Chapter

CHAPTER 15 Evaluating IMF-Supported Programs in the 1990s: The Importance of Taking Explicit Account of Stoppages

Author(s):
Alessandro Rebucci, and Ashoka Mody
Published Date:
April 2006
Share
  • ShareShare
Show Summary Details
Author(s)
Chuling Chen and Alun Thomas1 

This paper contributes to the literature on evaluating the impact of IMF-supported programs on three key macroeconomic variables: inflation, the budget position, and growth. The paper documents the importance of distinguishing between completed programs and those that terminated prematurely, showing that they have significantly different impacts on these targets. While stopped programs are associated with higher inflation and budget deficits, and lower growth relative to periods without a program, completed programs are marginally associated with increased growth three years after the termination of the program.

Introduction

During the past two decades, a number of studies have explored whether IMF programs are effective in improving participating countries’ current account, inflation, and growth outcomes. These studies developed different methodologies and used various datasets, and while coming to different conclusions on specific economic variables, the general thrust of the results is that IMF programs do not generally affect growth or inflation. This paper aims to provide some new insights on the effectiveness of IMF programs by using a new database for countries that did not participate in any IMF programs during the 1980s but were engaged in one or more programs in the 1990s. The sample characteristics of the data were chosen to better capture the independent effect of IMF programs on three target variables: inflation, budget conditions, and growth.

One of the major novelties of this approach is that it emphasizes the importance of implementing IMF programs by distinguishing between programs that were implemented successfully and those that were stopped prematurely. Surprisingly, this distinction has been largely absent from previous analyses of the effects of IMF programs. Indeed, stoppages are fairly prevalent in IMF programs, amounting to almost 40 percent of all programs during the 1992–2001 period. Distinguishing between IMF programs that were implemented successfully and those that broke down is essential in properly evaluating the effectiveness of IMF programs. The only study that previously addressed this issue was Killick (1995), which distinguished between these two groups by using the ratio of disbursements to committed amounts as controls.

The question of whether IMF-supported programs have significant independent effects on the macroeconomic outcomes of particular countries is difficult to answer because it requires the construction of a counterfactual indicating what policies and outcomes would have resulted in the absence of an IMF program. Since the mid-1980s, several papers have considered how to construct a counterfactual for such exercises through differentiating the effects of counterfactual policies from exogenous developments, initial conditions, and IMF support. The methodology that has been most widely applied, developed by Goldstein and Montiel (1986), uses policy reaction functions estimated for countries that did not have support from a particular international financial institution (IFI) to approximate the counterfactual for countries that did have IFI backing for their programs. Unfortunately, several recent studies have cast doubt on the appropriateness and reliability of this method. For instance, Dicks-Mireaux, Mecagni, and Schadler (2000) did diagnostic tests to show that the assumptions underlying the methodology may not hold.

Another method of evaluating the effects of IMF programs is the before-after approach, in which countries are evaluated for a number of years before and after the onset of an IMF program to identify whether IMF advice is able to improve the macroeconomic situation. The major problem with the simple before-after approach is that the economic condition of the country might have improved anyway without the presence of an IMF-supported program. Failure to take this into account leads to biased estimation of the coefficient associated with IMF programs.

This paper uses the two-step Heckman (1979) procedure to control for the bias caused by participation in an IMF program. In the first step, a probit model is used to capture the choice of participating in a program based on changes in initial economic and political conditions. A composite variable, the inverse Mill’s ratio (IMR), summarizing this choice is then introduced into the second-stage least-squares regression of the determinants of the various macroeconomic targets. This variable gives an estimated probability of program participation and therefore controls for nonrandom selection into IMF programs by holding fixed an estimated probability for this selection. Similar treatment is also given to the stoppage variable to control for the possible endogeneity problem—that is, the possibility that a program might be terminated prematurely owing to unsatisfactory performance of the macroeconomic targets that are under examination. This distinction separates out the effects of IMF programs from the timing of a program, and provides insights on evaluating whether stopped programs are any different.

Since the sample splits between the 1980s and 1990s, it is possible that any improvement in the macroeconomic targets during the 1990s could be related to a more stable economic environment rather than to the availability of IMF advice. To control for this effect, time dummies are included in the specification so that any improvement associated with an IMF-supported program is independent of the strength of the world economic cycle.

Our findings demonstrate significant differences between successful and stopped programs, although the macro targets associated with successful programs are no different from their preprogram values. While stopped programs raise inflation and the budget deficit, and lower growth relative to periods without a program, completed programs are marginally associated with increased growth three years after the termination of the program.

Our paper is organized as follows: the second section reviews the related literature; the third section describes the data, model specification, and estimates; and the fourth section presents our conclusion.

Related Studies

Studies that have tried to identify the factors that induce countries to initiate IMF-supported programs have found that deteriorating external conditions such as the balance of payments, the debt position, and rapid exchange rate movements play an important role in the timing of programs. Conway (1994) used both probit and tobit models to analyze the participation of 74 countries in Stand-By Arrangements (SBAs) and Extended Fund Facilities (EFFs) during 1976–86 and found that past participation of IMF programs, real GDP growth, and external factors (terms of trade, current account, long-term external debt) were significant determinants of the timing of IMF involvement. Joyce (2002) used a panel dataset for 45 countries for 1980–84 and identified the ratio of government expenditures to GDP and the reserves-import ratio as significant factors. Edwards and Santaella (1992) used 48 devaluation periods in developing countries and identified GDP per capita and the ratio of net foreign assets to the money supply as the most important factors. Knight and Santaella (1997) employed a bivariate probit model to examine 91 non-oil countries for 1973–91. They concluded that the level of international reserves and GDP per capita were the most important factors in influencing a county’s decision to negotiate programs while revenue and expenditure changes, domestic credit, and exchange rate movements were highlighted by the IMF.

