Journal Issue
Share
Article

Do IMF-Supported Programs Catalyze Donor Assistance to Low-Income Countries?1

Author(s):
Yasemin Bal-Gunduz, and Masyita Crystallin
Published Date:
November 2014
Share
  • ShareShare
Show Summary Details

I. Introduction

While the catalytic effect of IMF-supported programs on donor flows is potentially an important channel for the economic impact of the IMF engagement in low-income countries (LICs) it received scant attention in the literature. Empirical research on catalytic financing effects has focused overwhelmingly on private capital flows to emerging market economies and findings on the IMF catalytic impact are at best mixed. Nevertheless, this strand of literature appears to have reached an agreement that the catalytic effect is not uniform across different types of IMF-supported programs or across recipient countries.

When assessing the catalytic impact of IMF-supported programs in LICs, stark differences in the financing landscape in comparison to emerging market economies need to be taken into account. LICs tend to depend on donor financing owing to their limited access to international capital flows and low domestic savings. Therefore, in the LIC context the relevant test of the catalytic impact of IMF-supported programs is whether such programs lead to significantly higher donor assistance. Catalysis could occur through two channels: (i) IMF conditionality, aiming at restoring macroeconomic stability and advancing economic reform in program countries to increase resilience to shocks; and (ii) the IMF financing, easing the burden of adjustment to shocks, thereby, supporting economic stabilization and near-term growth. On both fronts IMF engagement may provide some assurances to donors that resources allocated to program countries could be utilized more effectively to support better macroeconomic outcomes.

The major conceptual and methodological challenge in estimating the catalytic impact of IMF-supported programs is selection bias because initial economic conditions will differ systematically for a program versus a non-program country. Countries that approach the Fund tend to already face economic difficulties or expect to experience problems in the near future. If both donors and the IMF respond to economic circumstances of countries, obviously a positive association between programs and donor assistance will be observed. However, causality could only be examined by comparing donor assistance to program and non-program countries experiencing similar prior economic conditions.

The key step to address selection bias is to estimate determinants of IMF-supported programs (participation or selection model). The literature has reached a consensus on the need to improve selection models in order to properly assess program effects and suggested looking into subsets of programs, distinguishing the traditional current account crisis, capital account crisis, and LICs (survey articles by Steinwand and Stone, 2008, and Bird, 2007). Recognizing the wide spectrum of IMF programs tailored to specific needs of the membership, factors affecting participation may vary across different subsets of IMF programs. Therefore, focusing on more homogenous subsets of programs to estimate the participation model, and thereby address selection bias more effectively, is a promising avenue in assessments of program effects.

Interestingly, even programs specifically designed for the LIC membership differ significantly in terms of the nature of balance of payments needs that they address. Some programs deal with immediate needs arising from policy and exogenous shocks while others address more protracted balance of payments needs associated with lack of diversification in economic structures and scarce domestic savings that could be addressed over time through structural transformation. The economic conditions prior to these two types of programs are likely to be quite distinct; suggesting that further disaggregation within LIC programs could improve the performance of the selection model.

More recently a handful of papers have looked into subsets of Fund programs. Bal Gündüz (2009) examines the participation in a subset of IMF-supported programs with LICs addressing policy and exogenous shocks. She reports significant effects from various economic variables (reserve coverage, current account balance to GDP, real GDP growth, macroeconomic stability, and terms of trade shocks) and global conditions (real growth in oil and non-oil commodity prices and world trade). She highlights two factors that are likely to account for higher explanatory power and better model specification, capturing the impact of economic conditions on participation: (i) studying a more homogenous subset of Fund financing addressing immediate balance of payments needs for LICs; and (ii) accounting for observable “supply-side” constraints that would preclude a member’s access to Fund financing.

Building on the empirical strategy of examining participation and impact using more homogenous subsets of IMF-supported programs (Bal Gündüz, 2009Bal Gündüz and others, 2013, and Mumssen and others, 2013) this study investigates the existence of the catalytic impact of the IMF engagement with LICs addressing policy or exogenous shocks. This paper makes several contributions to the empirical literature on the catalytic impact of IMF-supported programs. First, it is the only study to explore the catalytic impact of a unique set of financing arrangements with LICs addressing the policy and exogenous shocks. Second, it implements the propensity score matching (PSM) technique to address selection bias. So far only a handful of papers have used PSM to examine the economic impact of IMF-supported programs but not in the context of the catalytic financing impact.2 Third, we examine the catalytic impact through not only the amount but also the modality of Official Development Assistance (ODA). Furthermore, we test a comprehensive set of ODA measures, including gross and net disbursements (both including and excluding debt relief), net commitments, and untied disbursements. Fourth, we explore heterogeneity of the catalytic impact by donors, inspired by findings in the literature that bilateral aid may be more responsive to political and strategic considerations of donors. Fifth, we explicitly account for the implementation of programs in estimating the catalytic impact. Although the literature recognizes that the impact of programs would depend on how successfully they are implemented previous empirical work has rarely accounted for the implementation record.3

Our results highlight that IMF-supported programs with LICs lead to significantly higher ODA and affect donors’ preferences in terms of the modality of aid. Countries with IMF-supported programs tend to have an increase in gross disbursements (excluding debt relief) amounting to 1.9 percent of GDP. Interestingly, the size of the estimated catalytic impact does not vary much for program countries experiencing substantial prior macroeconomic imbalances or large exogenous shocks. Moreover, the catalytic impact on commitments appears to be larger than the impact on disbursements, likely suggesting some room to improve both the utilization of aid by recipients and the predictability of aid disbursements. Finally, IMF-supported programs are associated with significantly higher ODA from multilateral donors while the estimated impact is, albeit positive, weaker for bilateral donors.

In terms of aid modality, countries with IMF-supported programs tend to receive a higher proportion of aid in general budget support from International Development Association (IDA) and European Commission (EC). Furthermore, the proportion of untied aid (excluding technical cooperation and humanitarian aid) in total aid is higher for countries with IMF-supported programs.

The paper is structured as follows: Section II reviews the literature on catalytic effects of IMF-supported programs. Section III presents some stylized facts on the evolution of ODA. Section IV introduces the methodology followed by empirical results in section V. Finally, conclusions are summarized in section VI.

II. Literature Review

The literature on the catalytic financing effect of IMF-supported programs focuses on the effect on private capital flows in emerging market economies.4 Although this body of research could not support a uniform and significantly positive catalytic impact, they did report some non-monotonic positive impact depending on the initial economic conditions of recipient countries and type of private flows. Steinwand and Stone (2008), in their review article, point out that one clear finding of the literature is that the catalytic effects of IMF lending are not uniform across countries. Studies that investigate the possibility of non-monotonic effects find positive catalytic effects only for countries in a middle range of economic indicators for wealth or financial stability. Specifically, Mody and Saravia (2006) report that IMF program participation lowers the bond spread for countries with medium levels of foreign reserves, while countries with higher level of reserves experience negative catalytic effects (higher bond spread) and at the lower end have neither positive nor negative catalytic effect. They also observe that the sign of the catalytic impact of IMF financing may change when selection bias is explicitly accounted for.

Bird and Rowlands (2007) is the only paper examining the catalytic impact of IMF-supported programs on donor assistance to LICs. Their results indicate a strong positive association and also suggest that this effect may have more to do with conditionality than with the provision of IMF resources. However, this paper differs from our study in two important ways: (i) it does not correct for selection bias; and (ii) BR includes all programs with LICs regardless of the nature of balance of payments needs: Stand-By Arrangements (SBA) addressing immediate balance of payments needs, and programs supported by Extended Fund Facility (EFF), and Poverty Reduction and Growth Facility (PRGF) addressing longer-term balance of payments needs.

Previous research on aid has found that determinants of bilateral and multilateral aid are different. Alesina and Dollar (2000) find that political and strategic considerations are more significant factors explaining bilateral aid than economic variables. Clist (2011) reports that aid allocation is influenced by donors’ commercial and strategic ties with recipients and by the needs of the recipients. Similarly, Berthélemy (2006) found that most bilateral donors target their assistance to their most significant trading partners while they also respond to recipients’ needs and merit. Based on these findings, we will also explore whether IMF-supported programs affect assistance from bilateral and multilateral donors differently.

