A disaggregated empirical analysis of the determinants of IMF arrangements: Does one model fit all?

Journal of International Development

J. Int. Dev. 21, 915–931 (2009)

Published online 4 November 2008 in Wiley InterScience

(www.interscience.wiley.com) DOI: 10.1002/jid.1520

A DISAGGREGATED EMPIRICALANALYSIS OF THE DETERMINANTS

OF IMF ARRANGEMENTS:DOES ONE MODEL FIT ALL?

GRAHAM BIRD1* and DANE ROWLANDS2

1Surrey Centre for International Economic Studies, University of Surrey, Guildford, Surrey, UK2Norman Paterson School of International Affairs, Carleton University, Ottawa, Ontario, Canada

Abstract: Does one model fit all when it comes to the determinants of IMF programs?

Certainly claims have been made by the IMF that capital account crisis (CAC) countries are

discernibly different in terms of the characteristics that lead them to borrow from it, while

other research has claimed that it is only Asian economies that are different from the rest. This

paper sets out to examine these issues. It tests various forms of a fairly conventional model to

see whether some forms better fit certain groups of countries than others. It then uses the

favoured models to estimate the probability of countries having an IMF arrangement. In

particular it examines countries that have been identified by the Fund as CAC countries, but it

also looks at a number of comparator countries. The findings suggest that there are some

differences between low income and middle income countries. Pressures in the foreign

exchange market are significant for the latter but not for the former. The paper also discusses

differences between regions and within regions. Broadly speaking the findings confirm that

Asian economies around the time of the 1997/1998 crisis tended to turn to the IMF for financial

support more quickly than would have been anticipated on the basis of the existing best-fitting

models. The paper also discusses the implications of the findings for policy and for the reform

of the IMF. Copyright # 2008 John Wiley & Sons, Ltd.

Keywords: IMF arrangements

JEL: F33, F3

1 INTRODUCTION

In seeking to explain the probability of countries negotiating arrangements with the

International Monetary Fund (IMF) the claim has sometimes been made that capital

*Correspondence to: Graham Bird, Surrey Centre for International Economic Studies, University of Surrey,Guildford, Surrey, UK. E-mail: [email protected]

Copyright # 2008 John Wiley & Sons, Ltd.

916 G. Bird and D. Rowlands

account crisis (CAC) countries are different. The implicit assumption is that, whereas

countries have traditionally chosen to turn to the IMF when they encounter current account

deficits associated with fiscal and monetary excesses or external terms of trade shocks,

CAC countries have been pushed towards the Fund by a loss of market confidence which

strategically weakens the capital account. The implication is that these arrangements

would not have been predicted on the basis of fiscal imbalances or current account

imbalances, or more broadly on the basis of conventional models of the determinants of

IMF lending.

In recent reports, the IMF has itself sought to distinguish between countries in terms of

the economic circumstances in which they turn to it for assistance (IMF, 2004). The Fund

classifies its clients as ‘classic’, ‘capital account crisis’ and ‘low income’ countries. The

inference is that there are discernible differences between each of these categories. The

‘classic’ country borrower is one that encounters current account balance of payments

difficulties largely as a consequence of fiscal deficits that are either financed by rapid

monetary expansion leading to inflation, or excessive debt accumulation. These countries

tend to have ‘twin’ fiscal and current account deficits. ‘Low income’ countries also tend to

borrow from the IMF as a result of current account deficits. But although macroeconomic

disequilibria may have a part to play in explaining IMF arrangements in these cases, they

may also be significantly associated with adverse trade shocks.

In contrast, it has been suggested that CAC countries may have only relatively modest

fiscal imbalances and current account deficits. What drives them to the Fund is a relatively

sudden loss of market confidence and the resulting pressure on their currency in foreign

exchange markets. Although this may be reflected by a precipitous fall in reserves, and

while low holdings of reserves is a conventional reason for countries to turn to the IMF, the

suggestion is that a conventional model of IMF arrangements which focuses on fiscal

deficits, monetary expansion and current account deficits, alongside reserve levels, will

perform less well in predicting IMF arrangements in CAC countries. Some empirical

studies have suggested that in attempting to explain IMF programs during the 1990s, an

‘Asian dummy’ is significant and is needed to capture the experience of these countries

(Anderson et al., 2006). The implication is that while Asia is different, CAC countries in

general are not.

