Journal of Development Economics Vol. 62 2000. 385421 www.elsevier.

comrlocatereconbase

The effect of IMF programs on economic growth

Adam Przeworski a , James Raymond Vreeland b, ) ,1
b a Department of Politics, New York Uniersity, 715 Broadway, New York, NY 10003, USA Department of Political Science, Yale Uniersity, 124 Prospect Street, P.O. Box 208301, New Haen, CT 06520-8301, USA

Abstract Using a bivariate, dynamic version of the Heckman selection model, we estimate the effect of participation in International Monetary Fund IMF. programs on economic growth. We find evidence that governments enter into agreements with the IMF under the pressures of a foreign reserves crisis but they also bring in the Fund to shield themselves from the political costs of adjustment policies. Program participation lowers growth rates for as long as countries remain under a program. Once countries leave the program, they grow faster than if they had remained, but not faster than they would have without participation. q 2000 Elsevier Science B.V. All rights reserved.
JEL classification: C1; O4 Keywords: Economic growth; IMF; Selection

Our primary objectie is growth. In my iew, there is no longer any ambiguity about this. It is toward growth that our programs and their conditionality are aimed. It is with a iew toward growth that we carry out our special responsibility of helping to correct balance of payments disequilibria and, more generally, to eliminate obstructie macroeconomic imbalances.
)

Corresponding author. Tel.: q 1-203-432-6196; fax: q 1-203-432-6196. E-mail addresses: ap3@is2.nyu.edu A. Przeworski., james.vreeland@yale.edu J.R. Vreeland.. 1 Tel.: q 1-203-432-5230; fax: q 1-203-432-6196.

0304-3878r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved. PII: S 0 3 0 4 - 3 8 7 8 0 0 . 0 0 0 9 0 - 0

386

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Michel Camdessus, IMF Managing Director Statement before the United Nations Economic and Social Council in Geneva, July 11, 1990..

1. Introduction International Monetary Fund IMF. programs are controversial. Governments that enter into agreements with the IMF claim that it is for the better, that opposition to them is uninformed or badly intentioned. Yet general strikes, riots, and ransacking of supermarkets manifest that IMF programs mobilize popular resistance. And scholarly opinion is also divided: statistical findings range all over the spectrum of possible conclusions. Hence, our question: What is the effect of IMF programs on growth? The immediate goals of the IMF concern exchange rate stability and balance of payments, and evaluations of IMF programs tend to concentrate on these objectives. Thus, Reichmann and Stillson 1978. and Connors 1979. found that IMF programs had no effect on balance of payments, while Pastor 1987b., Gylfason 1987., Khan 1990., and Bird 1996. reported improvements. Most studies find that Fund programs have no effect on inflation Bird, 1996; Edwards and Santaella, 1993; Pastor, 1987b; Gylfason, 1987; Connors, 1979., although Reichmann and Stillson 1978. reported an unclear effect, and Killick 1995. reported reduced inflation. And while Connors 1979., Killick 1995., and Pastor 1987b. found no effect on current account, Khan 1990. and Edwards and Santaella 1993. discover that it improves. Yet whether or not the IMF programs have positive effects on these short-term goals, what ultimately matters is whether they induce economic growth and do not concentrate incomes.2 Indeed, the Articles of Agreement state that the mission of the IMF is to facilitate the expansion and balanced growth of international trade, and to contribute thereby to the promotion and maintenance of high levels of employment and real income and to the development of the productive resources of all members as primary objectives of economic policy. As we have seen above, the former managing director of the Fund has placed economic growth as the primary objective. But here again the results are ambivalent: while Reichmann and Stillson 1978., Connors 1979., Pastor 1987b., and Gylfason 1987. reported no effect, Killick 1995. found ambiguous effects, and Conway 1994. argued that while growth declines in the first year of a program, the negative effects diminish thereafter.
According to Camdessus 1990., the Fund seeks high quality growth, not merely growth for the privileged few, leaving the poor with nothing but empty promises. Pastor 1987a,b., however, found that IMF programs sharply redistribute incomes from labor to capital in Latin America. We intend to pursue the distributional effects in a separate paper.
2

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

387

The standard difficulty in evaluating effects of any policy or program is nonrandom selection Heckman 1988..3 What we observe in the real world are not experiments, which would match the treatment and the control groups, thus permitting direct inferences about the experimental effects. Indeed, one would hope that governments do not enter into the IMF programs as an experiment. These treatments are costly: at least in the short run, they limit national sovereignty and inflict economic pain. In fact, governments often claim that they are going under only because the situation is dire and no choice is left but to swallow the bitter pill, undergo radical surgery, take a horse treatment: the lexicon is medical and the operation delicate. Going to the IMF is an act of courage, a demonstration of political will. But if countries enter and remain under agreements only when governments recognize that the situation so demands and have the courage to swallow the consequences, the conditions of countries participating in IMF programs are not the same as of those which abstain. And if these conditions are not the same, then the effects of the programs may depend in some part on their inherent consequences and in some part on these conditions. To evaluate the consequences of the programs, one must therefore distinguish the two components. The procedure is inextricably counterfactual: the task is to compare the performance if countries had participated and not participated in the programs under the same conditions and, if selection is nonrandom, we cannot always match the observed cases for these conditions. Hence, we must proceed differently. Moreover, while some of these conditions may be observable, as most economic circumstances are, some may not be. Political will is one example. A methodology failing to account for this unobservable variable would overstate the value of participation by attributing the positive effects of political will to the IMF program. Furthermore, if such selection occurs, merely controlling for observed variables can increase the bias Achen ,1986; Przeworski and Limongi, 1996.. Thus, we need to understand first why countries enter and leave the IMF programs. Once we do, we can use the methods developed by Heckman 1976, 1988. to estimate their effects independently of selection. In the appendices, we present the statistical model of selection and a method for analyzing the effect of programs on performance. These appendices are lengthy and technical but the issues are complex.

Evaluations of IMF programs have long paid attention to selection, from early beforeafter studies Reichmann and Stillson, 1978; Connors, 1979; Pastor, 1987a,b; Killick, 1995. and work comparing countries with and without IMF programs Gylfason, 1987; Edwards and Santaella, 1993., to more recent work which corrects for observable determinants of nonrandom selection of program countries Khan, 1990, Conway, 1994.. Yet, none of these studies allowed for the possibility of unobserved factors affecting selection and performance.

388

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

The article is organized as follows. The first part provides background information about our data. The second part analyzes why countries enter and stay under IMF programs. The third part provides the results concerning growth. A brief conclusion follows.

2. Background: the data set In 1944, 44 countries signed the Bretton Woods agreement which established the IMF to maintain exchange rates for international free trade. When the US went off the gold standard in 1973, the old exchange system collapsed. The new system did not need the IMF, and the organization faced a crisis of purpose Pastor, 1987a, 14; de Vries, 1986, 222.. However, the IMFs purposes also included providing wmembersx with opportunity to correct maladjustments in their balance of payments without resorting to measures destructive of national or international prosperity Articles of Agreement .. Fulfilling this function, the IMF transformed itself from a currency regulating institution to an international organization involved in the national policies of much of the third world, particularly since the onset of the debt crisis in 1982. According to our data, the number of countries under one or another IMF program grew from 21 in 1974 to 52 by 1983. Today there are 182 IMF members who are eligible to take out loans or technically make repurchases. from the Fund. Membership requires a contribution to the Fund, called a countrys quota, the size of which depends on the size of the members economy. A member-country can freely draw up to 25% of its quota to address balance of payments deficits. To draw on more than 25% requires a special agreement with the Fund. The IMF attaches to these loans conditions which oblige countries to undertake specific policies in order to receive the loan installments. These conditions typically entail fiscal austerity cutting government expenditures and increasing taxes., tight monetary policy raising interest rates and reducing credit creation., and currency devaluations. Upon entering an agreement, a member government collaborates with IMF officials to develop policy recommendations, but the IMF sets the tone Taylor, 1993, 412.. There are four main types of IMF agreements: the stand-by arrangement SBA., the extended fund facility EFF., the structural adjustment facility SAF., and the enhanced structural adjustment facility ESAF.. Polak 1991. describes the differences among these arrangements as they relate to the conditions, timing, and size of the loan disbursements, but he notes that the fundamental objectives of these programs do not differ. Thus, we consider only whether a country is under or not under agreement without differentiation between type of agreement. Note that in our full sample, which includes 678 agreements, 598 of them 88%. are SBAs. While SBAs are designed to last 12 to 18 months and the other agreements are supposed to cover 1 to 3 or at most 4. years, such time limits on participation are arbitrary. The vast experience of countries has been to sign consecutive agree-