Focusing only on the amount of ODA in assessing the catalytic impact of IMF-supported programs may give an incomplete picture. Other aspects of donor selectivity, i.e., how responsive aid allocation is to the needs and the policy environment of recipients, have also received attention in the literature. Clist, Isopi, and Morrissey (2012) argue that donors exercise selectivity over the aid modality. Specifically, multilateral donors (the EC and WB) cede more control to recipients over aid by granting more budget support to those recipients with better policies. Motivated by these findings we also explore the impact of IMF-supported programs on the aid modality of major multilateral donors.

III. Defining the Catalytic Impact on ODA

Cottarelli and Giannini (2002) define the catalytic impact of IMF-supported programs on private flows as follows: “… the IMF’s involvement in a country has a catalytic effect to the extent that the announcement of an economic program backed up by a limited amount of IMF resources increases the propensity of private investors to lend to the country concerned, thereby reducing the adjustment burden falling on the debtor country with respect to the no-catalysis scenario.” In the case of private flows an IMF-supported program is mostly an exogenous factor that would feed into investment/lending decisions of private investors.

The catalytic impact of programs on ODA from multilateral and bilateral donors, on the other hand, has more of a simultaneous and collaborative meaning. In a way an IMF-supported program likely acts as a coordination device among donors also motivated by the needs of the recipient countries among other more peculiar factors. The simultaneity aspect is also ingrained in the IMF’s policy on financing assurances. All IMF financing arrangements requires that IMF-supported programs can only be approved (and reviews can only be completed) when the program is fully financed. This means that donors and creditors have furnished assurances that they will provide the necessary financial support to meet the program financing requirements on terms consistent with the member’s return to external viability.5 Given the predominant role of official flows in LICs programs are approved only when such assurances are in place from multilateral and bilateral donors. If this device works well in garnering significant additional ODA, it could ease and smooth the required policy adjustment and alleviate the associated output costs. The question, therefore, is how effective the IMF-supported programs in LICs as coordinating devices for donors support compared to the non-program countries experiencing similar economic difficulties.

IV. Methodology

A. Data

Over the last three decades despite significant increases especially in Foreign Direct Investment (FDI) flows to LICs the ODA remains the most prominent source of financing by a large margin (Figure 1). Therefore, we prefer to focus on estimating the catalytic impact of programs on ODA flows. Furthermore, FDI flows reached meaningful levels only in the last decade, providing an extremely small sample to study that is not suited well to the PSM approach.

Figure 1.ODA and Foreign Direct Investment (FDI), medians over 1980–2010

We use databases of the Organization for Economic Co-operation and Development (OECD) Development Assistance Committee (DAC) disaggregated by donors and recipients over time.6 DAC definition of ODA includes official and concessional flows to developing countries granted with the objective of promoting the economic development and welfare of recipient countries.7 Although the coverage of donors is not comprehensive, the DAC database covers the majority of ODA flows.

We use four broad measures of ODA: gross disbursements, net disbursements, net commitments, and untied disbursements. In DAC statistics, repayments of loan principal (and any recoveries on grants or proceeds from equity sales) are subtracted from gross ODA to arrive at net ODA. Untied disbursements are derived by deducting humanitarian aid, technical cooperation grants, and food aid from gross disbursements.

Following Claessens, Cassimon, and Campenhout (2009) and Roodman (2012) we deduct debt forgiven and rescheduled from gross disbursements. Our motivation for removing debt relief is twofold: First, debt relief transactions do not represent actual current money transfers. Second, more importantly, the eventual debt relief is the outcome of a multi-year process and to a large extent predetermined with respect to the type of programs we examine.8 As such debt relief part of ODA cannot be attributed to the catalytic impact of these programs and may distort the results.

Any loan cancellation, ODA or other official financing (OOF) loans, increases gross ODA through “debt forgiveness grants.”9 Moreover, when donors and recipients reschedule debt, the capitalization of interest arrears is treated as a new aid flow, and is included in “ODA loans extended,” under “rescheduled debt.” Using these series our adjusted gross disbursement variable is derived as follows:

Gross disbursement= (total ODA grants– debt forgiveness grants) + (total gross loans extended – rescheduled debt).

The DAC definition of net ODA automatically removes grants for ODA loan forgiveness by deducting the offsetting entry for amortization recorded in loan repayments. However, the OECD definition of net disbursements overstates the amount of actual money transfers as the debt relief granted on OOF loans are recorded as debt forgiveness grants but the offsetting entries of OOF loan repayments are recorded under the original category not as ODA loan repayments. We remove these offsetting entries from our definition of net disbursements to refine our measure of net disbursements.

Finally, for gross commitments disaggregated data for debt relief on commitment basis is not available at OECD/DAC database; therefore, we use the original OECD definition including debt relief.

To assess the impact of IMF-supported programs on aid modality we use ODA commitments for general budget support by IDA and EU. We use Clist, Isopi, and Morrissey (2012) dataset covering 1997–2009 for EU and 1995–2007 for IDA.

B. Propensity Score Matching

The empirical analysis in this paper implements the PSM approach to control for selection bias, a relatively new and innovative class of statistical methods for impact evaluation using non-experimental or observational data.10 Participation in an IMF-supported program addressing policy or exogenous shocks is taken as the treatment status. The PSM involves a statistical comparison of program versus non-program countries in two steps: First, the probability of participating in IMF-supported programs is estimated conditional on observable economic conditions and country characteristics (selection model). At the second step, these probabilities, or propensity scores, are used to match program countries to non-program countries, and thereby, construct a statistical control group.11

The matching based on the probability of participation in IMF-supported programs assures similarity of initial macroeconomic conditions and country characteristics in the comparison, or control, group. The control group provides in effect a proxy for the counterfactual, that is, for the catalytic effect if program countries had not had a program. The catalytic effect of the IMF-supported programs is then calculated as the mean difference of relevant ODA measures (scaled by GDP) across these two groups.

The limitations of assumptions underlying the PSM should be noted. This approach is useful when only observed pre-treatment characteristics are believed to affect program participation. Two necessary assumptions for identification of the program effects are (i) conditional independence; and (ii) presence of a common support. Conditional independence, also called confoundedness, implies that the program participation is based entirely on observed pre-shock characteristics of LICs. If unobserved characteristics determine program participation, conditional independence will be violated, and PSM would not be an appropriate method. Using a rich set of pre-program data to estimate the probability of participation in IMF-supported programs addressing policy and exogenous shocks helps support the conditional independence assumption. In other words, a well-specified and comprehensive selection model explaining the participation in IMF-supported programs is the key to properly assess the catalytic impact of IMF programs. Moreover, we test the impact of programs based on both first differences and levels of ODA. The former transformation is the preferred to remove the unobserved heterogeneity arising from country-specific factors that are not controlled for in the participation equation, which should further buttress conditional independence. The second condition, i.e., presence of a common support, ensures that treatment observations have comparison observations “nearby” in the propensity score distribution. Again, a well specified participation equation differentiating the initial macro-economic conditions well and providing probability estimates well dispersed in the range of zero and one would help uphold this assumption.

IMF program countries are called the treatment group whereas the remainder of the sample constitutes the control group. In terms of policy research on estimating the catalytic impact of IMF-supported programs, the average treatment effect of IMF engagement on the treated group (ATT) would be of interest and given by:

where IMF is the dummy variable identifying LICs with IMF engagement in any given year. ΔODAi0|IMFi = 1 is the change in ODA (scaled by lagged GDP) that would have been observed if a LIC with IMF engagement had not experienced such an engagement, and ΔODAi1|IMFi = 1 is the change in ODA (scaled by lagged GDP) observed on the same country. The counterfactual outcome under no program is not observable for a program country. In order to derive the ATT based on observables equation [1] can be rearranged as follows:

The difference between the term on the left-hand side and the ATT is selection bias, i.e., the difference in average donor assistance to program countries under the condition of no program versus average donor assistance to nonprogram countries. Given that the initial macroeconomic conditions of program countries are substantially different than those of nonprogram countries, it is not plausible to assume that donor assistance would have been the same in the absence of IMF-supported programs, therefore, a sizeable selection bias would be present.