The purpose of this paper is to examine these issues empirically. Does one model of IMF

lending fit all countries? Are some determinants significant in some cases but not in others?

To what extent can the programs negotiated during the 1990s by countries identified by the

IMF as CAC countries be explained by one specific model, and in what ways does this

model differ from the one that might be used to explain drawings by low income countries?

The paper is organised as follows: Section 2 briefly summarises the existing literature on

the determinants of IMF arrangements. Section 3 goes on to present econometric models of

IMF programs which focus on a number of key economic characteristics intended to reflect

domestic economic imbalances as well as external factors. This exercise is designed to

explore whether the significance of certain determinants varies across countries. In

particular, a distinction between low income and middle income countries is investigated.

The models are then applied to calculate the probability that countries defined by the IMF

as CAC countries would have had IMF programs at the times that they did. As a point of

comparison, the section also explores the estimated probability of countries having

programs with the Fund that did not in fact draw from it. In particular while, in the context

of the Asian crisis in 1997/1998, Thailand, Korea and Indonesia negotiated programs with

the IMF, neither Malaysia nor China had a program. Can these differences be explained in

Copyright # 2008 John Wiley & Sons, Ltd. J. Int. Dev. 21, 915–931 (2009)

DOI: 10.1002/jid

Determinants of IMF Arrangements 917

terms of the economic characteristics conventionally seen as determining the incidence of

IMF programs? Section 4 briefly offers some concluding remarks and discusses the policy

implications of the findings.

2 THE DETERMINANTS OF IMF ARRANGEMENTS: A BRIEF REVIEW

OF THE LITERATURE

Empirical evidence does not allow factors affecting the demand for IMF arrangements to

be distinguished from those affecting the supply of them. What is observed is either a

program or no program, with these outcomes being the result of the interplay between

demand and supply side factors.

Having said this, the IMF can only formally respond to requests for assistance from

member countries; it cannot formally solicit applications. The existence of a program must

therefore mean that the government of the country concerned has requested assistance (the

demand side) and that the IMF has responded positively to the request (the supply side) and

has been facilitating. The complex configuration of potential economic and political

circumstances that results in IMF programs is a legitimate research topic in its own right,

and has been reviewed elsewhere by us (see, e.g. Bird and Rowlands, 2006). In this paper,

we merely seek to investigate whether the ‘conventional’ demand side determinants of IMF

programs (as revealed by the extant literature) apply consistently across and within sub-

groups of countries. Although this is a modest objective, it represents value added to the

existing literature.

Early research into IMF arrangements attempted to explain the amount of lending in

terms of key economic variables, making little distinction between the demand and supply

sides (Bird and Orme, 1981; Cornelius, 1987). Later research, while continuing to

emphasise the economic determinants, used probability approaches that sought to explain

whether or not countries had programs rather than to explain the amount of lending

involved in them (Joyce, 1992; Conway, 1994; Knight and Santaella, 1997). As studies

multiplied and became more sophisticated, they also encompassed a wider range of

potential explanatory variables, although these continued to emphasise the economic

dimensions of IMF borrowing and lending.

Over time, areas of consensus emerged. There were some economic variables that

appeared to be significant according to most studies (Bird, 1996). The mere existence of a

current account balance of payments deficit certainly did not appear, in itself, to make it

probable that a country would demand resources from the Fund. Beyond this, however,

arrangements with the Fund did seem to be linked to low levels of reserves, overvalued

exchange rates, a near-term record of past programs and low levels of income and

development. Some studies further suggested that programs were connected to external

debt and terms of trade shocks.

While previous studies have certainly identified important relationships and regularities,

the within sample and out-of-sample predictive capacity of the models has been limited.

