Macroeconomic policies and participation in IMF programs

Economic Systems 30 (2006) 264–281 www.elsevier.com/locate/ecosys

Macroeconomic policies and participation in IMF programs Ays¸e Y. Evrensel a,*, Jong Sung Kim b a

Department of Economics and Finance, Southern Illinois University Edwardsville, Edwardsville, IL 62026-1102, United States b Department of Statistics and Mathematics, Portland State University, Portland, OR 97006-0751, United States Received 25 December 2005; received in revised form 19 March 2006; accepted 1 May 2006

Abstract This paper poses the question whether and to what extent developing countries’ macroeconomic policies affect their participation in IMF programs. The data contain 91 developing countries that received four types of IMF programs during the period 1967–1996. Using survival analysis and generalized least squares, we examine the characteristics of interprogram years (years without any IMF program). The results suggest that the average number of years that countries spend without an IMF program is affected by their macroeconomic policies and exchange rate regimes. Policy combinations that prevent the reduction in reserves lengthen the average interprogram period. # 2006 Elsevier B.V. All rights reserved. JEL classification: C34; E63; F32; F33 Keywords: The International Monetary Fund; Balance of payments crisis; Exchange rate regimes; Moral hazard

1. Introduction Since the collapse of the Bretton Woods system in the early 1970s and the subsequent adoption of the flexible exchange rate regime in the developed countries, the International Monetary Fund has been providing balance of payments support mostly to the developing countries. Various studies evaluate the IMF’s involvement in the developing countries.1 The most common result of evaluation studies is that IMF programs improve program countries’ current account and international reserves. A recent evaluation of Fund-supported programs shows that these programs have three characteristics (Evrensel, 2002). First, improvements that are achieved * Corresponding author. Tel.: +1 618 650 2592; fax: +1 618 650 3047. E-mail address: [email protected] (A.Y. Evrensel). 1 See Ul Haque and Khan (1998) for a literature review on program evaluations. 0939-3625/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.ecosys.2006.05.002

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

265

during program years in current account and reserves disappear in the post-program years. Second, most of the conditionality-related variables do not improve significantly even during the program years. Third, Fund-supported programs exhibit a recurrent nature in that developing members of the IMF have spent on average more than one-third of the period 1971–1997 under the IMF’s care (Evrensel, 2002). Therefore, it has been suggested that the ineffective and revolving nature of IMF programs may create moral hazard (Dreher and Vaubel, 2004b; Edwards, 2000; Evrensel, 2002; Friedman, 1998; Niehans, 1985; Vaubel, 1983, 1991).2 Debtor moral hazard implies the provision of undesirable incentives to program countries’ governments to implement inconsistent macroeconomic policies, which may lead to further balance of payments problems. In this paper, we further investigate the relationship between program participation and macroeconomic policies through the examination of policy combinations during the interprogram years (years without any IMF program). By doing so, we identify the policy combinations that shorten the interprogram years and lead to an IMF program sooner. Quantitative studies that examine the relationship between macroeconomic policies and participation in IMF programs have been scant. Evrensel (2002) finds that the average program country pursues increasingly expansionary fiscal and monetary policies, as the number of IMF programs in the country increases. Dreher and Vaubel (2004b) conclude that program countries tend to have larger budget deficits as the interest subsidy associated with IMF loans increases. Additionally, the extent of expansionary fiscal and monetary policies becomes larger, as the amount of IMF loans increases. Finally, Gai and Taylor (2004) find that declining reserves and the appreciation in the real exchange rate increase the probability of an IMF program. We contribute to the existing literature in three areas. First, we use four different program types that reflect the income levels of program countries. Therefore, we are able to derive different policy implications for countries that are at different stages of development. Second, we employ survival analysis to describe the characteristics of program participation (failure) and identify the differences among various programs regarding failure. Third, we emphasize the relevance of foreign exchange regimes by including a regime-related variable in our estimations. The exchange rate absorbs any shock and it changes under a flexible exchange rate regime, which prevents currency crises. However, pegging the exchange rate in the presence of budget deficits and expansionary monetary policies leads to an appreciation of the real exchange rate, which eventually results in reserve loss and the subsequent involvement of the IMF in the country. The data contain 91 developing countries that received four types of IMF programs during the period 1967–1996. In our empirical analysis, we employ survival analysis and generalized least squares. While survival analysis describes the characteristics of failure, the results based on the generalized least squares examine the effects of macroeconomic policies during interprogram years on the length of interprogram periods. The results suggest that the average number of years that countries spend without an IMF program is significantly affected by their macroeconomic policies and foreign exchange regimes. The empirical results verify the relevance of exchange rate regimes regarding the participation in IMF programs that less-managed exchange rate regimes lengthen the interprogram period. 2

Even though studies on the IMF-induced moral hazard have emerged fairly recently, there exists a variety of definitions in the literature, which depends upon who receives implicit guarantees due to an IMF program. In recent years, the term moral hazard has been mostly used in the context of creditor moral hazard. This type of moral hazard may be created if investors altered their portfolio decisions and took excessive risks because of expected implicit guarantees associated with Fund-supported programs. See Evrensel and Kutan (2006) for a review of the literature on creditor moral hazard in emerging countries’ stock markets.

