World Development Vol. 87, pp.

3049, 2016
0305-750X/ 2016 Elsevier Ltd. All rights reserved.

www.elsevier.com/locate/worlddev
http://dx.doi.org/10.1016/j.worlddev.2016.04.012

The Economic Impact of Short-term IMF Engagement in

Low-Income Countries
YASEMIN _ BAL GUNDUZ*
International Monetary Fund, NW, DC, USA
Summary. This study estimates the economic impact of IMF-supported programs addressing policy and exogenous shocks in low-
income countries (LICs) using the propensity score matching approach. Focusing on a more homogenous group of nancing events
greatly improves the specication and predictive power of the selection model as compared to earlier studies, which is the key require-
ment to control for selection bias. This paper looks into a wide range of potential economic and political determinants of participation.
Lower reserve coverage, deterioration in the current account balance, a weaker real GDP growth, increased macroeconomic instability
(evident in higher scal decits, ination and exchange market pressures), and lagged elections would increase the probability of partic-
ipation, i.e., propensity scores. Moreover, global conditions, including changes in real oil and non-oil commodity prices and world trade,
are also signicant; therefore, adverse global shocks could create cycles in demand for IMF nancing. The short-term IMF engagement
is positively associated with a wide range of macroeconomic outcomes. Notably, the impact on the short-term growth is the greatest and
becomes signicant only for LICs facing substantial macroeconomic imbalances or large exogenous shocks. When matching accounts for
propensity scores and institutional quality, program countries in this group have changes in growth 1.5 and 3.5 percentage points higher
than the control group.
2016 Elsevier Ltd. All rights reserved.

Key words IMF-supported programs, low-income countries, economic growth, exogenous shocks, IMF nancing, aid eectiveness

1. INTRODUCTION policy or exogenous shocks have a signicant catalytic impact

Economic growth in low-income countries (LICs) has been
strong since the late 1990s thanks to a marked improvement
in macroeconomic policies and a favorable global environ-
ment until the onset of the global nancial crisis. The crisis
hit LICs hard primarily through adverse tail shocks to external
demand, foreign direct investments (FDI), and remittances.
Oil exporters also suered from a sharp decline in oil prices.
Nevertheless, unlike previous global downturns real per capita
growth stayed positive for most LICs and recovered quickly
after a sharp slowdown in 2009 (IMF, 2010, 2014). During this
period the IMFs concessional nancial support to LICs
surged; commitments increased from an average of $446 mil-
lion during 200507 to $1.5 billion in 2008 and reached a peak
of $3.8 billion in 2009. An IMF study argues (IMF, 2010) that
countercyclical policy response, a rst for LICs in contrast to
past crises, as well as substantial external support, including
from the IMF, were likely instrumental in supporting growth.
This paper explores how the short-term IMF supported pro-
grams responding to immediate nancing needs arising from
policy slippages (henceforth policy shocks) or adverse exoge-
nous shocks may have aected economic outcomes in LICs
over the last three decades. Such programs would often be
called for when a country faces a pressing balance of payments
problem, which would require a combination of macroeco-
nomic adjustment and external nancing. The IMF engage-
ment in these cases would typically involve understandings
on short-term macroeconomic adjustment accompanied by
IMF nancing. The potential channels of transmission would
likely include nancing to ease the burden of adjustment both
through the IMFs own nancing and the catalytic impact of
IMF-supported programs, policy advice, especially in case of
signicant prior policy slippages, and the short-term crisis-
related technical assistance. It is noteworthy that Bal-
Gunduz and Crystallin (2014) show that programs addressing
30
on both the size and the modality of ocial development assis-
tance (ODA), attributed primarily to multilateral ODA. 1
In the context of these short-term programs, the IMFs role
in facilitating the adjustment to policy and exogenous shocks
through nancing and macroeconomic stabilization has poten-
tial implications on long-term growth. It is well documented
that LICs are subject to larger, more frequent exogenous
shocks compared to advanced and emerging market countries.
The economic impact of shocks tends to get amplied by weak
coping mechanisms owing to insucient resources, weak pol-
icy buers, and shallow nancial markets. As such shocks are
more likely to induce breaks in trend growth rather than tran-
sitory cyclical uctuations around a trend, leading to signi-
cant welfare losses. Dabla-Norris and Bal Gunduz (2014)
develop an index which provides early warning signals of a
growth crisis in low-income countries to monitor individual
country risks of sharp growth declines arising from external
shocks. Their results show that macroeconomic policy buers,
exchange rate regimes, institutional quality, and the size of
shocks are important determinants of growth crises in low-
income countries. Findings of Becker and Mauro (2007) and
Perry (2009) indicate that the impact of exogenous shocks
on growth and consumption volatility is particularly pro-
nounced in LICs. Hnatkovska and Loayza (2005) nd a neg-
ative link between macroeconomic volatility and long-run
economic growth, which is exacerbated in countries that are
* I would like to thank Christian Mumssen, Linda Kaltani, Christian
Ebeke, Graham Bird, Dane Rowlands, Chris Lane, Hugh Bredenkamp,
Doris Ross, Alun Thomas, Juan Trevino, Felipe Zanna, and two
anonymous referees for helpful comments. Elizabeth Baxter and
Merceditas San Pedro-Pribram provided excellent editorial assistance.
The views expressed herein are those of the author and should not be
attributed to the IMF, its Executive Board, or its management. The usual
disclaimer applies. Final revision accepted: April 25, 2016.
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES
poor, institutionally underdeveloped, undergoing intermediate
stages of nancial development, or unable to conduct counter-
cyclical scal policies. They also report that the negative impact
on growth results mostly from large drops below the trend
rather than normal cyclical uctuations. Papageorgiou,
Pattillo, Spatafora and Berg (2010) suggest that adverse shocks
in LICs on average translate into substantial persistent output
losses over the medium-term. Higher volatility tends to reduce
investment in both physical and human capital in LICs particu-
larly given credit constraints. Moreover, Bruckner and Ciccone
(2010) nd that commodity price downturns have a robust eect
on the outbreak of civil wars in Sub-Saharan Africa.
Isolating the specic impact of an IMF-supported program
from the broader economic developments and the impact of
exogenous shocks in LICs is not an easy task. The vast empir-
ical literature on the impact of IMF-supported programs has
reported mixed results on economic growth, typically based
on samples including both LICs and middle-income econo-
mies. Moreover, results have been particularly sensitive to
the econometric methodology.
When assessing the economic impact of IMF-supported
programs selection bias, arising from the fact that participa-
tion in IMF-supported programs is not random, presents a
fundamental methodological challenge. In other words, initial
economic conditions or the external environment will dier
systematically for a program versus a non-program country.
Countries that approach the IMF tend to already face eco-
nomic diculties or expect to experience problems in the near
future. If countries that are experiencing balance of payments
crises owing to policy slippages or exogenous shocks are more
likely to participate in IMF-supported programs, failing to
correct for selection bias could lead to a awed conclusion that
programs cause these crises along with adverse eects on
macroeconomic outcomes.
The key step to control for selection bias and, thereby, disen-
tangle the economic impact of programs is to estimate a strong
selection (participation) model explaining the decision to request
a program. Despite the vast literature on determinants of IMF
arrangements, Steinwand and Stone (2008, p. 129) conclude that
the variety of models used to explain participation in IMF pro-
grams and the plethora of contradictory results they produce
indicates that existing models are far from denitive. This unn-
ished business is the strongest reason to urge caution in rushing
to judgment about the eects of IMF lending.
The heterogeneity of countries and types of programs pre-
sents a challenge to the estimation of a selection model. In
an attempt to improve the specication of participation mod-
els, recently just a handful of studies have disaggregated the
analysis of participation in IMF arrangements by country
income groups. Bird and Rowlands (2009), the only study that
looks separately into a LIC sample, report signicant dier-
ences between their regression specications for LICs and
MICs; however, the results for the LIC specication are
weaker than the results for the MIC specication. They nd
that only three variables are signicantly related to participa-
tion in IMF-supported programs in LICs: previous programs,
high ination, and the rescheduling of debt.
A key contribution of this paper to the literature is its focus
on a more homogenous subset of programs with LICs
addressing policy and exogenous shocks, which signicantly
improves the model specication and better distinguishes eco-
nomic circumstances leading up to such programs. A wide
range of economic variables, including reserve coverage,
current account balance to GDP, real GDP growth, and a
composite indicator assessing macroeconomic stability, turn
out to be signicant determinants of approval of IMF
31
arrangements with LICs. Moreover, global conditions, specif-
ically the change in real oil and non-oil commodity prices, and
the cyclical component of world trade, have signicant eects
on demand for programs.
This paper makes several contributions to the empirical lit-
erature on the economic impact of IMF-supported programs.
First, it is the only study to explore the impact of a unique set
of nancing arrangements with LICs addressing policy or
exogenous shocks. Second, it implements the propensity score
matching (PSM) technique to address selection bias and exam-
ines the impact on a wider range of macroeconomic and social
outcomes. So far, only a handful of papers have used PSM to
examine the economic impact of IMF-supported programs on
a few outcome variables. Third, it explicitly accounts for the
implementation of programs in estimating the impact.
Although the literature recognizes that the impact of programs
would depend on how successfully they are implemented, pre-
vious empirical work has rarely taken it into account.
Results indicate that stepped-up IMF nancing through aug-
mentations of existing programs or short-term and emergency
facilities are positively associated with short-term growth and
indicators of macroeconomic stability. However, the impact
on the short-term economic growth is the greatest and becomes
signicant only when LICs are faced with substantial short-term
macro-economic imbalances or large exogenous shocks.
Another noteworthy nding is that the implementation record
of programs does matter in impact assessments.
This paper is organized as follows: Section 2 briey reviews
the literature on participation in and the economic impact of
IMF-supported programs; an overview of the methodology is
provided in Section 3; Section 4 presents the results and robust-
ness checks; nally, conclusions are discussed in Section 5.
2. EMPIRICAL LITERATURE
Despite the vast literature on the macroeconomic conse-
quences of IMF-supported programs no consensus has yet
emerged on the impact of programs on growth, and very
few papers have focused exclusively on LICs. The treatment
of selection bias has been the fundamental methodological
challenge and has led to an equally vast literature on participa-
tion in IMF-supported programs. This section briey reviews
the body of empirical work on participation in programs and
then looks into the literature on the impact of IMF-supported
programs.
(a) Participation in IMF-supported programs
Numerous empirical studies have looked into determinants
of participation in IMF arrangements. Bird (2007) and
Steinwand and Stone (2008) provide comprehensive surveys.
Early research emphasized the economic determinants
(Conway, 1994; Joyce, 1992; Knight & Santaella, 1997;
Santaella, 1996). While most studies agreed on the signicant
positive association of program participation with low levels of
reserve holdings, previous participation in IMF programs, and
low levels of income, the evidence was at best mixed on a range
of other economic covariates. Moreover, the within sample and
out-of-sample predictive power of these models was limited.
The low predictive power of these models led researchers to
include political variables that would aect the supply or the
demand side of programs, such as the size of governments,
the quota at the IMF, the various instruments for U.S. and
European inuence, and the number of veto players. Evidence
is again mixed; some nd a role for U.S. inuence but its lim-
ited to the IMFs nonconcessional lending. Others suggest that
32 WORLD DEVELOPMENT
U.S. inuence has an impact on other aspects of the IMF lend-
ing rather than participation, such as the size of loans, nature
of structural conditionality, and record of program implemen-
tation (Andersen, Hansen, & Markussen, 2006, Barro & Lee,
2005; Dreher & Jensen, 2007; Dreher, Sturm, & Vreeland,
2006, 2009a, 2009b; Kuziemko & Werker, 2006; Oatley &
Yackee, 2004; Presbitero & Zazzaro, 2012; Stone, 2002,
2004). Bird and Rowlands (2001) note that including political
variables has not signicantly improved the predictive power.
Sturm, Berger, and de Haan (2005) use extreme bounds anal-
ysis (EBA) and nd it is mostly economic variables that are
robustly related to IMF lending. 2 Moser and Sturm (2011)
focus on the post Cold War period and look into robust eco-
nomic and political determinants of concessional versus non-
concessional loans using EBA. They report that in contrast
to the non-concessional loans, relatively few variables can
robustly explain the signing of concessional IMF loans (past
IMF programs, the ratio of international reserves to imports,
and an election in the year prior to participation).
Pointing to the lack of consensus in the literature on which
variables really matter for participation in programs Bird
(2007) conclude that the empirical evidence so far may imply
that important determining variables may still have been omit-
ted. . . or that there is no one overall explanation of IMF
arrangements. Rather certain things are important in some
cases but not in others, such that their signicance eectively
cancels out in large sample studies. Therefore, he suggests
that future research could look into subsets of country cases,
distinguishing between the traditional current account crisis,
capital account crisis, and LICs.
Recently only a handful of studies have looked into partic-
ipation in IMF-supported programs by country groups.
Ghosh, Goretti, Joshi, Thomas, and Zalduendo (2007) and
Cerutti (2007) examine IMF engagement with Middle Income
Countries (MICs) and emerging market economies and nd
several economic variables signicant. Bird and Rowlands
(2009) and Moser and Sturm (2011) are the only studies esti-
mating a LIC sample. Both report signicant dierences com-
pared to the MIC sample with only a few variables
signicantly associated with participation in programs.
The literature on the prolonged use of IMF resources high-
lights that repeated participation is associated with lower per
capita income, driven by LICs (Bird, Hussain, & Joyce,
2004; Independent Evaluation Oce, 2002; Joyce, 2005).
Conway (2007) reports that while the duration of program
spells is reduced signicantly by prior participation for MICs
it increases signicantly by prior participation for LICs.
The impact of global economic conditions on IMF-
supported programs has received scant attention in the litera-
ture. Using a mixed sample of LICs and MICs, Elekdag (2008)
reports that oil prices, world interest rates, and a measure of
the global business cycle have a signicant impact on the
approval of a stand-by (SBA) arrangement.
The participation model in this paper contributes to the
existing literature in three areas. First, it looks into a specic
subset of IMF-supported programs with LICs addressing
immediate balance of payments needs. More homogenous eco-
nomic circumstances leading up to signing of such arrange-
ments signicantly improve the model specication. Second,
in identifying normal episodes the model takes into account
observable supply-side constraints that would preclude a
members access to IMF nancing, which would further distin-
guish the eects of economic factors on demand for programs.
Third, it explicitly examines the eects of an extensive set of
global economic conditions on the demand for IMF-
supported programs addressing shocks.
(b) Economic impact of IMF-supported programs
The vast empirical literature has reached some consensus
that IMF-supported programs are associated with signicant
improvement in the balance of payments and have some eect
on ination; however, results are mixed regarding their impact
on growth (Table 1). 3
A few observations are noteworthy about this literature.
First, most of the previous research examines only non-
concessional programs, SBAs and Extended Fund Facilities
(EFFs), for a mixed sample of countries. A few studies that
have focused on just LICs or concessional programs have
identied some positive eects of IMF engagement on macroe-
conomic performance, but not in all areas (Bredenkamp &
Schadler, 1999; Bird & Mosley, 2006; Dicks-Mireaux,
Mecagni, & Schadler, 2000; Ghosh et al., 2007; IEO, 2004).
A number of studies have examined the social impact of
IMF programs with mixed results. Second, although the liter-
ature widely acknowledges that whether IMF-supported pro-
grams are fully implemented or not is a key issue in properly
assessing their impact, most studies do not take into account
compliance with programs. 4 Finally, correcting for selection
bias has become a standard component of the analysis only
more recently, while most studies have applied either the
Heckman two-stage methodology or instrumental variable
(IV) regressions (notable exceptions implementing the PSM
are Atoyan & Conway, 2006; Garuda, 2000; Hardoy, 2003;
Hutchison, 2004; IMF, 2012). 5 However, the key challenge
associated with these approaches was to identify exclusion
restrictions, i.e., nding variables strongly correlated with
the participation in an IMF-supported program but not corre-
lated with the macro-economic outcome of interest.
The more recent literature builds on techniques borrowed
from the microeconometric impact evaluation literature to
correct for selection bias. Under the PSM approach, each
IMF-supported program country observation is matched to
a counterfactual nonprogram-country observation with a sim-
ilar predicted probability of having a program, and their
macroeconomic outcomes are then compared. Using this tech-
nique, Atoyan and Conway (2006) found little statistical sup-
port that IMF programs contemporaneously improve real
economic growth in participating countries (though this was
the case for the scal and current account balances), but found
stronger evidence of an improvement in economic growth in
years following a program.
This paper is also related to the vast but mostly inconclusive
literature on aid eectiveness, which estimates the impact of
aggregate development assistance on economic growth in the
recipient countries. Most studies try to address the simultane-
ity of aid and growth using the instrumental variable
approach; however, nding strong and valid instruments for
aid to test causality has been a major challenge. Other, related,
complications arise from dierent types of aid ows dierenti-
ated by their purpose and the appropriate lag structure for the
aid growth relationship, i.e., when aid inows can be expected
to produce a measurable eect on economic growth. A few
recent studies implementing more innovative approaches
report positive eect of aid on growth (Bruckner, 2013,
Clemens, Radelet, Bhavnani, & Bazzi, 2012; Galiani, Knack,
Xu, & Zou, 2014). This paper nds a positive impact of
IMF-supported programs on contemporaneous annual
growth for LICs facing substantial macroeconomic imbal-
ances or large exogenous shocks. This nding, combined with
the results from Bal-Gunduz and Crystallin (2014) which
report a signicant increase in ocial development assistance
for the same subset of programs seems to suggest a positive
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES
Table 1. Summary of literature on the impact of IMF programs, 200014
Outcome variables GDP Ination Fiscal Current account Monetary Gini Education Health spending Poverty Period Countries Type of programs Selection correction; method
growth decit balance growth coecient spending
Bordo and Schwartz (2000)
+ + 197398 14 EMs SBA EFF No
*
Dicks-Mireaux et al. (2000) +
198691 61 LICs ESAF No
* * a
Garuda (2000) + /
197591 39 countries Mixed Yes; PSM
*
Przeworski and Vreeland (2000)
197090 79 countries Mixed Yes; Heckman
* *
Evrensel (2002) + b
+ +/
197197 91 countries Mixed&by program No
Hardoy (2003)
197090 109 countries Mixed Yes; PSM PSM(DID)
*
Hutchison (2003)
197597 67 countries SBA EFF Yes; Heckmanc
Lee and Rhee (2002) + 197394 88 countries Mixed No
*
Vreeland (2003)
196193 110 countries Mixed Yes; Heckman
* *
Dreher and Vaubel (2004)