In terms of the effects of IMF programs on key macroeconomic indicators, studies have found that IMF programs lead to immediate improvements in the current account and overall balance of payments, but do not have a consistent impact on inflation and economic growth. For example, Khan (1990), Schadler and others (1993), Conway (1994), Killick (1995), and Dicks-Mireaux, Mecagni, and Schadler (2000) all find a negative relationship between IMF programs and inflation, but the estimated effect is significant only in Killick’s analysis. Some studies find a significant positive relationship with respect to growth in the short term (Killick, 1995; Bagci and Perraudin, 1997; and Dicks-Mireaux, Mecagni, and Schadler, 2000) and in the long term (i.e., three years after the program—see, for example, Conway, 1994) whereas others, in particular, Khan (1990) and Przeworski and Vreeland (2000), find significant negative growth effects both in the short and long term. Concerning budgetary conditions, Schadler and others (1993) find that fiscal deficits fall during IMF programs, while Bulíř and Moon (2002) are unable to identify significant effects. A summary of the findings can be found in Table 1.

Table 1Summary of Recent Studies Analyzing the Impact of IMF Programs on Inflation, Budgetary Conditions, and Growth
StudySampleInflationGrowthBudget
Khan (1990)1973–88, 259 programs--*
Schadler and others (1993)1983–93, 55 programs-+-
Conway (1994)1976–86, 217 programs--, +*
Killick (1995)1979–85-*+*
Bagci and Perraudin (1997)1973–92-+*
Dicks-Mireaux, Mecagni, and Schadler (2000)1986–91, 88 programs-+*
Przeworski and Vreeland (2000)1951–90, 226 programs-*
Bulíř and Moon (2002)1993–96, 64 programs+
Note: A single asterisk (*) indicates significance at the 10 percent level.
Note: A single asterisk (*) indicates significance at the 10 percent level.

Data and Empirical Results

Data

The data choice to highlight differences between stopped and completed programs was based on requiring a fairly long period to evaluate the effectiveness of programs. In the past, a number of studies have isolated specific years in which an IMF program did or did not take place. However, since the macroeconomic objectives of IMF programs generally have a much longer duration than one year, it seemed appropriate to ensure a fairly long period between engaging and not engaging in IMF programs. With these objectives in mind, the countries used in this paper are those that had IMF programs during the 1990s but did not have any type of IMF program between 1980 and 1988: Algeria, Benin, Bulgaria, Burkina Faso, Cambodia, Cape Verde, Colombia, Djibouti, Indonesia, Jordan, Mongolia, Nicaragua, Papua New Guinea, Poland, Rwanda, República Bolivariana de Venezuela, and the Republic of Yemen. Since Nicaragua and Cambodia were involved in civil wars during the 1980s, and the Republic of Yemen became a republic only in 1990, data points for these countries during the 1980s are excluded from the sample. Djibouti was also excluded because of lack of data during the 1980s. In the end, our sample covers 17 countries and 35 programs for the period from 1980 to 1999.2

Since the sample is restricted, it is important to determine whether it is representative of all IMF programs. To aid in this process, Table 2 provides various benchmarks for this sample as well as for all IMF programs negotiated between 1992 and 2000. The average per capita income levels in the two samples are close, at about $1,000–$1,100. Moreover, the type of conditions that were applied in the programs are comparable in terms of number and use of prior actions, with conditions in the restricted sample slightly more numerous. Interestingly, the percentage of stoppages is slightly lower in the restricted sample. In terms of the target variables, the initial conditions are also broadly similar, with inflation averaging about 30–40 percent and the general government budget deficit averaging 3½ percent of GDP. The initial current account and growth estimates differ more significantly between the two samples, with the current account deficit in the restricted sample at 3.1 percent of GDP and the corresponding deficit in the full sample at 5.7 percent of GDP. Moreover, growth in the initial period was considerably higher in the restricted sample at 1.7 percent, compared with 1.1 percent for the full sample.

Table 2Comparing the Restricted and Full Sample Averages(standard errors in parentheses)
GDP Per Capita (In U.S. dollars)Total Number of ConditionsTotal Number of Prior ActionsPercentage of Stoppages
Restricted sample1,010.41259.936.4
(733.2)(21.7)(12.9)(48.9)
Full sample1,091.52226.641.8
(1,315)(23.2)(12.8)(49.5)
Values at Specific Points in Program Cycle
At beginning of programAt end of program3 years following program
Inflation (in percent)
Restricted sample32.413.211.3
(50.2)(15.1)(18.0)
Full sample42.322.411.8
(61.1)(38.2)(18.1)
Government Budget Balance (in percent of GDP)
Restricted sample-3.3-2.6-2.3
(3.6)(4.1)(5.0)
Full sample-3.9-3.5-3.9
(4.2)(4.0)(4.2)
Current Account (in percent of GDP)
Restricted sample-3.1-3.3-2.5
(9.7)(8.6)(11.1)
Full sample-5.7-5.9-5.7
(9.4)(11.1)(8.7)
Growth (in percent)
Restricted sample1.73.83.3
(6.1)(3.8)(3.8)
Full sample1.12.73.9
(7.2)(6.2)(5.8)

The behavior of the various target variables over time is also comparable between the two samples. The inflation rate declines through the program period and three years after the program to average about 11 percent in both samples, and the growth rate picks up to over 3 percent in both samples three years after the end of the program. In contrast, while the budgetary position improves in the reduced sample, the improvement shown in the full sample is temporary, since the budget deficit returns to its initial average value three years out. Similarly, while the current account position improves in the reduced sample, it initially deteriorates in the full sample and only returns to its initial value three years out.