More recent research has attempted to contribute to the literature by exploring the extent to

which our ability to explain IMF programs is improved by incorporating a series of

political variables that may influence the willingness of governments to turn to the IMF for

financial assistance and the willingness of the Fund to respond positively (Rowlands, 1995;

Thacker, 1999; Bird and Rowlands, 2001; Anderson et al., 2006). It is probably premature

to reach firm conclusions about the influence of political factors. While some studies find


DOI: 10.1002/jid


that there is a systematic ‘U.S. influence’ (Thacker, 1999; Anderson et al., 2006) others

conclude that, while individual political factors appear to be significant on occasions, the

inclusion of political variables—and institutional variables as well—does not significantly

improve our overall systematic ability to explain IMF programs (Bird and Rowlands, 2001,

2002).

Although the research summarised above has evolved to include a wide array of

potential determinants of IMF programs (both economic and political), it has tended not to

differentiate between sub-groups of countries and has instead included all countries

together in the samples used. It has generally not examined the extent to which programs in

different countries are affected by different factors. An exception may be found in Bird

et al. (2004) who investigated whether countries making prolonged use of IMF resources

exhibit particular and defining characteristics. But in general there is a gap between the

claims made by the IMF that low income and CAC countries are different from classic case

countries, and the available empirical evidence. It is this gap that we attempt to narrow in

this paper by empirically examining whether there are important differences to be found

amongst the countries that use IMF resources, as well as between those that do and do not

use IMF resources.

Again, however, in what follows we do not explore whether countries in the midst of a

CAC are in some way politically more inclined to approach the Fund for assistance, or

whether the Fund is more inclined (once more perhaps for political reasons) to offer such

assistance when asked for it. Our purpose is simply to test whether fairly conventional

models of IMF arrangements perform broadly equivalently across and within sub-groups of

countries.

3 METHODOLOGY, RESULTS AND INTERPRETATION

3.1 Methods and Results

In order to examine the probability of signing an agreement with the Fund, we use a probit

model in which the binary dependent variable indicates the signing of an IMF agreement

(under the Standby, EFF, SAF, ESAF or PRGF facilities) and the explanatory variables

reflect a country’s general aggregate economic performance, basic external balance

conditions and recent IMF activity. The explanatory variables are all lagged 1 year from the

signing, so that the model effectively generates within-sample predictions of a country’s

probability of signing an agreement in the following year. This lag appears appropriate

given the time taken to negotiate agreements. It is the predicted probabilities of signing that

are of interest to us here, as they indicate the degree to which an IMF program is expected

based on the economic circumstances of a country, as captured by the econometric model

summarised in Table 1 (and explained more fully in the Appendix).

The basic model used here has been derived from earlier research (summarised in Bird

and Rowlands, 2001) which also tested for a range of additional determinants. It therefore

represents a preferred parsimonious model. Thus, for example, the exclusion of terms of

trade and monetary variables reflects the fact that their inclusion did not improve the

explanatory power of the model. At the same time, the model does accommodate various

real and monetary factors that may lead to a loss of balance of payments sustainability that

then motivates countries to seek financial assistance from the IMF. The model incorporates

measures of indebtedness, debt difficulties and reserve adequacy as well as the size of


DOI: 10.1002/jid

Table

1.

Probitresultsexplainingthesigningofan

IMFagreem

entin

thefollowingyear1

Explanatory

variable

FullSam

ple

Low

Income

Middle

Income

Middle

Income

(excludingMPI)

Middle

Income(excluding

effectsofprevious

IMFprograms)

Constant

�1.12��(�

6.19)

�0.886�(�

2.14)

�1.27��(�

5.50)

�1.18��(�

6.35)

�0.843��(�

3.15)

GNPper

capita

0.0000256(0.76)

0.0000127(0.06)

0.0000468(1.15)

�0.0000121(�

0.37)

0.0000527(0.93)

GDPgrowth

�0.0357��(�

2.96)