197597 94 countries Mixed No
Hutchison (2004) + 197597 25 EMs SBA EFF Yes; PSM Heckman
*
Independent Evaluation Oce (2002)
/
d 197599 82 Prolonged vs temporary users Mixed&by program Yes; IV
*
Barro and Lee (2005)
197599 130 countries SBA EFF Yes; IV
*
Butkiewicz and Yanikkaya (2005)
197099 100 countries Mixed Yes; IV
Easterly (2005) 0e

198099 20 repeat users Mixed Yes; IV f
* *
Atoyan and Conway (2006) +/

+ 19932002 95 countries Mixed Yes; PSM Heckman
*
Dreher (2006)
19702000 98 countries SBA EFF Yes; IV
* *
Nooruddin and Simmons (2006)

19802000 98 countries Mixed Yes; Heckman
*
Eichengreen, Gupta, and Mody (2008)
19802003 24 countries; sudden stops SBA EFF Yes; Heckman
*
Marchesi and Sirtori (2011)
g 19822005 128 countries Mixed Yes; IV
IMF (2012)h
i
+ 0 200211 44 MICs SBA EFF Yes; PSM
* *
Clements, Gupta, and Nozaki (2012) + + 19852009 LICs and MICs Mixed&by program Yes; IV
* j * j
Oberdabernig (2013) + + 19822009 LICs and MICs Mixed Yes; IV
*
Bas and Stone (2014) + 19702008 104 countries Mixed Yes; Heckmank

Source: Draws on Steinwand and Stone (2008), expanded by the author to include selected key aspects of previous studies as well as recent literature.
Note: Heckman = Heckman two-step estimator for correcting selection bias; IV = Instrumental variable estimator; PSM = Propensity Score Matching; DID = Dierence-in-dierences;
EMs = Emerging Markets; EFF = Extended Fund Facility; ESAF = Enhanced Structural Adjustment Facility; LICs = Low-Income Countries; MICs = Middle Income Countries; SAF = Struc-
tural Adjustment Facility; SBA = Stand-By Arrangement. +*, Signicantly positive;
*, Signicantly negative; +, Positive but insignicant;
, Negative but insignicant; 0, Very close to zero.
a
Countries with low propensity scores show improvement, while for those with high propensity scores inequality deteriorates.
b
Signicant only for SAF/ESAF, positive but insignicant in mixed sample.
c
This study applies Heckman correction to growth equation, however, the inverse mills ratio (IMR) turns insignicant. The author notes that his participation equation is not strong. Therefore, it is
dicult to know whether the insignicance of the IMR is because a stable participation equation is not identied or participation is random.
d
Finds signicant negative eect for prolonged users only, while the impact on growth is insignicant for temporary users. GEE: General Evaluation Estimator.
e
Results from IV regressions are very close to zero.
f
Easterly (2005) notes that his instruments are weak.
g
They report a signicantly positive impact from the interaction of IMF and World Bank programs.
h
Based on descriptive comparison vis-a-vis the control group constructed by the PSM, therefore, signicance level is not reported.
i
Positive eect is reported for years following the initiation of programs.
j
The ndings are reversed for the period 200009 with IMF programs leading to lower poverty and lower inequality.
k
Selection model uses strategic probit with partial observability.

33
34 WORLD DEVELOPMENT

impact of aid on growth easing the burden of macroeconomic
adjustment when recipient countries are faced by large shocks.
This paper makes three key contributions to the existing lit-
erature on the impact of IMF-supported programs: First, by
focusing entirely on LICs, this paper contributes to a nascent
literature that distinguishes the impact of IMF-supported pro-
grams by country-income groups, explicitly recognizing LICs
unique characteristics. As discussed earlier this level of disag-
gregation also signicantly improves the explanatory power of
the participation model, which is the key step to correct for
selection bias. Second, based on a sample covering three dec-
ades, this paper examines a wide range of macroeconomic and
social outcomes using the PSM to correct for selection bias,
previously implemented by only a handful of studies for a
few outcome indicators for a mixed sample of countries.
Third, as opposed to the previous literature this study explic-
itly takes into account the implementation record of pro-
grams.
sentation of a balance of payments need is, indeed, the funda-
mental condition underlying the use of IMF resources. A
countrys balance of payments becomes unsustainable when
current policies cannot be continued indenitely and a drastic
policy shift becomes required. Sustainability of the current
account decit depends not only on the size of the decit
and factors inuencing it but also on the countrys willingness
and ability to borrow. This paper looks into a wide range of
potential economic and political determinants of participa-
tion. Most are already tested in previous studies (see Sturm
et al., 2005, and Moser & Sturm, 2011, for a summary).
(a) Economic variables
Table 2 presents the economic determinants of participation
tested in previous studies.
In addition to these variables, this study tests some new vari-
ables (Table 3), including a number of global variables to cap-
ture time-specic eects arising from common external shocks.