First-Stage Probit Specification

As mentioned in the introduction it is necessary to correct for the timing of an IMF program so that this decision can be isolated from the effects of IMF policies. To this end, a probit model was estimated with the dependent variable representing the observed 0-1 dummy capturing whether a program was initiated or not. The independent variables fall into two categories: macroeconomic indicators at the beginning of the program (GDP per capita, inflation, government budget position, debt/GDP ratio, exchange rate depreciation) and political economy variables (the democracy index, a dummy for the change of government within three years of the commencement of the program, and time in power). The reason for including political variables is that recent studies have found that they are closely related to the success of programs (Dollar and Svensson, 2000; Ivanova and others, 2002; and Thomas, forthcoming) and they are therefore suitable instruments for identifying the timing of programs. Barro and Lee (2002) also emphasize the need for appropriate instruments in conducting this type of analysis but rely on political economy variables to achieve this objective, including each country’s share of quotas and professional staff, as well as voting patterns in the United Nations.

The political economy variables used in this paper are constructed from the World Bank’s Data base of Political Institutions. The democratic index is coded from a democratic score on a 0–12 scale, with higher numbers associated with more democratic governments. For the democratic variable, the score was reclassified as a 1-0 dummy variable depending on whether the score was 9–12 or less; for the autocratic variable, the score was reclassified as a 1-0 dummy variable depending on whether the score was less than 3. The dummy variable for the change in government is 1 if there is a change in government within three years prior to the program and 0 otherwise. The variable, time in power, represents the log value of the years in power of the government at the time of program entry, assuming no change in the previous three years, and 0 otherwise. The dummy for the change in government is hypothesized to capture the raised likelihood of conditionality when a new government comes to power. In addition, separating out longer political regimes allows us to determine whether length of leadership influences the likelihood of having an IMF program. No dummy is included for past programs because the sample was identified on the basis that no program took place during the 1980s.

The results of the first-stage probit estimation can be found in Table 3. The overall performance of the various specifications is good in predicting the timing of an IMF program with at least 80 percent accuracy. Moreover, the accuracy of the predictions is quite stable across the specifications (Table 4). The pseudo R-square3 is also high compared with some studies, ranging from 53 percent to 65 percent.

Table 3Probit Model(maximum-likelihood estimates)Dependent Variable: l = 1 if there is an IMF programl = 0 if there is no IMF program
Independent variables(1)(2)
Constant4.15091.00721.65280.3607
(1.07)(0.45)(0.58)(0.19)
Economic variables
GDP per capita-1.4485*-0.6466**-0.9085*0.5697*
(-2.03)(-1.89)(-2.18)(-2.02)
Initial inflation0.0519-0.0164
(0.84)(-0.46)
Budget/GDP0.08600.0930
(0.77)(1.34)
Debt/GDP5.6267*4.9920*5.7400*4.8957*
(2.99)(3.48)(3.26)(3.75)
Exchange rate depreciation0.0630**0.0771*0.0741*0.0636*
(1.85)(2.72)(2.57)(2.99)
Reserves/imports0.30871.06240.10910.5228
(0.17)(0.86)(0.07)(0.48)
Political variables
Democratic-1.1778-1.1849
(-1.25)(-1.36)
Autocratic-1.3405-0.8158
(-1.18)(-0.87)
Change of government 3 years before program2.32331.2412
(1.44)(1.13)
Years in office (> 3 years)0.98920.4009
(1.26)(0.77)
Observations61666469
Pseudo R-square0.64080.54780.63030.5382
Log-likelihood-15.1833-20.6884-16.3871-22.0553
Lr test of political variables11.01***11.34***
Notes: The t-statistics are in parentheses. One asterisk (*) denotes significance at 5 percent; two asterisks (**) denote significance at 10 percent. The Lr test is for the joint hypotheses that the coefficients of the four political variables are all zero; the chi-square (4) test statistic at 5 percent is 9.49. Therefore, the political variables are significant, which is denoted by three asterisks (***).
Notes: The t-statistics are in parentheses. One asterisk (*) denotes significance at 5 percent; two asterisks (**) denote significance at 10 percent. The Lr test is for the joint hypotheses that the coefficients of the four political variables are all zero; the chi-square (4) test statistic at 5 percent is 9.49. Therefore, the political variables are significant, which is denoted by three asterisks (***).
Table 4Accuracy of Probit Predictions
(1)(2)
Total predicted l =128272727
Total actual l = 131333133
Accuracy (percent)90.3281.8287.1081.82
Total predicted l =127262830
Total actual l = 130333336
Accuracy (percent)9078.7884.8583.33
Overall correct predictions55535557
Overall actual programs61666469
Overall accuracy (percent)90.1680.3085.9482.61

Among the economic variables, we found that GDP per capita, the debt/GDP ratio, and the depreciation of the exchange rate were significant at the 5 percent level across all specifications, corroborating the results of previous studies. The higher the GDP per capita, the less likely that a country will enter a program, whereas higher debt/GDP ratios and exchange rate depreciations raise the likelihood that a country will come to the IMF for financial assistance. The initial inflation rate and the government budgetary position are insignificant in all cases.