266

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

The paper is organized as follows. Section 2 develops a framework in which the relationship between program participation and macroeconomic policies is developed. Section 3 presents the results of the empirical analysis. Section 4 provides concluding remarks. 2. Relationship between program participation and macroeconomic policies In this section, we provide a framework that describes the relationship between program participation and macroeconomic policies. This framework also allows us to identify the variables that will be used in the empirical analysis. The recurrent nature of the interaction between the IMF and its developing members in the form of program participation allows one to assume a repeated game. Such setups require an explicit definition of the participants with respect to their preferences, risk aversion, etc. However, this paper does not intent to formally examine the IMF’s or program countries’ preferences and describe their behavior.3 In the following, we focus on the players’ information about the general characteristics of their relationship to illustrate the link between program participation and macroeconomic policies. We assume that both the IMF and program countries have had perfect information on three areas since the beginning of their relationship. First, reserve loss constitutes the central issue concerning the IMF-program country relationship. The Fund defines the reason and the remedy for reserve loss based on the monetary approach to balance of payments (MBOP) (Polak, 1997). The MBOP implies that, in a small open economy with perfect capital mobility and fixed (or pegged) exchange rates, the supply of money cannot be regarded as an exogenous policy instrument.4 Therefore, a change in the base money implies a change in reserves in the opposite direction (Johnson, 1973). Based on the MBOP, incompatible macroeconomic policies refer to the policies that finance budget deficits by money creation at the given exchange rate peg. Because base money and reserves cannot be controlled simultaneously, an increase in base money will lead to a loss of reserves. Consequently, the IMF’s conditionality associated with its programs generally includes a reduction in the base money and budget deficit, along with the recommendation of the domestic currency’s devaluation. Second, the interest rate on the IMF loans is lower than that on a comparable loan from the private capital markets. In the past couple of years, the IMF has made its interest rates public.5 The basic rate on the loans associated with standby and Extended Fund Facility (EFF) is linked to the Special Drawing Rights (SDR) interest rate, which was 2.22% during the week of 6 December through 12 December 2004.6 The interest rate on Structural Adjustment Facility (SAF) and Extended Structural Adjustment Facility (ESAF) can be determined by a proxy, because the IMF established the Poverty Reduction and Growth Facility (PRGF) in 1999 and replaced ESAF to strengthen its anti-poverty focus. The current published interest rate on PRGF 3 Mosley (1987) discusses the bargaining process with respect to conditionality associated with structural adjustment lending by the IMF and the World Bank. 4 As opposed to fixed exchange rates that imply a commodity standard, pegged exchange rates imply unilateral exchange rate management without a commodity standard. In this paper, we use the term pegged exchange rates, because most developing countries peg their exchange rates unilaterally. We will discuss the classifications of foreign exchange regimes in the next section. 5 The following website contains information on the interest rates associated with various IMF programs: http:// imf.org/external/np/exr/facts/howlend.htm. 6 With respect to both standby and EFF, a surcharge is added to the SDR interest rate depending upon the extent to which the country is over its quota. The surcharge is 100 basis points for credit over 200% of quota, and 200 basis points for credit over 300% of quota. The most current weekly SDR rate can be found at http://imf.org/external/np/fin/rates/sdr_ir.cfm.

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

267

is only .5%. These rates can be compared to the riskless rate of international lending, London Interbank Offered Rate (LIBOR), that represents the widely used benchmark for short-term interest rates. On 10 September 2004, the LIBOR on a 12-month, US$-denominated loan was 2.36%. Compared to LIBOR, the interest rates on the IMF loans are lower. Additionally, by its Articles of Agreement, the IMF is not permitted to include a country risk component in its interest rate, which further decreases the interest rate on the IMF loans. Third, both the IMF and program countries have known that conditionality is not legally enforceable. Therefore, if a program country does not comply with the Fund’s conditionality, the maximum penalty that can be imposed by the Fund is to exempt this country from future IMF programs. While the above points must have been clear to the IMF and its developing members, imperfect information must have existed at the beginning of the IMF-program country relationship with respect to the players’ reaction to each others’ decisions. Assuming the IMF’s description of and remedy to the problem is correct, the question is whether program countries implement the IMF’s advice and how the IMF would react to program countries’ decisions regarding program implementation. When the IMF considers providing the second program to a country, it certainly has more information about the program country’s willingness to implement the IMF’s conditionality. As the number of programs increases, each player’s decision-making process should become clearer to the other. Because of the revolving nature of the IMF support to developing countries in the past three decades, it is plausible to suppose that initial imperfect information regarding the players’ reaction to each other’s decisions has been augmented. Suppose that, even though the program country did not implement the Fund’s conditionality in an earlier program and has been engaged in inconsistent macroeconomic policies, it receives a subsequent program. The continuing low-cost Fund support irrespective of the program country’s implementation of conditionality may give the program country the incentive to pursue policies that have led to balance of payments problems in the first place. Additionally, the country may follow even riskier macroeconomic policies, because IMF loans at subsidized interest rates reduce the relative price of these policies. Clearly, the implementation of conditionality is of great importance, because conditionality is supposed to make IMF loans relatively more expensive through the imposition of policy changes on the program country. Assuming that the Fund’s conditionality is both credible and correct in the sense that it can reduce the program country’s balance of payments problems, conditionality cannot fulfill its role as an implicit tax on inconsistent macroeconomic policies, if it is not effectively implemented. The above framework implies that pegged exchange rate regimes coupled with expansionary fiscal and monetary policies will lead to loss of reserves. Additionally, the provision of IMF programs irrespective of countries’ implementation of conditionality may generate recurrent program participation. Among others, there are three reasons to question the validity of the suggested framework: the use of the first-generation currency crisis setup, the counterfactual, and the role of external factors. As far as the currency crisis model is concerned, this paper views balance of payments crises in a first-generation (canonical) currency crisis framework (Krugman, 1979). One possible objection to the use of the canonical currency crisis model is the possibility of a currency crisis even under a credible peg, which is suggested by self-fulfilling (second-generation) currency crisis models (Obstfeld, 1986). In this setting, information-related problems may undermine an otherwise credible peg. Imperfect information regarding the size or the use of reserves to protect the currency may motivate players to engage in a reserve game that would create multiple equilibria under a given set of fundamentals. Therefore, according to the second-generation currency crisis models,

268

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

investor behavior that cannot be explained by program countries’ fundamentals leads to currency crises. If one believes that currency crises are self-fulfilling, topics such as the effectiveness of IMF programs or moral hazard associated with these programs would not be meaningful. In this case, the IMF would have provided financial support to countries that have not done anything wrong, such as implementing inconsistent macroeconomic policies at the given foreign exchange peg. The main reason for using the first-generation currency crisis framework and not the selffulfilling model in this paper is the fact that countries that were suspected of experiencing selffulfilling currency crises during the 1990s were emerging countries. These countries account for a small fraction of developing economies. Additionally, currency crises of the 1990s may not be classified as self-fulfilling after all.7 Still, the assumption of the first-generation currency crisis model leads to another question. Why would the government of a country repeatedly implement inconsistent macroeconomic policies that lead to balance of payments crises? Calvo (1987) suggested that politicians with low time preference would adopt such regimes. More generally, the answers provided in Alesina and Drazen (1991) and Cukierman et al. (1992) are related to the characteristics of the political environment in which policy makers may decide not to implement a credible and eventually successful stabilization program. In addition to the type of currency crisis, the second reason for questioning this paper’s framework is related to the counterfactual (the outcome in the absence of the IMF). It means that, despite the recurring nature of balance of payments problems in developing countries in the presence of the IMF, the outcome would have been worse in the absence of the IMF. In fact, it has been suggested that monetary and fiscal policies would have been inefficiently tight without the IMF (Jeanne and Zettelmeyer, 2004). The counterfactual argument is certainly valid in the IMFrelated research as well as in all economic policy-related research. However, the difficulty of estimating the counterfactual is obvious.8 Despite the validity of the counterfactual argument, the