�0.00528(�

0.31)

�0.0545��(�

3.28)

�0.0381��(�

3.24)

�0.0563��(�

3.69)

Reserve-to-importratio

�0.281(�

1.22)

�1.01(�

1.77)

�0.171(�

0.67)

�0.527(�

1.74)

�0.260(�

0.86)

%changein

reserves/imports

0.0000276(0.05)

0.0005(0.74)

�0.000936(�

0.71)

�0.0034�(�

1.97)

�0.000844(�

1.1)

Currentaccountbalance/GDP

0.424(0.53)

0.346(0.36)

�0.615(�

0.51)

�0.307(�

0.36)

0.477(�

0.38)

%changein

thecurrentaccount

0.0000384(0.79)

�0.000182(�

1.42)

0.000114(1.16)

0.000108(1.51)

0.000106(1.13)

Realexchangerate

depreciation

�0.000177(�

0.83)

0.00776(1.89)

�0.00498(�

0.88)

�0.000122(�

0.60)

�0.00904(�

1.57)

Debtservice-to-exportsratio

0.493(1.64)

0.752(1.14)

0.353(1.12)

0.432(1.38)

0.624(1.73)

%changein

debt–serviceratio

0.000378(0.7)

�0.000559(�

0.63)

0.000656(0.82)

�0.0001(�

0.13)

0.000916(1.19)

Publicexternal

debt/GDPratio

0.174(1.64)

0.155(0.85)

0.435�(2.26)

0.0425(0.28)

0.784��(3.23)

Currentrescheduling

�0.284(�

1.32)

�0.779�(�

2.15)

0.0496(0.19)

0.0502(021)

0.209(0.93)

Reschedulingsin

pastyears

0.119(1.02)

0.114(0.57)

0.0437(0.27)

0.0923(0.62)

0.204(1.30)

Inflation

0.000131(1.37)

0.000421��

(2.44)

0.0000745(0.74)

0.000105(1.07)

0.00000148(0.01)

Fixed

exchangerate

regim

e0.136(1.1)

0.276(1.47)

0.0312(0.18)

0.205(1.44)

0.0757(0.33)

Flexible

exchangerate

regim

e0.0134(0.09)

�0.244(�

0.78)

0.156(0.85)

0.430��(2.95)

0.238(1.20)

Market

pressure

index

0.0083��

(2.63)

0.00337(0.77)

0.00974�(2.09)

—0.0107��

(2.51)

CurrentIM

Fprogram

0.713��(4.46)

0.506�(2.04)

0.727��(3.39)

0.817��(4.47)

—

IMFprogram

inprioryear

0.423��(2.82)

0.437(1.95)

0.446�(2.27)

0.446��(3.08)

—

Number

ofobservations

922

291

631

1006

1006

Pseudo-R

20.176

0.135

0.233

0.232

0.112

Loglikelihood

�469.8

�166.9

�285.8

�409.9

�473.6

�� ,

��,�referto

statisticalsignificance

atthe1,2and5per

centlevelsfortwo-tailedtests,respectively.

1Theestimationwas

runonStata

usingtherobustprobitestimationprocedure

withcountriesidentified

asthecluster.


DOI: 10.1002/jid



current account imbalances. It also takes into account the nature of the exchange rate

regime and the extent of exchange rate adjustment.

We also examine the importance of pressure in the foreign exchange market and

previous engagement with the IMF. It is expected that market pressure, measured here

conventionally as a weighted average of interest rate, international reserves and exchange

rate movements will be more important in explaining the presence of IMF programs in

middle income countries, since they are far more integrated into the global economy than

are low income countries, particularly perhaps in terms of their access to international

capital markets.1 The presence of previous IMF agreements (either 1 or 2 years before the

recorded signing) may be expected to explain future agreements in both sets of countries,

and has been traditionally found to be perhaps the best single predictor of the pattern of

Fund arrangements.

While the basic model has shortcomings in terms of predictive power, as is the case with

all such models (Bird and Rowlands, 2001, 2002), it is found to perform reasonably well.