3. DETERMINANTS OF PARTICIPATION (i) A composite indicator for macroeconomic stability

This paper focuses on IMF-supported programs addressing
urgent nancing needs arising from policy or exogenous
shocks, therefore, potential determinants of participation are
closely associated with balance of payments sustainability
(Bird & Rowlands, 2006; Milesi-Ferretti & Razin, 1996). Pre-
When faced with increased macroeconomic instability
resulting from domestic policy slippages and/or exogenous
shocks, governments are more likely to turn to the IMF to
ease the burden of adjustment. Previous studies include
ination and budget balance to GDP with mixed results. In
order to assess the macroeconomic policy stance based on a

Table 2. Economic determinants of participation tested in previous studies

Variable Expected sign Findings in previous studies
Reserves (months of imports) Negative Invariably signicant
Real GDP growth Negative Mixed results
Debt Service/Exports Mixeda Mixed results
Current account balance/GDP Negative Mixed resultsb
Ination Positivec Mixed results
Government budget balance/GDP Negative Mixed results
Income per capita Negative Mostly signicant (in mixed samples of LICs and MICs)
Terms of trade shocksd Negative Mixed results
U.S. real interest rate Positive Mixed results
Paris Club debt reschedulinge Positive Signicant
Source: Draws on Sturm et al.(2005), expanded by the author.
a
While a heavy debt burden increases the demand for IMF nancing weaker capacity to repay may adversely aect the supply side. However, high debt
service could also signal the willingness to repay, leading to a positive association on the supply side as well.
b
Mixed results possibly reect samples mixing traditional current account crises and capital account crises. Moreover, various studies included pre-
cautionary SBAs as well as IMFs structural adjustment programs, mixing immediate and protracted nancing needs.
c
Positive on the demand side, but the IMF would not provide nancing without a strong adjustment program.
d
Both the change in price of exports to price of imports and overall terms of trade losses scaled by lagged GDP.
e
An IMF-supported program has usually been a pre-requisite for a PC debt rescheduling. This study uses a dummy variable taking the value of 1 if a PC
rescheduling takes place in the current or the following year, and zero otherwise.

Table 3. Economic determinants of participation: new variables

Variable Expected sign
a
A composite indicator for macroeconomic stability (mitot) Positive
Net resource transfers/GDP Negative
Real export growth Negative
Common global shocks
Global demand growthb Negative
Change in real non-oil commodity pricesc Negative (for commodity exporters)
Change in real oil pricesd Positive (for oil importers)
Source: Author.
a
Higher levels of the indicator correspond to increased macroeconomic instability.
b
The cyclical components of World GDP and real World trade.
c
Interacted with a dummy variable indicating commodity exporters.
d
Interacted with a dummy variable indicating net oil importers. The ratio of oil imports to GDP is also tested.
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES
comprehensive set of complementary indicators, this study
uses a composite indicator derived as:
   
ln cpicpiit ln xrxrit
1
it resit
resit
1
mgst
1
gbalit
gdpit
mitotit


it
1
rD lncpi rD lnxr rDres=mgst
1 rgbal=gdp
ln1 blackprit
rD lnxr
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 (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 r is the standard deviation of each
variable. Weights are inverse of standard deviation for each
component for all countries over the full sample after remov-
ing the outliers. 6 Higher levels of mitot indicate increased
macroeconomic instability. A variant of this composite indica-
tor was introduced by Jaramillo and Sancak (2009). This study
adds the black market premium as a separate component of
the index, inspired by Fischer (1993). 7 Components of the
index are also tested.
(ii) Net resource transfers to GDP
This variable reects combined and complementary factors
of the debt service burden, creditors willingness to nance
the balance of payments need, and donor inows that may o-
set nancial outows. Using the denition in the World
Banks Global Development Finance (GDF) database:
nrt D
P
I short fdi
profit port grant
where nrt is net resource transfers; D is disbursements on long-
term debt and IMF purchases/loans; P is repayments on long-
term debt and IMF repurchases; I is total interest payments;
short is the change in short-term debt; fdi is the foreign direct
investment; prot is fdi prot remittances; port is portfolio
investment ows; and grant is ocial grants excluding techni-
cal cooperation. 8 Components are also tested individually.
(b) Institutional and political variables
Institutional quality is tested by the World Bank CPIA and
the ICRG index as measured by the International Country
Risk Guide. 9 Political variables tested in this study cover
the extensive set tested in Moser and Sturm (2011).
(i) Previous IMF engagement
This variable was signicantly positive in many studies, indi-
cating the persistence in IMF engagement. This study uses two
versions: (i) The lag of a three-year moving average of a
dummy indicating whether or not a country had an IMF-
supported program; and (ii) A dummy variable taking the
value of one if a country had an IMF-supported program in
any of the previous ve years. In addition to the programs
addressing shocks, ECF programs providing medium-term
support to LICs with protracted balance of payments prob-
lems and programs supported by the Policy Support Instru-
ment (PSI), the Funds nonnancial policy support tool for
LICs, are included when creating the dummy variable for pre-
vious IMF engagement.
(ii) Election dummy
New governments are more likely to request an IMF
arrangement and governments may be reluctant to request a
35
program with upcoming elections. This study constructs two
versions of an election dummy separately for executive and
legislative elections: lagged elections marking elections in the
previous year and upcoming elections marking elections in
the following year. Some studies found this variable signicant
(Dreher, Sturm, & Vreeland, 2006; Moser & Sturm, 2011;
Przeworski & Vreeland, 2000).
(iii) Other political variables
Using the Moser and Sturm (2011) dataset a number of
political variables are tested: political instability, social unrest,
a measure of political rights and civil liberties (Freedom House
Index), political globalization, temporary membership on the
UN Security Council, relative size (share in world GDP and
population), and variables capturing US inuence (share of
a countrys bilateral trade with the United States relative to
the countrys GDP and how often countries vote in line with
the United States at the United Nations General Assembly).
4. METHODOLOGY
(a) Identication of the dependent variable: approval of IMF
arrangements
The dependent variable is a panel dummy that indicates par-
ticipation in an IMF-supported program addressing policy or
exogenous shocks. It takes the value of one if such a program
is approved, and zero for normal episodes. The latter is iden-
tied as the initial year of two successive years with no IMF
nancing for such shocks when the member is eligible to access
IMF resources. Programs include those addressing an immedi-
ate balance of payments need arising from policy or exogenous
shocks: SBAs, Exogenous Shocks Facility (ESF), Standby
Credit Facility (SCF), Rapid Credit Facility (RCF), Compen-
satory Financing Facility (CFF) and augmentations of struc-
tural adjustment programsStructural Adjustment Facility
(SAF), Enhanced Structural Adjustment Facility (ESAF),
and Extended Credit Facility (ECF). 10 Over the sample per-
iod of 19802010, 289 nancing events and 437 normal epi-
sodes are identied for 55 countries.
Two important modications to this set are made: (i) Pre-
cautionary SBAs as well as SBAs and ECF augmentations
addressing natural disasters are excluded; and (ii) Some
SAF/ESAF/ECF arrangements are added when they address
immediate balance of payments needs arising from policy
shocks. The latter cases are determined by program interrup-
tions preceding SAF/ESAF/ECF programs. 11 Similar to the
Mecagni (1999, chap. 9), a delay of more than six months in
completing a review owing to noncompliance with macro per-
formance criteria is taken as an interruption, often accompa-
nied by undrawn balances. First time SAF/ESAF/ECF
arrangements addressing shocks are identied by narratives
in relevant sta reports. A drastic shift in macroeconomic poli-
cies under a program to close a nancing gap indicates an
immediate nancing need.
Normal episodes are identied as the initial year of two suc-
cessive years with no IMF nancing for policy or exogenous
shocks when the member is eligible for IMF nancing. 12
Years with no programs that are immediately followed by
IMF nancing programs are excluded from the set of normal
episodes as negotiations may have taken place in these years.
Therefore, economic circumstances in these years may resem-
ble those of program years. Allowing a safe distance away
from program episodes helps better distinguish economic cir-
cumstances of program versus normal episodes. Table 4 pre-
36 WORLD DEVELOPMENT

Table 4. Observations excluded from normal episodes

Countries ineligible to use IMF resources
Non-membersa
Members with overdue obligations to the IMF
Signicant arrears to other bilateral and multilateral creditorsb
Other observable supply constraints that would deny access to IMF resources
Program interruptions or break-up of negotiations for a program
Lack of technical capacity or willingness to implement upper credit tranche (UCT) quality adjustment programsc
Dierent nature of the immediate nancing need (unpredictable ahead of time)
IMF nancing for natural disasters through ENDA or ECF augmentations
Source: Author.
a
Countries are included only after they became members.
b
The absence of a cooperative debt relief/rescheduling from other creditors prohibits the use of IMF resources.
c
A member-country can freely draw up to 25% of its quota to address its balance of payments need. To draw on more than 25% requires a Fund program
with UCT conditionality. Sta-Monitored Program (SMP), Emergency Post-Conict Assistance (EPCA), and three years leading up to EPCAs are
excluded. SMPs are used to establish a track record of willingness and ability to implement necessary policy adjustments following program interruptions.
EPCA cases lack the ability to implement a UCT-quality program. EPCA loans are usually quick-disbursing and do not involve adherence to performance
criteria.