The key assumption to eliminate selection bias from equation (2) through matching methods is conditional independence which requires that, conditional on some control variables X, the catalytic impact be independent of the IMF engagement dummy, i.e. E[ΔODAi0|IMFi = 1, Xi] − E[ΔODAi0|IMFi = 0, Xi] would be zero. Under this assumption, equation (2) can be rewritten as

Rosenbaum and Rubin (1983) propose that one can match the treated units and control units on their propensity scores, which can be estimated by simple probit or logit models. A further assumption needed to apply PSM is the common support assumption (p(Xi) < 1), which requires the existence of some comparable control units for each treated unit. When PSM is used, the ATT now can be estimated as

The strategy then consists in computing the differences in ODA(Yi) for observations with similar propensity scores (the probability of participating in IMF-supported programs addressing policy or exogenous shocks). Various methods have been proposed in the literature to match observations. In this study, we present results using the nearest neighbor technique. The nearest neighbor matching estimator sorts all records by the estimated propensity score, and then searches forward and backward for the closest control units.

Selection model for IMF-supported programs addressing policy and exogenous shocks

The selection model adopted in this study draws on Bal Gündüz (2009). This is the only study looking into determinants of LIC participation in IMF arrangements addressing immediate balance of payments needs in response to domestic policy and/or external shocks. Examining this more homogenous subset of IMF arrangements significantly improves the specification of the selection model, which is the key to counter selection bias to properly assess the impact of IMF-supported programs.12

The dependent variable is a panel dummy variable, taking the value of 1 if a new IMF shock financing is approved, and 0 otherwise, indicating a normal episode. The set of arrangements include those addressing an immediate balance of payments need arising from policy and/or exogenous shocks. SBA, Structural Adjustment Facility (SAF)/ Enhanced Structural Adjustment Facility (ESAF)/PRGF/ Extended Credit Facility (ECF) augmentations, Exogenous Shocks Facility (ESF), Standby Credit Facility (SCF), Rapid Credit Facility (RCF) and Compensatory Financing Facility (CFF) are included in this set.13 The following refinements are made to this basic set: (i) precautionary SBA/SCF and SBA/PRGF/ECF augmentations addressing natural disasters are excluded,14 and (ii) some SAF/ESAF/PRGF/ECF arrangements are added if they address immediate balance of payments needs arising from policy shocks. In order to systematically determine the latter cases, this study relied heavily on program interruptions preceding SAF/ESAF/PRGF/ECF arrangements. For first time SAF/ESAF/PRGF arrangements, narratives from IMF staff reports are used to identify programs that envisaged a drastic shift in macroeconomic policies to address an immediate financing gap. Normal episodes are identified as the initial year of two successive years with no IMF financing for shocks when the member is eligible to access IMF resources.15 Several refinements are made to normal episodes to identify cases where supply constraints are binding.16

The effects of various economic variables on the probability of a LIC requesting IMF financing in response to shocks are assessed by estimating a binary response model for panel data. The general specification for panel probit models is given by

where, y is the observed outcome, Φ is the cumulative normal density function (c.d.f.), xit is the 1xk vector of explanatory variables, and β is kx1 vector of coefficients associated with xit. Different estimators are constructed depending on their assumptions for the panel heterogeneity, i.e., how they treat ci.17 The estimations are carried out step-by-step under different estimators and a correlated random effects probit model is preferred based on the econometric tests for the significance of both the individual specific effect and the sample average for covariates.

Bal Gündüz (2009) finds that a number of economic variables are significantly associated with increased probability of IMF financing, including reserve coverage, the ratio of current account balance to GDP, real GDP growth, macroeconomic stability indicator and terms of trade shocks (Table 1).18 Moreover, adverse global shocks to the change in real oil and non-oil commodity prices, and the cyclical component of world trade increase the participation in IMF arrangements. Therefore, the demand for IMF resources by LICs is likely to be cyclical in response to global conditions with its intensity depending on the magnitude and persistence of adverse external shocks.19

Table 1.Demand for IMF Financing in Response to Policy and/or External Shocks
Current account balance to GDP (t-1)−0.076 ***

(−4.61)
Reserve coverage in months of imports (CFA) (t-1)−0.478 ***

(−6.08)
Reserve coverage in months of imports (non-CFA) (t-1)−0.769 ***

(−8.71)
Macroeconomic stability indicator (t-1)0.068 ***

(2.89)
Real GDP growth (t-1)−0.113 ***

(−4.24)
Change in terms of trade (t-1)−0.022 ***

(−2.8)
Change in real oil prices in previous two years0.009 ***

(2.85)
Real world trade, cyclical component−0.099 **

(−2.53)
Change in real non-oil commodity prices−0.020

(−1.58)
Real growth of goods exports (t-1)−0.009 *

(−1.79)
Paris Club dummy0.774 ***

(3.24)
Constant0.551

(1.23)
Country-specific averages

Total debt service to exports
0.044 ***

(2.63)
FDI to GDP−0.105 *

(−1.76)
Pseudo R20.58
LR test: β2 = … = βk = 0 χ2 (Prob)376 (0.00)
LR test: ρ = 0 χ2 (Prob)11 (0.00)
Number of observations532
Sample probability0.44
Number of countries55
Source: Bal Gunduz (2009).Note: Demand for IMF financing in response to policy and/or exogenous shocks excluding natural disasters is estimated by a correlated random effects probit model. Significant at 10 percent: *; 5 percent: **; and 1 percent: ***, t-statistics in paranthesis. Country-specific averages are calculated as the sample average of variables for each country. FDI = foreign direct investment; LR = likelihood ratio test.1 The CFA franc zone consists of 14 countries in sub-Saharan Africa, each affiliated with one of two monetary unions maintaining the same currency, the CFA Franc.
Source: Bal Gunduz (2009).Note: Demand for IMF financing in response to policy and/or exogenous shocks excluding natural disasters is estimated by a correlated random effects probit model. Significant at 10 percent: *; 5 percent: **; and 1 percent: ***, t-statistics in paranthesis. Country-specific averages are calculated as the sample average of variables for each country. FDI = foreign direct investment; LR = likelihood ratio test.1 The CFA franc zone consists of 14 countries in sub-Saharan Africa, each affiliated with one of two monetary unions maintaining the same currency, the CFA Franc.

The ultimate objective is to distinguish the short-term impact of IMF-supported programs when a country has an immediate external financing need. The treatment variable is identified mostly symmetrically to the one used in the selection equation. A panel dummy variable taking the value of 1 for the approval of IMF-supported programs with LICs addressing immediate balance of payments needs, and 0 for non-program episodes, is constructed as the treatment variable.20 Refinements to the program and non-program episodes are made similar to those for the dependent variable in the selection equation. Within the set of program countries, a higher propensity score will identify the IMF-supported programs addressing a clear financing need. Severe state failure events are excluded from both program and non-program sets as the macroeconomic outcomes in these episodes will be frail, independent of the impact of IMF-supported programs.21 Furthermore, as state failures could lead to a disruption of all donor involvement including these episodes could potentially bias results on the catalytic impact of programs. Finally, in order to take account of program implementation, years of program interruptions are excluded from the sample.

V. Results

Using the PSM we examine the catalytic impact of IMF-supported programs both on the size and the modality of ODA flows to LICs during 1980–2010.22 Our outcome variables include four measures of ODA: gross and net disbursements, net commitments, and untied disbursements. For the first three measures, we further explore heterogeneity in the catalytic impact on ODA flows from bilateral versus multilateral donors. We take first differences of all outcome variables scaled by lagged GDP to eliminate persistent country-specific differences in aid allocation and focus on increases in aid that could be attributed to the catalytic impact of IMF-supported programs.23 Furthermore, we present results on disbursements both excluding and including debt relief to examine the effect of this adjustment on the estimated catalytic impact.