The political economy variables also provide significant explanatory power. Indeed, including them raises the pseudo R-square by about 10 percentage points and a likelihood ratio test rejects the joint hypothesis that they are all equal to zero. Interestingly, highly democratic and autocratic countries are less likely to enter IMF programs. A possible rationale for democratic countries not coming to the IMF is that considerable bargaining between political interest groups might be needed before an agreement is reached and this takes time. While this is not an issue for autocratic regimes, they may try to dissuade foreign institutions from closely scrutinizing their economic policies. Interestingly, Thomas (forthcoming) finds that once autocratic governments acquiesce to an IMF program, the program is much less likely to stop. New governments are more likely to enter an IMF program; and for those not changing office, longer incumbents are more likely to come to the IMF, although this effect tapers off after some time.

The stoppage variable is also an endogenous variable and therefore may require instruments. If a program is stopped because of unsatisfactory macroeconomic performance, it is inappropriate to use program stoppages as an exogenous variable when evaluating the effects of IMF programs on these same target variables.4Mecagni (1999) argues based on country report documentation that, in most cases, program interruptions were not associated with IMF conditionality that was too stringent. Rather, the interruptions were associated with domestic dissatisfaction with existing institutional and political arrangements or with preexisting problems. The robustness of this assumption is evaluated by comparing the estimates assuming that program stoppages are exogenous with those obtained from instrumenting for this variable.

Instruments for the stoppage variable comprise dummies for autocratic regimes and countries engaging in guerrilla warfare, a dummy representing a change in government within three years of the program, the total number of conditions, the real exchange rate depreciation during the year prior to the program, the initial inflation level, and the logarithm of GDP per capita. The estimation results can be found in Table 5. On the whole, we find that among the instruments used, the political variables have stronger effects on the likelihood of a stoppage than the economic variables. The hypothesis that the political variables are all equal to zero cannot be accepted at the 10 percent level by the likelihood ratio test. However, the hypothesis that the economic variables are all equal to zero can be accepted. In terms of the coefficients, the economic variables are all insignificant, while the democracy variable is consistently significant. The political variables suggest that the higher probability of stoppages is associated with less democracy and wars, but, perhaps surprisingly, not a change in government. The relatively weak effects of the economic variables mitigate somewhat concerns about the endogeneity of the stoppage variable. However, to make our analysis more robust, we experiment with both exogenous and endogenous stoppage variables.

Table 5Probit Model for StoppageDependent Variable: l = 1 if there is an IMF programl = 0 if there is no IMF program
Independent Variables(1)(2)(3)
Constant-0.72450.37330.0034
(-0.31)(0.20)(0.01)
GDP per capita0.1637-0.0349
(0.2671)(-0.16)
Initial inflation-0.0099-0.0008
(-0.94)(-0.12)
Exchange rate depreciation0.02930.0169
(1.32)(1.10)
Total number of conditions-0.1875-0.3023
(-0.48)(-0.82)
Democratic-1.2279-1.0714
(-2.05)(-2.03)
Guerrilla wars1.14450.6922
(0.92)(0.72)
Change of government 3 years before program-0.3559-0.1881
(-0.61)(-0.39)
Observations353535
Pseudo R-square0.21770.05020.1404
Log-likelihood-16.92-20.54-18.59
Restricted log-likelihood-21.63-21.63-21.63
Lr test7.25*3.34
Note: For the joint test that all institution variables are zero, the chi(3) statistic at 10 percent is 6.25; for the joint test that all macroeconomic variables are zero, the corresponding chi(4) statistic is 7.78. Therefore, the institution variables are significant, which is denoted by an asterisk (*).
Note: For the joint test that all institution variables are zero, the chi(3) statistic at 10 percent is 6.25; for the joint test that all macroeconomic variables are zero, the corresponding chi(4) statistic is 7.78. Therefore, the institution variables are significant, which is denoted by an asterisk (*).

Second-Stage Specification

One of the main subjects of interest in IMF programs is how they succeed in improving macroeconomic target variables during the duration of programs and following their termination. To analyze this issue effectively, the dependent variable is defined in two ways for each target variable. First, it is defined as the average change in the value of the target between the first year of the program and the final year of the program. Second, it is defined as the average change in the value of the target between the first year of the program and three years after the end of the program. For periods during which no program was conducted (1980–92), three- and six-year intervals are used for the dependent variable. This time interval roughly corresponds to the average duration of IMF programs during the 1990s of 2½ years.

In many studies, the effects of IMF involvement in particular countries are identified through the inclusion of a dummy variable for the macroeconomic programs that were supported by the IMF. However, as indicated previously, it is also necessary to account for stoppages. In our sample of 35 programs, 12, or 36 percent, were stopped. Killick (1995) differentiates between countries that completed programs and those that terminated early by using a threshold value of 80 percent of the initial committed loan that was disbursed. He argues that this cutoff point is likely to be closely associated with successful completions based on a survey made of programs over the 1980–92 period. Consistent with his hypothesis, he finds that the inflation rate and the current account improved significantly for countries with completed programs relative to countries with noncompleted programs. This paper controls for the possible different impacts of completed programs and those that terminated early with a stoppage variable. Although as recorded most of the stoppages happened as a result of the program countries’ failure to follow IMF recommendations, to account for the possible endogeneity problem, we consider both exogenous and endogenous stoppages.5 Moreover, following Killick (1995), we construct the disbursement ratio as another proxy for stoppages. By doing so, the effectiveness of IMF programs on the target macroeconomic variables is evaluated more accurately, and the effects of stoppages are also analyzed. For the sample of observations in the 1980s when no program took place, the dummy variables are recorded as zero.

Since timing can make a large difference in the interpretation of results, annual time dummies were included in each specification to control for this factor. Owing to insufficient degrees of freedom, both the dependent and time-varying independent variables were purged of the timing effect before being introduced into the second-stage equation.