7 Frankel (1996), for example, argues that currency crises of the 1990s were largely the fault of governments’ macroeconomic policies, not of the markets. Additionally, first-generation crises models are sometimes dismissed in favor of the second-generation models, because a particular crisis cannot be explained based on the changes in usual fundamentals, such as the growth rate of base money, change in the exchange rate, etc. The interaction of weakening capital controls and undercapitalization of the banking sector, expected budget deficits because of possible bailouts in the banking sector, and the inefficient use of funds that were poured into emerging markets may have triggered the currency crises of the 1990s (Chang and Velasco, 1998; Krugman, 1996). 8 There may be three proxies for the counterfactual: macroeconomic policies implemented by the nonmembers of the IMF, the nonprogram members of the IMF, and by the current members of the IMF during their nonmembership years. There are various problems associated with using the nonmembers of the IMF as the counterfactual. Monaco, Liechtenstein, Andorra, and Vatican City do not qualify to be in the sample, because these countries do not have any control over their monetary policies. Additionally, they are considered developed countries. No data are available for Nauru, Palau, and Tuvalu. Zaire became a nonmember country in 1997 that marks the end of the sample period. Taiwan is left as the only nonmember country that could be used as the counterfactual sample, which is not enough to draw any conclusions (Evrensel, 2002). As to the nonprogram members, there are 18 developing members of the IMF that have not received any IMF program during the sample period (Angola, Antigua & Barbuda, Bhutan, Botswana, Cape Verde, Kiribati, Lebanon, Malaysia, Maldives, Namibia, Paraguay, Seychelles, St. Kitts and Nevis, St. Lucia, Suriname, Swaziland, Tonga, Vanuatu). However, the use of the nonprogram countries as the counterfactual for program countries is not appropriate. Evrensel (1999) shows that program and nonprogram countries constitute two distinct groups with statistically significantly different macroeconomic policies. Finally, regarding the nonmembership years in current program countries, the macroeconomic policies these countries implemented as a nonmember would reflect their policy decisions in the absence of the IMF. However, there are two problems with this approach. First, for most developing countries, the data regarding relevant macroeconomic variables are not available for the nonmembership years. Second, the nonmembership years coincide with the gold exchange standard and later dollar exchange standard era, which is not consistent with the unilaterally pegged foreign exchange regimes of the post-Bretton Woods period.

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

269

use of the actual data should not be dismissed too quickly, if the question is whether program countries tend to implement crisis-causing macroeconomic policies, everything else constant. The third reason for questioning this paper’s framework is the possibility of external shocks, which means that currency crises may be initiated by external shocks, largely unrelated to program countries’ macroeconomic policies. However, it is unlikely that a large number of program countries would be struck by exogenous, adverse shocks in a systematic fashion during a period of, say, 30 years. Additionally, there may be issues related to international lending. First, one would argue that the observed frequent program participation may not be a negative phenomenon, because IMF programs increase program countries’ access to private capital markets. However, the existing empirical research finds either no or only weak evidence that IMF programs provide better access to private capital markets (Evrensel, 2004). Second, credit rationing in international lending and the associated difficulty of obtaining funds in private capital markets may motivate developing countries to seek IMF assistance more frequently. However, Evrensel (2004) shows that, independent of their income level, higher creditworthiness of borrowing countries increases their access to private capital markets. Therefore, we argue that, despite the above-mentioned points, the approach in this paper is expected to deliver informative results regarding the relationship between program participation and macroeconomic policies. The variables used in the empirical analysis are based on the models that identify unsustainable monetary and fiscal policies under a pegged exchange rate regime as the triggering condition for reserve loss (Burnside et al., 2000; Calvo, 1987; Chang, 1994; Helpman and Razin, 1987; Krugman, 1979). They illustrate the costs of a pegged exchange rate regime that is allowed to have real effects. An exchange rate management of this sort will have an effect on variables, such as the real exchange rate, current account, and reserves. A reduction in the depreciation rate (or, in the extreme case, an exchange rate freeze) without an accompanying reduction in the private disposable income, will lead to the appreciation of the real exchange rate, loss of reserves, and a reduction in the country’s international reserves. 3. Empirical analysis In this section, we describe our data and variables, provide descriptive statistics of our data using survival analysis, present the results of the generalized least squares regression, and discuss the empirical results. 3.1. Data and variable description The data contain 91 developing members of the IMF that received IMF programs during the period 1967–1996.9 Annual macroeconomic data are obtained from the IMF’s International Financial Statistics CD-ROM of December 2000. Information regarding the type of IMF programs is based on the IMF’s Annual Reports. Four structural adjustment programs are considered: standby, EFF, SAF, and ESAF.10 9 The term developing country refers to the countries in Latin America, Africa, the Middle East, and Asia, excluding the high-income, oil-producing Middle Eastern countries, the former Soviet Union and Yugoslavian republics, and the Eastern European countries. Additionally, the nonprogram developing members of the IMF are not included in our data. 10 While SB and EFF provide balance of payments support to the middle- and high-income developing countries, SAF and ESAF are designed for the low-income developing countries. Standby arrangements are provided for a year with a possible extension up to 3 years. SAF and ESAF last longer and imply longer periods of repayment. See Schadler et al. (1995) for more information on these programs.

270

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

Fig. 1. Interprogram periods.