Countries with higher propensity scores are more likely to request an IMF-supported program to address large exogenous shocks or substantial prior macroeconomic imbalances arising from policy slippages. Therefore, for each outcome variable we further examine the heterogeneity of the catalytic impact with respect to initial economic difficulties of recipients by testing the impact separately for three sub-groups of propensity scores: low (less than 0.3), medium to high (between 0.3 and 0.7) and very high (higher than 0.7). Table 2 shows the distribution of propensity scores across the treatment (programs) and the control groups (nonprograms). As noted earlier for nonprogram episodes we introduced some asymmetries compared to the dependent variable in the participation equation to increase the common support for the PSM by including nonprogram years followed immediately by an IMF-supported program or episodes without IMF membership. As a result of this strategy we significantly increased the size of the control group for high propensity scores.24

Table 2.Sample Description: Distribution of Propensity scores across the Treatment and the Control Groups
PS<0.30.3PS>0.7Total
IMF-supported programs2947112188
Nonprograms20371124398
Total232118236586
Control-treated ratio7.01.51.12.1
Source: Authors’ calculations. IMF-supported programs include those addressing immediate balance of payments needs addressing policy or exogenous schocks. PS stands for propensity scores.
Source: Authors’ calculations. IMF-supported programs include those addressing immediate balance of payments needs addressing policy or exogenous schocks. PS stands for propensity scores.

Simulation studies for the PSM report that the control to treated ratio is an important parameter affecting the level of difficulty for the matching estimators: When this ratio is smaller than 1:1 the mean square error (MSE) of the estimate of mean outcome in the nontreated population conditional on the propensity score becomes larger indicating worsening performance of the matching estimator (e.g., Frolich, 2004). Conversely the higher the ratio the more efficient the estimation becomes. In our study, the control to treated ratio is safely above one for both the full sample and the sub-groups by propensity scores.

Gross and Net Disbursements

Benchmark results on disbursements suggest that countries with IMF-supported programs receive significantly higher ODA (Table 3 top panel):

  • Contemporaneous increases in both gross and net disbursements (excluding debt relief) are significantly higher by about 2.0 and 2.4 percent of GDP respectively for countries with IMF-supported programs. Interestingly, the impact is, while positive, not significant for countries with low propensity scores, nevertheless, it is substantially higher and becomes significant for medium to high and very high propensity scores. We observe the same pattern for untied ODA disbursements. This finding seems to suggest that donors respond to economic difficulties of recipients and within the group of countries experiencing similar levels of economic hardship, i.e., substantial macroeconomic imbalances or large shocks, they tend to favor those with IMF-supported programs.
  • Both multilateral and bilateral donors significantly raise their gross and net disbursements to countries with IMF-supported programs. While increases are highly significant across the board for multilateral flows, except for the net disbursements to the low propensity group, the significance of bilateral flows is driven by medium to high propensity scores. Net disbursements excluding debt relief are most likely to affect near-term economic outcomes as they represent net current cash transfers to recipients. It is noteworthy that changes in net bilateral disbursements are only slightly lower than those of net multilateral disbursements, except for low propensity scores.
Table 3.The Catalytic Impact of IMF-supported Programs on Change in Disbursements of Official Development Assistance (ODA)
Variables (first-differenced)All LICsPS<0.30.3PS>0.7
Excluding Debt Relief
Gross disbursement1.992***

(0.572)

584
0.964

(0.898)

232
2.087***

(0.629)

118
2.165**

(0.906)

234
Net disbursement2.405***

(0.871)

584
1.978

(2.118)

232
2.635***

(0.998)

118
2.359*

(1.301)

234
Untied ODA disbursement1.688***

(0.498)

584
0.776

(0.758)

232
1.840***

(0.549)

118
1.812**

(0.790)

234
Bilateral gross disbursement0.854**

(0.420)

567
0.419

(0.335)

226
1.099**

(0.529)

114
0.875

(0.656)

227
Multilateral gross disbursement1.394***

(0.308)

567
1.604***

(0.394)

226
0.964***

(0.373)

114
1.492***

(0.477)

227
Bilateral net disbursement1.231*

(0.639)

567
0.931

(0.660)

226
1.323*

(0.780)

114
1.280

(1.000)

227
Multilateral net disbursement1.463***

(0.500)

567
2.176

(2.086)

226
1.320**

(0.654)

114
1.317**

(0.595)

227
Including Debt Relief
Gross disbursement2.347**

(0.967)

584
−0.173

(2.449)

232
1.747**

(0.734)

118
3.211**

(1.474)

234
Net disbursement1.975***

(0.689)

584
0.525

(1.039)

232
1.998***

(0.507)

118
2.288**

(1.117)

234
Untied ODA disbursement2.044**

(0.898)

584
−0.361

(2.373)

232
1.500**

(0.666)

118
1.812**

(0.790)

234
Bilateral gross disbursement1.198

(0.770)

567
−0.457

(0.906)

226
1.148**

(0.534)

114
1.643

(1.242)

227
Multilateral gross disbursement1.459***

(0.452)

567
1.212

(1.908)

226
0.543

(0.548)

114
1.871***

(0.542)

227
Bilateral net disbursement1.460*

(0.767)

557
0.196

(2.501)

224
0.613

(0.642)

113
2.137**

(1.087)

220
Multilateral net disbursement1.017**

(0.414)

557
1.411

(2.036)

224
1.283**

(0.559)

113
0.788*

(0.410)

220
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (Xt-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (Xt-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.

The bottom panel of Table 3 presents results for the same ODA measures but including debt relief. As argued earlier, we prefer the ODA measures excluding debt relief to eliminate the noise arising from the debt relief process that is predetermined for the type of programs we examine. Interestingly, even with the unadjusted measures results are still broadly in line with our benchmark results. However, a noteworthy difference is that the increase in gross ODA disbursements from bilateral donors becomes insignificant. Furthermore, the significance of the catalytic impact for gross ODA from multilaterals is driven only by the very high propensity scores.

ODA in first differences is conceptually a better measure to assess the catalytic impact of IMF-supported programs as it removes country-specific and persistent differences in aid allocation, a phenomenon well documented in the literature for bilateral donors. Table 4 presents results for ODA measures in levels to examine whether our results for benchmark first-differenced measures would hold. Results are broadly similar except that the catalytic impact of programs on bilateral flows becomes insignificant for both gross and net disbursements across both the full sample and the sub-groups by propensity scores. On the other hand, aggregate disbursements including all donors remain significant thanks to multilateral flows. This finding is consistent with the literature highlighting that political and strategic factors are more influential on the allocation of bilateral aid. Nevertheless, within the group of countries experiencing similar economic problems bilateral donors appear to favor those with the IMF-supported programs and raise their base aid, which is determined mainly by their political and strategic considerations, more for this group.

Table 4.The Catalytic Impact of IMF-supported Programs on Disbursements of Official Development Assistance (ODA)
Variables (level in % of GDP)All LICsPS<0.30.3PS>0.7
Excluding Debt Relief
Gross disbursement2.224**

(0.992)

584
3.970**

(1.778)

232
3.530*

(2.035)

118
1.165

(1.370)

234
Net disbursement2.240**

(1.056)

584
3.509*

(1.919)

232
3.953*

(2.205)

118
1.130

(1.444)

234
Untied ODA disbursement2.217***

(0.707)

584
3.340**

(1.394)

232
3.481**

(1.493)

118
1.341

(0.956)

234
Bilateral gross disbursement0.804

(0.634)

584
1.806

(1.259)

232
1.585

(1.344)

118
0.186

(0.853)

234
Multilateral gross disbursement1.420***

(0.477)

584
2.163***

(0.677)

232
1.945**

(0.870)

118
0.979

(0.697)

234
Bilateral net disbursement0.960

(0.681)

584
1.385

(1.351)

232
2.035

(1.527)

118
0.367

(0.888)

234
Multilateral net disbursement1.280**

(0.534)

584
2.124*

(1.091)

232
1.918**

(0.911)

118
0.762

(0.764)

234
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.