sents other cases excluded from normal episodes based on sup-
ply constraints and the nature of the balance of payments
need.
For 55 countries over the sample period of 19802010, the
frequency of the approval of programs addressing policy or
exogenous shocks (excluding programs addressing natural dis-
asters and precautionary SBAs) is 26%. Two hundred and
eighty-nine programs were approved out of 1,124 observations
of which all were countries that were eligible to use the IMF
resources for these types of programs (including years after
a country became a member of the IMF and excluding the
years with arrears to the IMF, program interruptions, and
SMP and EPCA cases).
(b) Econometric methodology for the participation model
The eects of various economic variables on the probability
of a LIC requesting IMF nancing in response to shocks are
assessed by estimating a binary response model for panel data.
The general specication for panel probit models is given by:
y it 1 if Fund financing is requested
y it 0 normal episodes
P y it 1jxit ; ci Ux0it b ci i 1; . . . ; n and t 1; :::; T
where, y is the observed outcome, U is the cumulative normal
density function (c.d.f.), xit is the 1xk vector of explanatory
variables, and b is kx1 vector of coecients associated with
xit . Dierent estimators are constructed depending on their
assumptions for the panel heterogeneity, i.e., how they treat
ci . 13 The estimations are carried out step-by-step under dier-
ent estimators and a correlated random eects probit model is
preferred based on the econometric tests for the signicance of
both the individual-specic eect and the sample average for
covariates.
(c) Propensity Score Matching (PSM) methodology
This paper uses the PSM approach to control for selection
bias. The PSM is a relatively new and innovative class of sta-
tistical methods for impact evaluation using macroeconomic
data. It involves a statistical comparison of country groups
based on two steps. First, the probability of participating in
IMF-supported programs is estimated conditional on observ-
able economic and political conditions (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. 14
The matching based on the likelihood of participation in
IMF-supported programs assures similarity of initial macroe-
conomic conditions and country characteristics in the control
group. The control group provides in eect a proxy for the
counterfactual, that is, for macroeconomic outcomes if pro-
gram countries had not had a program. The eects of pro-
grams are then calculated as the mean dierence in a range
of macroeconomic outcomes across these two groups.
The PSM is useful when only observed pre-treatment char-
acteristics are believed to aect program participation. Two
necessary assumptions for identication of the program eects
are (i) conditional independence and (ii) presence of a common
support. Conditional independence, also called confounded-
ness, 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.
A well-specied and comprehensive selection model explain-
ing the participation in IMF-supported programs helps sup-
port the conditional independence assumption. The second
condition, i.e., presence of a common support, ensures that
1 treatment observations have comparison observations
nearby in the propensity score distribution.
In the analyses of this paper, IMF engagement is taken as a
treatment status. Countries that have engagement with the
IMF are called the treatment group whereas the others in
the sample are called the control group. The average treatment
eect of IMF engagement on the treated group (ATT) is given
by:
ATT EY i1 jDi 1
EY i0 jDi 1 2
where D is the dummy variable identifying LICs with IMF-
supported programs, Y i0 jDi 1 is the value of the macroeco-
nomic outcome that would have been observed if a LIC with
a program had not had such a program, and Y i1 jDi 1 is the
outcome value observed on the same country. Conditional on
some control variables X, if outcomes are independent of the
IMF engagement:
ATT EY i1 jDi 1; X i 
EY i0 jDi 0; X i  3
where EY i0 jDi 1; X i  is replaced with EY i0 jDi 0; X i ,
which is observable. Rosenbaum and Rubin (1983) propose
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES
that one can match the treated units and control units on their
propensity scores conditional on X, estimated by simple probit
or logit models. When PSM is used, the ATT now can be esti-
mated as:
ATT EY i1 jDi 1; pX i 
EY i0 jDi 0; pX i  4
The strategy then consists in computing the dierences in
the outcomes Y i for observations with similar propensity
scores. This study uses the nearest neighbor technique using
four nearest neighbors.
(d) Identication of the treatment status
The treatment variable is identied mostly symmetrically to
the one used in the participation model. A panel dummy vari-
able taking the value of one for the approval of IMF-
supported programs with LICs addressing immediate balance
of payments needs, and zero for non-program episodes, is con-
structed as the treatment variable.
Severe state failure events are excluded from both program
and non-program sets as the macroeconomic outcomes in
these episodes will be frail and donor involvement will be dis-
rupted, independent of the impact of IMF-supported pro-
grams. 15 Moreover, years of program interruptions are
excluded from the sample.
Some asymmetries compared to the dependent variable in
the participation equation are introduced for nonprogram epi-
sodes to increase the common support for the PSM. The treat-
ment variable includes nonprogram years followed
immediately by a program, and nonprogram episodes without
IMF membership. On the other hand, on-track IMF-
supported programs addressing protracted balance of pay-
ments needs (ECFs and its predecessors and Extended Fund
Facility (EFF) arrangements) or PSIs providing non-
nancial policy support are excluded from the sample to con-
struct the control group purely from the countries that have
no IMF engagement.
5. RESULTS
This section presents rst the participation model followed
by the PSM estimates of the short-term macroeconomic eects
of IMF-supported programs addressing policy slippages or
external shocks.
(a) Participation model
The benchmark participation model was derived through a
general-to-specic approach, starting from a wide range of
potential variables. The sample covers 55 countries over
19802010. All explanatory variables, except exogenous
shocks, are lagged one period for IMF nancing events to
account for potential endogeneity between macroeconomic
outcomes and programs. An RE specication is strongly sup-
ported since the correlation of error terms for the same coun-
try, q, is signicant. Moreover, country-specic averages of
some explanatory variables are signicant, indicating that a
correlated RE model is appropriate to account for the correla-
tion between unobserved heterogeneity and these explanatory
variables.
A deterioration in macroeconomic conditions and adverse
external shocks signicantly increases the probability of par-
ticipation in programs addressing policy or exogenous shocks
(Table 5). Compared to earlier studies, examining more
homogenous nancing events greatly improves model speci-
37
cation. A lower reserve coverage, a deterioration in the current
account balance, a weaker real GDP growth, and increased
macroeconomic instability (evident in higher scal decits,
ination and exchange market pressures) would signicantly
increase the likelihood of IMF nancing. Moreover, global
conditions, including changes in real oil and non-oil commod-
ity prices and world trade, are also signicant determinants of
participation; therefore, adverse global shocks could create
cycles in demand for IMF nancing. Finally, persistent dier-
ences in debt service burden and FDI inows among LICs
seem to be signicantly associated with unobserved country
heterogeneity.
When the composite indicator for macroeconomic stability
is replaced by its individual components, ination and depre-
ciation plus the black market premium turn out signicant
while the change in reserves and government balance come
out with the correct sign but are not signicant. 16 When the
CPIA is added to the model its coecient is negative, sizeable,
but not signicant. It becomes signicant only when ination
or the composite depreciation indicator is omitted from the
model. This nding suggests that the IMF tends to have pro-
grams addressing shocks more often with countries with weak
institutions, highlighting once more the importance of control-
ling for selection bias.
The Paris Club dummy is also highly signicant reecting a
history of prolonged debt distress in the sample. However,
both the Paris Club dummy and the debt service to exports
ratio (averaged over the sample period) become insignicant
in the sub-sample of 19902010 while having much higher
and signicant coecient estimates for the sub-sample of
19802000. Both variables appear to capture the impact of
LIC debt crises of the 1980s and the ensuing debt restructuring
on the approval of more IMF-supported programs addressing
macroeconomic imbalances as a pre-requisite for debt relief. 17
Reserve coverage has a larger impact on participation for
non-CFA countries vis-a-vis CFA countries. 18 The likelihood
ratio test rejects equality of coecients at the 5% signicance
level. The reserve pooling arrangement among CFA countries
as well as the French Treasurys guarantee for the convertibil-
ity of the CFA franc might have provided extra buers.
The variable two-year change in real non-oil commodity
prices (over t and t
1) interacted with a dummy variable
for non-oil commodity exporters (whose non-oil commodity
exports exceed 66% of merchandise exports, median over
19802010) has a signicantly negative impact on participa-
tion suggesting potential bunching of IMF nancing in
response to adverse price shocks. 19
Among political variables, only the dummy variable captur-
ing lagged legislative elections is signicant. Previous IMF
engagement, although insignicant, has a sizeable and nega-
tive eect and becomes signicant in the post-1989 sample.
This result is at odds with earlier ndings in the literature. It
suggests that previous IMF engagement, including through
structural programs, tends to reduce the likelihood of pro-
grams addressing immediate nancing needs arising from pol-
icy and exogenous shocks. Examining transmission channels
for this eect, including through policy advice and capacity
building, is left for future research.
Country-specic averages of external debt service to exports
and FDI to GDP over the sample period are signicant, there-
fore, correlated with the unobserved country heterogeneity.
Persistent divergences in debt burden and FDI inows among
LICs could explain why some countries use IMF resources
more often. Except for FDI, other indicators of investors
and donors willingness to meet nancing needs prior to pro-
grams are not signicant.
38 WORLD DEVELOPMENT

Table 5. Demand for the fund nancing in response to policy or external shocks
Lagged legislative elections 0.367**
(0.168)
Reserve coverage in months of imports (non-CFA) (t
1)a
0.506***
(0.0512)
Reserve coverage in months of imports (CFA) (t
1)a
0.34***
(0.0528)
Current account balance to GDP (t
1)
0.0505***
(0.0109)
Macroeconomic stability indicator (t
1) 0.0543***
(0.0207)
Real GDP growth (t
1)
0.0835***
(0.0194)
Two-year change in real non-oil commodity prices (t) interacted with non-oil commodity exporter dummy
0.0142**
(0.00585)
Real world trade, cyclical component
0.0345*
(0.0184)
Two-year change in real oil prices (t
1) interacted with oil importer dummy 0.00466**
(0.00202)
Paris Club dummy 0.47***
(0.175)
Country-specic averages
Total debt service to exports 0.0227*
(0.0118)
FDI to GDP
0.117**
(0.0471)

Pseudo R2 0.43
LR test : b2 = . . . = bk = 0 v2 (Prob) 377.1 (0.00)
LR test : q = 0 v2 (Prob) 11.8 (0.00)
Number of observations 719
Financing events 285
Normal episodes 434
Sample probability 0.40
Number of countries 55
Source: Authors calculation.
Note: Demand for IMF nancing in response to policy and/or exogenous shocks excluding natural disasters is estimated by a correlated random eects
probit model. Signicant at 10%:*; 5%: **; and 1%: ***; standard errors in parenthesis. Country-specic averages are calculated as the sample average of
variables for each country. FDI = foreign direct investment; LR = likelihood ratio test.
a
The CFA franc zone consists of 14 countries in sub-Saharan Africa, each aliated with one of two monetary unions maintaining the same currency, the
CFA Franc.

Income per capita is not signicant. Its signicance in previ-
ous studies may have been driven by mixed samples of LICs
and MICs, and also including structural arrangements (ECFs
and its predecessors) in the sample.
As the model is non-linear the eect of a variable on the
probability of participation depends not only on its own coef-
cient but also on the values of other explanatory variables
and their coecients. Therefore, estimated coecients have
no direct interpretation. A useful measure of the relative
impact of covariates is the average marginal eects of a specic
covariate on the response probability averaged across the dis-
tribution of all covariates (Table 6). Reserve coverage is the
most inuential covariate, followed by the real GDP growth,
current account balance, and the macroeconomic stability
indicator.
The loss function approach suggested by Demirguc-Kunt
and Detragiache (1999) determines the threshold probabil-
itya predicted probability exceeding this threshold signals
a nancing event. The expected loss function is the weighted
average of type I and type II errors, with weights reecting
costs attached to each type of error. Minimization of the loss
function yields the threshold probability. This study calculates
threshold probabilities for 3:1 ratio for missed nancing events
(type I error) versus false alarms (type II error). In order to
capture countries that might need IMF nancing at an early
stage, missed nancing events are assigned a higher weight.
Figure 1 plots missed nancing events (type I error), false
alarms (type II error), and the weighted loss function versus
the threshold probability.
An increase in the threshold probability raises type I error
while reducing type II error. The loss function is minimized
at the threshold probability of 0.27, corresponding to about
27% false alarms versus 10% missed nancing events in the
sample.
(b) Results: impact of the short-term IMF engagement
Variables aecting participation in programs also aect
growth and other macroeconomic outcomes, leading to selec-
tion bias. Table 7 illustrates this point with regressions esti-
mating the impact of the propensity scores and IMF
program dummy on growth. As expected, the coecient esti-
mates for propensity scores are negative and highly signicant.
The IMF program dummy is not signicant in the full sample
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES 39

Table 6. Selection model: average marginal eects

Average marginal eects
Lagged legislative elections 0.070**
Reserve coverage in months of imports (non-CFA) (t
1)a
0.096***
Reserve coverage in months of imports (CFA) (t
1)a
0.065***
Current account balance to GDP (t
1)
0.010***
Macroeconomic stability indicator (t
1) 0.010***
Real GDP growth (t
1)
0.016***
Two-year change in real non-oil commodity prices (t) interacted with non-oil commodity exporter dummy
0.003**
Real world trade, cyclical component
0.007*
Two-year change in real oil prices (t
1) interacted with oil importer dummy 0.001**
Paris Club dummy 0.089***
Country-specic averages
Total debt service to exports 0.004*
FDI to GDP
0.022**
Source: Authors calculation.
Note: Demand for IMF nancing in response to policy and/or exogenous shocks excluding natural disasters is estimated by a correlated random eects
probit model. Signicant at 10%:*; 5%:**; and 1%:***. Country-specic averages are calculated as the sample average of variables for each country.
FDI = foreign direct investment.
a
The CFA franc zone consists of 14 countries in sub-Saharan Africa, each aliated with one of two monetary unions maintaining the same currency, the
CFA Franc.