The variables that are used in the empirical analysis are based on the MBOP that is discussed in the previous section: nominal and real exchange rates, reserves, base money, and current account. Except for the nominal and real exchange rates, all variables are expressed as a ratio of GDP. Nominal exchange rate implies the amount of domestic currency necessary to buy US$ 1. Real exchange rate is calculated as the ratio of the domestic currency price of the US basket to that of the domestic basket. Nominal and real exchange rates are used as log differences.11 In this paper, we are particularly interested in the interprogram periods, during which a country does not receive any IMF program.12 Because interprogram periods are the periods between the two consecutive programs, we do not include the period between the start of IMF membership and the first IMF program. Fig. 1 illustrates the meaning of interprogram periods for a country that has received three IMF programs during the sample period. Interprogram periods (IPPs) are described in a way that the first interprogram period is the period between the first and the second IMF program.13 We create a variable that measures the mean IPP years for country j by dividing the sum of all IPP years by the number of interprogram periods. That is: Pk j Pn l IP j ¼

l¼1

i¼1

kj

yearsli

;

where IP j is the mean interprogram years in country j; kj the number of interprogram periods in country j; yearsli is the ith year within the lth interprogram period. 3.2. Survival analysis We use survival analysis to describe our data regarding program participation. If program participation is viewed as a failure, the data can be described as a multiple-failure data, because countries seem to enter IMF programs recurrently. Therefore, being in an IMF program can be 11

Even though the IMF’s International Financial Statistics reports nominal and real effective exchange rates for some countries, they are infrequent. Because their use leads to a considerable loss of observations, we use the nominal and real exchange rates with respect to the US dollar. 12 Countries may receive technical support or be under the IMF’s surveillance during interprogram years. Because neither technical support nor surveillance is associated with loans or conditionality, they do not constitute an IMF program. 13 Fig. 1 shows the interprogram periods in a hypothetical country, where the observation period ends in a nonprogram year. Clearly, the period following the third IMF program cannot be identified as an interprogram period. Even though we have countries in our data that are similar to the hypothetical country in Fig. 1, some countries were in an IMF program when the observation period ended.

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

271

Fig. 2. Determination of the origin of risk and failure.

interpreted as a failure, much like hospitalization. Survival time refers to the period, during which a country is under no IMF program. The following discussion of survival analysis provides an overview of this technique and is based on Tableman and Kim (2004). We will use Fig. 2 to illustrate the characteristics of our survival data. First, in survival analysis, it is important to determine the origin of survival time, because the origin implies the time when the clock starts ticking for a possible failure. In our case, the determination of origin is closely related to IMF membership, because only member countries receive IMF-supported programs. Therefore, IMF-membership represents the exposure event at which countries become at risk of being in an IMF program. In our sample, 70 countries became members of the IMF during the period between 1946 and 1967. Because comprehensive data are not available for this period, our data start in 1967. Therefore, for these 70 countries, we consider the origin of survival time as 1967.14 For the remaining 21 countries that became IMF members in the 1970s and 1980s, survival time starts in the year of their membership. Second, we want to discuss censoring, which represents an important characteristic of survival time data. In fact, the main difference between linear and logistic regressions and survival analysis lies in the fact that the latter analysis implies censoring or incomplete observations of subjects. Censoring refers to observations that are incomplete due to random factors affecting the subjects. Our data can be described as right-censored, because the observation period terminates at some point in time (in 1996), after which program participation cannot be observed. As of the last observation year, some countries may be in an IMF program and some are not known to be in a program (Fig. 2 illustrates the later possibility). Therefore, survival can be observed only partially, because countries that are not known to be in a program at the end of the observation period may or may not receive IMF programs in the future. Table 1 describes the survival time data. When all countries are considered, there are 91 countries in the dataset and 2455 observations. We have on average about 27 years of observation for each country. Two hundred and seventy-five observations out of 2730 total observations were taken out because they imply the time before the start of risk (IMF membership). Therefore, the 14 Even though a small number of developing countries received IMF programs prior to 1967 (such as Argentina, Brazil, Turkey, etc.), prior to the early 1970s, the IMF provided balance of payments support to mostly developed countries. Therefore, we assume that the IMF’s financial support provided to developing countries prior to 1967 may be negligible.

272

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

Table 1 Description of survival time data Category

Total

Mean

Minimum

Median

Maximum

26.98 11.01

5 1

29 11

29 26

Countries that received only standby or EFF (no. of countries: 47) Time at risk 1275 27.13 6 Failures 448 9.53 1

29 9

29 24

Countries that received SAF or ESAF and standby or EFF (no. of countries: 44) Time at risk 1180 26.82 5 Failures 554 12.59 3

29 13.5

29 26

All countries (no. of countries: 91) Time at risk 2455 Failures 1002

number of observations at risk is 2455. The median number of observations is 29, with a minimum of 5 and a maximum of 29 observations. One thousand and two observations are associated with failures, where failure is described as an IMF program. Countries receive an average of 11 years of IMF programs during the observation period. During the period 1967–1996, countries were under the IMF care between (minimum) 1 year and (maximum) 26 years. We also make a distinction between two groups of countries: countries that received either standby or EFF and countries that received SAF or ESAF (although a small number of these countries received standby or EFF as well). When only standby and EFF programs are considered, 47 countries received any one of these two programs. We have an average of 27 years for each country. Countries spent an average of almost 10 years under a standby or EFF program. Table 1 also shows that 44 countries received all four types of programs. We have on average 27 years of observations for each country. Countries spent an average of almost 13 years under any of the four IMF programs. Table 2 summarizes the survival data. When all countries are considered, the incidence rate was .41, which means that 41% of 2455 observations at risk are associated with one of the four types of IMF programs. When only countries that received standby or EFF are considered, the incidence rate is .35, whereas it is .47 for countries that received a combination of SAF or ESAF, along with standby or EFF. It means that the probability of failure is higher in countries that receive a combination of long- and short-term programs. Next, based on Tableman and Kim (2004), we discuss the nonparametric representation of the survival time data using survival functions. The probability distribution of survivor time duration, T, is shown by the distribution function: FðtjXÞ ¼ P½T  tjX;

(1)

where X refers to the vector of covariates. The probability distribution function indicates the probability that the random variable T is less than or equal to some value t. Then, the Table 2 Summary of survival time data

All countries Countries with standby and EFF only Countries with SAF or ESAF and standby or EFF

Time at risk

Incidence rate

No. of countries

Survival time 25%

50%

75%

2455 1275 1180

.4082 .3514 .4695

91 47 44

4 5 4

6 6 5

8 7 8

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

273

corresponding density, f(t), and survivor functions, S(t), are described by f ðtjXÞ ¼

@FðtjXÞ ; @t

(2)

SðtjXÞ ¼ P½T  tjX ¼ 1  FðtjXÞ;