Although we collected a large amount of data for several countries over many years, some

observations, including those prior to 1977, are suppressed due to missing data. In addition,

countries that were not clearly eligible to sign agreements (since they were already

operating under an IMF program) were also removed from the sample. Consequently, the

full-sample model is estimated using an unbalanced panel of 922 observations covering

88 low and middle income countries from 1977 to 2000. Table 1 presents the estimation

results. For the full sample, these appear in the first column of the table. The subsequent

two columns re-estimate the same equation for low income and middle income countries,

respectively. The final two columns repeat the middle income estimation first without the

market pressure index (column 4) and then without the indicator of previous IMF programs

(column 5).

The results are generally sensible, and align with the findings of other studies. The full

sample is dominated by the middle income group of countries. Focusing on the estimated

coefficients that are statistically significant, the results indicate that countries are more

likely to sign an agreement with the IMF in the following year if they have weak economic

growth and, for middle income countries, if they have relatively high debt-to-GDP ratios.

The market pressure index is also statistically significant. When it is removed (in column 4)

from the estimating equation, its absence is partially offset by the observed effects on the

significance of reserves and exchange rate arrangements. Finally, previous IMF agreements

are strongly linked to the presence of subsequent ones in the near term.2 When these

agreement indicators are removed, the estimated coefficient for external indebtedness

becomes statistically significant.

By way of contrast, the estimation of future program signing for low income countries

has weaker overall explanatory power. IMF programs in this sub-group of countries are

significantly associated with the presence of previous agreements or high inflation, and

negatively related to the rescheduling of debt in the current year and therefore the presumed

absence of such requirements in the immediate future. The results reported in Table 1

therefore indicate that there are important differences as between middle and low income

1We would like to thank Alex Mandilaras for allowing us to use his data set on the market pressure index.2The greater statistical significance of the indicator of previous IMF agreements is perhaps unexpected, as lowincome countries have a reputation for being more dependent on consecutive IMF programs than their middleincome counterparts. Offsetting this in the estimation, however, is the fact that their agreements tend to last longer,so that estimations on signing new agreements emphasise more heavily the consecutive signing of shorter termarrangements.


DOI: 10.1002/jid

Table 2a. Predicted probabilities of signing IMF agreements: sample averages

Countries and model sample Countries thatdid sign (number of

observations)

Countries thatdid not sign

(number of observations)

All, full sample 0.448 (285) 0.245 (637)

Low income countries, full sample 0.466 (110) 0.324 (181)

Middle income countries, full sample 0.437 (175) 0.214 (456)

Low income countries and sample 0.484 (110) 0.315 (181)

Middle income countries and sample 0.463 (175) 0.203 (456)

Asia, full sample 0.350 (24) 0.166 (110)

Latin America, full sample 0.418 (40) 0.250 (65)

Asia, middle income sample 0.355 (21) 0.117 (89)

Latin America, middle income sample 0.469 (40) 0.255 (65)

Statistical tests (t-tests) confirm that in all cases the sample average for signing countries is significantly differentthan the sample average for non-signing countries.


countries in terms of the circumstances under which they turn to the IMF for financial

assistance and sign IMF arrangements.3

Next, we go on to explore whether there are differences amongst those middle income

countries that have been classified by the IMF as experiencing capital account crises. For

this purpose, the model summarised in Table 1 forms an adequate basis since we wish to

examine whether the arrangements between the IMF and various CAC countries are

consistently less well explained by the standard factors reported in that table than

arrangements with middle income countries in general.

To do this, we estimate the predicted probabilities of signing an agreement based on

these standard factors. The probabilities are calculated for a number of Asian and Latin

American countries, as well as for Russia. Our sample includes countries that signed

programs, as well as some which did not, and contains all those countries identified as

‘crisis countries’ by the IMF (IMF, 2004) for the mid to late 1990s. This list includes

Mexico (1995), Argentina (1995 and 2000), Thailand (1997), Korea (1997), Indonesia

(1997) and Brazil (1998, 2001 and 2002).