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Threshold Probability
Wrong Calls (Type II error) Missed Events (Type I error)
Weighted Loss Function
Figure 1. Selection model: threshold probability analysis
(columns 1 and 2), however, it becomes signicant in the sub-
sample with propensity scores exceeding 0.5 (columns 36),
adding about 1.3% to growth (column 4, the lagged GDP
growth is not signicant in the sub-sample). After the onset
of concessional nancing, for the sample 19862010, the
growth impact of IMF-supported programs increases to
1.9% (column 6).
The PSM results suggest that programs lead to signicantly
better macroeconomic outcomes particularly for LICs experi-
encing substantial prior macroeconomic imbalances or severe
adverse external shocks (Table 8).
While growth is estimated to be 0.4% higher than the con-
trol group for the full sample, the impact rises to 11.7%
and becomes signicant only for countries with high propen-
sity scores experiencing immediate nancing needs owing to
existing macroeconomic imbalances or external shocks. Fur-
100% thermore, change in growth is, though not signicant, positive
and increases with propensity scores.
90% Overall, program countries attain signicantly higher cur-
rent account balances, and reserve coverage, as well as lower
80% ination and scal decits compared to their control groups.
70% Moreover, these variables tend to post signicant improve-
ments during the program, with the impact especially pro-
60% nounced for the sub-sample with high propensity scores.
Although program countries tend to have more depreciated
50% real exchange rates, dierences with the control groups
40% are not signicant. The change in real health spending per cap-
ita is signicantly higher for program countries for the full
30% sample while positive but not signicant for those with high
propensity scores. The change in real education spending per
20% capita is not statistically dierent from those of the control
10% group.
A relevant question is whether the growth impact could be
0% biased upward for program countries reecting a cyclical
recovery from a weak base. Table 9 presents PSM results that
match program countries to their control groups on both
propensity scores and the lagged GDP growth to ensure sim-
ilar pre-program growth performance. In the participation
model, a weaker pre-program GDP growth strongly raises
the probability of participation. Matching on both the lagged
growth and the propensity score would also control for the
direct impact of pre-program growth performance on
growth. 20 Program countries with high propensity scores
attain signicantly higher growth by about 1.252% compared
to the control group.
As shown earlier, the IMF tends to have programs address-
ing shocks more often with countries having weak institutions.
Therefore, it is worthwhile to explore if the estimated growth
impact could be biased against programs when the estimation
does not control for institutional quality. Table 10 presents the
PSM results that condition matching on both the propensity
score and a variable measuring institutional quality (lagged
ICRG and lagged CPIA). For comparability, the results of
matching only on the propensity score for the comparable
sample are also presented. The key nding is that the change
in real GDP growth becomes highly signicant for countries
with high propensity scores when matching also accounts for
the institutional quality. Using lagged ICRG, program coun-
40 WORLD DEVELOPMENT

Table 7. Growth regressions: propensity scores and IMF-supported programs

Full sample Propensity score > 0.5
(1) (2) (3) (4) (5) (6)
Propensity score
7.314 ***

5.895 ***

11.36 ***

12.05 ***

10.28 ***

11.38***
(0.911) (0.936) (2.496) (2.393) (3.612) (3.490)
IMF program dummy 0.311 0.216 1.291* 1.318* 1.840** 1.930**
(0.538) (0.527) (0.684) (0.683) (0.897) (0.894)
Real GDP growth (t
1) 0.174*** 0.0441 0.0655
(0.0348) (0.0449) (0.0559)
Constant 7.399*** 6.488*** 12.95*** 13.50*** 11.36*** 12.15***
(0.986) (0.983) (2.803) (2.748) (3.423) (3.361)
Observations 644 644 378 378 235 235
R-squared 0.226 0.260 0.189 0.186 0.234 0.228
Number of countries 55 55 53 53 51 51
Time dummies Yes Yes Yes Yes Yes Yes
Estimator FE FE FE FE FE FE
Sample period 19802010 19802010 19802010 19802010 19862010 19862010
Source: Authors calculation.
Note: Signicant at 10%:*; 5%:**; and 1%:***, standard errors in parenthesis.

Table 8. Results: impact of short-term IMF engagement by propensity score matching

All LICs LICs with weaker fundamentals
PS > 0.5 PS > 0.7 PS > 0.8
Macroeconomic outcomes
Real GDP growth (%) dgdpog 0.416 1.030* 1.348** 1.715**
(0.487) (0.581) (0.678) (0.814)
Ination (%) in100og
17.51***
22.39***
25.84***
33.35***
(4.171) (5.467) (6.530) (7.743)
Reserve coverage (in months of imports) impcovog 0.643*** 0.668*** 0.647*** 0.481***
(0.142) (0.151) (0.156) (0.171)
Current account balance plus FDI (% of GDP) cabfdiy 1.672** 2.672*** 4.614*** 4.585***
(0.838) (1.014) (1.142) (1.295)
Government balance (% of GDP) gbaltogdp100og 2.448*** 2.864*** 3.515*** 4.016***
(0.582) (0.728) (0.853) (0.968)
Change in real health spending per capita (%) heal_dlnrpop 8.001** 7.641 8.325 9.240
(3.943) (5.271) (6.132) (8.573)
Change in real education spending per capita (%) educ_dlnrpop 1.781
0.183
0.251
2.432
(3.165) (4.064) (4.972) (5.754)
Change in REER dreerog
2.402
1.820
2.138
1.993
(6.644) (7.739) (8.911) (10.71)
Change in macroeconomic outcomes (X(t)
X(t
1))
Real GDP growth (%) ddgdpog 0.269 0.813 1.475 2.102
(0.747) (0.907) (1.070) (1.295)
Ination (%) dinf
7.946*
12.76**
17.25**
20.92**
(4.714) (5.976) (7.076) (8.631)
Reserve coverage (in months of imports) drescv 0.606*** 0.772*** 0.796*** 0.658***
(0.104) (0.114) (0.125) (0.137)
Current account balance plus FDI (% of GDP) dcabfy 1.530** 2.043** 2.805*** 3.113***
(0.669) (0.822) (0.919) (1.087)
Government balance (% of GDP) dgovbaly 0.895** 1.173** 1.696*** 1.846***
(0.416) (0.506) (0.569) (0.654)
Number of observationsa 644 378 296 249
Source: Authors calculations.
Note: PS stands for the propensity score indicating the likelihood of IMF programs addressing immediate balance of payments needs. Changes in
macroeconomic outcomes refer to rst dierences of the variables in the top panel. Signicant at 10%:*; 5%:**; and 1%:***. Standard errors in parenthesis.
FDI = Foreign Direct Investment.
a
All variables except for health and education spending and change in Real Eective Exchange Rate (REER). The sample size is 267 for health and
education spending and 477 for REER.

tries with high propensity scores record improvements in
growth performance 1.83.3% higher than the control group.
Despite very dierent samples, the PSM results using lagged
CPIA indicate a similar range of 1.53.5% for the impact of
programs on the change in growth for the same group.
The estimated positive impact on growth could be attributed
to IMF nancing along with its catalytic eect easing the bur-
den of the short-term adjustment as well as the restoration of
macroeconomic stability, especially for countries experiencing
signicant levels of instability prior to the program. 21
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES 41

Table 9. Impact of short-term IMF engagement by propensity score matching, matching on propensity score and lagged GDP growth
All LICs LICs with weaker fundamentals
PS > 0.5 PS > 0.7 PS > 0.8
Macroeconomic outcomes
Real GDP growth (%) dgdpog 0.603 1.250** 1.691** 1.953**
(0.510) (0.592) (0.663) (0.802)
Ination (%) in100og
19.15***
23.57***
31.66***
35.12***
(5.937) (7.512) (8.712) (10.49)
Reserve coverage (in months of imports) impcovog 0.655*** 0.730*** 0.674*** 0.483***
(0.136) (0.151) (0.153) (0.173)
Current account balance plus FDI (% of GDP) cabfdiy 2.328*** 2.823** 4.474*** 5.191***
(0.876) (1.100) (1.253) (1.371)
Government balance (% of GDP) gbaltogdp100og 2.719*** 2.910*** 3.729*** 4.363***
(0.630) (0.787) (0.959) (1.144)
Change in real health spending per capita (%) heal_dlnrpop 6.867** 7.638 6.628 12.99
(3.135) (4.867) (6.100) (8.371)
Change in real education spending per capita (%) educ_dlnrpop 1.305
0.377
1.910
3.517
(2.584) (3.210) (4.497) (6.236)
Change in REER (%) dreerog
2.642 5.164 7.357 2.736
(9.353) (14.33) (16.92) (21.54)
Change in macroeconomic outcomes (X(t)
X(t
1))
Real GDP growth (%) ddgdpog 0.603 1.250** 1.691** 1.953**
(0.515) (0.601) (0.671) (0.819)
Ination (%) dinf
8.484*
11.33*
18.45**
21.13**
(4.962) (6.318) (7.882) (9.223)
Reserve coverage (in months of imports) drescv 0.651*** 0.777*** 0.837*** 0.684***
(0.106) (0.126) (0.142) (0.158)
Current account balance plus FDI (% of GDP) dcabfy 1.346** 1.998** 2.662*** 2.909***
(0.679) (0.846) (0.882) (1.003)
Government balance (% of GDP) dgovbaly 0.670 0.881* 1.393** 1.390**
(0.437) (0.533) (0.666) (0.770)
Number of observationsa 644 378 296 249
Source: Authors calculations.
Note: PS stands for the propensity score indicating the likelihood of IMF programs addressing immediate balance of payments needs. Changes in
macroeconomic outcomes refer to rst dierences of the variables in the top panel. Signicant at 10%:*; 5%:**; and 1%:***. Standard errors in parenthesis.
FDI = Foreign Direct Investment.
a
All variables except for health and education spending and change in Real Eective Exchange Rate (REER). The sample size is 267 for health and
education spending and 477 for REER.

(i) Robustness checks
This section rst explores the robustness of results to two
sensitivity analyses: (i) relaxing the adjustment for the imple-
mentation record of programs and (ii) setting the sample to
198099 for the comparability of results to earlier research.
Results are robust to these adjustments with some noteworthy
changes highlighted below. The sensitivity of results to hidden
bias is then examined.
When the IMF program dummy is not adjusted for the pro-
gram implementation the estimated impact on macroeconomic
outcomes remains qualitatively similar (Table 13). However,
the size of the impact tends to get lower, especially for govern-
ment balance and to a lesser extent for current account bal-
ance, ination, reserve coverage, and growth. The measure
of implementation accounts only for serious interruptions in
programs (e delay of six months or more in completing a
review) and does not map compliance with program condi-
tionality on a continuous scale. Therefore, the estimates may
reect the lower bound for the impact of a strongly imple-
mented program.
For comparability of results to earlier research, the impact is
estimated for the 198099 sample studied by most of the ear-
lier research. The results turn out to be qualitatively similar
and estimated quantitative impact on growth is even stronger
(Table 14). Therefore, the ndings are not driven by dier-
ences in the sample.
Sensitivity analysis to hidden bias: The key assumption
behind the PSM is conditional independence, implying that
the program participation depends entirely on observed char-
acteristics of LICs. However, hidden bias may arise from the
omission of important covariates or unobserved variables. A
strong participation model, encompassing a number of highly
signicant variables, should alleviate the hidden bias. More-
over, examining the impact of programs on rst-dierences
in outcomes should remove the eects of unobserved
country-specic factors.
In addition, Table 11 presents sensitivity of ndings to hidden
bias, Rosenbaums sensitivity analysis (Rosenbaum, 2002), for
the PSM results with matching on both the propensity score
and the lagged GDP growth. This analysis manipulates the esti-
mated 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 remain robust. The
parameter C is a measure of how much hidden bias can be pre-
sent before results of the study begin to change. A variable is
highly sensitive to hidden bias if the conclusions change for C
just barely larger than 1, and it is insensitive if the conclusions
change only for larger values of C. 22
42 WORLD DEVELOPMENT

Table 10. Impact of short-term IMF engagement on growth by propensity score matching, matching on institutional quality
All LICs LICs with weaker fundamentals
PS > 0.5 PS > 0.7 PS > 0.8
Matching on propensity score and lagged ICRG
Real GDP growth (%) dgdpog 0.750 1.918** 1.830** 2.555**
(0.651) (0.838) (0.908) (1.237)
Change in real GDP growth (%) ddgdpog 1.059 1.839** 2.091** 3.259**
(0.772) (0.915) (1.051) (1.427)

Number of observations 373 207 160 127

Matching on propensity score, comparable sample
Real GDP growth (%) dgdpog 0.428 1.295** 1.558** 2.151**
(0.573) (0.658) (0.771) (1.038)
Change in real GDP growth (%) ddgdpog 0.279 0.674 0.994 1.761
(0.844) (0.986) (1.167) (1.528)
Number of observations 373 207 160 127
Matching on propensity score and lagged CPIA
Real GDP growth (%) 0.412 1.035* 1.244* 1.841**
(0.535) (0.627) (0.689) (0.774)
Change in real GDP growth (%) 0.753 1.548* 2.106** 3.545***
(0.719) (0.849) (0.945) (1.102)
Number of observations 535 320 250 210
Matching on propensity score, comparable sample
Real GDP growth (%) dgdpog 0.512 1.219** 1.523** 1.746**
(0.470) (0.548) (0.626) (0.740)
Change in real GDP growth (%) ddgdpog 0.245 0.926 1.476 2.151*
(0.695) (0.824) (0.943) (1.112)
Number of observations 535 320 250 210
Source: Authors calculations.
Note: PS stands for the propensity score indicating the likelihood of IMF programs addressing immediate balance of payments needs. Signicant at 10%:*;
5%:**; and 1%:***. Standard errors in parenthesis.