(3)

respectively. The survivor function implies the probability that a randomly selected country will have a survival time greater than or equal to some value t. The Kaplan–Meier estimator of the survivor function considers survival up to a particular point in time as a series of steps defined by the observed survival and failure times. It estimates the conditional probabilities of survival at the current time point given survival up to the previous time point and multiplies them to provide an estimate of the survival function. That is: Y SðtÞ ¼ PðT > tÞ ¼ pi ; (4) yðiÞt

where y(i) denotes the ith distinct ordered censored or uncensored observation and pi represents the conditional probability of survival at each time point. If ni and di indicate the number of observations where countries are at risk and the observed number of failures, respectively, the estimated survival function can be shown as ˆ ¼ SðtÞ

 k  Y  ni  d i  Y ni  d i ¼ ; ni ni i¼1 yðiÞt

(5)

where yðkÞ  t < yðkþ1Þ . Fig. 3 shows two survival functions depending upon the types of programs considered. While Fig. 3a indicates the survival function for all program types, Fig. 3b shows two survival functions that are associated with the standby–EFF and SAF–ESAF combination. As the summary of survival data indicates, compared to countries with only standby or EFF arrangements, countries that received mainly SAF or ESAF, have a shorter time to survive. It seems that once low-income countries enter SAF or ESAF, they tend to receive successive programs without any interprogram periods between these programs. The tests for the equality of survivor functions (log-rank, Wilcoxon, and Cox) indicate that the null hypothesis of the equality of survivor functions associated with different program types can be rejected at better than 1% significance level. Survival analysis provides two important pieces of information. First, it confirms the recurrent nature of IMF programs. Second, it demonstrates the differences in incidence (failure) rates associated with various program types. While the first point justifies the further empirical examination of the relationship between program participation and macroeconomic policies, the second point motivates the inclusion of program types in the estimations. 3.3. Generalized least squares results We apply generalized least squares (GLS) to our data to examine the nature of the relationship between the length of the interprogram period and macroeconomic policies during this period in program countries. Because the Cook and Weisburg test indicated the presence of heteroscedasticity, we employ the GLS model with heteroscedastic panels. Additionally, we incorporated the presence of a panel-related AR(1) process to account for autocorrelation within panels.

274

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

Fig. 3. Survival functions: (a) all countries; (b) countries that received either the standby–EFF or SAF–ESAF combination.

We have three program type dummies. PTYP1 is 1 for standby or EFF and zero otherwise. PTYP2 is 1 for the combination of all four programs and zero otherwise. PTYP3 is 1 for SAF and ESAF and zero otherwise, which is included in the intercept. Regarding foreign exchange regimes, we have five dummy variables associated with foreign exchange regimes. ER1 takes the value of 1 if the country has multiple exchange rates and zero otherwise. ER2 takes the value of 1 if the country pegs its currency to a single foreign currency and zero otherwise. ER3 takes the value of 1 if the country pegs its currency to a basket of currencies and zero otherwise. ER4 takes the value of 1 if the foreign currency regime is called managed floating and zero otherwise. ER5 takes the value of 1 if the foreign currency regime is called floating and zero otherwise, which is included in the intercept. The above classification of exchange rate regimes is based on the IMF’s annual reports. Annual reports tabulate exchange rate arrangements in member countries, which should be

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

275

understood as declarations to the IMF by its member countries regarding their exchange rate regime. As Broda (2001) argues, exchange rate regimes published by the IMF reflect de jura, not de facto classifications. The above dummies for exchange rate regimes are from Evrensel (1999), who uses the basic information published in the IMF’s Annual Reports and creates a more differentiated classification than Annual Reports suggest, especially with respect to the various types of pegged exchange rate regimes. Additionally, we employed the foreign exchange regime classification by Levy-Yeyati and Sturzenegger (2003).15 Because Evrensel (1999) and LevyYeyati and Sturzenegger (2003) classifications are similar, especially in the pegged exchange rate regime category, the reported results are based on the classification in Evrensel (1999). We use interaction variables between continuous and indicator variables as well as between two continuous variables. The inclusion of interaction variables is important for two reasons. First, the interaction terms between macroeconomic variables and program type dummies can give us insights whether the effects of the macroeconomic variables on the mean interprogram years differ among the different program types and exchange rate regimes. Second, the relevance of the interaction terms between the continuous macroeconomic variables is based on the discussion of the monetary approach to balance of payments in Section 2. The macroeconomic variables that we use are relevant under the pegged foreign exchange regimes. For example, the interaction between the reserves and the real exchange rate is important because we would expect that the lower rates of real depreciation accompanied by lower reserves will make the country ask for an IMF program rather sooner than later (shorten the mean interprogram years). The interaction between the base money and the real exchange rate is expected to illustrate that the higher rates of base money creation and lower rates of real depreciation will shorten the mean interprogram years as well. Additionally, lower reserves and a higher current account deficit are expected to bring an IMF program in the country rather sooner than later. The GLS results shown in Table 3 are obtained after running a series of models, starting with the basic model (without nominal exchange rate, inflation rate, and their interactions) and excluding variables with high p-values. We kept a variable whose main effect is statistically insignificant in our model, as long as at least one of its interaction terms is statistically significant. To interpret the coefficients on the interaction terms between two continuous variables, we use the percentiles of the variables (the 25th, 50th, and 75th). Table 4 shows the percentiles of the continuous macroeconomic variables in the data. Using the estimated coefficients in Table 3 and the percentiles of the relevant variables in Table 4, Table 5 summarizes the effects of the explanatory variables on the mean interprogram years. When interpreting our results in Table 5, we use the term shorter or longer interprogram years to express the effects of the explanatory variables on the mean interprogram years. Shorter interprogram years indicate that the country will receive the next IMF program sooner. The opposite is true for longer interprogram years. Now, we summarize our GLS results. Higher reserves lead to longer interprogram years; however, the coefficient is rather small. A 1% increase in reserves extends the interprogram period by 2 months (Table 3). Because the following results include interaction variables, their effects on interprogram years are shown in Table 5 (with the use of Tables 3 and 4). The effect of the real exchange rate on the mean interprogram years interacts with the program types SAF or ESAF (PTYP3). Considering the fact that these programs are provided to the low-income 15 Levy-Yeyati and Sturzenegger (2003) data regarding the exchange rate classifications are available at http:// www.utdt.edu/ely/base_2002.xls. The accompanying methodological note is available at http://www.utdt.edu/ely/ AppendixAER.pdf.