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

389

ments. Out of the 226 separate spells of one or more consecutive IMF agreements during the period from 1951 to 1990, 111 of them spanned four or more years. South Korea, for example, spent 13 years from 1965 to 1977 under consecutive agreements, Zaire spent 14 straight years in IMF programs 19761989., Liberia spent 15 years straight 19631977., Peru was under consecutive agreements from 1954 to 1971, and Panama from 1968 to 1987. And, after a brief stint of only seven straight years of agreements 1961 to 1967., Haiti went under again from 1970 to 1989 27 out of 29 years.. During the entire period between 1951 and 1990, an average completed spell lasted 4.7 years; between 1971 and 1990 these spells were longer, lasting on average 5.3 years. Consecutive agreements are thus the rule, not the exception. We consider such consecutive agreements as a single spell of remaining under. While our basic data set covers the period between 1951 and 1990 135 countries, for the total of 4126 country-years., some information is not available for several countries and years.4 Most statistical analyses reported below are based on 1024 annual observations of 79 countries, all for the post-1970 period. Thus, the availability of data is yet another source of nonrandom selection: it turns out that countries for which information is available differ in some systematic ways from those for which it is not. To test the robustness of our findings, we examined them using various stripped versions of the selection model. These tests are reported in Appendix C. 3. Selection Since a balance of payments deficit or a reserves crisis is the prerequisite for signing an IMF agreement, one might not expect controversy in predicting participation. According to the Conditions governing use of the Funds general resources, found in Article V, Section 3 of the IMF Articles of Agreement : b. A member shall be entitled to purchase the currencies of other members from the Fund in exchange for an equivalent amount of its own currency subject to the following conditions: . . . ii. the member represents that it has a need to make the purchase because of its balance of payments or its reserve position or developments in its reserves. Yet a balance of payments deficit is not sufficient to explain agreements. While Santaella 1996., as well as Goldstein and Montiel 1986., discovered that a balance of payments deficit predicts participation, Knight and Santaella 1997. found it not to be important. Other recent studies have also failed to report
We list the countries, years, and IMF agreements in our sample in Appendix D. Most of the economic data used in this article come from the World Bank. The exception is growth, measured as total annual growth of output, taken from the Penn World Tables. The political data come from the ACLP World PoliticalrEconomic Database see Alvarez et al., 1996 for details..
4

390

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

significant findings on balance of payments Bird, 1996; Conway, 1994; Edwards and Santaella, 1993.. According to our data, while the IMF signs most of its agreements with countries facing balance of payments deficits, it signed 32% of them with countries running an overall balance of payments surplus the year preceding the agreement. In 20% of agreements, countries had surpluses both the year before and the year of the agreement. We find the same with regard to the ratio of international reserves to monthly import requirements. Typically countries that participate in IMF programs have low reserves, averaging 2.54 times monthly import requirements, and countries without agreements hold stronger reserves, on the average 3.64 times monthly imports. However, many countries enter and remain under agreements with strong reserves. Ghana, for example, entered an IMF agreement in 1983 when its foreign reserves reached their strongest position ever: 4.8 times monthly import requirements. Uruguay entered a 1-year arrangement in 1990 when its reserves averaged 7.7 times monthly imports, more than double the average reserves of the rest of Latin America. Portugal entered into a 3-year IMF program in 1983 with foreign reserves averaging 9.6 times monthly imports. And after 8 years of consecutive participation in IMF programs, Turkey signed an IMF agreement in 1968 with reserves averaging 9.7 times monthly imports. The Turkish government signed again in 1969 when reserves were 11.7 times monthly imports, and then signed again in 1970.5 Why do countries enter into agreements with the IMF when they do not need foreign exchange? It is often assumed that, because IMF agreements limit national sovereignty by imposing specific conditions, governments will avoid the Fund if they do not need a loan. As Stanley Fischer notes, Policy conditionality can be interpreted as a . . . penalty, as seen from the viewpoint of the borrower countrys policy makers Fischer 1999.. In this view, governments accept conditions only when they need the IMF loan. Yet governments may enter into an IMF agreement not necessarily for a loan, but because they want conditions to be imposed Spaventa, 1983; Remmer, 1986; Vaubel, 1986, 1991; Putnam, 1988; Stein, 1992; Edwards and Santaella, 1993;

Other economic variables have also failed to provide a complete picture. Regarding inflation, Edwards and Santaella 1993. and Goldstein and Montiel 1986. found that countries entering IMF programs had experienced increasing inflation, but Santaella 1996., Knight and Santaella 1997., and Conway 1994. found inflation not to be a significant predictor of agreements. The importance of terms of trade is also a subject of divergent findings, with Conway 1994. and Santaella 1996. showing that it is significant, but Knight and Santaella 1997. finding it not important. Bird 1996. summarizes the consensus in the literature about variables that increase the likelihood of an IMF program: low levels of development, an overvalued exchange rate, and low GDP growth. Yet, as reported below, we do not find these variables to have significant effects. We do, however, find that high debt plays a role in entering programs as do Knight and Santaella 1997., Santaella 1996., and Conway 1994..

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

391

Bjork, 1995; Dixit, 1996.. Suppose that a government wants to restructure public finances but faces tough opposition see Vreeland, 1999 for details.. Tying the governments budget proposal to the conditions of an IMF agreement raises the costs for the domestic constituencies of rejecting the proposal. Turning down the policy is no longer a rejection of the government, but a rejection of the IMF, which is costly because it sends a negative signal to creditors and investors. Even though the government risks that the opposition will accuse it of selling out to the Fund and thus faces political sovereignty costs, the IMF agreement may enable the government to push through policies that otherwise would have been rejected. Note that in order for this strategy to be effective, the IMFs decision must be independent from the government. If the government and the Fund always collude, the threat that the IMF will impose rejection costs upon a country is not credible, and without this threat, the government gains no bargaining leverage. Thus, the IMF must make decisions separately from governments. We see the Fund as motivated by both technocratic and bureaucratic concerns. We agree with Vaubel 1986, 1991., who argues that the Fund seeks to maximize its budget, but we do not neglect the fact that the IMF is constrained by its mandate to promote international financial stability. Because the decisions of the government and the Fund are separate, participation in an IMF program is a joint decision Knight and Santaella 1997.. For a country to enter and to remain under IMF agreements, both the government and the IMF must want to do so, and they may have different reasons. A second complication is that the decisions to enter an agreement and the decisions to terminate it may have different determinants. As we noted above, contrary to what the IMF advertises that SBAs are intended to last 1 to 2 years and the other programs 3 to 4 countries remain under agreements an average of 4 to 5 years. Why does participation continue and why does it end? To distinguish between these questions, we study participation in IMF programs as a Markov process Amemiya, 1985.. Hence, four decisions are entailed in the interaction between a particular country and the Fund: 1. governments decision to seek an agreement, 2. IMFs decision to seek an agreement, 3. governments decision to terminate an agreement, and 4. IMFs decision to terminate it. Our statistical model, presented in Appendix A, identifies the parameters of the variables that affect these decisions Table 1.. To measure a governments need for a loan, we include RESERVES, the average annual foreign reserves in terms of monthly imports. RESERVES has a significant negative effect on the probability that the government will enter into an IMF agreement. Governments are more likely to participate in IMF arrangements when foreign reserves are low. To test whether governments turn to the Fund to lower the budget deficit, we include DEFICIT, the annual government budget surplus as a proportion of GDP.

392 A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Table 1 Determinants of program participation Determinants of entering Goernment Variable

CONSTANT

Determinants of remaining

Coefficient y2.27 y0.83 y0.95 1.38 y6.06 0.36 0.44 0.87 IMF

S.E. 0.611 0.424 0.277 0.516 1.789 0.212 0.178 0.288

Coefficient y0.01 y0.26 y0.29 0.65 y0.17 y0.36 0.38 y0.01

S.E. 0.592 0.475 0.329 0.678 1.922 0.266 0.190 0.352

Loan Conditions Rejection costs Sovereignty costs

RESERVES DEFICIT DEBT SERVICE INVESTMENT YEARS UNDER NUMBER UNDER LAGGED ELECTION

Variable
CONSTANT

Coefficient 2.14 y0.91 y0.73 0.43

S.E. 1.241 0.370 0.268 0.260

Coefficient 2.84 y0.41 y0.39 0.33

S.E. 2.016 0.230 0.429 0.273

Mandate Budget Negotiation costs

BOP NUMBER UNDER REGIME

Frequencies of actual and predicted outcomes Actual Predicted 0 0 1 TOTAL Number of observations Log likelihood function Restricted log likelihood Chi-squared Degrees of freedom Significance level 484 66 550 1024 y355.77 y705.46 699.39 23 0.0000 1 75 399 474 Total 559 465 1024 A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421 393

Variables: CONSTANT: to enter, 1; to remain, dummy variable lagged participation status; RESERVES: average annual foreign reserves in terms of monthly import requirements; DEFICIT: measured as surplus of government budget as a proportion of GDP; DEBT SERVICE: debt service as a proportion of GDP; INVESMENT: real gross domestic investment private and public. as a proportion of GDP; LAGGED ELECTION: dummy variable coded 1 for a lagged legislative election and 0 otherwise; YEARS UNDER: sum of past years under agreements for a country; NUMBER UNDER: number of other countries currently under IMF agreements; BOP: overall balance of trade in constant 1987 US$; REGIME: Dummy variable coded 1 for dictatorships and 0 for democracies. All variables are lagged 1 year.