We begin by examining the expected probabilities of signing an IMF agreement for the

whole sample, for sub-samples separating Asian and Latin American countries and for sub-

samples separating low income and middle income countries. These probabilities are

presented in Table 2a, divided into countries that signed agreements and those that did not

sign them.4

3As helpfully suggested by one referee, we formally tested whether the full sample, and low income, middleincome, Asian and Latin American sub-sample estimations differ significantly from each other. We compared themodel where the estimated coefficients were restricted to be the same across the sample with the unrestrictedmodel that allowed the coefficients to vary for each sub-group. Likelihood ratio tests confirmed that both the lowincome countries sub group and the middle income sub group differed significantly from the full sample (at the2.5 per cent level of significance). In addition, the Latin American sub-group differed significantly from the fullsample and from the Asian sub group (both at less than a 0.5 per cent level of significance). The Asian sub-group,however, was found to differ from the full sample at only the 20 per cent level of significance (approximately). Inall cases, the degrees of freedom for the chi-squared statistic was taken as 19, though step-wise regressions wouldhave reduced the number of restrictions and made the groups appear even more significantly distinct.4As requested by a referee we also generated conditionality tables for actual and predicted signings and non-signings amongst program-eligible countries. The results from this exercise are reported later in the text.


DOI: 10.1002/jid

Table 2b. Contingency table for the full-sample estimation

Predicted to sign Predicted not to sign Totals

Signed 161 124 285

Did not sign 124 513 637

Total 285 637 922

The resulting chi-squared statistic is 126.46, which is statistically significant at <0.5 per cent, with one degree offreedom.


Three interesting results emerge. First, the findings are consistent with prior beliefs that

the predicted probability of signing an agreement should be significantly higher for

countries that actually do sign them than for those that do not. The model, therefore,

certainly has some merits. Second, the gaps between the average predicted probabilities of

IMF programs when comparing those countries that actually signed them with those that

did not are found to be wider when the estimation is based on a sub-sample model than

when based on a full-sample model. This result suggests that the models are more

discriminating when estimated for specific sub-samples and are performing better at a

disaggregated level at distinguishing between countries that sign and do not sign programs.

A third observation is that the average predicted probabilities of signing for the Asian sub-

sample are noticeably lower than for the sample as a whole and for the Latin American

countries. However, the estimated probability for those Asian economies that did not sign

programs is also lower and this mitigates the size of the gap between the estimated

probabilities for those signing and not signing programs.

As noted above, we also generated contingency tables for actual and predicted signings

and non-signings amongst program-eligible countries. Since the estimation procedure

generates a probability rather than a predicted ‘yes’ or ‘no’, we set the number of predicted

agreements equal to the number of actual agreements and ranked the probability estimates

in descending order. If there were ‘n’ agreements in the sample, we classified the ‘n’ cases

with the highest predicted probabilities as having been predicted to have an agreement. We

then tested for independence in the tables. Table 2b provides one example of a contingency

table for the full sample, and for which the chi-squared test statistic of 126.5 is highly

significant at less than the 0.5 per cent level. The contingency tests for all of the groups in

Table 2a were all similarly significant, indicating that the models were generating

predictions significantly different than, or independent of, the simple percentages of each

category.5

The predicted probabilities of signing an IMF agreement that are derived from the basic

model are presented individually in separate sub-tables (Table 3) for each country of

interest. The overall probabilities associated with the model allow for the fact that countries

may be ineligible to draw resources from the Fund if they are in arrears, and that they may

not sign a program in a particular year because they are already under an arrangement. The

methodology adopted in this paper is different from others that have been used to test for

region-specific effects or effects pertaining to a particular time period where relevant

dummy variables have usually been included. The advantage of our methodology is that it

provides a richer source of information and is better suited to the broader research

questions that we pose.

5The results for the various sub-samples are available from the authors on request.