Table 11. Short-term IMF engagement: Rosenbaum sensitivity analysis to hidden selection bias
All LICs LICs with Weaker
Fundamentals PS > 0.8
C Probability C Probability
Macroeconomic outcomes
Real GDP growth (%) 1.06 0.043 1.52 0.048
Ination (%) 1.03 0.047 1.24 0.047
Reserve coverage (in months of imports) 1.90 0.048 1.85 0.047
Current account balance plus FDI (% of GDP) 1.13 0.046 2.09 0.051
Government balance (% of GDP) 1.86 0.045 2.41 0.050
Change in macroeconomic outcomes (X(t)
X(t
1))
Real GDP Growth (%) 1.08 0.041 1.62 0.048
Ination (%) 1.02 0.051 1.23 0.045
Reserve coverage (in months of imports) 2.44 0.048 2.90 0.044
Current account balance plus FDI (% of GDP) 1.13 0.046 1.43 0.048
Government balance (% of GDP) 1.24 0.050 1.50 0.047
Source: IMF sta calculations.
Note: PS stands for the propensity score indicating the likelihood of Fund programs addressing immediate balance of payments needs. Changes in
macroeconomic outcomes refer to rst dierences of the variables in the top panel. The sample is composed of 55 LICs and covers 19802010. C is a
measure of how much hidden bias can be present. i.e., how much C can deviate from 1, before results of the study begin to change. FDI = Foreign Direct
Investment.

The reserve coverage and the change in reserve coverage are
the least sensitive while results for ination are highly sensitive
to hidden bias for both the full sample and the sample of LICs
with high propensity scores. In general, results are less sensi-
tive to hidden bias for countries with high propensity scores.
As a nal robustness check for sensitivity to hidden bias,
Table 12 presents the results from dierence in dierences
(DID) regressions. In the DID approach, rst dierences in
growth are regressed on the rst dierences in all time-
varying controls and the IMF program dummy for all obser-
vations that had no program in t
1. Consistent with the
ndings from the PSM, the growth impact of programs is
insignicant in the full sample but becomes signicant for
the very high propensity group. Results also highlight the
importance of exploring non-linearities in the program impact
on growth using the hybrid PSM-DID approach. 23
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES
Table 12. Impact of short-term IMF engagement on growth, dierence in dierences regressions
IMF program dummy
DReserve coverage in months of imports (non-CFA) (t
1)
DReal GDP growth (t
1)
Number of observations
Source: Authors calculations.
Note: Program impact estimated by the dierence in dierences estimator. Both the dependent variable and covariates are rst dierenced. PS stands for
the propensity score indicating the likelihood of IMF programs addressing immediate balance of payments needs. Changes in macroeconomic outcomes
refer to rst dierences of the variables in the top panel. Signicant at 10%:*; 5%:**; and 1%:***. Standard errors in parenthesis.
6. CONCLUSIONS
This paper assesses the impact of IMF-supported programs
addressing the urgent nancing needs of LICs arising from
policy or exogenous shocks on a wide range of macroeco-
nomic outcomes. Such shocks tend to be associated with
immediate and substantial output costs owing to policy tight-
ening required to restore macroeconomic stability. To the
extent that programs provide the IMFs own nancing and
catalyze donor assistance they can help ease the pace of adjust-
ment, thereby, alleviate the adverse impact on growth and
social outcomes.
Results show that these programs lead to better macroeco-
nomic outcomes, particularly for countries experiencing sub-
stantial prior macroeconomic imbalances or severe adverse
external shocks. Bal-Gunduz and Crystallin (2014) report a
signicant increase in ocial development assistance for the
same subset of programs. These ndings taken together sug-
gest that the catalytic nancing impact of programs on
ODA may be an important channel of transmission for the
impact of programs on growth.
While program countries achieve higher short-term growth
compared to the control group, the impact becomes signicant
and rises to about 1.01.7% only for countries with high
propensity scores (i.e., experiencing immediate nancing needs
owing to macroeconomic imbalances or external shocks).
Moreover, program countries attain signicantly higher
NOTES
1. Bird and Rowlands (2007) 2 explore the extent to which the IMF has
had a catalytic eect on ocial development assistance and the potential
channels through which catalysis might work. They nd strong evidence of
a positive association between participation in IMF-supported programs
and ODA and suggest that this may have more to do with conditionality
than with the provision of IMF resources.
2. Steinwand and Stone (2008), on the other hand, argue that the
usefulness of such analysis is limited since randomly chosen, instead of
theoretically guided, model specications are compared, which could
introduce omitted variable bias.
3. For an extensive survey of earlier literature on eects of Fund
programs see Ul Haque and Khan (1998).
4. Mercer-Blackman and Unigovskaya (2000) report a positive
association between growth and compliance in transition economies.
Dreher (2006) nds that increased compliance reduces the negative impact
of programs on growth.
43
All LICs LICs with weaker fundamentals
PS > 0.5 PS > 0.7 PS > 0.8

0.0932 1.079 2.049 2.802**
(0.879) (1.096) (1.254) (1.416)
0.472** 0.512** 0.521** 0.512**
(0.186) (0.233) (0.246) (0.257)

0.233***
0.230***
0.217***
0.214***
516 352 310 286
current account balances and reserve coverage, as well as
lower ination and scal decits compared to their control
groups. When matching also controls for institutional quality,
program countries with high propensity scores record changes
in growth 1.53.5% higher than the control group.
A key methodological contribution of this paper is to imple-
ment the PSM approach to address selection bias, i.e., system-
atic dierences in initial economic conditions of a program
versus a non-program country. Although a handful of papers
used the PSM approach in assessing the impact of IMF-
supported programs none of them examined this specic sub-
set of programs. The success of the PSM approach in address-
ing selection bias hinges on constructing a control group of
nonprogram countries experiencing similar economic dicul-
ties, which requires a well specied selection model with high
explanatory power to estimate the likelihood of programs,
i.e., propensity scores. Dierently from the previous literature,
this study focuses on a homogenous subset of programs with
LICs addressing immediate nancing needs and explicitly
takes into account observable supply constraints precluding
a members access to IMF nancing. Results show that this
empirical strategy leads to a well-specied selection model
with high predictive power, encompassing a wide range of eco-
nomic determinants of program participation. Overall, exam-
ining the impact of IMF-supported programs with more
granularities appears to be a promising avenue for future
research.
5. For early studies Ul Haque and Khan (1998) conclude that those that
do not correct for selection bias (beforeafter and withwithout
approaches) yield less favorable results compared to studies using the
General Evaluation Estimator (GEE). However, Dicks-Mireaux et al.
(2000) largely discredit the validity of the GEE owing to many restrictive
assumptions necessary to dene the counterfactual based on policy
reaction functions.
6. Observations above the 95th percentile for ination and depreciation,
above the 97.5th percentile or below the 2.5th percentile for the change in
reserve coverage and below 5th percentile for government balance to GDP
are considered as outliers.
7. Moreover, reserves are scaled by lagged imports of goods and services,
rather than by base money as in the original index, to assure comparability
across countries.
8. This series is available in the World Banks GDF only 1985 onwards.
44 WORLD DEVELOPMENT

9. The World Banks CPIA (the Country Policy and Institutional 16. As depreciation and ination are highly correlated only one of them
Assessment) is a broad indicator of the quality of a countrys present enters into the model.
policy and institutional framework. It is based on 16 criteria which are
grouped into four clusters: economic management, structural policies, 17. The Paris Club dummy is not a proxy for the past use of IMF
policy for social inclusion and equity, and public sector management and resources in the context of the IMF-supported programs examined in this
institutional quality. It is strongly correlated with other indices of paper. The correlation between the past engagement and the Paris club
institutional quality. dummy is low at 14% in the estimation sample. Including variables
capturing the past IMF engagement does not aect the size and
10. SAF/ESAF/ECF augmentations and CFFs explicitly address exoge- signicance of the coecient estimated for the Paris club dummy.
nous shocks. However, they may overlap with policy slippages since in
many cases CFFs were approved in parallel to SBAs. ECF augmentations 18. Estimating the model with only non-CFA countries gives qualita-
are more likely to address purely exogenous shocks as they are approved tively similar results.
when a member with an on-track ECF program is hit by an exogenous
shock. Finally, CFFs and ECF augmentations do not necessarily include
19. Two-year changes in real non-oil commodity prices over the current
all exogenous shocks leading to Fund nancing since SBAs could address
and the previous year is also signicant at 10%. As expected interaction
policy shocks combined with exogenous shocks.
with the non-oil commodity exporter dummy increases the size and the
signicance of the estimated coecient.
11. Anna Ivanova kindly shared the Ivanova, Mayer, Mourmouras, and
Anayiotos (2003) dataset on program interruptions. This dataset identies
20. With matching on lagged growth, the impact on the level and the rst
interruptions by programs, therefore, it is extended to identify specic
dierenced series are equivalent.
years of interruptions within a program. Mecagni (1999, chap. 9) dataset is
used to identify SAF/ESAF program interruptions. IMF Monitoring of
Fund arrangements (MONA) database, Ex-Post Assessments (EPAs), 21. Countries treated by the World Bank and included in non-
specic program documents and sta reports are used to identify specic program episodes could potentially weaken the estimated economic
years of interruptions and ll the gaps in the sample vis-a-vis the impact. On the other hand, countries treated by both the IMF and
aforementioned studies. the World Bank do not pose a problem for the interpretation of results
because the paper recognizes the catalytic nancing eect as an important
channel of transmission for the IMF-supported programs. For programs
12. By this denition ECF arrangements that are on-track and not
addressing policy or exogenous shocks Bal-Gunduz and Crystallin (2014)
augmented to address immediate balance of payments needs are also
report a signicant increase in multilateral ODA as well as signicantly
included in normal episodes when other conditions are met as well. Only
higher proportion of aid allocated as general budget support from the
ECF augmentations or program interruptions during an ECF program
IDA and the EC. As highlighted in that paper the catalytic impact of
signal the presence of an immediate nancing need. Many LICs have
programs on ODA from multilateral and bilateral donors has more of a
back-to-back on-track ECF arrangements during the sample period.
simultaneous and collaborative meaning in the context of programs with
LICs as opposed to the exogenous nature of programs for catalyzing
13. Pooled probit models assume independence of observations over private ows. Further disentangling the impact of the World Bank- versus
both t and i. A random eects (RE) probit model treats the individual- the IMF-engagement could be an area for further research.
specic eect, ci , as an unobserved random variable with ci jxit  INlc ; r2c
if an overall intercept is excluded, and imposes independence of ci and xit .
22. Robins (2002) expressed skepticism about the usefulness of sensitivity
A xed eects (FE) probit model treats ci as parameters to be estimated
analysis as he proved that Rosenbaums C t the criteria of a paradoxical
along with b, and does not make any assumptions about the distribution
measure: its magnitude increases as the analyst decreases the amount of
of ci given xit . This can be problematic in short panels as both b and ci are
hidden bias by measuring some of the unmeasured covariates. As such,
inconsistently estimated owing to an incidental parameters problem.
this measure could be useful only if experts could provide a plausible and
Finally, a correlated random eects model relaxes independence between
logically coherent range of C.
covariates and the individual-specic eect using the Chamberlain (1982)
and Mundlak (1978) device under conditional normality. In this
specication, the time average is often used to save on degrees of freedom. 23. Cerulli (2015) highlights that hybrid program evaluation methods are
generally more robust than stand alone approaches. Dierence-in-
dierences (DID) is suitable to address both observable and
14. This study uses the nearest neighbor matching approach, which
unobservable selection in a parametric model making use of the
constructs a control group of countries by choosing those four non-
longitudinal data structure where outcome data as well as covariates are
program countries with probability of requesting a program as close as
available both before and after treatment. The disadvantage of this
possible to that of the specic program country in question.
approach is that it imposes a linear parametric form for the program
impact on growth. The hybrid use of the PSM and the DID has the
15. The severe state failure events are identied from Political Instability advantage of relaxing the linear-in-parameters form of the outcome
Task Force (PITF) dataset. Four types of political crises are included in equation for the full sample.
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.