For the inflation equation, the economic determinants include the level of inflation at the beginning of the period and the change in the local currency/U.S. dollar exchange rate. The variables representing IMF involvement include a dummy for IMF programs, a dummy for stoppages, and the IMR calculated from the first-stage probit estimation.

The first two columns of Table 6 present the results without correcting for the timing of IMF involvement. In this case, distinguishing between completed and stopped programs makes no difference because both coefficients are insignificant. Similar results hold when a dummy variable for programs with disbursed amounts over 80 percent of the committed total (comparable to Killick’s definition) is substituted for the stoppage variable. Turning to the other variables, high-inflation countries have difficulty reducing inflation because the coefficient on the initial inflation level is significantly positive. Moreover, the change in the exchange rate is positive although significantly below unity. The inclusion of the IMR has little effect on the other variables in this specification (columns (3) and (4)). The major change in results occurs when the stoppage variable is instrumented, because the coefficient becomes significantly positive at the 10 percent level. The coefficient estimate implies that the inflation rate rises by 6½ percent a year during the program period in stopped programs.

Table 6Inflation Equation During Program
Independent Variables(1)(2)(3)(4)(5)
Initial inflation0.3304*0.3469*0.3255*0.3422*0.3202*
(5.92)(5.89)(5.48)(5.51)(5.40)
Change in exchange rate0.6243*0.6246*0.6158*0.6175*0.6169*
(6.59)(6.67)(6.03)(6.15)(6.08)
IMF program-0.7319-0.0217-0.45740.1047-1.9763
(-0.48)(-0.01)(-0.21)(0.04)(-0.85)
Stoppage11.99622.21456.6876**
(1.12)(1.09)(1.75)
Disbursement ratio0.04070.3952
(0.02)(0.19)
Inverse Mill’s ratio-0.5546-0.4308-0.3682
(-0.37)(-0.29)(-0.24)
Number of observations6969606060
R-square0.75690.75440.74850.74570.7525
Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using the following variables: dummies for autocracy, guerrilla warfare, changes in government, the total number of structural conditions, the change in the real exchange rate one year prior to the program, the initial inflation level, and the log of GDP per capita.

Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using the following variables: dummies for autocracy, guerrilla warfare, changes in government, the total number of structural conditions, the change in the real exchange rate one year prior to the program, the initial inflation level, and the log of GDP per capita.

In the regression of inflation changes over a longer horizon (Table 7), the coefficients on the program dummies are broadly similar to the previous specification, with the instrumented stoppage variable the only significant variable. The coefficient estimate of IMF programs is comparable in magnitude to that presented in the short-run equation, suggesting that the inflation rate keeps declining after the termination of the program. Both the initial inflation rate and the change in the exchange rate remain significant, with the exchange rate coefficient now insignificantly different from unity.

Table 7Inflation Equation Three Years After Program
Independent Variables(1)(2)(3)(4)(5)
Initial inflation0.1717*0.1789*0.1209*0.1354*0.1245*
(3.95)(4.24)(3.93)(4.58)(4.98)
Change in exchange rate0.8509*0.8503*0.9124*0.9119*0.9037*
(11.09)(10.98)(13.61)(13.47)(13.44)
IMF program-0.4122-0.1270-0.09460.1684-1.6231
(-0.26)(-0.08)(-0.06)(0.10)(-0.95)
Stoppage11.06002.68095.8508**
(0.61)(1.55)(1.76)
Disbursement ratio0.28571.4460
(0.18)(0.99)
Inverse Mill’s ratio-1.7229-1.5511-1.5551
(-1.45)(-1.28)(-1.34)
Number of observations5757505050
R-square0.91440.91410.93900.93810.9397
Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

For the equation explaining the change in the budgetary position, the independent variables comprise the initial budget position, the change in inflation, and the change in the terms of trade. The change in the terms of trade proxies strong developments in GDP, which would likely lower the budget deficit. The change in the inflation rate is included to control for the Tanzi-Olivera effect, which postulates that as the inflation rate is lowered, the budgetary position would be expected to improve because the real value of conventional tax revenue rises on account of collection lags. The variables representing IMF involvement include the dummy for IMF programs, a dummy for IMF programs that terminated prematurely, and the IMR calculated from the first-stage probit estimation.

In the regression for the budgetary change during the program (Table 8), the initial budget level is consistently significantly negative, implying that the budgetary improvement is greater the larger the initial size of the deficit. But the budgetary position also weakens if the initial position is in surplus. The change in the inflation rate is significantly negative, suggesting that the Tanzi-Olivera effect holds. Indeed, a 10 percent decline in the inflation rate would raise the budget surplus by about ½ percent of GDP. In contrast, the change in the terms of trade does not play a significant role in budgetary developments. For the IMF dummies, while the coefficient on the completed IMF program is significantly positive in the specification without the inverse Mill’s ratio, the coefficient is insignificant when the IMR is included. In contrast, the coefficient on the stoppage variable is significantly negative in each specification, with a particularly high value when instrumented.

Table 8Budgetary Positions During Program
Independent Variables(1)(2)(3)(4)(5)
Initial budget level-0.1782*-0.1657*-0.2242*-0.2108*-0.2220*
(-3.02)(-2.74)(-4.85)(-4.20)(-4.58)
Change in inflation-0.0429*-0.0529*-0.0471*-0.0572*-0.0495*
(-3.12)(-3.89)(-3.61)(-4.31)(-3.74)
Change in terms of trade0.00550.00960.00940.01370.0132
(0.41)(0.67)(0.75)(0.97)(0.99)
IMF program0.6094**0.38720.36080.18580.6476
(1.78)(0.96)(0.84)(0.38)(1.27)
Stoppage1-1.1169*-1.1380*-2.0330*
(-2.58)(-2.25)(-2.43)
Disbursement ratio-0.3753-0.4550
(-0.86)(-0.92)
Inverse Mill’s ratio-0.0779-0.1371-0.1389
(-0.26)(-0.49)(-0.44)
Number of observations6565575757
R-square0.46690.43240.63310.59940.6255
Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

Three years after the program has terminated (Table 9), initial conditions remain significant, although the impact is less strong. The dummy for completed IMF programs is insignificant in all specifications, so that the budgetary position is not affected by an IMF program. In contrast, for programs that stop, the budgetary position deteriorates relative to periods without an IMF program. Similar results hold when the dummy variable reflecting the disbursement ratio is used.

