 (November), pp. 1211-36.

101

Rogoff, Kenneth, 1996, "The Purchasing Power Parity Puzzle," Journal of Economic Literature, Vol. 34 (June), pp. 647-68. Shiller, Robert J. and Pierre Perron, 1985, "Testing the Random Walk Hypothesis: Power Versus Frequency of Observation," Economics Letters, Vol. 18, pp. 381-86. Stockman, Alan C, 1987, "The Equilibrium Approach to Exchange Rates," Federal Reserve Bank of Richmond Economic Review, Vol. 73 (March-April), pp. 12-30. Vaubel, Roland, 1978, "Real Exchange Rate Changes in the European Community: A New Approach to the Determination of Optimum Currency Areas," Journal of International Economics, Vol. 8 (May), pp. 319-39. Williamson, John, 1987, "Exchange Rate Management: The Role of Target Zones," American Economic Review, Papers and Proceedings, Vol. 77 (May), pp. 200-204.

102

IMF Working Papers

IMF Staff Papers draws on IMF Working Papers, which are research studies by members of the IMF's staff. A list of Working Papers issued in 1998:4 follows:

"Real Exchange Rate Volatility: Does the Nominal Exchange Rate Regime Matter?" by Hong Liang [98/147] "Purchasing Power Parities in Five East African Countries: Burundi, Kenya, Rwanda, Tanzania, and Uganda," by Noureddine Krichene [98/148] "Time Series Analysis of Export Demand Equations: A Cross-Country Analysis," by Abdelhak Senhadji and Claudio Montenegro [98/149] "European Trade and Foreign Direct Investment U-Shaping Industrial Output in Central and Eastern Europe: Theory and Evidence," by Alexander Repkine and Patrick P. Walsh [98/150] "Pension Developments and Reforms in Transition Economies," by Marco Cangiano, Carlo Cottarelli, and Luis Cubeddu [98/151] "The Effects of Tax Wedges on Hours Worked and Unemployment in Sweden," by Alun Thomas [98/152] "Monetary Policy in a Small Open Economy with Credit Goods Production," by Jorge A. Chan-Lau [98/153] "Are Currency Crises Predictable? A Test," by Andrew Berg and Catherine Pattillo [98/154] "Financial Market Contagion in the Asian Crisis," by Taimur Baig and Ilan Goldfajn [98/155] "Soft Exchange Rate Bands and Speculative Attacks: Theory, and Evidence from the ERM since August 1993," by Leonardo Bartolini and Alessandro Prati [98/156] "Demand for Money in Mozambique: Was There a Structural Break?" by Marco Pinon- Farah [98/157] "Fiscal Effects of the 1993 Colombian Pension Reform," by Sergio Clavijo [98/158] "Capital Flows with Debt- and Equity-Financed Investment: Equilibrium Structure and Efficiency Implications," by Assaf Razin, Efraim Sadka, and Chi-Wa Yuen [98/159] "Determinants of Inflation, Exchange Rate, and Output in Nigeria," by Louis Kuijs [98/160] "Inflation, Uncertainty, and Growth in Colombia," by Henry Ma [98/161]

103

"The Impact of Human Capital on Growth: Evidence from West Africa," by Emilio Sacerdoti, Sonia Brunschwig, and Jon Tang [98/162] "The Volatility of the Relative Price of Commodities In Terms of Manufactures Across Exchange Regimes: A Theoretical Model," by Hong Liang [98/163] "The Quality of Governance: "Second-Generation" Civil Service Reform in Africa," by Nadeem Ul Haque and Jahangir Aziz [98/164] "Migration and Pension," by Assaf Razin and Efraim Sadka [98/165] "Economic Determinants of Government Subsidies," by Benedict Clements, Hugo Rodriguez, and Gerd Schwartz [98/166] "The Decline of Traditional Sectors in Isreal: The Role of the Exchange Rate and the Minimum Wage," by Eric V. Clifton [98/167] "Managing Capital Flows: Lessons from the Experience of Chile," by Bernard Laurens and Jaime Cardoso [98/168] "Do IMF-Supported Programs Work? A Survey of the Cross-Country Empirical Evidence," by Nadeem Ul Haque and Mohsin S. Khan [98/169] "Monetary Policy in the Aftermath of Currency Crises: The Case of Asia," by Ilan Goldfajn and Taimur Baig [98/170] "Private Saving in Colombia," by Alejandro Lopez-Mejia and Juan Ricardo Ortega [98/171] "International Capital Flows and National Creditworthiness: Do the Fundamental Things Apply as Time Goes By?" by Paul Cashin and C. John McDermott [98/172] "Fixed-Income Markets in the United States, Europe, and Japan: Some Lessons for Emerging Markets," by Garry J. Schinasi and R. Todd Smith [98/173] "The Wage Bargaining Structure in Norway and Sweden and its Influence on Real Wage Developments," by Alun Thomas [98/174] "Export Credit Agencies, Trade Finance, and South East Asia," by Malcolm Stephens [98/175] "Will Fiscal Policy Be Effective Under EMU?" by Marco Cangiano and Eric Mottu [98/176] "Terms of Trade Shocks and the Current Account," by Paul Cashin and C. John McDermott [98/177] "Fundamental Determinants of Inequality and the Role of Government," by Vito Tanzi [98/178] "Correlations Between Real Interest Rates and Output in a Dynamic International Model; Evidence from G-7 Countries," by Jahanara Begum [98/179] "The Intragenerational Redistributive Effects of Unfunded Pension Programs," by Luis Cubeddu [98/180] "Why Do Countries Use Capital Controls?" by R. Barry Johnston and Natalia T. Tamirisa [98/181] "The Dynamic Macroeconomic Effects of Tax Policy in an Overlapping Generations Model," by Ben J. Heijdra and Jenny E. Ligthart [98/182]

104

Papers on Policy Analysis and Assessment

Papers on Policy Analysis and Assessment are intended to make work by the IMF staff in the area of policy design available to a wide audience. A list of PPAAs issued in 1998:4 follows. These Papers may be considered for publication in the journal.

"Corporate Debt Restructuring in East Asia: Some Lessons from International Experience," by Mark R. Stone [98/13] "Economic Transition and Social Protection: Issues and Agenda for Reform," by Sanjeev Gupta [98/14]

105

dots (...) indicate that the data are not available;

a dash (Ð) indicates that the figure is zero or less than half the final digit shown, or that the item does not exist;

a single dot (.) indicates decimals;

a comma (,) separates thousands and millions;

"billion" means a thousand million; and "trillion" means a thousand billion;

a short dash (-) is used between years or months (for example, 1998-99 or January-June) to indicate a total of the years or months inclusive of the beginning and ending years or months;

a slash (/) is used between years (for example, 1998/99) to indicate a fiscal year or a crop year;

a colon is used between a year and a number indicating a quarter within that year (for example, 1998:1); and

components of tables may not add to totals shown because of rounding.

The term "country," as used in this publication, may not refer to a territorial entity that is a state as understood by international law and practice; the term may also cover some territorial entities that are not states but for which statistical data are maintained and provided inter- nationally on a separate and independent basis.