394

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

DEFICIT which is measured as budget surplus, following the World Bank convention. has a significant effect on a governments decision to enter. Governments with high deficits are more likely to turn to the Fund. To capture rejection costs, the sensitivity of a country to the decisions of creditors and investors, we include DEBT SERVICE, the annual debt service as a proportion of GDP, and INVESTMENT, real gross domestic investment as a proportion of GDP. Rejection costs also matter. When DEBT SERVICE is high, a government is more likely to enter an IMF arrangement, and governments typically turn to the IMF when INVESTMENT is low. By bringing in the IMF, a government risks being accused of selling out. Thus we include sovereignty cost variables: YEARS UNDER, the number of past years a country has spent under IMF agreements; NUMBER UNDER, the number of other countries around the world currently participating in an IMF agreement, and LAGGED ELECTION, a dummy variable coded 1 if the previous year had a legislative election and 0 otherwise. When many governments in a countrys past have been under IMF agreements and when many other countries around the world are participating, a government is more likely to enter into an IMF program. Governments are more likely to enter agreements early in their electoral terms, hoping that the stigma of signing an agreement will be forgiven or forgotten before the next elections. The effect of the above variables on the governments decision to remain under agreements is surprising. None of the economic variables matter. The only variable which has a significant effect on the governments decision to continue participation in an IMF agreement is NUMBER UNDER. It has a positive effect. A government is more likely to continue participation when others do so.6 For the IMF, we include three variables: BOP, NUMBER UNDER, and REGIME. BOP, a countrys overall balance of payments in constant 1987 US dollars, is included to test whether the Fund is concerned with fulfilling its mandate of promoting world financial stability. The larger a balance of payments deficit in absolute terms, the more likely the Fund will provide a country with an arrangement. NUMBER UNDER is a proxy for the IMF budget.7 The significant effect of this variable shows that the IMF seeks to sign more countries when few are participating, when the budget constraint is less binding. REGIME, a dummy variable coded 1 for dictatorships and 0 for democracies, indicates the IMF is more likely to sign with dictatorships. This may be because they are easier to negotiate with. Governments have private information about the
6

In this specification, no other variables we tested affected the governments decisions to enter or leave: TERMS OF TRADE, BOP, CURRENT ACCOUNT, INFLATION, FOREIGN DIRECT INVESTMENT, EXPORTS as a capacity to import., EXCHANGE RATE, PER CAPITA INCOME, WORLD PER CAPITA INCOME GROWTH, OUTPUT GROWTH, CAPITAL STOCK GROWTH, LABOR FORCE GROWTH, and a dummy variable for post-1982 years. 7 There are, of course, better measures for the IMF budget, but the Fund keeps such information secret.

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

395

policies domestic constituencies will accept and can use this information in IMF negotiations to have their preferred policies imposed through the agreement. Such a strategy is more effective in democracies, where opposition may have more power. As Schelling 1960, 28. argues, the ability of a democratic government to get itself tied by public opinion may be different from the ability of a totalitarian government to incur such a commitment. Dictatorships are less constrained by public opinion and competitive elections, and thus they make easier negotiation partners. Regarding the decision of the Fund to retain countries under its programs, only BOP has a significant effect. This is not surprising. Once negotiations have been concluded, the IMFs costs of negotiation have been met. All that matters is whether balance of payments deficit continues to be large.8 Hence, governments typically seek IMF programs when they need a loan. But some governments also want conditions. Governments with high deficits enter into IMF arrangements to have fiscal discipline imposed upon them. They are likely to pursue this strategy when the costs of rejecting the IMF are high when they are sensitive to decisions of creditors and investors provided sovereignty costs are low. The Fund cares about the absolute size of a balance of payments deficit. Beyond this, it seeks to maximize its budget by signing as many countries as its resources allow.

4. The effect of programs on growth The average rate of growth of total income among the 1,024 annual observations for which we can specify the full selection model was 3.33%. Countries participating in IMF programs grew at the rate of 2.04% N s 465., while countries not under the programs at 4.39% N s 559.: a difference of 2.35%. The question is whether this difference is due to the conditions the countries faced or to program participation. Growth observed under IMF programs was lower regardless of the conditions under which countries participated. In Table 2 we classify these conditions by the size of the domestic deficit and of foreign reserves. Note that in 97 cases countries remained under IMF programs even though they had decent reserves and low deficits: those countries grew by 1.02% slower than countries which enjoyed the same conditions while not being subject to these programs. But even countries

Other variables that we tested were not significant in this specification:

TERMS OF TRADE, CURRENT

ACCOUNT, INFLATION, DEBT, FOREIGN DIRECT INVESTMENT, EXPORTS, EXCHANGE RATE, PER CAPITA INCOME, WORLD ECONOMIC GROWTH, CAPITAL STOCK GROWTH, LABOR FORCE GROWTH,

and the post-1982 dummy

variable.

396 A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Table 2 Growth according to observable conditions reserves and deficit. Reserves, deficit Good, good Good, bad Bad, good Bad, bad Total Not under Growth 5.22 4.65 4.00 2.19 4.39 Under Good, good Good, bad Bad, good Bad, bad Total 4.20 3.14 1.95 0.40 2.04 y2.25 y9.35 y2.07 y11.87 y7.34 4.26 3.34 1.06 0.89 2.10 6.08 5.46 6.65 7.54 6.65 97 89 97 182 465 Philippines 1994, Ecuador 1970 Solomon 1982, Ghana 1979 Mauritania 1988, Uganda 1988 Jordan 1989, Panama 1977 Togo 1989, Uruguay 1980 Mexico 1988, India 1981 Ivory Cst 1989, Gabon 1981 G. Bisau 1987, Guyana 1982 Deficit y0.84 y11.99 y1.96 y12.51 y5.30 Reserves 5.47 4.36 1.19 1.09 3.76 Debt service 3.50 3.83 3.76 5.29 3.90 N 248 121 102 88 559 Examples Typical Columbia 1988, Iran 1977 Rwanda 1988, Nepal 1984 Nigeria 1979, Dom. Rep. 1975 Nicaragua 1982, Zambia 1989 Extreme Botswana 1987, Malta 1975 C. Verde 1984, Iran 1981 Cameroon 1987, Benin 1979 Guyana 1986, Congo 1989

Good reserves: foreign reserves ) 2 times monthly imports. Bad reserves: foreign reserves F 2 times monthly imports. Good deficit: government budget surplus )y5% of GDP. Bad deficit: government budget surplus Fy5% of GDP.

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

397

with low reserves and high deficits did better if they did not participate: their growth was 1.79% faster. Thus, while countries facing bad conditions grew slower, participation in IMF programs lowered growth under all conditions. Yet matching the observed cases can be misleading. For one, the distribution of conditions is not the same for cases observed under and not IMF agreements and those observed not under. For two, unobserved variables may be responsible for the observed differences in performance. Hence, we proceed as follows. First, a growth model is estimated separately for countries observed under and those not under the agreements, with instruments to control for the effects of selection. The parameters resulting from this estimation are not biased by selection. Second, the vector of independent variables characterizing each country at each time is multiplied by the parameters characterizing the under and the not under samples. This procedure generates values of growth expected in the two situations independently of selection, for each country during each year. Finally, these selection-unbiased values are averaged over all countries, so that the difference between them is the net effect of IMF programs. A technical justification of these procedures, due originally to Heckman 1976, 1988., is included in Appendix B. The growth model used is a simple production function of the form Y s AF K , L. s AK a L b , to which we refer as barebones Table 3.. The model is estimated in growth form separately for j s 1 if the country is observed under agreement and j s 0 otherwise,
I rY s CONSTANTj q a j K rK . q b j L rL . q uG j lG Y j q uI j lj . I . where lG i and l 1 called hazard rates are the instruments used to control for the effects of selection. When selection is taken into account, IMF programs still appear to reduce growth. The average effect of program participation predicted by the full selection model and the barebones growth model in the 1,024 sample is y1.53. If all countries in this sample were under IMF programs every year, they would have grown at the average rate of 2.00. If none of the countries ever had an agreement, they would have grown at the rate of 3.53. Hence, of the observed difference of y2.35, y0.82 is due to the economic conditions and y1.53 to the effect of IMF programs. The countries that had IMF agreements experienced less favorable conditions, but the effect of IMF programs was adverse to growth independently of these conditions. Indeed, this difference is very close for the cases actually observed under and not under:

Observed as Under Not All

Hypothetically as Under 0.70 3.08 2.00 Not 2.33 4.52 3.53 Program effect y1.63 y1.44 y1.53

398

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Table 3 Barebones growth regression by participation status Under Selection-corrected

CONSTANT

Not under Observed means 1.00 2.01 2.80 0.14 0.12 Selection-corrected y0.13 0.38 0.44 0.02 0.56 0.02 0.07 0.23 0.09 0.29 3.53 5.50 559 1.89 0.59 1.68, p s 0.19. Observed means 1.00 7.15 2.69 y0.73 y0.82

Standard error

y1.73 0.44 0.47 0.01 0.53 0.01 4.31 1.48 6.17 2.23 2.00 5.93 465 1.75 0.71 0.00, p s 0.99.

rK K Standard error rL L Standard error lG Standard error lI Standard error r Y . E Y Standard deviation
Observations DurbinWatson Adjusted R 2 F-test

2.04 6.68

4.39 7.15

Note: F-test is for the restriction a q b s 1.

Experiments with other specifications and different samples, reported in Appendix C, generate the same qualitative results. Under different selection mechanisms, model specifications and samples, the effect ranges from y1.07 to y3.88. The difference in the predicted growth rates is due almost exclusively to the constants, indicating that the effects of unobserved variables operate through channels other than the supply of capital and labor see Heckman, 1979, 155.. When growth equations are estimated without selection instruments, the constants are almost the same for the observations not under y0.26. and under y0.38.. Yet, when countries are observed under agreements, the coefficients on G I the selection instruments, l1 and l1 , are large and significant. Hence, omitting these instruments strongly biases the constant. When corrected for selection, the difference between the constants becomes large. Hence, there are grounds to suspect that some unobserved factors influence both program participation and growth. One candidate is the political will of countries participating in IMF programs.9 Several students of the IMF have argued
9

Political will is merely one possible unobserved variable that affects both selection and performance. The reputation of a governments officials or advisors, its negotiation posture, or the populations trust in government may all lead a government to seek an IMF agreement and also affect growth.

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421 Table 4 Growth according to transition type Observations of countries Never under a spell Before spells Before and between spells Between spells Between and after spells After spells During spells N 82 142 346 204 335 131 465 YG 5.973 4.135 4.301 4.416 4.118 3.654 2.042 KSG 9.619 7.270 7.107 6.994 6.493 5.714 2.010 LFG 2.356 2.654 2.711 2.750 2.793 2.860 2.802 Reserves 5.895 2.786 3.219 3.521 3.653 3.858 2.101 Deficit y6.979 y4.792 y4.943 y5.048 y5.100 y5.180 y7.339

399

that agreements may break down because of a lack of political will Edwards, 1999; Killick, 1995; Nelson, 1991.. Political will may make countries more likely to continue participation and to grow. Moreover, not only are highly motivated governments more likely to continue IMF programs but the IMF is also predisposed to keep them. Yet while political will makes countries that participate grow faster than would participating countries not endowed with this will, or whatever else the unobserved factors may be, the same countries would have grown faster not participating in the IMF programs. Participation in IMF programs calls for several measures that reduce growth in the short run. But the standard argument in favor of these programs is that once the economy is stabilized, deficits are eliminated or reduced, and the balance of payments is improved, growth will resume. This argument is often only implicit and it is ideological: there are no good theoretical reasons to believe that a balanced budget and foreign account are sufficient for growth to occur. And it appears to be false. Contrary to Conway 1994., countries that emerge from IMF

Fig. 1. Observed growth before and after participation.

400 A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Table 5 The experience over time of countries that participated in IMF programs Consecutive years participating Years since leaving the IMF program 0 1 2 Growth First Second Third Fourth Fifth Sixth Seventh Eighth Cumulative annual growth for 8 years weighted average. 0.92 y1.63 1.90 y1.62 2.61 y1.63 2.50 y1.59 1.06 y1.72 3.62 y1.62 3.90 y1.50 1.47 y1.66 2.04 y1.63 N 82. 82. 66. 49. 37. 29. 25. 25. 5.81 y1.39 4.80 y1.61 4.26 y1.56 6.29 y1.50 5.51 y1.29 9.76 y1.04 6.53 y1.55 5.60 y1.47 Growth N 0. 14. 7. 11. 7. 4. 2. 5. 3.96 y1.52 5.19 y1.49 5.99 y1.44 5.18 y1.39 5.73 y1.37 5.38 y1.57 3.85 y1.47 5.03 y1.48 Growth N 0. 16. 17. 14. 9. 4. 7. 5. 3.71 y1.50 5.60 y1.42 4.45 y1.50 4.80 y1.43 6.02 y1.56 3.47 y1.50 2.22 y1.82 4.53 y1.50 3 Growth N 0. 23. 18. 15. 9. 9. 8. 3. 5.48 y1.42 4.81 y1.50 3.51 y1.47 6.66 y1.48 4.78 y1.46 5.39 y1.49 10.42 y0.85 5.43 y1.43 4 Growth N 0. 24. 18. 14. 15. 10. 6. 5. 4.70 y1.46 5.15 y1.49 4.57 y1.49 5.84 y1.45 5.44 y1.46 5.10 y1.48 6.15 y1.38 Cumulative growth over next 4 years weighted average. Observed If participation had continued 2.09 y1.63 2.42 y1.63 2.60 y1.61 2.40 y1.64 2.82 y1.59 2.38 y1.59 1.76 y1.64 1.93 y1.65

Ninth Tenth Eleventh Twelfth Thirteenth Fourteenth Fifteenth Cumulative annual growth for 15 years weighted average.

1.64 y1.56 1.95 y1.63 2.56 y1.70 1.67 y1.81 2.75 y1.71 2.68 y1.60 4.25 y1.47 2.04 y1.63

14. 10. 9. 7. 4. 6. 5. 465. 4.90 y1.49

0. 0. 0. 0. 0. 0. 0. 61. 4.06 y1.50

0. 0. 0. 0. 0. 0. 0. 43. 3.16 y1.55

0. 0. 0. 0. 0. 0. 0. 38. 6.76 y1.27

0. 0. 0. 0. 0. 0. 0. 32.

2.17 y1.70 2.37 y1.71 2.73 y1.66

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Top row for each consecutive year of participation: the observed rate of growth. Bottom row for each consecutive year of participation: the estimated effect of IMF programs on growth if the country participated.

401

402

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

programs do not grow faster than they did before having entered them or faster than countries that never entered. Again, consider first the observed patterns. In Table 4, we classify the observations according to their experience of participation. Note that these observations are right-hand, but not left-hand, censored: they reflect all the prior experience but end in 1990. Countries which never experienced IMF programs grew the fastest. But the most relevant comparison is of growth before and after program participation. And, whether or not we include cases in which a country would turn to the IMF again those cases observed between spells. program participation certainly does not accelerate growth. Indeed, for any combination of the before and after growth rates, the latter are somewhat lower. Neither can we detect any trend once countries leave a program. Fig. 1 shows a stylized picture of the experience with participation. The three horizontal lines show the average growth rates of countries that never participated, countries that were currently not participating, and countries currently participating which we stylized to last 5 years, about the average spell.. Both before and after growth rates exhibit wide swings and no trend emerges once countries leave programs. Statistical analysis confirms these visual impressions. Does it matter for future growth how long countries stay under the programs? Table 5 shows the observed rates of growth of all countries that ever entered and left IMF programs and the hypothetical differences described above for each consecutive year during and after program participation. Since this table is somewhat complicated, it bears an explanation. Take the second year under. There were 82 episodes of participating at least 2 years, the mean rate of growth during the second year of participation was 1.90, and if these countries had not participated, they would have grown 1.62 faster during this year. The next four columns of this row show the rates of growth observed during the subsequent 4 years for countries that left the program after 2 years and the hypothetical effect if participation had continued. Eye inspection of these numbers appears to indicate that countries that emerge from longer spells grow somewhat faster. But their observations are few, so that little can be inferred from these numbers. When a variable measuring the length of the past participation spell is added to the growth model for countries which are not under, it is never significant. Hence, it seems that duration of programs does not affect growth once a country leaves them. Perhaps the best way to conclude is to point to the last two columns of Table 5. They show the weighted average rate of growth during the subsequent 4 years of countries that left the program after different numbers of years with that of countries that continued to participate four more years. Their comparison shows that for any number of years already under the program, one through eight, the observed rates of growth during the next four years were lower if a country remained under than if it left. The selection-corrected, hypothetical effect of the program is also always negative. Thus both the observed and the hypothetical

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

403

differences indicate that countries are better off, at least during the next 4 years, leaving the program rather than continuing to participate. Hence, our conclusions are the following. When matched for exogenous conditions, participation in IMF programs reduces growth while a country remains under and has no salutary effect once a country leaves.