Table 9Budgetary Positions Three Years After Program
Independent Variables(1)(2)(3)(4)(5)
Initial budget level-0.1303*-0.1264*-0.1456*-0.1400*-0.1419*
(-8.03)(-7.26)(-12.44)(-9.43)(-7.70)
Change in inflation-0.0050-0.0076-0.0060-0.0090-0.0074
(-0.82)(-1.19)(-0.98)(-1.37)(-1.09)
Change in terms of trade0.01020.00910.01850.01850.0176
(0.68)(0.60)(1.26)(1.10)(1.10)
IMF program0.38310.35590.28190.27030.3330
(1.56)(1.26)(1.05)(0.87)(0.99)
Stoppage1-0.8651*-0.9318*-0.8644
(-3.10)(-3.03)(-1.60)
Disbursement ratio-0.6930*-0.7108*
(-2.22)(-2.07)
Inverse Mill’s ratio-0.0988-0.1308-0.1440
(-0.44)(-0.62)(-0.76)
Number of observations5353474747
R-square0.57730.56070.68970.66680.6359
Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

While improvements in the inflationary environment and in the budgetary position are important targets in themselves, they are not the ultimate goals of economic policy. The ultimate objective of a country’s economic policy is to improve the welfare of its citizens, which is normally measured as the change in per capita income. There is a huge literature on the determinants of growth, and therefore this paper has chosen to be selective in deciding which controls to include in the specification. The paper takes as its point of departure the analysis of Doppelhofer, Miller, and Sala-i-Martin (2000), who has tried to detect the variables that are the most robust determinants of growth using the Bayesian updating technique. He finds that a dummy variable for sub-Saharan Africa, the initial level of GDP per capita (the convergence effect), and the primary school enrollment rate (measuring human capital) are robust to changes in specification. This paper includes the first two variables and substitutes the illiteracy rate for the primary school enrollment rate. We also include the change in the budget balance and the inflation rate to capture direct effects from the policy changes, and export market growth to capture exogenous effects.

In the regression for the change in growth during the program (Table 10), the initial convergence term is generally significant, indicating that each year poor countries make up 1 percent of the disparity in per capita GDP with the richest country, or that, holding all other factors constant, it would take them 100 years to fully catch up. Perhaps surprisingly, both policy variables represented by the change in the inflation rate and the budgetary position are insignificant, although the dummy variable for countries in sub-Saharan Africa is significantly negative in the specifications with the IMR. The other two macroeconomic variables, the export market growth and the illiteracy rate, are insignificant.

Table 10Growth Equation During Program
Independent Variables(1)(2)(3)(4)(5)
Change in inflation-0.0353-0.0489-0.0261-0.0380-0.0375
(-0.77)(-1.02)(-0.51)(-0.72)(-0.70)
Change in budget position0.08690.12470.14830.17860.1833
(0.49)(0.69)(0.73)(0.90)(0.96)
GDP per capita-0.6307**-0.5604-0.9826*-1.0287*-0.9078**
(-1.65)(-1.30)(-2.03)(-2.04)(-1.76)
Export market GDP growth0.24310.25820.19990.22730.1683
(0.84)(0.82)(0.61)(0.62)(0.48)
Sub-Saharan region-0.7704-0.3896-2.5127**-2.4202*-2.1550**
(-0.74)(-0.39)(-1.94)(-1.96)(-1.84)
Illiteracy-0.0070-0.01370.01230.00680.0048
(-0.30)(-0.58)(0.44)(0.26)(0.18)
IMF program0.59040.25741.06760.88661.0665
(0.92)(0.29)(1.46)(0.88)(1.22)
Stoppage1-1.6959**-1.7196**-1.6672
(-1.78)(-1.66)(-1.13)
Disbursement ratio-0.4399-0.8266
(-0.48)(-0.86)
Inverse Mill’s ratio-0.1578-0.2179-0.2613
(-0.24)(-0.33)(-0.40)
Number of observations6565575757
R-square0.18430.14840.25720.22460.2243
Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

Notes: The t-statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

Turning to the direct effects of IMF programs, the dummy variable for completed programs is insignificant in all specifications whereas the stoppage variable is significantly negative in all specifications except the instrumented equation.

Over the longer run (Table 11), the effect of the change in inflation on growth is significantly negative, with a 10 percentage point decline in the inflation rate leading to a 0.5–0.6 percentage point improvement in the growth rate. This is comparable to the findings of Ghosh and Phillips (1998). GDP per capita is generally significantly negative, indicating that a lower growth rate is associated with a higher level of GDP per capita. The dummy for the sub-Saharan region is significantly negative at 5 percent when the inverse Mill’s ratio is included, while the illiteracy rate is significantly negative when the IMR is excluded. It appears therefore that sub-Saharan countries and countries with high illiteracy rates have difficulty in raising their standard of living over the longer run.