276

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

Table 3 Results of the GLS estimations (dependent variable: mean interprogram years) Explanatory variables

Coefficient

z

P > jzj

Constant Real exchange rate Reserves Base money Current account Standby or EFF All four programs Multiple exchange rates Single currency peg Standby or EFF  current account Standby or EFF  currency basket peg Standby or EFF  managed floating All four programs  currency basket peg All four programs  current account SAF or ESAF  real exchange rate Single currency peg  base money Currency basket peg  base money

7.813 .00046 .143 .029 .1103 5.169 4.393 1.427 1.003 .152 1.225 3.417 2.109 .229 .026 .136 .172

19.172 .474 5.094 3.124 1.238 18.848 11.501 1.936 2.643 1.686 -2.797 9.451 5.318 2.485 1.856 1.937 3.866

.0000 .634 .0000 .002 .216 .0000 .0000 .0089 .009 .092 .005 .0000 .0000 .013 .065 .053 .0000

No. of observations No. of groups Log likelihood Wald x2 (9) Prob > x2

191 43 322.68 81.57 .0000

countries, the results indicate that the higher rates of depreciation in the real exchange rate in the low-income countries is associated with longer interprogram years. However, the coefficient is small. The effect of the base money on the mean interprogram years interacts with foreign exchange regimes. If a country pegs its currency to a single currency (ER2) or to a currency basket (ER3), this leads to longer interprogram years. However, the relevant coefficients are small, indicating an extension of 1 or 2 months. The effect of the current account on the mean interprogram years depends upon the program type. If a country receives the standby–EFF combination or all four programs, higher percentiles of the current account (smaller current account deficits) lead to longer interprogram years. Again, the relevant coefficients are small, indicating an extension of 2 or 3 months. Table 4 Percentiles of the continuous explanatory variables Explanatory variablesa

Real exchange rate Reserves Base money Current account a

Percentiles 25th

50th

75th

.88 1.67 10.77 6.79

9.67 4.06 13.49 3.93

29.32 7.37 18.99 .64

Real exchange rate is expressed in log difference. Reserves, base money, and current account are expressed as a percentage of GDP.

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

277

Table 5 Interpretation of estimated coefficients Explanatory variablea

Estimated effect of the explanatory variable on mean interprogram years, accounting for interaction effects

Real exchange rate (RER) If PTYP3 = 1

.026 PTYP3 .026 years

Base money (BM) If ER2 = 1 If ER3 = 1

.029 + .136 ER2 + .172 ER3 .107 years .143 years

Current account (CA) If PTYP1 = 1 If PTYP2 = 1

.152 PTYP1 + .229 PTYP2 .152 years .229 years

Standby–EFF (PTYP1) If ER3 = 1 25th percentile of current account 50th percentile of current account 75th percentile of current account

5.169  1.225 ER3 + 3.417 ER4 + .152 CA 7.4 years 6.9 years 6.5 years

If ER4 = 1 25th percentile of current account 50th percentile of current account 75th percentile of current account

2.8 years 2.4 years 1.9 years

Standby–EFF–SAF–ESAF (PTYP2) If ER3 = 1 25th percentile of current account 50th percentile of current account 75th percentile of current account

4.393 + .2109 ER3 + .229 CA

SAF–ESAF (PTYP3) 25th percentile of real exchange rate 50th percentile of real exchange rate 75th percentile of real exchange rate Pegged to a single currency (ER2) 25th percentile of base money 50th percentile of base money 75th percentile of base money

7.813 + .026 RER 7.8 years 8.1 years 8.6 years 1.003 + .136 BM 2.5 years 2.8 years 3.6 years

Pegged to a currency basket (ER3) If PTYP1 = 1 25th percentile of base money 50th percentile of base money 75th percentile of base money

1.225 PTYP1 + 2.109 PTYP2 + .172 BM .63 years 1.1 years 2.1 years

If PTYP2 = 1 25th percentile of base money 50th percentile of base money 75th percentile of base money

3.9 years 4.4 years 5.4 years

Managed floating 1 (ER4) If PTYP1 = 1

5.7 years 5.1 years 4.2 years

3.417 PTYP1 3.417 years

a Percentiles are associated with the continuous variables. For example, the effect of the standby–EFF combination on the mean interprogram years, assuming the exchange rate regime is pegged to a basket of currency (ER = 3) and the current account deficit is in the 25th percentile is: [5.169  1.225 + (.152  6.798)] = 7.427 years.

278

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

The effect of program types on the mean interprogram years interacts with exchange rate regimes, current account, and real exchange rate. For example, the effect of the standby–EFF combination (PTYP1) interacts with the current account and the foreign exchange regime. The results indicate that, if a country has larger current account deficits (25th and 50th percentiles) and pegs its currency to a currency basket (ER3), interprogram years are shortened by about 7 years. Even with a smaller current account deficit, interprogram years are shorter by 6.5 years. If the country has a managed floating foreign exchange regime (ER4), the interprogram years are shortened by a smaller number of years. As the current account deficit becomes smaller, interprogram years are shortened from 2.8 to 1.9 years. The effect of all four programs on the mean interprogram years interacts with foreign exchange regime (ER3) and current account. If a country pegs its currency to a currency basket (ER3) and has the 25th percentile current account deficit (largest deficit), the mean interprogram years are shortened by 5.7 years. An improvement in the current account makes only a small difference. If the current account deficit becomes smaller, interprogram years decline down to 4.2 years. The effect of the SAF–ESAF combination (PTYP3) on the mean interprogram years interacts with the real exchange rate. The mean interprogram years become longer (from 7.8 to 8.6 years), if the country allows its real exchange rate to depreciate at a higher rate. As to the effects of the foreign exchange regimes on the mean interprogram years, base money and some program types are the relevant interaction variables. If a country pegs its currency to a single currency (ER2), increases in the base money are associated with longer interprogram years (from 2.5 to 3.6 years). If a country pegs its currency to a currency basket, the effects of this foreign exchange regime on the mean interprogram years interact with program types and base money. Independent of the type of the program (the standby–EFF combination or all four programs), the mean interprogram years increase with base money. For example, if a country receives all four programs and the base money creation is at the 75th percentile, this will lengthen the mean interprogram years by 5.4 years. Finally, if a country adopts a managed floating regime and receives the standby–EFF combination, the mean interprogram years get longer by 3.4 years. The results indicate that the effects of the changes in the real exchange rate, base money, and current account on the length of interprogram years are rather small. Even though the effects of exchange rate regimes on the length of interprogram years are larger, the results seem to be counterintuitive. Exchange rate regimes interact with base money and the results imply that higher growth rates in base money increases the length of interprogram years under pegged currencies. The effect of the exchange rate regime ER3 (pegged to a currency basket) on interprogram years provides some insight into this result. The ER3 regime interacts with base money as well as two program types (standby or EFF and all four programs). Because program types can be used as a proxy for income levels of program countries, the results suggest that lowincome countries can increase the length of interprogram years under a pegged exchange rate regime (pegged to a currency basket) and an expansionary monetary policy much more than high-income countries. One can argue that the countries with higher income levels and probably more established financial markets cannot survive with inconsistent policies as long as lowincome countries do. This point becomes clear when observing the effects of program types on the length of interprogram periods. When high-income countries adopt a managed floating regime, the reduction in the length of interprogram periods is much less than in the case of highincome countries that peg their currency to a currency basket. This means that it pays of for highincome countries to adopt less managed exchange rate regimes. Regarding the low-income countries, the depreciation in the real exchange rate leads to longer interprogram periods.