5. Conclusions Our findings appear robust to the selection mechanism, specification of growth equations, and samples. Governments facing economic difficulties adopt IMF programs either because they are desperate for foreign reserves, or because they want to use the IMF as a foil to reduce budget deficits, or both. These programs reduce growth while countries remain under and do not return benefits that would compensate the losses once they leave. These conclusions are not intended as a blanket indictment of the IMF. Balance of payment crises and exchange instability are facts of life, so the IMF has an important role to play. The question is whether coping with these crises must necessarily reduce growth. This is a broader, and more fundamental, question than the effect of IMF programs. Our results indicate that countries that do not enter into IMF programs grow faster than those that do even when both groups face high domestic deficits or foreign reserves crises. Hence, if soliciting IMF conditionality is just a way to impose domestically unpopular austerity policies, then the culprit is austerity policies per se, rather than the fact that they result from IMF agreements. Political will may just lead governments astray. The question whether or not austerity policies are necessary to restore growth is beyond the scope of this paper. We believe to have shown, however, that if growth is the primary objective then IMF programs are badly designed. Indeed, some of the sharpest critiques of these programs are intramural. Tanzi 1989. argued that IMF programs induce governments to save on public investment see also Tanzi and Davoodi, 1998., with nefarious consequences for growth. Blejer and Cheasty 1989. pointed out that the high real interest rates induce good firms to shut down along with bad ones. These effects seem to be avoidable. However, it appears that programs that stabilize inflation, reduce deficits, improve balance of payments, and at the same time do not hurt growth, are yet to be developed. Since our analysis ends in 1990, one may wonder whether the IMF has modified the designs of its programs since then. We could find no evidence that it has. In the last official statement we discovered, dated October 1993 IMF Surey: Supplement on the IMF 1993, 13 ., the Fund reported that The Executive Board undertakes periodic reviews of conditionality and, on many occasions, it has adjusted the policies and practices relating to the use of the IMFs resources. In its most recent discussion of issues related to conditionality and program design, in

404

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

July 1991, the Executive Board affirmed that the current guidelines on conditionality, which the Board adopted in 1979, remain broadly appropriate.

Acknowledgements We appreciate comments by Jess Benhabib, David Denoon, Martin Edwards, Jennifer Gandhi, Manuel Pastor, David Rowe, Alastair Smith, Elisabeth Wood, and the anonymous reviewer.

Appendix A. Dynamic bivariate probit model with partial observability Let Si t denote participation status of country i at time t , Si t sUnder if country i is under agreement at time t , and Si t sNot Under otherwise. To simplify notation, write the probability that Si t s Under as pUi t . and the probability that Si t s Not Under as p Ni t .. Let pN U , i t denote the probability that country i enters an agreement at time t i.e., that it goes from not under at time t y 1 to under at time t .. The probability that country i does not sign at time t is pN N , i t s 1 y pN U , i t . Similarly, pU U , i t denotes the probability that country i stays under at time t. The probability that participation ends at time t is thus pU N , i t s 1 y pU U , i t . These transition probabilities vary across time and country, and thus are subscripted i , t to designate individual country-years. With lagged participation status Si , ty 1 . and these transition probabilities, one can describe participation as a first-order Markov process:10 p Ui t N Si , ty 1 . p Ni t N Si , ty 1 . s pN U , i t pN N , i t pU U , i t Ni , ty 1 pU N , i t Ui , ty 1

where Ui , ty 1 is an indicator variable set equal to 1 if country i was under at time t y 1, and 0 otherwise. Similarly, Ni , ty 1 is a dichotomous variable set to 1 if country i was not under at time t y 1 and 0 otherwise, Ni , ty 1 s 1 y Ui , ty 1. The probability of participating at time t , pUi t ., is thus p Ui t N Si , ty 1 . s pN U , i t Ni , ty 1 q pU U , i t Ui , ty 1 s pN U , i t q pU U , i t y pN U , i t . Ui , ty 1 ,
10

A1.

This characterization of a first-order Markov process follows Amemiya 1985, Chapter 11.

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

405

and the probability of not being under agreement at time t is the complement of the above to 1. The decisions to enter and to continue IMF agreements are joint decisions between a government and the Fund. For the governments decision, we write the value of participation as the latent regression:
X G X G ) G d iG t s g x i , ty 1 q k x i , ty 1Ui , ty 1 q i t ,

A2.

where the effects of the vector of variables determining the value of participation, x iG , ty 1 , are captured by g if the country was not under an agreement at time t y 1 and by if the country was under. We write g q a . as a shift for convenience, as will be seen below. The effect of unobserved variables determining the value of participation for the government is captured by iG t , which we assume to be normally distributed. The government wants to be under an IMF agreement if and ) only if the value of participation is positive, d iG t ) 0. We define the value of an agreement to the IMF with a similar equation, d iIt) s mX x iI, ty 1 q b X x iI, ty 1Ui , ty 1 q iIt ,

A3.

where the effect of the vector of variables x iI, ty 1 is captured by the vector m if Ui , ty 1 s 0 and by m q b . if Ui , ty 1 s 1. Unobserved effects are captured by iIt , also assumed to be normally distributed. If we assume the unobserved variables that influence the government and the IMF are independent, we can write the probability of entering an IMF agreement as
X I pN U , i t s F g X x iG , ty 1 . F m x i , ty 1 . ,

A4.

where F P. is the CDF of the standard normal distribution. The probability of continued participation can be written as:
I pU U , i t s F g q k . x iG , ty 1 F m q h . x i , ty 1 . X X

A5.

Substituting Eqs. A4. and A5. into Eq. A1. yields Eq. A6.:
X I p Ui t N Si , ty 1 . s F g X x iG , ty 1 . F m x i , ty 1 . I q F g q k . x iG , ty 1 F m q h . x i , ty 1 X I yF g X x iG , ty 1 . F m x i , ty 1 . 4 Ui , ty 1 X X

A6.

The likelihood function can be formed by using the above Eq. A6.. However, after some tedious algebra, Eq. A6. can be rewritten in a simplified form:
X G p Ui t N Si , ty 1 . s F g X x iG , ty 1 q Ui , ty 1 k x i , ty 1 .

= F mX x iI, ty 1 q Ui , ty 1hX x iI, ty 1 . .

A7.

406

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

The likelihood function is thus

X G X I X I L s F g X x iG , ty 1 q Ui , ty 1 k x i , ty 1 . F m x i , ty 1 q Ui , ty 1h x i , ty 1 . Ui t

i,t

X G = 1 y F g X x iG , ty 1 q Ui , ty 1 k x i , ty 1 .

=F mX x iI, ty 1 q Ui , ty 1hX x iI, ty 1 .

1 y U i t .

A8.

which can be estimated by a canned version of the Abowd and Farber 1982. bivariate probit model with partial observability as is found in LIMDEP.. If the vectors X G and X I contain exactly the same set of variables, then the model cannot be identified. To identify the parameters g , m , k , and h , it is sufficient that there be at least one variable not in common between X G and X I.

Appendix B. Correcting for selection bias Let Y stand for the rate of growth and X for some vector of observable variables. Assume that y is a function of x , with the population density f y N x .. Write the regression equations as suppressing the i, t subscripts. yj s b jX x q e j ,

B1.

where j s 1 if the country-year observation is under agreement, j s 2,3,44 otherwise. The expected value of y 1 for the observed sample is E yu N x , Y1 is observed. s E y 1 N x , d G ) )0, d I ) )0 .
X s b1 x q E e1 N d G ) ) 0, d I ) ) 0 . ,

B2.

where E e1 N d G ) ) 0, d I ) ) 0. is the conditional expectation of e1 given that y 1 is observed. On the other hand, the expected value for the population is
X x. E y1 N x . s b1

B3.

Hence, the value expected in the sample differs from the population value by the conditional expectations of the error term given what we observe. If selection I is not random, then e1 is correlated with either or both. iG t or i t , the effects of unobserved variables on the respective decisions of the government and the IMF to participate from Eqs. A2. and A3. in Appendix A..

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

407

The expected value for the selected sample can be derived as follows see Poirier, 1980, 216 and Greene, 1993.. Assuming that V G and V I are independent, we can write the expected value of E e1 N d G ) ) 0, d I ) ) 0. as:
X X G G) I I) E e1 N d G ) ) 0, d I ) ) 0 . s u 1 ) 0. q u 1 ) 0. , G E N d I E N d

B4.
where E N d and
I E I N d I ) ) 0 . s l1 s G G)

) 0.