Table 11Growth Equation Three Years After Program
Independent Variables(1)(2)(3)(4)(5)
Change in inflation-0.0606*-0.0668*-0.0543*-0.0603*-0.0582*
(-4.04)(-4.17)(-4.20)(-4.50)(-3.92)
Change in budget position-0.0331-0.0265-0.0553-0.05050.0307
(-0.13)(-0.10)(-0.22)(-0.20)(0.13)
GDP per capita-0.7701**-0.7734**-1.3182*-1.3717*-1.3237*
(-1.93)(-1.82)(-2.73)(-2.78)(-2.64)
Export market GDP growth0.35440.36780.31260.32130.2352
(1.08)(1.10)(0.93)(0.89)(0.66)
Sub-Saharan region-0.2987-0.2148-2.6695*-2.5925*-2.4166*
(-0.34)(-0.25)(-2.57)(-2.53)(-2.28)
Illiteracy-0.0347**-0.0391*-0.0083-0.0136-0.0136
(-1.82)(-2.07)(-0.39)(-0.66)(-0.64)
IMF program0.80970.78881.0505**1.02831.1199
(1.35)(1.27)(1.71)(1.52)(1.26)
Stoppage1-1.7342*-1.6221*-1.4884
(-2.57)(-2.23)(-1.03)
Disbursement ratio-1.4352*-1.2934*
(-2.23)(-2.00)
Inverse Mill’s ratio-0.3443-0.3510-0.4348
(-0.72)(-0.71)(-0.93)
Number of observations5353474747
R–square0.40780.39610.51600.50350.4852
Notes: The t–statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

Notes: The t–statistics appear in parentheses; one asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

In column (5), the stoppage variable is instrumented using variables specified in footnote 1 to Table 6.

In the long run, the IMF dummy for completed programs remains insignificant except in the specification including the IMR in which it yields positive effects on growth (Table 12). In contrast, the stoppage variable is significant except in the instrumented equation, and the coefficient implies negative effects on growth. The different results point to the possibility that a failure in macroeconomic performance could trigger a stoppage.

Table 12IMF Dummy Only Versus Dummies Distinguishing Stoppage
InflationBudgetGrowth
During programThree years laterDuring programThree years laterDuring programThree years later
Completed programs--++++**
Stopped programs++-*-*-**-*
Stopped programs (using instruments)+**+**-*---
Notes: One asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.
Notes: One asterisk (*) denotes significance at the 5 percent level; two asterisks (**) denote significance at the 10 percent level.

Conclusion

This paper has presented empirical evidence relevant to evaluating whether IMF programs have been effective in the 1990s in influencing three major macroeconomic variables: inflation, the budget position, and growth. To properly identify the effect of IMF policies on these aggregates, variables were included to control for the timing of IMF programs. The paper finds that the timing of an IMF program can be represented well by the level of GDP per capita, the debt/GDP ratio, and the magnitude of the exchange rate depreciation, with over 90 percent of the timing decisions correctly identified. Although the inverse Mill’s ratio is insignificant in all specifications, its inclusion does change the significance of the dummy for completed IMF programs in some cases, demonstrating the importance of its inclusion in the analysis.

The paper highlights the importance of distinguishing between completed and stopped programs because they are associated with significantly different macroeconomic outcomes. Since stoppages are likely endogenous, they were instrumented with both political and economic variables. Our analysis shows that political variables seem to have more impact than do economic variables on stoppages. On the whole, endogenizing the stoppage variable does not change our conclusion that stopped programs raise inflation, worsen the budgetary position, and impede growth, but it deepens the negative effects that a stopped program can have on inflation while it lessens the negative impact on budget conditions and growth. On IMF programs that end successfully, the only variable that appears to be affected in a positive way is growth, and the significance of this variable breaks down when instruments are found for stoppages.

One of the limitations of this analysis is the sample size. We avoid the problems with cross-country comparisons by making a clear split between program periods and nonprogram periods, but the trade-off is a relatively short sample that might make our results less convincing and less robust. We hope to tackle this problem in the future when more data become available. We also intend to dig deeper into whether IMF programs in the 1990s have succeeded in fostering growth by distinguishing among types of programs and between normal and prolonged users of IMF resources.

Appendix. List of Countries and Programs Considered in 1990s
CountryType of ProgramStartEnd
AlgeriaSBA19941995
AlgeriaEFF19951998
BeninESAF19931996
BeninESAF19961999
BulgariaSBA19921993
BulgariaSBA19941995
BulgariaEFF19961998
BulgariaSBA19971998
BulgariaEFF19982001
Burkina FasoESAF19931996
Burkina FasoESAF19961999
Burkina FasoESAF19992002
CambodiaESAF19941997
CambodiaPRGF19992002
Cape VerdeSBA19981999
ColombiaEFF19992002
DjiboutiESAF19961999
DjiboutiESAF19992002
IndonesiaSBA19971998
IndonesiaEFF19982000
JordanSBA19921993
JordanEFF19941996
JordanEFF19961999
JordanEFF19992002
MongoliaESAF19931996
MongoliaESAF19972000
NicaraguaESAF19941997
NicaraguaESAF19982001
Papua New GuineaSBA19951997
PolandSBA19931994
PolandSBA19941996
RwandaESAF19982001
Venezuela, República Bolivariana deSBA19961997
Yemen, Republic ofSBA19961997
Yemen, Republic ofESAF19972000
Notes: SBA denotes a Stand-By Arrangement; EFF denotes the Extended Fund Facility; ESAF denotes the Enhanced Structural Adjustment Facility; and PRGF denotes the Poverty Reduction and Growth Facility.
Notes: SBA denotes a Stand-By Arrangement; EFF denotes the Extended Fund Facility; ESAF denotes the Enhanced Structural Adjustment Facility; and PRGF denotes the Poverty Reduction and Growth Facility.
References

    BagciPinar and WilliamPerraudin1997“The Impact of IMF Programs,” CEPR Working Paper No. 24 (London: Center for Economic Policy Research).