REFERENCES

Andersen, T. B., Hansen, H., & Markussen, T. (2006). US politics and Atoyan, R., & Conway, P. (2006). Evaluating the impact of IMF
World Bank IDA-lending. Journal of Development Studies, 42(5), programs: A comparison of matching and instrumental-variable
772794. estimators. The Review of International Organizations, 1(2), 99124.
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES
Bal-Gunduz, Y., & Crystallin, M. (2014). Do IMF-supported programs
catalyze donor assistance to low-income countries?, IMF Working
Paper No. 14/202. Washington: International Monetary Fund.
Barro, R., & Lee, J.-W. (2005). IMF programs: Who is chosen and what
are the eects?. Journal of Monetary Economics, 52(7), 12451269.
Bas, M. A., & Stone, R. W. (2014). Adverse selection and growth under
IMF programs. The Review of International Organizations, 9(1), 128.
Becker, T., & Mauro, P. (2007). Output drops and the shocks that matter,
IMF working paper 07/172. Washington, DC: International Monetary
Fund.
Bird, G. (2007). The IMF: A birds eye view of its role and operations.
Journal of Economic Surveys, 21(4), 683745.
Bird, G., Hussain, M., & Joyce, J. P. (2004). Many happy returns?
Recidivism and the IMF. Journal of International Money and Finance,
23(2), 231251.
Bird, G., & Rowlands, D. (2001). IMF lending: How is it aected by
economic, political and institutional factors?. Journal of Policy Reform,
4, 243270.
Bird, G., & Mosley, P. (2006). In G. Ranis, J. R. Vreeland, & S. Kosack
(Eds.), Globalization and the nation state: Should the IMF discontinue
its long-term lending role in developing countries (pp. 231262). New
York: Routledge.
Bird, G., & Rowlands, D. (2007). Should it be curtains for some of the
IMFs lending windows?. The Review of International Organizations, 2,
281299.
Bird, G., & Rowlands, D. (2009). A disaggregated empirical analysis of
the determinants of IMF arrangements: Does one model t all?.
Journal of International Development, 21, 915931.
Bird, G., & Rowlands, D. (2006). In G. Ranis, J. R. Vreeland, & S.
Kosack (Eds.), Globalization and the nation state: The impact of the
IMF and the World Bank (pp. 231262). New York: Routledge.
Bordo, M. D., & Schwartz, A. J. (2000). Measuring real economic eects
of bailouts: Historical perspectives on how countries in nancial
distress have fared with and without bailouts. Carnegie-Rochester
Conference Series on Public Policy, 53(1), 81167.
Bredenkamp, H., & Schadler, S. (Eds.) (1999). Economic adjustment and
reform in low-income countries: Studies by the sta of the International
Monetary Fund. Washington: International Monetary Fund.
Bruckner, M. (2013). On the simultaneity problem in the aid and growth
debate. Journal of Applied Econometrics, 28(1), 126150.
Bruckner, M., & Ciccone, A. (2010). International commodity prices,
growth, and the outbreak of civil war in Sub-Saharan Africa. The
Economic Journal, 120(544), 535550.
Butkiewicz, J. L., & Yanikkaya, H. (2005). The eects of IMF and World
Bank lending on long-run economic growth: An empirical analysis.
World Development, 33(3), 371391.
Cerulli, G. (2015). Econometric evaluation of socio-economic programs:
Theory and applications. New York: Springer.
Cerutti, E. (2007). IMF drawing programs: Participation determinants
and forecasting, IMF working paper no. 07/152. Washington: Inter-
national Monetary Fund.
Chamberlain, G. (1982). Multivariate regression models for panel data.
Journal of Econometrics, 18(1), 546.
Clemens, M. A., Radelet, S., Bhavnani, R. R., & Bazzi, S. (2012).
Counting chickens when they hatch: Timing and the eects of aid on
growth. The Economic Journal, 122(561), 590617.
Clements, B., Gupta, S., & Nozaki, M. (2012). What happens to social
spending in IMF-supported programs?. Applied Economics, 45(28),
40224033.
Conway, P. (1994). IMF lending programs: Participation and impact.
Journal of Development Economics, 45(2), 365391.
Conway, P. (2007). The revolving door: Duration and recidivism in IMF
programs. Review of Economics and Statistics, 89(2), 205220.
Dabla-Norris, E., & Bal Gunduz, Y. (2014). Exogenous shocks and
growth crises in low-income countries: A vulnerability index. World
Development, 59, 360378.
Demirguc-Kunt, A., & Detragiache, E. (1999). Monitoring banking sector
fragility: A multivariate logit approach, IMF working paper 99/147.
Washington: International Monetary Fund.
Dicks-Mireaux, L., Mecagni, M., & Schadler, S. (2000). Evaluating the
eect of IMF lending to low-income countries. Journal of Development
Economics, 61, 495526.
Dreher, A. (2006). IMF and economic growth: The eects of programs,
loans, and compliance with conditionality. World Development, 34(5),
769788.
45
Dreher, A., & Jensen, N. M. (2007). Independent actor or agent? An
empirical analysis of the impact of U.S. interests on IMF conditions.
Journal of Law and Economics, 50, 105124.
Dreher, A., Sturm, J.-E., & Vreeland, J. R. (2006). Does membership on
the UN Security Council inuence IMF decisions? Evidence from
panel data, Swiss Institute for Business Cycle Research no. 151.
Zurich: Institute for Business Cycle Research.
Dreher, A., Sturm, J. E., & Vreeland, J. R. (2009a). Global horse trading:
IMF loans for votes in the United Nations Security Council. European
Economic Review, 53, 742757.
Dreher, A., Sturm, J.-E., & Vreeland, J. R. (2009b). Development aid and
international politics: Does membership on the UN Security Council
inuence World Bank decisions?. Journal of Development Economics,
88, 118.
Dreher, A., & Vaubel, R. (2004). Do IMF and IBRD cause moral hazard
and political business cycles? Evidence from panel data. Open
Economies Review, 15, 522.
Easterly, W. (2005). What did structural adjustment adjust? The associ-
ation of policies and growth with repeated IMF and World Bank
adjustment loans. Journal of Developments Economics, 76(1), 122.
Eichengreen, B., Gupta, P., & Mody, A. (2008). Sudden stops and IMF-
supported programs. In Financial markets volatility and performance in
emerging markets (pp. 219266). University of Chicago Press.
Elekdag, Selim. (2008). How does the global economic environment
inuence the demand for IMF resources?. IMF Sta Papers, 55(4),
624653.
Evrensel, A. (2002). Eectiveness of IMF-supported stabilization pro-
grams in developing countries. Journal of International Money and
Finance, 21(5), 565587.
Fischer, S. (1993). The role of macroeconomic factors in growth, NBER
working paper no. 4565. Cambridge, Massachusetts: National Bureau
of Economic Research.
Galiani, S., Knack, S., Xu, L. C., & Zou, B. (2014). The eect of aid on
growth: Evidence from aquasi-experiment, World Bank policy research
working paper (6865). .
Garuda, G. (2000). The distributional eects of IMF programs: A cross-
country analysis. World Development, 28(6), 10311051.
Ghosh, A., Goretti, M., Joshi, B., Thomas, A., & Zalduendo, J. (2007).
Modeling aggregate use of fund resourcesAnalytical approaches and
medium-term projections, IMF working paper 07/70. Washington:
International Monetary Fund.
Hardoy, I. (2003). Eect of IMF programmes on growth: A reappraisal
using the method of matching. Institute for Social Research Paper,
2003, 040.
Hnatkovska, V., & Loayza, N. (2005). Volatility and growth in managing
economic volatility and crises: A practitioners guide. In J. Aizenman,
& B. Pinto (Eds.), Cambridge. MA: Cambridge University Press.
Hutchison, M. M. (2003). A cure worse that the disease? Currency crises
and the output costs of IMF-supported stabilization programs. In M.
P. Dooley, & F. A. Jerey (Eds.), Managing currency crises in emerging
markets (pp. 321359). Chicago: University of Chicago Press.
Hutchison M. M., 2004. Selection Bias and the Output Costs of IMF
Programs, EPRU Working Paper Series No. 2004-15, University of
Copenhagen.
Independent Evaluation Oce (2002). Evaluation of prolonged use of
IMF resources. Washington: International Monetary Fund.
Independent Evaluation Oce (2004). Evaluation of the IMFs role in
Poverty Reduction Strategy Papers and Poverty Reduction and
Growth Facility. Washington: International Monetary Fund.
International Monetary Fund (2010). Emerging from the global crisis:
macroeconomic challenges facing low-income countries. (Washing-
ton).
International Monetary Fund (2012). 2011 review of conditionality
background paper: Outcomes of imf-supported programs. Washing-
ton: International Monetary Fund.
International Monetary Fund (2014). Macroeconomic developments in
low-income developing countries. (Washington).
Ivanova, A., Mayer, W., Mourmouras, A., & Anayiotos, G. (2003). What
determines the implementation of IMF-supported programs?, IMF
working paper no. 03/8, Published. In A. Mody, & A. Rebucci (Eds.),
IMF-supported programs: Recent sta research 2006. Washington:
International Monetary Fund.
Jaramillo, L., & Sancak, C. (2009). Why has the grass been greener on one
side of Hispaniola? A comparative growth analysis of the Dominican
46 WORLD DEVELOPMENT
Republic and Haiti. IMF Sta Papers, 56, 323349 (Washington:
International Monetary Fund).
Joyce, J. P. (1992). The economic characteristics of IMF program
countries. Economic Letters, 38(2), 237242.
Joyce, J. P. (2005). Time present and time past: A duration analysis of
IMF program spells. Review of International Economics, 13, 283297.
Knight, M., & Santaella, J. A. (1997). Economic determinants of IMF
nancial arrangements. Journal of Development Economics, 54(2),
405436.
Kuziemko, I., & Werker, E. (2006). How much is a seat on the Security
Council worth? Foreign aid and bribery at the United Nations. Journal
of Political Economy, 114, 905930.
Lee, J. W., & Rhee, C. (2002). Macroeconomic impacts of the Korean
nancial crisis: Comparison with the cross-country patterns. The
World Economy, 25(4), 539562.
Marchesi, S., & Sirtori, E. (2011). Is two better than one? The eects of
IMF and World Bank interaction on growth. The Review of Interna-
tional Organizations, 6(3), 287306.
Mecagni, M. (1999). The causes of program interruptions. In H.
Bredenkamp, & S. Schadler (Eds.), Economic adjustment and reform
in low-income countries, studies by the sta of the International
Monetary Fund (pp. 215276). Washington: International Monetary
Fund.
Mercer-Blackman, V., & Unigovskaya, A. (2000). Compliance with IMF
program indicators and growth in transition economies, IMF working
paper no, 00/47. Washington: International Monetary Fund.
Milesi-Ferretti, G. M., & Razin, A. (1996). Current account sustainability:
Selected East Asian and Latin American experiences, NBER working
paper no. 5791. Cambridge, Massachusetts: National Bureau of
Economic Research.
Moser, C., & Sturm, J.-E. (2011). Explaining IMF lending decisions after
the cold war. Review of International Organizations, 6(3), 307340.
Mundlak, Y. (1978). On the pooling of time series and cross section data.
Econometrica, 46(1), 6985.
Nooruddin, I., & Simmons, J. W. (2006). The politics of hard choices:
IMF programs and government spending. International Organization,
60(4), 10011033.
Oatley, T., & Yackee, J. (2004). American interests and IMF lending.
International Politics, 41, 415429.
Oberdabernig, D. A. (2013). Revisiting the eects of IMF programs on
poverty and inequality. World Development, 46, 113142.
Papageorgiou, C., Pattillo, C., Spatafora, N., & Berg, A. (2010). The end
of an era? The medium- and long-term eects of the global crisis on
growth in low-income countries, IMF working papers 10/205. Wash-
ington, DC: International Monetary Fund.
APPENDIX A
Data
Variable
Paris Club debt rescheduling and cancellations
Principal arrears on long-term debt outstanding
Interest arrears long-term debt outstanding
Long-term public and publicly guaranteed debt
Total debt service paid
Overdue obligations to the IMF principal
Overdue obligations to the IMF interest and charges
Gross domestic product, current and constant prices
Current account balance
GDP per capita (constant 2000 US$)
Non-oil commodity exports
Population
Exchange rate, national currency per US$
Exports of goods and services
Export of goods (volume)
Imports of goods and services
Value of oil imports
General government balance
Presbitero, A. F., & Zazzaro, A. (2012). IMF lending in times of crisis:
Political inuences and crisis prevention. World Development, 40(10),
19441969.
Perry, G. (2009). Beyond lending: How multilateral banks can help
developing countries manage volatility. Washington, DC: Center for
Global Development.
Przeworski, A., & Vreeland, J. R. (2000). The eect of IMF programs on
economic growth. Journal of Development Economics, 62(2), 385421.
Reinhart, C. M., & Rogo, K. (2004). The modern history of exchange
rate arrangements: A reinterpretation. The Quarterly Journal of
Economics, CXIX(1), 148.
Robins, J. M. (2002). Comment on Covariance adjustment in random-
ized experiments and observational studies. Statistical Science, 17(3),
309321.
Rosenbaum, P. R. (2002). Observational studies. Springer series in
statistics. (New York).
Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the
propensity score in observational studies for causal eects. Biometrika,
70(1), 4155.
Santaella, J. A. (1996). Stylized facts before IMF-supported macroeconomic
adjustment, IMF sta papers no. 43 (pp. 502544). Washington:
International Monetary Fund.
Steinwand, M., & Stone, R. (2008). The International Monetary Fund: A
review of the recent evidence. Review of International Organizations, 3
(2), 123149.
Stone, R. W. (2002). Lending credibility: The International Monetary Fund
and the post-communist transition. Princeton: Princeton University
Press.
Stone, R. W. (2004). The political economy of IMF lending in Africa.
American Political Science Review, 98, 577592.
Sturm, J.-E., Berger, H., & de Haan, J. (2005). Which variables explain
decisions on IMF credit? An extreme bounds analysis. Economics &
Politics, 17(7), 177213.
Thorsten, B., Clarke, G., Gro, A., Keefer, P., & Walsh, P. (2007). New
tools and new tests in comparative political economy: The database of
political institutions, The Development Research Group of the World
Bank. Washington: World Bank.
Ul Haque, N., & Khan, M. S. (1998). Do IMF-supported programs work?
A survey of the cross-country empirical evidence, IMF working paper
no. 98/169. Washington: International Monetary Fund.
Vreeland, J. R. (2003). The IMF and economic development. Cambridge:
Cambridge University Press.
Source
Paris Club webpage
WB Global Development Finance (GDF)
WB GDF
WB GDF
WB GDF
IMF Fund Accounts database
IMF Fund Accounts database
IMF World Economic Outlook (WEO) database
IMF WEO
WB World Development Indicators (WDI) and IMF WEO
WB WDI
WB WDI
IMF WEO
IMF WEO
IMF WEO
IMF WEO
IMF WEO
IMF WEO
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES 47