G s l1 s

f g q k . x iG , ty 1 F g q k . x iG , ty 1
X X

f m q h . x iI, ty 1 F m q h . x iI, ty 1
X

I . Hence the variables lG 1 and l 1 are the hazard rates from Eq. A8 . We can thus write the expected value in the observed sample as X X X G I E y 1 N d G ) ) 0, d I ) ) 0 . s b 1 x qu1 G l1 q u 1 I l1 .

B5.

Note that if B5. is estimated on the basis of the observed sample, the variables lG and l I are omitted from the specification. Hence, selection is a source of omitted variable bias Heckman, 1979.. We can now also understand why controlling matching. for the variables that enter both into selection and outcome equations may in fact exacerbate, rather than attenuate, the selection bias Achen, 1986.. Following Heckman 1988., distinguish first between selection on observables and on unobservables. Selection on observables occurs when the expected covariance E e1 j . / 0 j g G, I 4., but once the observed variables X j are controlled the covariance vanishes, so that E e1 j N X j . s 0. Selection is on unobservables when E e1 j N X j . / 0, so that controlling the factors observed by the investigator does not remove the covariance between the errors in the outcome and the selection equations. Now, note that the regression coefficient u 1 s cov e1 , j .rvar e1 .. If selection is on unobservables, controlling for some variable x in the outcome equation may reduce the error variance e1 without equally reducing the covariance e1 and j. Hence, the coefficient on the omitted variable will be larger and the bias will be exacerbated. Note that correcting for selection bias for country-years observed not under is not straightforward. There are a total of four states of the world. In state 1., where country-years observed under an IMF agreement, correction is straightforward: 1. E y 1 N x , observe an agreement. s
X E y 1 N x , d G ) ) 0, d I ) ) 0 . s b 1 x q E e1 N d G ) ) 0, d I ) ) 0 . ,

where both the government and the IMF want the agreement.

408

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

For country-years observed not under , however, there are three possible states of the world: X 2. E y 2 N x , observe no agreement. s E y 2 N x , d G ) - 0, d I ) - 0. s b 2 x q E e 2 N G) I) d - 0, d - 0., X 3. E y 3 N x , observe no agreement. s E y 3 N x , d G ) - 0, d I ) ) 0. s b 3 x q E e 3 N G) I) d - 0, d ) 0., X 4. E y4 N x , observe no agreement. s E y4 N x , d G ) ) 0, d I ) - 0. s b4 x q E e 4 N G) I) d ) 0, d - 0.. In 2., neither the IMF nor the government wants to be under. In 3., the IMF wants an agreement, but the government does not want to be under. In 4., the government wants the agreement, and the IMF does not. The problem is that since we do not observe the actors individual decisions, we do not know how to break up the not under observations respectively between states 2. through 4.. We know that E G N d G ) - 0 . s lG 0s and
I E I N d I ) - 0. s l0 s

yf g X x iG , ty 1 . 1 y F g X x iG , ty 1 .

yf mX x iI, ty 1 . 1 y F mX x iI, ty 1 .

but we do not observe when to use which pair of instruments when an observation is not under:
I G I G I lG 0 , l 0 . , l 0 , l1 . , or l1 , l 0 . .

We proceed by trying three different assumptions, assuming that all observations belong to state 2., all observations belong to state 3., or all observations belong to state 4.:
X X X G I E y1 N d s 1. s b1 x q uG 1 l1 q u I 1 l1 , X X X G I E y2 N d s 0. s b 2 x q uG 2 l0 q u I 2 l0 , X X X G I E y3 N d s 0. s b 3 x q uG 3 l 0 q u I 3 l1 , X X X I G E y4 N d s 0 . s b4 x q uG 4 l1 q u I 4 l 0 .

This produces four sets of b s with which to estimate growth according to four possible states of the world. The expected average rates of growth are

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

409

1. Growth under both want agreement. 2. Growth not under because both 3. Growth not under because of government 4. Growth not under because of IMF

2.00 3.53 3.66 3.19

The effect of the IMF ranges from y1.19 to y1.66. In the main body of the paper, we report the effect of IMF programs as the difference between state 1. and state 2.. Appendix C. Tests for robustness Our robustness tests are best presented as a table, in which the entries are the selection-corrected differences between average growth expected under and not under agreements: Growth model Selection modelrsample Full Exogenous N s 1024 Barebones rK With endogenous K Panel one-way fixed. y1.53 y3.88 y2.29 y1.33 y3.68 y1.96 Endogenous Stripped Post-1970 N s 2607 y1.33 y2.67 y1.07 Large N s 3991 y1.95 y3.05 y1.29

The Full selection model is the one presented in Table 1 in the main body of the text. Since several observations necessary to estimate this model are not available, we also estimate a Stripped selection model, which includes only those variables that are available for the entire sample. RESERVES, DEBT SERVICE and BOP are not available and instead of DEFICIT we use government consumption as a proportion of GDP. Other variables are the same as in the full selection model. Selection is exogenous if it is not driven by the expected consequences: hence, the predetermined variables in the growth equation do not enter into the reduced form of the selection equation. If the expected consequences do enter into the structural form of the model, selection is endogenous, and the right-hand side variables of the growth equation enter into the reduced form of the selection equation. Obviously, some variables must appear in the selection but not in the growth equation for the model to be identified. Samples differ according to the availability of data and periods. No data concerning the basic selection variables are available for the pre-1970 period and more than one half observations are missing for the post-1970 period. Hence, the Full selection model is estimated for the smallest, only post-1970, number of observations N s 1024.. To test whether our results depend on period and data availability, a Stripped selection model combined with the Barebones

410

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

growth model is estimated for the same period N s 2607. and for the entire 19511990 period N s 3991.. This is particularly important because our sample is strongly biased by the availability of data: the 466 observations of countries under agreements which enter into our Full selection sample had an average rate of growth of 2.05, while the 614 observations of countries under for which other data are not available had an average rate of 4.11. Statistical tests both F-test and x 2 . indicate for all samples and model specifications that the not under sample is not heterogeneous but that the under sample exhibits fixed, country, effects. In the Panel line of the table, we follow these tests, that is, we use OLS for the not under observations and one-way fixed effects panel for the under. If we were to calculate the average effect if both groups were estimated by panels, the result would be y3.29 for the 1024 sample. To test the robustness of the specification of the growth model, we experimented by adding to the basic model the balance of payments as a proportion of GDP, reserves, deficit, debt service, terms of trade, and the rate of growth of education of an average member of the labor force, per capita income, the average rate of growth in the world during the year, the percentage difference in output per worker between a given country and the maximum product per worker during this year, total number of years a country was under agreements in the past and, for countries not currently under, the duration of their last stint under IMF. None of these variables is significant when the appropriate estimators are used. Finally, note that the procedure in which we match observations for the values of the predetermined variables assumes that they are exogenous with regard to the states under which these observations are made. If participation affects the values of these variables, then the counterfactual had the countries been observed under the same conditions is no longer valid. Having repeated the same procedure with regard to the predetermined variables, we learned that the rate of growth of capital rK , is lower when countries are under an agreement. We cannot tell, stock, K however, whether this is an effect of programs or of the sample for which other relevant variables are available. Again, our sample is strongly biased against rK : the mean rate of growth of capital stock for all the countries with higher K under observations in the large, 19511990, sample is 7.15 for the subset for which other data are missing and only 2.01 for the observations for which other information is available. Hence, we do not know if the rate of growth of capital rK are based stock can be treated as exogenous. The lines with endogenous K on the assumption that participation affects this variable and the expected rates of rK at its mean for each participation status. growth are calculated by taking K When the rate of growth of capital stock is considered to be an endogenous effect of IMF programs, the effect of these programs appears much larger. Hence, if anything, the results reported in the text are conservative. Indeed, what is striking is that IMF programs have a negative effect on growth even if we assume investment to be exogenous with regard to participation.