Commitments

Table 5 reports results for ODA commitments including debt relief.25 The short-term IMF engagement leads to significantly higher commitments from multilaterals while its impact on bilateral commitments is insignificant. A comparison of these results with those for gross disbursements (including debt relief) reveals that the catalytic impact of programs on commitments is somewhat higher than the impact on gross disbursements. For multilateral flows all except countries with middle to high propensity scores commitments are higher than gross disbursements, possibly suggesting room to raise the utilization of aid either through increasing the technical implementation capacity of recipients or predictability of aid by multilateral donors. Interestingly, for bilateral donors the catalytic impact on gross disbursements is higher than the impact on commitments for both the full sample and the sub-groups by propensity scores.

Table 5.The Catalytic Impact of IMF-supported Programs on Change in Commitments of Official Development Assistance (ODA)
Variables (first-differenced)All LICsPS<0.30.3PS>0.7
Including debt relief
Commitment2.632**

(1.074)

567
1.869

(1.218)

226
0.925

(1.360)

114
3.495**

(1.674)

227
Bilateral commitment0.815

(0.719)

567
−0.469

(0.915)

226
0.178

(0.814)

114
1.371

(1.129)

227
Multilateral commitment1.817***

(0.566)

567
2.337***

(0.738)

226
0.747

(0.944)

114
2.124**

(0.840)

227
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (Xt-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (Xt-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.

Aid Modality

In addition to the quantity based-ODA measures above, we also assess the catalytic impact of IMF-supported programs on aid modality. Specifically, we look into whether programs are associated with an increase in the proportion of general budget support from IDA and EC.26Table 6 presents results.

Table 6.The Catalytic Impact of IMF-supported Programs through Aid Modality
Variables (first differenced)ALL LICS
Proportion general budget support from IDA18.58***

(5.318)

146
Proportion general budget support from EC18.59***

(3.776)
Observations212
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.

Our findings suggest that IMF-supported programs have a significant catalytic impact through aid modality. For all LICs in our sample, Fund Programs tend to induce significantly higher proportion of aid allocated as general budget support from the IDA and the EC.

Owing to the relatively small sample size the PSM may not perform well, therefore, our results related to the aid modality should be interpreted with caution. For that reason we also do not report results by sub-groups of propensity scores.

Robustness Checks

Rosenbaum sensitivity analysis to hidden bias

The key assumption behind the PSM is conditional independence which means that program participation depends only on the observed characteristics of LICs. However, hidden bias may arise from important omitted covariates. The strong specification for the participation equation, encompassing a number of highly significant variables, should tend to alleviate the hidden bias. Moreover, this study looks into the impact of programs on changes in outcomes as well as their levels, which should help remove the unobserved heterogeneity arising from time-invariant country-specific factors not controlled in the participation equation.

In addition to these safeguards, we conducted Rosenbaum’s sensitivity analysis to test sensitivity of our findings to hidden bias. This analysis manipulates the estimated odds of having a program versus not having a program to see how much it can deviate from 1, the expected odds ratio for a randomized experiment, while results still remaining robust. Table 7 presents results for the Rosenbaum sensitivity analysis to hidden bias. The parameter Γ is a measure of how much hidden bias can be present before results of the study begin to change. A variable is highly sensitive to hidden bias if the conclusions change for Γ just barely larger than 1, and it is insensitive if the conclusions change only for quite large values of Γ.27

Table 7.Rosenbaum Sensitivity Analysis (Γ parameters)
VariablesAll LICsPS<0.30.3PS>0.7
Excluding Debt Relief
Gross Disbursement1.741.001.791.77
Net Disbursement1.661.461.761.73
Bilateral Gross Disbursement1.241.31.441.22
Multilateral Gross Disbursement2.181.001.862.53
Bilateral Net Disbursement1.241.001.441.27
Multilateral Net Disbursement1.941.001.782.23
Flexible Disbursement1.951.002.821.97
Including Debt Relief
Gross Disbursement1.761.001.831.79
Net Disbursement1.731.001.821.7
Bilateral Gross Disbursement1.281.381.51.24
Multilateral Gross Disbursement2.51.031.872.6
Bilateral Net Disbursement1.581.001.431.3
Multilateral Net Disbursement2.391.002.582.83
Flexible Disbursement1.981.022.091.94
Commitment2.071.492.032.42
Bilateral Commitment1.771.241.862.04
Multilateral Commitment2.561.032.113.11
Source: Authors’ calculations. The parameter Γ is a measure of how much hidden bias can be present, i.e. how much the estimated odds of having a program versus not having a program can deviate from 1 before results of the study begin to change. A variable is highly sensitive to hidden bias if the conclusions change for Γ just barely larger than 1, and it is insensitive if the conclusions change only for quite large values of Γ.
Source: Authors’ calculations. The parameter Γ is a measure of how much hidden bias can be present, i.e. how much the estimated odds of having a program versus not having a program can deviate from 1 before results of the study begin to change. A variable is highly sensitive to hidden bias if the conclusions change for Γ just barely larger than 1, and it is insensitive if the conclusions change only for quite large values of Γ.

Sensitivity analysis shows that our results across different measures of ODA are robust to hidden bias except for the results for the bilateral ODA, with Γ ranging from 1.7 to 3.11. While results for sub-groups with medium to high and very high propensity scores are robust, except for the bilateral disbursements, the results are very sensitive to hidden bias for the group with low propensity scores.

Sensitivity of results to immediate versus protracted balance of payments problems

In this section, we examine whether our results change if programs supported under ECF, addressing protracted balance of payments needs, are also included in our IMF program dummy used in the second stage for the PSM analysis.28Table 8 presents the results.

Table 8.Robustness Checks: IMF-Supported Programs including Extended Credit Facility
Variables (first-differenced)All LICsPS<0.30.3PS>0.7
Excluding Debt Relief
Gross disbursement0.724***

(0.280)

1,149
0.358

(0.333)

627
0.886*

(0.525)

209
1.185

(0.808)

313
Net disbursement0.121

(0.572)

1,149
−0.575

(0.818)

627
1.069

(0.729)

209
0.773

(1.269)

313
Untied ODA disbursement0.658***

(0.243)

1,149
0.352

(0.283)

627
0.806*

(0.437)

209
1.034

(0.709)

313
Bilateral gross disbursement0.223

(0.182)

1,121
0.141

(0.145)

617
0.266

(0.366)

203
0.406

(0.600)

301
Multilateral gross disbursement0.627***

(0.144)

1,121
0.495***

(0.156)

617
0.636*

(0.351)

203
0.878**

(0.378)

301
Bilateral net disbursement0.384

(0.305)

1,121
0.351

(0.299)

617
0.233

(0.501)

203
0.593

(0.964)

301
Multilateral net disbursement−0.153

(0.505)

1,121
−0.671

(0.818)

617
0.873*

(0.487)

203
0.271

(0.665)

301
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (X t-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (X t-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.

As expected observations with programs under ECFs predominantly add to the group with low propensity scores, thereby, results for the full sample gets weaker for all ODA measures but it becomes insignificant only for net disbursements (both aggregate and by donors) and gross disbursements for bilateral ODA. The results for multilateral gross disbursements are the most robust to the inclusion of ECFs. For net disbursements the catalytic impact is significant only for the group with medium to high propensity scores.

Weakening in results for the high to medium and very high propensity scores is noteworthy and could be an area for further research. It should be noted that we take all three years with ECF programs, not only the approval year of a program which may induce more donor aid than the interim years. Moreover, propensity scores are still estimated from the participation equation for programs addressing immediate balance of payments needs to flag the economic needs of countries, however, determinants of ECF programs are likely to differ substantially from those that matter for the subset of programs we focus on. Therefore, this sensitivity analysis should not be interpreted as weaker catalytic impact for ECF programs.