    • Search Google Scholar
    • Export Citation

    BarroRobert and J. W.Lee2002“IMF Programs: Who Is Chosen and What Are the Effects?” NBER Working Paper No. 8951 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    BulířAleš and SoojinMoon2002“The Composition of Fiscal Adjustment and Structural Conditionality Under IMF-Supported Programs”(unpublished; Washington:International Monetary Fund). This paper also appears, in somewhat different form, as Chapter 14 in this volume.

    • Search Google Scholar
    • Export Citation

    ConwayPatrick1994“IMF Lending Programs: Participation and Impact,”Journal of Development EconomicsVol. 45 (December) pp. 36591.

    • Search Google Scholar
    • Export Citation

    Dicks-MireauxLouisMauroMecagni and SusanSchadler2000“Evaluating the Effect of IMF Lending to Low-income Countries,”Journal of Development EconomicsVol. 61 (April) pp. 495526.

    • Search Google Scholar
    • Export Citation

    DollarDavid and JakobSvensson2000“What Explains the Success or Failure of Structural Adjustment Programmes?”Economic JournalVol. 110 (October) pp. 894917.

    • Search Google Scholar
    • Export Citation

    DoppelhoferGernotRonald I.Miller and XavierSala-i-Martin2000“Determinants of Long-Term Growth: A Bayesian Averaging of Classical Estimates (BACE) Approach,”NBER Working Paper No. 7750 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    EdwardsSebastian and JulioSantealla1992“Devaluation Controversies in the Developing Countries: Lessons from the Bretton Woods Era,”NBER Working Paper No. 4047 (Cambridge, Massachusetts: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    GhoshAtish and S.Phillips1998“Inflation, Disinflation and Growth,”Staff Papers International Monetary FundVol. 45 (December) pp. 672710.

    • Search Google Scholar
    • Export Citation

    GoldsteinMorris and PeterMontiel1986“Evaluating Fund Stabilization Programs with Multi-Country Data: Some Methodological Pitfalls,”Staff Papers International Monetary FundVol. 33 (June) pp. 30444.

    • Search Google Scholar
    • Export Citation

    HaqueNadeem Ul and MohsinKhan1998“Do IMF-Supported Programs Work? A Survey of the Cross-Country Empirical Evidence,”IMF Working Paper 98/169 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    HeckmanJames1979“Sample Selection Bias as a Specification Error,”EconometricaVol. 47 (January) pp. 15362.

    IvanovaAnnaWolfgangMayerAlexMourmouras and GeorgeAnayiotos2002“What Determines the Success or Failure of Fund-Supported Programs?” (unpublished: Washington:International Monetary Fund). This paper also appears, in somewhat different form, as Chapter 10 in this volume.

    • Search Google Scholar
    • Export Citation

    JoyceJames2002“Through a Glass Darkly: What We Know (and Don’t Know) About IMF Programs,” Wellesley College Working Paper No. 2002-04 (Wellesley, Massachusetts: Wellesley College).

    • Search Google Scholar
    • Export Citation

    KhanMohsin1990“The Macroeconomic Effects of Fund-Supported Adjustment Programs,”Staff Papers International Monetary FundVol. 37 (June) pp. 195231.

    • Search Google Scholar
    • Export Citation

    KillickTony1995IMF Programmes in Developing Countries: Design and Impact (London: Routledge).

    KnightMalcolm and Julio A.Santaella1997“Economic Determinants of IMF Financial Arrangements,”Journal of Development EconomicsVol. 54 (December) pp. 40536.

    • Search Google Scholar
    • Export Citation

    MecagniMauro1999“The Causes of Program Interruptions” in Economic Adjustment and Reform in Low-Income Countries (Washington: International Monetary Fund) pp. 21576.

    • Search Google Scholar
    • Export Citation

    PrzeworskiAdam and James R.Vreeland2000“The Effect of IMF Programs on Economic Growth,”Journal of Development EconomicsVol. 62 (August) pp. 385421.

    • Search Google Scholar
    • Export Citation

    SchadlerSusanFranekRozwadowskiSiddharthTiwari and DavidRobinson1993Economic Adjustment in Low-Income Countries: Experience Under the Enhanced Structural Adjustment Facility IMF Occasional Paper No. 106 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    ThomasAlunforthcoming“Prior Actions-True Repentance? An Evaluation Based on Active Programs over the 1992–99 Period,”IMF Working Paper (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation
1The authors are grateful for helpful comments from the participants of various seminars held at the IMF and at the Western Economic Association International Annual Meetings held in Denver, Colorado in July 2003. Any errors that remain, however, are their sole responsibility.
2The appendix gives a list of program countries, years, and program types.
3Pseudo R-square is calculated as 1−[log(Lur) − log(Lr)], where Lur is the maximum-likelihood value of the unrestricted-likelihood function and Lr is the maximum-likelihood value of the restricted-likelihood function (constant only). Conway (1994) reports a 90 percent ratio whereas Dicks-Mireaux, Mecagni, and Schadler (2000) only records a 3.5 percent ratio for their best fitted probit model.
4The stoppage variable used in this paper represents programs that were abandoned because the authorities did not follow IMF policy recommendations. If IMF targets were not met because of exogenous shocks, this would not warrant a stoppage of the program but would be accommodated through the granting of a waiver.
5In the endogenous version, the variable is instrumented by the predicted stoppage probability from the probit regression in Table 5.

    Other Resources Citing This Publication