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

411

Appendix D. Country-years in sample D.1. 4126 Obseration-sample: 135 countries 19511990 Country Algeria Angola Benin Botswana Burkina Faso Burundi Years in sample 1962 1975 1960 1966 1960 1962 1990 1989 1990 1989 1990 1990 Spell of agreements Start End 1989 1990 Never under 1989 1990 Never under Never under 1965 1971 1976 1977 1986 1989 1988 1990 Never under 1980 1981 1983 1990 1987 1990 Never under 1967 1968 1977 1977 1979 1980 1986 1988 1990 1990 Never under 1977 1981 1987 1988 1981 1982 1978 1982 1986 1990 1977 1980 1982 1990 1966 1970 1979 1980 1983 1985 1987 1990 1982 1983 1986 1990 1987 1990 1981 1990 1975 1986 1988 1990

Cameroon Cape Verde Island Central African Republic Chad Comoro Island Congo

1961 1975 1961 1961 1975 1961

1990 1990 1990 1990 1990 1990

Djibouti Egypt Ethiopia Gabon Gambia Ghana

1977 1951 1951 1961 1965 1957

1987 1990 1986 1990 1990 1990

Guinea Guinea-Bisau Ivory Coast Kenya

1960 1974 1961 1963

1990 1990 1990 1990

412

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Lesotho Liberia Madagascar Malawi Mali

1966 1961 1961 1964 1961

1990 1986 1990 1990 1990

Mauritania

1961

1990

Mauritius Morocco

1968 1956

1990 1990

Mozambique Niger Nigeria Rwanda Senegal Seychelle Sierra Leone

1975 1961 1960 1962 1961 1976 1962

1990 1989 1990 1990 1990 1990 1990

Somalia South Africa

1961 1951

1989 1990

Sudan Swaziland Tanzania

1971 1968 1961

1990 1989 1988

Togo Tunisea

1961 1961

1990 1990

1988 1963 1979 1977 1980 1979 1988 1964 1967 1982 1977 1980 1985 1979 1959 1965 1980 1987 1983 1987 1966 1979 1979 Never under 1966 1969 1977 1984 1964 1980 1958 1961 1976 1982 1972 1979 Never under 1975 1980 1986 1979 1964 1986

1990 1977 1986 1978 1990 1986 1990 1965 1972 1990 1978 1982 1990 1986 1960 1972 1990 1990 1989 1990 1970 1980 1990 1967 1970 1982 1989 1971 1989 1959 1962 1977 1983 1975 1985 1976 1982 1988 1990 1970 1990

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

413

Uganda

1962

1990

Zaire Zambia Zimbabwe Bahamas Barbados Belize Canada Costa Rica Dominican Republic El Salvador

1960 1964 1965 1978 1966 1981 1951 1951 1951 1951

1989 1990 1990 1987 1989 1990 1990 1990 1990 1990

Grenada Guatemala

1985 1951

1990 1990

Haiti Honduras

1961 1951

1989 1990

Jamaica

1962

1990

Mexico

1951

1990

Nicaragua

1951

1990

Panama Trinidad and Tobago

1951 1962

1990 1990

1971 1980 1987 1976 1973 1976 1981 Never under 1982 1984 Never under 1980 1964 1983 1958 1980 1990 Never under 1960 1966 1981 1988 1961 1970 1957 1968 1979 1990 1963 1973 1977 1954 1959 1961 1977 1983 1956 1963 1968 1979 1965 1968 1989

1972 1984 1990 1989 1974 1987 1984 1984 1986 1990 1965 1986 1973 1983 1990 1962 1973 1984 1990 1967 1989 1966 1973 1983 1990 1964 1974 1990 1955 1959 1962 1979 1990 1961 1965 1973 1979 1966 1987 1990

414

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

USA Argentina

1951 1951

1990 1990

Bolivia

1951

1990

Brazil

1951

1990

Chile

1951

1990

Columbia Ecuador

1951 1951

1990 1990

Guyana Paraguay Peru

1966 1951 1951

1990 1990 1990

Suriname Uruguay

1975 1951

1989 1990

Venezuela Bangladesh

1951 1971

1990 1990

China India

1961 1951

1990 1990

1963 1958 1967 1976 1983 1956 1973 1980 1986 1958 1961 1965 1983 1988 1956 1961 1974 1983 1957 1961 1969 1983 1967 1990 1957 1954 1977 1982 Never under 1961 1966 1975 1990 1960 1989 1974 1979 1985 1981 1986 1957 1962 1981

1965 1963 1969 1977 1990 1970 1974 1981 1990 1959 1962 1973 1986 1990 1959 1970 1976 1990 1974 1967 1973 1990 1982 1990 1969 1971 1980 1985 1963 1973 1987 1990 1961 1990 1976 1983 1990 1981 1987 1958 1966 1984

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

415

Indonesia Iran Iraq Israel Japan Jordan South Korea Laos Malaysia Mongolia Myanmar

1961 1956 1954 1954 1952 1955 1954 1985 1957 1985 1951

1990 1990 1987 1990 1990 1990 1990 1990 1990 1990 1989

Nepal Pakistan

1961 1951

1986 1990

Philippines

1951

1990

Singapore Sri Lanka

1965 1951

1990 1990

Syria Taiwan Thailand

1961 1952 1951

1990 1990 1990

Yemen Arab Republic Austria

1970 1951

1989 1990

1961 1968 1956 1960 Never under 1974 1962 1989 1965 1980 1989 Never under Never under 1969 1973 1977 1981 1976 1985 1958 1965 1968 1972 1977 1980 1988 1962 1973 1983 Never under 1965 1974 1977 1983 1988 1962 1964 Never under 1978 1981 1985 Never under Never under

1964 1974 1956 1962 1977 1965 1990 1977 1987 1990

1970 1975 1979 1982 1977 1986 1959 1966 1969 1975 1978 1983 1990 1965 1981 1990 1972 1975 1981 1984 1990 1962 1964 1979 1983 1986

416

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Belgium Bulgaria Czechoslovakia Denmark Finland

1951 1981 1961 1951 1951

1990 1990 1990 1990 1990

France Germany East Germany Greece Hungary Iceland Ireland Italy Luxembourg Malta Netherlands Norway Poland Portugal Romania Spain Sweden Switzerland Turkey UK

1951 1951 1971 1951 1971 1951 1951 1951 1951 1964 1951 1951 1971 1951 1961 1951 1951 1951 1951 1951

1990 1990 1988 1990 1990 1990 1990 1990 1990 1989 1990 1990 1990 1990 1989 1990 1990 1990 1990 1990

USSR Yugoslavia

1961 1961

1989 1990

1952 Never under Never under Never under 1953 1967 1975 1956 1969 Never under Never under Never under 1982 1988 1960 Never under 1974 1977 Never under Never under 1957 Never under 1990 1977 1983 1975 1981 1959 1978 Never under Never under 1961 1978 1956 1961 1967 1975 Never under 1961 1965 1971 1988

1957

1953 1968 1976 1959 1970

1985 1990 1963 1975 1978

1958 1990 1979 1985 1978 1984 1961 1979

1971 1985 1959 1965 1970 1979 1961 1967 1986 1990

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

417

Australia Fiji New Zealand Papua New Guinea Solomon Islands Vanuatu Western Samoa

1951 1970 1951 1975 1981 1984 1980

1990 1990 1990 1990 1988 1990 1990

1961 1974 1967 1990 1981 Never under 1980 1983

1961 1975 1968 1990 1984 1980 1985

D.2. 1024 Obseration-sample: 79 countries 19711990 Country Benin Botswana Burkina Faso Burundi Cameroon Cape Verde Central African Republic Chad Congo Egypt, Arab Republic Ethiopia Gabon Years in sample 1978 1977 1974 1986 1976 1986 1979 1981 1973 1984 1972 1981 1976 1984 1974 1980 1990 1974 1973 1987 1980 1973 1989 1987 1989 1989 1981 1990 1989 1988 1977 1990 1972 1989 1990 1986 1977 1988 1990 1990 1983 1990 1990 1990 Agreements Start End 1989 1989 Never under Never under 1986 1989 1988 1990 Never under 1981 1981 1983 1988 1987 1990 1986 1988

1977 1981 1987 1988 Never under 1980 1982 1986 1990 1977 1982 1979 1983 1987 1981 1975 1988 1980 1990 1980 1983 1990 1990 1986 1990

Gambia, The Ghana Guinea-Bissau Cote dIvoire Kenya

418

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Lesotho Liberia Madagascar Malawi Mali Mauritania

1983 1976 1973 1981 1972 1978 1976 1983 1977 1985 1977 1973 1983 1974 1971 1976 1978 1975 1973 1973 1975 1984 1978 1973 1984 1972 1973 1978 1973 1973 1973 1971 1973 1990

1990 1986 1975 1990 1989 1989 1980 1989 1990 1989 1987 1979 1990 1990 1974 1976 1980 1990 1989 1990 1989 1988 1989 1990 1989 1989 1990 1990 1989 1988 1990 1982 1986 1990

1988 1976 1979 1981 1979 1988 1982 1977 1980 1985 1979 1987 1983 1987 1979 1979

1990 1977 1986 1990 1986 1989 1989 1978 1980 1989 1986 1989 1987 1990 1980 1980

Mauritius Mozambique Niger Nigeria Rwanda Senegal

Sierra Leone Somalia Sudan Swaziland Tanzania Togo Tunisia Uganda Zaire Zambia Zimbabwe Barbados Costa Rica Dominican Republic El Salvador Guatemala