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

279

These results stress the relevance of income levels and related characteristics of countries (such as financial market development) on the effects of macroeconomic policies. Table 4 reports the results of the diagnostic tests. In the presence of heteroscedasticity-robust standard errors, the Wald statistic is used to demonstrate the overall explanatory power of the model (Wooldridge, 2002). Our results indicate that the null hypothesis is rejected at a significance level better than 1%. Additionally, we ran other GLS and OLS models to check the sensitivity of our results to alternative model specifications. In an alternative GLS model, we define the dependent variable as the number of interprogram years in each country. In this case, we used the means of the independent variables over time by collapsing the pooled data into the cross-section data. The results of this alternative GLS model were largely similar to those of the reported GLS model. As to the results of the OLS estimation, the effects of reserves, real exchange rate, base money, and current account on the mean interprogram years are fairly similar to the GLS results, even though the coefficients of the OLS model are smaller. However, the OLS model picks up additional interaction variables related to foreign exchange regimes and the real exchange rate. For example, even if a country has multiple exchange rates, higher rates of depreciation in the real exchange rate increase the mean interprogram years by up to 5.7 years. 3.4. Discussion of empirical results The results of this paper are consistent with those of other studies. For example, our results associated with the real exchange rate confirm the results of Conway (1994), which indicate that the depreciation of the real exchange rate leads to less frequent program participation. Our results are also consistent with those of Dreher and Vaubel (2004b), who showed that, as the amount of IMF loans increases, program countries tend to have larger budget deficits and implement more expansionary monetary policies. In this paper, we look at the issue from a different point of view and find out that expansionary monetary policies coupled with pegged foreign exchange regimes shorten the average length of the interprogram periods. The main point in all these papers is the fact that, to a great extent, countries are in charge of their macroeconomic policies and we should be watchful of the circumstances in which they choose one set of policies over another one. As Dreher and Vaubel (2004b) show, the IMF assistance may change countries’ macroeconomic policies in an undesirable direction. A similar point has also been made by Easterly (2005), who states that the lending by the IMF and the World Bank has not succeeded in adjusting macroeconomic policies and growth outcomes. Extreme imbalances in inflation, budget deficit, current account, inflation, and real overvaluation of the exchange rate are not affected by the time spent under the structural adjustment programs and the number of programs received by countries (Easterly, 2005). In terms of policy implications, there are two issues. First, the IMF’s conditionality associated with a program and the possible effects of conditionality on the program country’s policies during the nonprogram years should be considered. In recent years, the IMF’s conditionality has been criticized for being too broad. Vaubel (1991) and Dreher and Vaubel (2004a) point out that the current practice of conditionality may gain from a more focused approach that targets the macroeconomic variables that can be effectively controlled. In fact, the IMF’s current rethinking of program design along the lines of streamlining conditionality constitutes an effort to increase the effectiveness of IMF programs and prevent frequent program participation (Erbas¸, 2004).16 16 Streamlining conditionality implies that conditionality should not aim as many as possible target variables. It should cover only the most critical macroeconomic measures that are necessary for the success of a program. The ultimate objective of streamlining conditionality is to increase the effectiveness of Fund-supported programs.

280

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

Additionally, we would like to emphasize the content of the conditionality. If one wants the IMF’s conditionality to be effective during the program years and to observe its continuing positive effects during the nonprogram years, then conditionality should include institutional reforms and not solely the nominal values of some target macroeconomic variables. The IMF’s experience in the developing countries in the past 30 years indicates that, for example, the IMF’s expectation of the reduction in the program country’s inflation rate from 25% to 10% may not even be materialized during the program years and certainly not after the program is over. The main reason for this outcome lies in the institutional characteristics of the program country, in which inflation is created. Second and related to the first point, the IMF’s decision regarding the provision of funds to a country is of great relevance. As discussed in Section 2, the provision of cheaper funds irrespective of the country’s implementation of the IMF’s conditionality may motivate countries to implement inconsistent macroeconomic policies. Bird (2001) suggests that the IMF identify indicators of success and failure on the basis of its own institutional objectives to reduce recidivism in program participation and improve program completion. Similarly, the Independent Evaluation Office (2002) recommends that the IMF should review the program countries’ performance during the previous programs and provide a candid estimation of the nonimplementation risk in the prospective program, especially in the case of the frequent or prolonged use of the Fund’s resources. 4. Conclusion This paper examines the relationship between program participation and macroeconomic policies of program countries during the interprogram years. The theoretical literature predicts that a country with a pegged exchange rate regime, an expansionary monetary policy, and a real appreciation of the domestic currency will lose reserves and is more likely to ask for IMF support. Our results verify this prediction for countries of all income levels. However, the results also indicate that the contribution of inconsistent macroeconomic policies to program participation is especially strong in the high-income countries. Therefore, it may be possible to drastically reduce program participation especially in this country group. This paper’s as well as related studies’ suggestion focuses on the IMF’s treatment of its own conditionality. Reforming program countries’ institutional structure should be a priority, because inconsistent macroeconomic policies stem from the institutional framework. Therefore, the Fund’s conditionality should include institutional reforms in program countries. However, conceptual changes in conditionality will remain ineffective if they are not integrated in the IMF’s decision-making process. The IMF should rather enforce the content of its conditionality by evaluating the prospective program country’s record regarding its implementation of conditionality in previous programs before providing the new program. The future research should consider the effects of subsequent programs on program countries’ macroeconomic policies. The possibility of a learning curve implies that, as countries receive subsequent programs, the extent of their inconsistent macroeconomic policies may decline. It is important to investigate whether there is a learning curve and its variation among the different country groups, so that the IMF’s conditionality is designed to be more effective. References Alesina, A., Drazen, A., 1991. Why are stabilizations delayed? Am. Econ. Rev. 81, 1170–1188. Bird, G., 2001. IMF programs: do they work? Can they be made to work better?. World Dev. 29, 1849–1865.