Sensitivity of results to controlling for “donor favorites”

In our benchmark results, we focus on the catalytic impact of IMF-supported programs measured by the first-differenced ODA for program versus nonprogram countries to eliminate the time-invariant or highly persistent political and strategic considerations of donors affecting the level of their ODA allocation. In this section, we examine the robustness of our results to matching on both the propensity score and the lagged level of ODA to see whether the change in ODA could also be systematically different for countries attracting similar donor assistance before the shock. For that purpose, we match program and nonprogram countries having a similar propensity score and receiving a similar level of ODA at t-1. With this strategy we aim to identify “donor favorites”, which attract sizeable donor resources before the shock, and match them with other “donor favorites” experiencing a similar shock but not having a program. Results (Table 9) are qualitatively similar to our benchmark results for total and multilateral disbursements. A noteworthy result is that the estimated impact of programs on the bilateral gross and net disbursements becomes insignificant for the full sample while it turns out to be significant for the sub-group with low propensity scores. This finding may suggest that bilateral donors tend to step up assistance to their favorites experiencing severe economic hardship regardless of the program status of these countries. Nevertheless, after controlling for donor favorites, bilateral donors appear to increase in their support significantly for LICs with IMF-supported programs when recipients experience only mild economic distress (low propensity scores).

Table 9:Robustness Checks: Matching on Propensity Scores and Lagged ODA Disbursements
Variables (first-differenced)All LICsPS<0.3• 0.3• PS>0.7
Excluding Debt Relief
Gross disbursement2.086***

(0.739)

584
0.959

(0.912)

232
2.112***

(0.620)

118
2.518**

(0.995)

234
Net disbursement2.677**

(1.129)

584
2.570

(1.860)

232
2.760**

(1.081)

118
2.755*

(1.518)

234
Untied ODA disbursement2.049***

(0.588)

584
0.697

(0.728)

232
2.034***

(0.588)

118
2.421**

(1.052)

234
Bilateral gross disbursement0.927

(0.588)

567
0.730**

(0.291)

226
1.017*

(0.608)

114
0.956

(1.006)

227
Multilateral gross disbursement1.803***

(0.290)

567
1.606***

(0.371)

226
1.190***

(0.297)

114
2.055***

(0.438)

227
Bilateral net disbursement1.252

(1.016)

567
1.544***

(0.501)

226
1.351

(1.118)

114
1.377

(1.489)

227
Multilateral net disbursement1.899***

(0.416)

567
3.068**

(1.385)

226
1.545***

(0.365)

114
1.908***

(0.579)

227
Including Debt Relief
Gross disbursement2.390**

(1.184)

584
−0.266

(2.087)

232
1.901***

(0.691)

118
3.238**

(1.570)

234
Net disbursement2.026***

(0.730)

584
0.532

(1.097)

232
2.065***

(0.463)

118
2.433**

(1.220)

234
Untied ODA disbursement2.371**

(1.021)

584
−0.304

(2.048)

232
1.607**

(0.742)

118
3.251*

(1.802)

234
Bilateral gross disbursement0.945

(1.004)

567
−0.117

(0.717)

226
1.057*

(0.626)

114
1.302

(1.667)

227
Multilateral gross disbursement1.829***

(0.429)

567
0.871

(1.680)

226
0.798

(0.497)

114
2.444***

(0.520)

227
Bilateral net disbursement1.352*

(0.703)

557
−0.122

(1.921)

224
0.782

(0.598)

113
2.112*

(1.224)

220
Multilateral net disbursement1.405***

(0.323)

557
1.484

(1.309)

224
1.393***

(0.296)

113
1.279***

(0.388)

220
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (Xt-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.
Source: Authors’ calculations. Using the propensity score matching (PSM) the average treatment effect for the treated is reported. PS stands for the propensity score. Each variable is first differenced and scaled by lagged GDP: (Xt-X t-1)/ GDP t-1. Standard errors are in parentheses, followed by number of observations. * Significant at 10%; ** Significant at 5%; *** Significant at 1%.

VI. Conclusion

This paper examines the catalytic impact of IMF-supported programs with LICs. We focus on a special subset of programs addressing immediate balance of payments needs of countries arising from policy or exogenous shocks. The premise of examining this group is twofold: First, regardless of the program status, these countries would need to adjust their policies to restore macroeconomic stability. To the extent that the IMF provides its own resources and catalyzes donor assistance in the context of IMF-supported programs it can help ease the pace of adjustment, thereby, alleviate the concomitant immediate output costs. Therefore, the catalytic impact of these programs is potentially an important channel of transmission for the impact of programs on near-term output. The second premise is related to the methodological challenge of addressing selection bias, i.e., systematic differences in initial economic conditions of a program versus a non-program country. Focusing on a homogenous subset of IMF-supported programs with LICs addressing immediate balance of payments needs is critical to identify the economic determinants of IMF programs, i.e., a strong participation equation to estimate the likelihood of such programs (propensity scores), thus, assess the catalytic impact of programs on donor support for countries experiencing similar economic difficulties using the PSM approach.

Using a comprehensive set of ODA measures, including gross and net disbursements (both including and excluding debt relief), net commitments, and untied disbursements we find that the IMF-supported programs in LICs have a significant catalytic impact on both the change in ODA, the primary source of financing to LICs, and the modality of ODA. Results are primarily driven by countries experiencing sizeable initial macroeconomic imbalances or large exogenous shocks (high propensity scores) while the catalytic impact is not significant for countries with low propensity scores. In other words, donors seem to respond to economic difficulties of recipients and within the group of countries experiencing substantial economic problems they tend to favor those with IMF-supported programs. Moreover, both multilateral and bilateral donors significantly raise their ODA (excluding debt relief) to countries with IMF-supported programs. We also assess the catalytic impact of IMF-supported programs on aid modality and find that programs tend to induce significantly higher proportion of aid allocated as general budget support from the IDA and the EC.

In order to remove the unobserved heterogeneity arising from country-specific factors we choose to test the catalytic impact on the change in ODA, rather than on its level. Nevertheless, results using levels offer some key insights as well. The catalytic impact of programs on the level of bilateral flows is insignificant while the impact on aggregate disbursements is significant, driven by multilateral flows. This finding is consistent with the aid allocation literature highlighting that political and strategic factors are more prominent for bilateral donors. Nonetheless, our results highlight that within the group of countries experiencing similar economic difficulties bilateral donors appear to favor, at the margin, those with the IMF-supported programs by raising the level of base aid, which is mainly determined by their political and strategic considerations, more for this group. We further examine if the impact would remain significant when we control for “donor favorites” by matching program and nonprogram countries having a similar propensity score and receiving a comparable level of ODA in the previous year. With this strategy we aim to identify “donor favorites”, which attract sizeable donor resources before the shock, and match them with other “donor favorites” experiencing a similar shock but not having a program. While our results remain qualitatively similar for total and multilateral disbursements the estimated catalytic impact of programs on bilateral disbursements (gross and net), though still positive, become insignificant. This finding as well as the high sensitivity of results for bilateral flows to hidden bias indicates that the catalytic impact of programs with LICs is primarily attributed to multilateral flows.

As another robustness check we examine how results would change if programs supported under ECF, addressing protracted balance of payments needs of LICs, are added to the set of programs while still using the participation equation estimating the likelihood of programs addressing urgent financing needs. Overall, results get weaker for all ODA measures as programs under ECFs predominantly add to the observations with low propensity scores. Nonetheless, results for the high to medium and very high propensity scores gets weaker as well, suggesting that further research could usefully explore the catalytic impact of programs supported under ECFs using a participation model explaining these programs.