1977 1984 1980 1973 1979 Never under 1986 1979 1986 1984 1987 1976 1973 1976 1981 1982 1980 1983 1971 1980 1973 1981 1990

1982 1989 1989 1975 1985 1988 1989 1990 1984 1989 1989 1974 1987 1984 1984 1988 1986 1973 1982 1973 1984 1990

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

419

Haiti Honduras Jamaica Mexico Nicaragua Panama Trinidad and Tobago Brazil Chile Colombia Ecuador Guyana Paraguay Peru Uruguay Venezuela Bangladesh

1981 1973 1976 1973 1971 1974 1977 1979 1981 1973 1972 1974 1971 1973 1971 1973 1971 1974

1988 1988 1988 1990 1984 1990 1990 1979 1990 1983 1990 1990 1986 1990 1981 1983 1990 1990

India Indonesia Iran, Islamic Republic Jordan Malaysia Nepal Pakistan

1975 1973 1975 1984 1973 1977 1974

1990 1983 1983 1990 1984 1986 1990

Philippines Sri Lanka

1973 1971 1974

1990 1972 1990

1981 1973 1979 1977 1977 1983 1971 1979 1974 1989 1983 1988 1974 1983 1972 1983 1971 Never under 1971 1977 1973 1975 1989 1974 1979 1985 1981 1973 Never under 1989 Never under 1977 1985 1974 1977 1980 1988 1973 1983 1971 1977 1983 1988

1988 1973 1983 1988 1979 1990 1973 1979 1987 1990 1986 1990 1976 1983 1974 1990 1982 1971 1980 1973 1983 1990 1976 1983 1990 1984 1974 1990 1977 1986 1975 1978 1983 1990 1981 1990 1975 1981 1984 1990

420

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

Syrian Arab Republic Thailand

1986 1973

1989 1987

Malta Turkey Fiji Papua New Guinea Solomon Islands Vanuatu

1973 1981 1971 1971 1976 1982 1985

1979 1987 1987 1987 1987 1987 1987

Never under 1978 1979 1981 1983 1985 1986 Never under 1971 1978 1974 Never under 1982 Never under 1971 1985 1975 1984

References
Abowd, J.M., Farber, H.S., 1982. Job queues and the union status of workers. Industrial and Labor Relations Review 35, 354367. Achen, C.H., 1986. The Statistical Analysis of Quasi-Experiments. University of California Press, Berkeley. Alvarez, M., Cheibub, J.A., Limongi, F., Przeworski, A., 1996. Classifying political regimes for the ACLP data set. Studies in International Comparative Development 31, 336. Amemiya, T., 1985. Advanced Econometrics. Harvard University Press, Cambridge, MA. Bird, G., 1996. Borrowing from the IMF: the policy implications of recent empirical research. World Development 24, 17531760. Bjork, J., 1995. The uses of conditionality. East European Quarterly 29, 89124. Blejer, M.I., Cheasty, A., 1989. Fiscal policy and mobilization of savings for growth. In: Blejer, M.I., Chu, K.-Y. Eds.., Fiscal Policy, Stabilization, and Growth in Developing Countries. IMF, Washington DC. Camdessus, M., 1990. Statement before the United Nations Economic and Social Council in Geneva, July 11. IMF Survey 19 15., 235. Financial support for member countries complements economic policy changes. In: Cheney, D.M. Ed.., IMF Survey: Supplement on the IMF. 1993. pp. 1315. Connors, T.A., 1979. The Apparent Effects of Recent IMF Stabilization Programs. International Finance Discussion Papers 135. Board of Governors of the Federal Reserve System. de Vries, M., 1986. The IMF in a changing world: 19451985. IMF, Washington, DC. Dixit, A.K., 1996. The Making of Economic Policy: A Transaction-Cost Politics Perspective. MIT Press, Cambridge. Edwards, M.S., 1999. Things Fall Apart: Why Do IMF Agreements Break Down? Paper Prepared for the Duke University Center for International Studies Conference International Institutions: Global ProcessesrDomestic Consequences, Durham April 911. Edwards, S., Santaella, J.A., 1993. Devaluation controversies in the developing countries: lessons from the Bretton Woods Era. In: Bordo, M.D., Eichengreen, B. Eds.., A Retrospective on the Bretton Woods System. University of Chicago, Chicago, pp. 405455. Fischer, S., 1999. On the Need for an International Lender of Last Resort. Paper prepared for delivery at the joint luncheon of the American Economic Association and the American Finance Association, New York, January 3 1999. Available at www.imf.org. Goldstein, M., Montiel, P.J., 1986. Evaluating fund stabilization programs with multicountry data: some methodological pitfalls. IMF Staff Papers 33, 304344.

A. Przeworski, J.R. Vreeland r Journal of Deelopment Economics 62 (2000) 385421

421

Greene, W.H., 1993. Econometric Analysis. 2nd edn. Prentice Hall, Englewood Cliffs, NJ. Gylfason, T., 1987. Credit Policy and Economic Activity in Developing Countries with IMF Stabilization Programs. Studies in International Finance 60, Princeton. Heckman, J.J., 1976. The common structure of statistical models of truncation, sample selection, and limited dependent variables and a simple estimator for such models. Annals of Economic and Social Measurement 5, 475492. Heckman, J.J., 1979. Sample selection bias as a specification error. Econometrica 47, 153161. Heckman, J.J., 1988. The microeconomic evaluation of social programs and economic institutions. In: Chung-Hua Series of Lectures by Invited Eminent Economists, No. 14. The Institute of Economics, Academia Sinica, Taipei. Khan, M.S., 1990. The macroeconomic effects of fund-supported adjustment programs. IMF Staff Papers 37 2., 195234. Killick, T., 1995. IMF Programs in Developing Countries: Design and Impact. Routledge, London. Knight, M., Santaella, J.A., 1997. Economic determinants of fund financial arrangements. Journal of Development Economics 54, 405436. Nelson, J. Ed.., Economic Crisis and Policy Choice Princeton University Press, Princeton. Pastor, M., 1987a. The International Monetary Fund and Latin America: Economic Stabilization and Class Conflict. Westview Press, Boulder. Pastor, M., 1987b. The effects of IMF programs in the third world: debate and evidence from Latin America. World Development 15, 365391. Poirier, D.J., 1980. Partial observability in bivariate probit models. Journal of Econometrics 12, 209217. Polak, J.J., 1991. The Changing Nature of IMF Conditionality. Princeton University, International Finance Section, Department of Economics, Princeton. Przeworski, A., Limongi, F., 1996. Selection, Counterfactuals and Comparisons. Working Paper, March 11. Putnam, R.D., 1988. Diplomacy and domestic politics: the logic of two-level games. International Organization 42, 427460. Reichmann, T.M., Stillson, R.T., 1978. Experience with programs of balance of payments adjustment: stand-by arrangements in the highest tranches, 196372. IMF Staff Papers 25, 292310. Remmer, K.L., 1986. The politics of economic stabilization, IMF standby programs in Latin America, 19541984. Comparative Politics 19, 124. Santaella, J.A., 1996. Stylized facts before IMF-supported adjustment. IMF Staff Papers 43 3., 502544. Schelling, T.C., 1960. The Strategy of Conflict. Harvard University Press, Cambridge. Spaventa, L., 1983. Two letters of intent: external crises and stabilization policy, Italy, 197377, In: Williamson, J. Ed.., IMF Conditionality Institute for International Economics, Washington, DC. Stein, H., 1992. Economic policy and the IMF in Tanzania: conditionality, conflict, and convergence. In: Campbell, H., Stein, H. Eds.., Tanzania and the IMF: The Dynamics of Liberalization. Westview, Boulder, pp. 5983. Tanzi, V., 1989. Fiscal policy, stabilization and growth. In: Blejer, M.I., Chu, K.-Y. Eds.., Fiscal Policy, Stabilization, and Growth in Developing Countries IMF, Washington, DC. Tanzi, V., Davoodi, H., 1998. Roads to Nowhere: How Corruption in Public Investment Hurts Growth. IMF, Washington, DC. Taylor, L. Ed.., The Rocky Road to Reform: Adjustment, Income Distribution, and Growth in the Developing World The MIT Press, Cambridge. Vaubel, R., 1986. A public choice approach to international organization. Public 51. Vaubel, R., 1991. Problems at the IMF. Swiss Review of World Affairs 40 12., 2022. Vreeland, J., 1999. The IMF: lender of last resort or scapegoat? Prepared for the International Studies Association Conference, Omni Shoreham Hotel, Washington, DC, February 1621.