CPI ination IMF WEO

Terms of trade, goods IMF WEO
Gross international reserves at year end IMF WEO
Black market premium Reinhart and Rogo (2004) dataset
FDI IMF WEO and WB
Portfolio investment net WB GDF and IMF
Net transfers on debt WB GDF
Short-term debt outstanding WB GDF
Prot remittances on FDI WB GDF
Grants, excluding technical cooperation WB GDF
Real eective exchange rate IMF Information Notice System
Real oil prices (ASPS) Oil prices deated by US CPI (WEO)
Real non-oil commodity prices Non-oil commodity prices deated by US CPI (WEO)
World trade (index number 2000=100) IMF WEO
World GDP (index number 2000=100) IMF WEO
US interest rate IMF WEO
SBA, CFF, PRGF/SAF/ESAF arrangements, EPCA, SMP IMF databases: Fund Accounts, International Financial
Statistics (IFS) and the Fund Arrangements database of the
IMF Finance Department
Program interruptions Ivanova et al. (2003) dataset, Mecagni (1999, chap. 9),
various country reports and IMF MONA database
Elections WB the Database on Political Institutions (DPI, Thorsten,
Clarke, Gro, Keefer, & Walsh, 2007)
Population aected by natural disasters Centre for Research on the Epidemiology of Disasters
Database
Dummy variable for oil exporting LICs IMF WEO

Table A1. Country list: estimation sample

1 Albania 29 Lao PDR

2 Armeni 30 Madagascar
3 Azerbaijan 31 Malawi
4 Bangladesh 32 Mali
5 Benin 33 Mauritania
6 Bolivia 34 Moldova
7 Burkina Faso 35 Mongolia
8 Burundi 36 Mozambique
9 Cambodia 37 Nepal
10 Cameroon 38 Nicaragua
11 Central African Republic 39 Niger
12 Cha 40 Nigeria
13 Comoros 41 Pakistan
14 Republic of 42 Papua New Guinea
15 Cote dlvoire 43 Rwanda
16 Democratic Republic of Congo 44 Senegal
17 Ethiopia 45 Sierra Leone
18 The 46 Sri Lanka
19 Georgia 47 Sudan
20 Ghan 48 Tajikistan
21 Guinea 49 Tanzania
22 Guinea 50 Tog
23 Guyana 51 Uganda
24 Haiti 52 Uzbekistan
25 Honduras 53 Vietnam
26 Indi 54 Zambia
27 Kenya 55 Zimbabwe
28 Kyrgyz Republic
48 WORLD DEVELOPMENT

Table 13. Impact of short-term IMF engagement with no adjustment for program implementation
All LICs LICs with weaker fundamentals
PS > 0.5 PS > 0.7 PS > 0.8
Macroeconomic outcomes
Real GDP growth (%) dgdpog 0.323 0.854* 1.163** 1.495**
(0.437) (0.505) (0.577) (0.663)
Ination (%) in100og
18.16***
22.32***
25.67***
30.28***
(3.949) (4.935) (5.601) (6.259)
Reserve coverage (in months of imports) impcovog 0.454*** 0.396*** 0.349** 0.242*
(0.125) (0.130) (0.138) (0.144)
Current account balance plus FDI (% of GDP) cabfdiy 1.988** 2.862*** 4.500*** 5.035***
(0.824) (0.985) (1.110) (1.220)
Government balance (% of GDP) gbaltogdp100og 2.170*** 2.479*** 3.011*** 3.450***
(0.556) (0.674) (0.771) (0.849)
Change in real health spending per capita (%) heal_dlnrpop 5.016 3.980 3.573 0.697
(3.794) (4.820) (5.659) (7.634)
Change in real education spending per capita (%) educ_dlnrpop
1.578
3.547
4.962
8.073
(3.333) (4.165) (5.368) (6.411)
Change in REER (%) dreerog
5.793
5.940
6.542
6.823
(4.726) (5.401) (6.148) (7.250)
Change in macroeconomic outcomes (X(t)
X(t
1))
Real GDP growth (%) ddgdpog 0.350 0.859 1.449 2.131*
(0.682) (0.806) (0.933) (1.093)
Ination (%) dinf
6.618*
10.15**
13.62***
14.69**
(3.636) (4.416) (5.044) (5.908)
Reserve coverage (in months of imports) drescv 0.534*** 0.603*** 0.618*** 0.534***
(0.0918) (0.0971) (0.104) (0.110)
Current account balance plus FDI (% of GDP) dcabfy 1.151* 1.399* 1.961** 2.635***
(0.675) (0.812) (0.914) (1.017)
Government balance (% of GDP) dgovbaly 0.400 0.514 0.876 0.939
(0.410) (0.488) (0.554) (0.630)
Number of observationsa 749 468 376 320
Source: Authors calculations.
Note: PS stands for the propensity score indicating the likelihood of IMF programs addressing immediate balance of payments needs. Changes in
macroeconomic outcomes refer to rst dierences of the variables in the top panel. Signicant at 10%:*; 5%:**; and 1%:***. Standard errors in parenthesis.
FDI = Foreign Direct Investment.
a
All variables except for health and education spending and change in Real Eective Exchange Rate (REER). The sample size is 281 for health and
education spending and 572 for REER.
 THE ECONOMIC IMPACT OF SHORT-TERM IMF ENGAGEMENT IN LOW-INCOME COUNTRIES 49

Table 14. Impact of short-term fund engagement (198099)

All LICs LICs with weaker fundamentals
PS > 0.5 PS > 0.7 PS > 0.8
Macroeconomic outcomes
Real GDP growth (%) dgdpog 1.096* 1.386** 1.681** 2.032**
(0.643) (0.706) (0.788) (0.926)
Ination (%) in100og
20.47***
23.37***
24.38***
30.93***
(6.596) (7.526) (8.428) (9.810)
Reserve coverage (in months of imports) impcovog 0.399** 0.450*** 0.508*** 0.383**
(0.165) (0.166) (0.167) (0.184)
Current account balance plus FDI (% of GDP) cabfdiy 3.024*** 3.391*** 4.994*** 4.892***
(1.150) (1.230) (1.316) (1.458)
Government balance (% of GDP) gbaltogdp100og 3.040*** 3.387*** 3.979*** 4.550***
(0.803) (0.906) (1.002) (1.114)
Change in real health spending per capita (%) heal_dlnrpop 9.630 8.490 7.102 5.073
(6.287) (6.861) (7.446) (10.78)
Change in real education spending per capita (%) educ_dlnrpop 1.621 0.404
0.195
6.214
(4.167) (4.843) (5.355) (6.066)
Change in REER (%) dreerog
0.821
0.332
0.171 0.395
(7.319) (8.324) (9.353) (11.29)
Change in macroeconomic outcomes (X(t)
X(t
1))
Real GDP growth (%) ddgdpog 0.940 1.364 2.127* 2.654*
(0.985) (1.076) (1.207) (1.407)
Ination (%) dinf
9.974
13.78*
16.98**
20.42**
(6.975) (7.368) (8.253) (9.720)
Reserve coverage (in months of imports) drescv 0.552*** 0.637*** 0.689*** 0.617***
(0.118) (0.123) (0.133) (0.146)
Current account balance plus FDI (% of GDP) dcabfy 2.196** 2.155** 2.869*** 2.969**
(0.915) (1.007) (1.053) (1.212)
Government balance (% of GDP) dgovbaly 1.124** 1.410** 2.047*** 2.106***
(0.552) (0.616) (0.640) (0.722)
Number of observationsa 438 313 258 222
Source: Authors calculations.
Note: PS stands for the propensity score indicating the likelihood of IMF programs addressing immediate balance of payments needs. Changes in
macroeconomic outcomes refer to rst dierences of the variables in the top panel. Signicant at 10%:*; 5%:**; and 1%:***. Standard errors in parentheses.
FDI = Foreign Direct Investment.
a
All variables except for health and education spending and change in Real Eective Exchange Rate (REER). The sample size is 133 for health and
education spending and 369 for REER.

Available online at www.sciencedirect.com

ScienceDirect