A.Y. Evrensel, J.S. Kim / Economic Systems 30 (2006) 264–281

281

Broda, C., 2001. Coping with terms-of-trade shocks: pegs versus floats. Am. Econ. Rev. 91, 376–380. Calvo, G.A., 1987. Balance of payments crises in a cash-in-advance economy. J. Money Credit Bank. 19, 19–32. Chang, R., 1994. Endogenous currency substitution, inflationary finance, and welfare. J. Money Credit Bank. 26, 903–916. Chang, R., Velasco, A., 1998. Financial crises in emerging markets: a canonical model. Federal Reserve Bank of Atlanta Working Paper 98-10. Federal Reserve Bank of Atlanta, Atlanta, GA. Conway, P., 1994. IMF lending programs: participation and impact. J. Dev. Econ. 45, 365–391. Cukierman, A., Edwards, S., Tabellini, G., 1992. Seignorage and political instability. Am. Econ. Rev. 82, 537–555. Dreher, A., Vaubel, R., 2004a. The causes and consequences of IMF conditionality. Emerg. Markets Finance Trade 40, 26–54. Dreher, A., Vaubel, R., 2004b. Do IMF and IBRD cause moral hazard and political business cycles? Evidence from panel data. Open Econ. Rev. 15, 5–22. Edwards, S., 2000. Exchange rate systems in emerging economies. Paper prepared for the Meltzer Commission on the New Financial Architecture, available at: http://www.anderson.ucla.edu/faculty/sebastian.edwards/systems.pdf. Easterly, W., 2005. What did structural adjustment adjust? The association of policies and growth with repeated IMF and World Bank adjustment loans. J. Dev. Econ. 75, 1–22. Erbas¸, N.S., 2004. IMF conditionality and program ownership. Emerg. Markets Finance Trade 40, 10–25. Evrensel, A.Y., 1999. Effectiveness of IMF-supported stabilization programs in developing countries. Unpublished Dissertation. Clemson University, Clemson, SC. Evrensel, A.Y., 2002. Effectiveness of IMF-supported stabilization programs in developing countries. J. Int. Money Finance 21, 565–587. Evrensel, A.Y., Kutan, A.M., 2006. Creditor moral hazard in equity markets: empirical evidence from Indonesia and Korea. J. Int. Money Finance 25, 640–654. Evrensel, A.Y., 2004. Lending to developing countries revisited: changing nature of lenders and payment problems. Econ. Syst. 28, 235–256. Frankel, J.A., 1996. Recent exchange-rate experience and proposals for reform. Am. Econ. Rev. 86, 153–158. Friedman, M., 1998. Markets to the rescue. The Wall Street J., Tuesday, October 13. Gai, P., Taylor, A., 2004. International financial rescues and debtor-country moral hazard. Bank of England Working Paper 217. Bank of England, London. Helpman, E., Razin, A., 1987. Exchange rate management: intertemporal tradeoffs. Am. Econ. Rev. 77, 107–123. IMF, 1971–1997. Annual Report. IMF, Washington, DC. Independent Evaluation Office, 2002. Evaluation of Prolonged Use of IMF Resources. IMF, Washington, DC. Jeanne, O., Zettelmeyer, J., 2004. The Mussa theorem (and other results on IMF-induced moral Hazard). IMF Working Paper 192. IMF, Washington, DC. Johnson, H.J., 1973. Further Essays in Monetary Economics. Harvard University Press, Cambridge, MA. Krugman, P., 1979. A model of balance of payments crises. J. Money Credit Bank. 11, 311–325. Krugman, P., 1996. Are currency crises self-fulfilling? NBER Macroecon. Annu. 11, 345–378. Levy-Yeyati, E., Sturzenegger, F., 2003. To float or to fix: evidence on the impact of exchange rate regimes on growth. Am. Econ. Rev. 93, 1173–1193. Mosley, P., 1987. Conditionality as Bargaining Process: Structural Adjustment Lending, 1980–1986. Essays in International Finance 168. Princeton University, Princeton, NJ. Niehans, J., 1985. International debt with unenforceable claims. Economic Review (Winter), Federal Reserve Bank of San Francisco, 65–79. Obstfeld, M., 1986. Rational and self-fulfilling balance-of-payments crises. Am. Econ. Rev. 76, 72–81. Polak, J., 1997. The IMF monetary model. A hardy perennial. Finance Dev. 34, 16–19. Schadler, S., Bennett, A., Carkovic, M.V., Dicks-Mireaux, L., Mecagni, M., John, J.H., Savastano, M.A., 1995. IMF conditionality: experience under stand-by and extended arrangements. Occasional Paper 128. IMF, Washington, DC. Tableman, M., Kim, J.S., 2004. Survival Analysis using S. Analysis of Time-to-event Data. Chapman & Hall/CRC, Boca Raton, FL. Ul Haque, N., Khan, M.S., 1998. Do IMF-supported programs work? A survey of the cross-country empirical evidence. IMF Working Paper 169. IMF, Washington, DC. Vaubel, R., 1983. The moral hazard of IMF lending. In: Meltzer, A.H. (Ed.), International Lending and the International Monetary Fund. The Heritage Foundation, Washington, DC, pp. 65–79. Vaubel, R., 1991. The political economy of the International Monetary Fund: a public choice analysis. In: Vaubel, R., Willett, T.D. (Eds.), The Political Economy of International Organizations: A Public Choice Approach. Westview Press, Boulder, CO, pp. 204–244. Wooldridge, J.M., 2002. Econometric Analysis of Cross-section and Panel Data. The MIT Press, Cambridge, MA.