Annex 1. List of Countries and Average Annual ODA Disbursements to GDP (1980–2010)
CountryGrossNetUntied ODABilateral GrossMulti Gross
1Albania9.039.206.575.323.71
2Armenia5.645.661.500.375.27
3Azerbaijan1.021.030.55(0.56)1.58
4Bangladesh2.422.711.610.302.12
5Benin10.3410.027.695.674.67
6Bolivia5.145.312.842.362.78
7Burkina Faso13.7913.649.748.175.62
8Burundi25.8324.4816.0312.1613.67
9Cambodia5.695.783.192.313.38
10Cameroon3.523.542.562.381.14
11Central African Republic12.6312.198.677.255.38
12Chad12.5913.018.846.436.16
13Comoros20.0420.5813.4511.348.70
14Congo, Republic of5.295.823.734.091.20
15Cote Divoire4.354.833.422.621.72
16Democratic Republic of Congo8.158.205.194.114.04
17Ethiopia9.598.845.974.954.64
18Gambia16.1616.4511.827.708.46
19Georgia3.863.871.66(0.26)4.11
20Ghana6.125.805.183.083.04
21Guinea9.019.846.624.514.49
22Guinea-Bissau21.9221.5215.4511.4910.43
23Guyana3.214.061.87(4.23)7.44
24Haiti13.9214.007.449.514.41
25Honduras6.416.164.603.862.55
26India0.400.540.310.070.32
27Kenya5.946.774.093.872.06
28Kyrgyz Republic5.085.102.36(1.29)6.37
29Laos PDR6.286.353.900.655.63
30Madagascar10.659.958.025.385.27
31Malawi21.7619.2915.8910.7910.98
32Mali17.0716.8112.6310.286.78
33Mauritania19.9220.4315.1310.389.55
34Moldova6.396.513.783.353.04
35Mongolia6.376.503.622.993.63
36Mozambique25.1024.1718.6416.808.30
37Nepal4.904.982.971.243.66
38Nicaragua15.6114.4111.4210.505.11
39Niger14.5413.7810.168.615.93
40Nigeria0.680.740.460.360.32
41Pakistan1.201.460.820.021.18
42Papua New Guinea7.808.105.786.561.23
43Rwanda19.1818.0912.4510.828.36
44Senegal10.9910.857.737.153.84
45Sierra Leone19.0818.1112.969.229.87
46Sri Lanka3.944.592.901.842.11
47Sudan8.528.854.935.333.19
48Tajikistan4.254.281.07(1.33)5.62
49Tanzania14.0713.8511.159.224.86
50Togo10.6512.047.946.054.60
51Uganda12.0611.229.416.105.96
52Uzbekistan0.750.770.430.570.17
53Vietnam1.952.071.450.841.11
54Zambia16.0115.5112.529.306.71
55Zimbabwe9.579.664.357.282.30
References

    Alesina, Alberto, and DavidDollar, 2000, “Who Gives Foreign Aid to Whom and Why?” Journal of Economic Growth, Vol. 5 (1), pp. 33–63.

    Atoyan, Ruben, and PatrickConway, 2006, “Evaluating the Impact of IMF Programs: A Comparison of Matching and Instrumental-Variable Estimators,” The Review of International Organizations, Vol. 1 (2), pp. 99–124.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bal Gündüz, Yasemin, 2009, “Estimating Demand for IMF Financing by Low-Income Countries in Response to Shocks,” IMF Working Paper 09/263 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bal-Gündüz, Yasemin, ChristianEbeke, BurcuHacibedel, LindaKaltani, VeraKehayova, ChrisLane, ChristianMumssen, NkundeMwase, and JosephThornton, 2013, “The Economic Impact of IMF-Supported Programs in Low-Income Countries,” Occasional Paper No. 277 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Berthélemy, Jean-Claude, 2006, “Bilateral Donors’ Interest vs. Recipients’ Development Motives in Aid Allocation: Do All Donors Behave the Same?” Review of Development Economics, Vol. 10 (2), pp. 179–94.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bird, Graham, 2007, “The IMF: A Bird’s Eye View of its Role and Operations,” Journal of Economic Surveys, Vol. 21 (4), pp 683–745.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bird, Graham, and DaneRowlands, 2002, “Do IMF Programmes Have a Catalytic Effect on Other International Capital Flows?” Oxford Development Studies, Vol. 30 (3), pp. 229–49.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bird, Graham, and DaneRowlands, 2007, “The IMF and the Mobilisation of Foreign Aid,” Journal of Development Studies, Vol. 43 (5), pp. 856–70.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bird, Graham, and DaneRowlands, 2009a, “The IMF’s Role in Mobilizing Private Capital Flows: Are There Grounds for Catalytic Conversion?” Applied Economics Letters, Vol. 16 (17), pp. 1705–08.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Bird, Graham, and DaneRowlands, 2009b, “A Disaggregated Empirical Analysis of the Determinants of IMF Arrangements: Does One Model Fit All?” Journal of International Development, Vol. 21 (7), pp. 915–31.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Claessens, Stijn, DannyCassimon, and BjornVan Campenhout, 2009, “Evidence on Changes in Aid Allocation Criteria,” The World Bank Economic Review, Vol. 23 (2), pp. 185–208.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Clist, Paul, 2011, “25 Years of Aid Allocation Practice: Whither Selectivity?” World Development, Vol. 39 (10), pp. 1724–34.

    Clist, Paul, AlessiaIsopi, OliverMorrissey, 2012, “Selectivity on Aid Modality: Determinants of Budget Support From Multilateral Donors,” The Review of International Organizations, Vol. 7 (3), pp. 267–84.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Cottarelli, Carlo, and CurzioGiannini, 2002, “Bedfellows, Hostages, or Perfect Strangers? Global Capital Markets and the Catalytic Effect of IMF Crisis Lending,” IMF Working Paper 02/193 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    Chamberlain, Gary, 1982, “Multivariate Regression Models for Panel Data,” Journal of Econometrics, Vol. 18 (1), pp. 5–46.

    Dehejia, Rajeev H., and SadekWahba, 1999, “Causal Effects in Nonexperimental Studies: Reevaluating the Evaluation of Training Programs”, Journal of the American statistical Association, Vol. 94 (448), pp. 1053–62.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Frölich, Markus, 2004, “Finite-Sample Properties of Propensity-Score Matching and Weighting Estimators”, Review of Economics and Statistics, 86 (1), pp. 77–90.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Heckman, James J., HidehikoIchimura, and PetraTodd, 1998, “Matching as an econometric evaluation estimator”, The Review of Economic Studies, Vol. 65 (2) pp. 261–294.

    • Crossref
    • Search Google Scholar
    • Export Citation

    International Monetary Fund, 2012a, “2011 Review of Conditionality—Overview Paper” (Washington).

    • Export Citation

    International Monetary Fund, 2012b, “2011 Review of Conditionality—Background Paper: Outcomes of IMF-Supported Programs” (Washington).

    • Export Citation

    International Monetary Fund, 2014, “The Handbook of IMF Facilities for Low-Income Countries” (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    Jaramillo, Laura, and CemileSancak, 2009, “Why Has the Grass Been Greener on One Side of Hispaniola? A Comparative Growth Analysis of the Dominican Republic and Haiti,” IMF Staff Papers, Vol. 56, pp. 323–49 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Mercer-Blackman, Valerie, and AnnaUnigovskaya, 2000, “Compliance with IMF Program Indicators and Growth in Transition Economies,” IMF Working Paper 00/47 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    Mody, Ashoka, and DiegoSaravia, 2006, “Catalysing Private Capital Flows: Do IMF Programmes Work as Commitment Devices?” Economic Journal, Vol. 116 (513), pp. 843–67.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Morris, Stephen, and Hyun SongShin, 2006, “Catalytic Finance: When Does It Work?” Journal of international Economics, Vol. 70 (1), pp. 161–77.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Mumssen, Christian, YaseminBal Gündüz, ChristianEbeke, LindaKaltani, 2013, “IMF-Supported Programs in Low-Income Countries: Economic Impact over the Short and Long Term,” IMF Working Paper 13/273 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    Mundlak, Yair, 1978, “On the Pooling of Time Series and Cross Section Data,” Econometrica, Vol. 46 (1), pp. 69–85.

    Robins, James M., 2002, Comment on “Covariance Adjustment in Randomized Experiments and Observational Studies,” Statistical Science, Vol. 17 (3), pp. 309–21.

    • Search Google Scholar
    • Export Citation

    Roodman, David, 2012, “An Index of Donor Performance,” Center for Global Development, Working Paper No 67.

    Rosenbaum, Paul R., 2002, “Observational Studies,” Springer Series in Statistics (New York).

    Rosenbaum, Paul R., and Donald B.Rubin, 1983, “The Central Role of the Propensity Score in Observational Studies for Causal Effects,” Biometrika, Vol. 70 (1), pp. 41–55.

    • Crossref
    • Search Google Scholar
    • Export Citation

    Steinwand, Martin, and RandallStone, 2008, “The International Monetary Fund: A Review of the Recent Evidence,” Review of International Organizations, Vol. 3 (2), pp. 123–49.

    • Crossref
    • Search Google Scholar
    • Export Citation
1

The authors would like to thank Chris Lane, Catherine Pattillo, Olaf Unteroberdoerster, Marco Arena, Wendell Daal, Ermal Hitaj, Mumtaz Hussain, Paulo Lopez, Henry Moone, Nkunde Mwase, Saad Quayyum, Frank Dong Wu, seminar participants at the IMF, and OECD colleagues Brenda Killen, Elena Bernaldo de Quirós, Olivier Bouret, Andrzej Suchodolski for very helpful comments and suggestions. The usual disclaimer applies.

2

See Bal Gündüz and others (2013) for a comprehensive survey of findings in the literature on the economic impact of IMF-supported programs.

5

Specifically there should be no unfilled financing gaps over the 12 months immediately following the approval of the arrangement (and the completion of each review), and that there is a clear expectation that the program will be fully financed through the remainder of the arrangement period.

6

DAC is a specialized committee of aid donors that includes 29 member countries, plus the European Union as a full member.

7

The minimum concessionality requirement is 25 percent, calculated using a discount factor of 10 percent.

8

Although IMF-supported programs are required for debt relief, therefore, by definition catalyze donor support the subset of programs we examine tend not to overlap with those pre-requisite programs. In order to reach the decision and the completion points under the Heavily Indebted Poor Countries (HIPC) initiative and receive Multilateral Debt Relief Initiative (MDRI) a minimum six-month track record of policy performance under an IMF-supported program is required preceding both points. Programs that count toward qualification include those supported by ECF, SCF, or Extended Arrangements, on a case-by-case basis as determined by the Board, SBA, Rights Accumulation Program (RAP), Rapid Financing Instrument (RFI), Rapid Credit Facility (RCF), and Staff-Monitored Program (SMP) (when the Executive Board agrees that its policies meet the policy standard of an Upper Credit Tranche (UCT) arrangement). We do not typically have these programs in our subset of financing events addressing shocks. Only RCFs, and first years of some ECFs, if they address policy or exogenous shocks, are included. Furthermore, CFFs and augmentations of access under ECFs, included in our set, do not overlap with pre-requisite programs.

9

In general, official loans that do not meet the concessionality requirement or allocated for non-developmental objectives such as military aid are classified as OOF. Loans with maturity of less than one year are also not counted as ODA.

10

The interest in PSM accelerated after Heckman and others (1998) assessed the validity of using propensity matching to characterize selection bias using experimental data, and Dehejia and Wahba (1999) used PSM to approximate the experimental results from the National Supported Work Demonstration. In the context of evaluating the impact of IMF-supported programs, only Atoyan and Conway (2006), IMF (2012b), and Bal Gündüz and others (2013) implemented the PSM.

11

This study uses the nearest neighbor matching approach, which constructs a control group of countries by choosing those four non-program countries with probability of requesting a program as close as possible to that of the specific program country in question.

12

Before Bal Gündüz (2009), only Bird and Rowlands (2009b) looked into determinants of Fund arrangements with LICs, albeit without much success in improving the model specification. Only three variables turned significant: the presence of previous Fund arrangements, high inflation, and the rescheduling of debt in the current year.

13

Bal Gündüz and others (2013) provides an extensive discussion on the evolution of the IMF’s concessional facilities since 1986. For the IMF’s current toolkit of facilities with LICs please refer to “Handbook of IMF Facilities for Low-Income Countries” (2014).

14

The exclusion was based on the lack of immediate balance of payments need for precautionary SBAs and different nature of the shock for SBAs/PRGF augmentations addressing natural disasters. Specifically, it is quite unlikely that one could predict Fund financing addressing natural disasters, conditional on a similar set of explanatory variables as in policy and other exogenous shocks. In that regard, please see Bal Gündüz (2009) presenting a robustness check which shows that the participation equation estimated for programs addressing natural disasters are substantially different than programs addressing other immediate financing needs.

15

Years with no programs that are immediately followed by IMF financing programs are excluded from the set of normal episodes as depending on the timing of programs negotiations may have taken place in these years. Therefore, economic circumstances in these years may resemble to those of program years. Allowing a safe “distance” away from program episodes helps better distinguish economic circumstances of program versus normal episodes which should improve the model specification.

16

Members with overdue obligations to the Fund are ineligible to use Fund resources, therefore, observations with arrears to the Fund are excluded from normal episodes. Observations with Fund financing for natural disasters through Emergency Natural Disaster Assistance (ENDA) or PRGF augmentations, program interruptions or break-up of negotiations for a program, SMP, Emergency Post-Conflict Assistance (EPCA), and three years leading up to EPCAs are also excluded. Finally, episodes during which members incurred arrears to other bilateral and multilateral creditors and did not have adjustment programs that would garner the Fund support and rescheduling by their major creditors are excluded from normal episodes.

17

Pooled probit models assume independence of observations over both t and i. A random effects (RE) probit model treats the individual specific effect, ci, as an unobserved random variable with ci|xitIN(μc,σc2) if an overall intercept is excluded, and imposes independence of ci and xit. A fixed effects (FE) probit model treats ci as parameters to be estimated along with β, and does not make any assumptions about the distribution of ci given xit. This can be problematic in short panels as both β and ci are inconsistently estimated owing to an incidental parameters problem. Finally, a correlated random effects model relaxes independence between covariates and individual-specific effect using the Chamberlain (1982) and Mundlak (1978) device under conditional normality. In this specification, the time average is often used to save on degrees of freedom.

18

In order to assess the macroeconomic policy stance based on a comprehensive set of complementary indicators, this study used a variant of the composite indicator introduced by Jaramillo and Sancak (2009). The version of this index that includes the black market premium was first used in Bal Gündüz (2009). The formula for the indicator is given by:

where mitot is the macroeconomic stability index for country i at time t, cpi is the consumer price index, xr is the exchange rate of national currency to U.S. dollar (an increase indicates a nominal depreciation), res is the stock of international reserves, mgs is the imports of goods and services, gbal is the government balance, gdp is the nominal GDP, blackpr is the black market premium, and σ is the standard deviation of each variable. Weights are inverses of the standard deviation of each component for all countries over the full sample after removing the outliers. Higher levels of mitot indicate increased macroeconomic instability.

19

See Bal Gündüz (2009) for other variables that could be significantly associated with the participation in IMF-supported programs but do not turn out to be significant, including variables capturing investors/donors’ willingness to meet financing needs, i.e., access to alternative financing, prior to financing events.

20

Some asymmetries compared to the dependent variable in the participation equation are introduced for nonprogram episodes to increase the common support for the PSM. The treatment variable includes nonprogram years followed immediately by an IMF-supported program and nonprogram episodes without IMF membership. The dependent variable in the participation equation excludes these observations from the sample.

21

The severe state failure events are identified from Political Instability Task Force (PITF) dataset. Four types of political crises are included in this dataset: revolutionary wars, ethnic wars, adverse regime changes, and genocides and politicides. From this dataset the variable SFTPMMAX, which presents the maximum magnitude of all events in a year, exceeding 3.9 is taken as a severe state failure event.

22

Our definition of LICs is defined as those countries that were eligible to receive the IMF’s subsidized resources as of January 1st, 2010. Please see the annex for the list of countries included in the sample.

23

We preferred to use a common denominator to put the emphasis on the change in ODA.

24

Owing to the small sample for the sub-group with medium to high propensity scores, the PSM may not perform well, therefore, our results related to this sub-group should be interpreted cautiously.

25

It is not possible to calculate ODA commitments excluding debt relief as disaggregated data on debt relief on commitment basis is not available in the OECD/DAC database.

26

Data availability is limited, from 1997 to 2009 for the EC and from 1995 to 2007 for the IDA.

27

Robins (2002) expressed skepticism about the usefulness of sensitivity analysis as he proved that Rosenbaum’s Γ fit the criteria of a paradoxical measure: its magnitude increases as the analyst decreases the amount of hidden bias by measuring some of the unmeasured covariates. As such, this measure could be useful only if experts could provide a plausible and logically coherent range of Γ.

28

The ECF provides financial assistance to LICs with protracted balance of payments problems. Assistance under an ECF arrangement is provided usually for three years.

Other Resources Citing This Publication