DOI: 10.1002/jid

Table 3. Expected probabilities of selected countries signing an IMF agreement in the next year

Year Predictedprobabilityof signingnext year

Marketpressureindex

Predictedprobability

excluding theeffect of pastIMF programs

Indicatorof signingnext year

Eligibleto signnext year

(a) Mexico

1991 0.307 �30.67 0.107 0 0

1992 0.460 �5.39 0.224 0 1

1993 0.468 �4.36 0.220 0 1

1994 0.518 48.14 0.531 1 1

1995 0.628 29.85 0.553 0 0

1996 0.443 �17.33 0.242 0 1

1997 0.343 �25.19 0.154 0 1

1998 0.207 3.93 0.218 1 1

1999 0.291 �12.22 0.220 0 1

2000 0.392 �14.82 0.171 0 1

(b) Brazil

1991 0.314z n/a 0.345z 1 1

1992 0.348z n/a 0.309z 0 1

1993 0.523z n/a 0.337z 0 1

1994 0.156 32.32 0.283 0 1

1995 0.020 �57.80 0.089 0 1

1996 0.098 �15.63 0.295 0 1

1997 0.149 10.48 0.337 1 1

1998 0.476 15.25 0.495 0 0

1999 0.618 21.83 0.371 0 0

2000 0.523 3.01 0.335 1 1

(c) Argentina

1991 0.176 �37.34 0.051 1 1

1992 0.211 �42.98 0.079 0 0

1993 0.464 �20.03 0.293 0 0

1994 0.503 �6.84 0.313 0 0

1995 0.731 5.97 0.574 1 1

1996 0.450 �17.91 0.227 0 0

1997 0.464 �10.24 0.248 1 1

1998 0.576 �3.43 0.347 0 0

1999 0.721 �1.70 0.511 1 1

2000 0.693 2.47 0.495 0 0

(d) Chile

1991 0.119 �13.34 0.164 0 1

1992 0.026 �11.37 0.092 0 1

1993 0.073 4.22 0.148 0 1

1994 0.050 �16.25 0.106 0 1

1995 0.051 �4.16 0.124 0 1

1996 0.070 �0.02 0.142 0 1

1997 0.052 �5.67 0.100 0 1

1998 0.109 11.63 0.187 0 1

1999 0.115 �5.69 0.203 0 1

2000 0.063 4.61 0.131 0 1

(e) Uruguay

1991 0.322 14.08 0.226 1 1

(Continues)


DOI: 10.1002/jid


Table 3. (Continued)


Marketpressureindex





1992 0.417 �11.57 0.222 0 1

1993 0.434 �14.54 0.251 0 1

1994 0.231 0.06 0.282 0 1

1995 0.286 5.91 0.543 1 1

1996 0.265 �8.07 0.192 1 1

1997 0.427 �15.48 0.194 0 0

1998 0.455 �15.07 0.221 1 1

1999 0.656 �0.08 0.399 1 1

2000 0.595 �8.31 0.339 0 0

(f) Thailand

1991 0.046 �6.43 0.100 0 1

1992 0.044 �14.97 0.098 0 1

1993 0.048 �6.72 0.108 0 1

1994 0.048 �7.21 0.108 0 1

1995 0.055 1.02 0.117 0 1

1996 0.073 �3.62 0.144 1 1

1997 0.562 38.75 0.465 0 0

1998 0.633 �13.72 0.383 0 0

1999 0.326 �24.22 0.157 0 1

2000 0.383 �0.56 0.176 0 1

(g) Korea

1991 0.082 4.11 0.173 0 1

1992 0.116 �2.15 0.234 0 1

1993 0.110 �7.26 0.231 0 1

1994 0.075 �5.52 0.164 0 1

1995 0.093 �3.60 0.199 0 1

1996 0.126 �2.62 0.239 1 1

1997 0.591 54.53 0.486 0 0

1998 0.649 �24.45 0.477 0 0

1999 0.434 �27.32 0.390 0 0

2000 0.491 0.392 0.389 0 1

(h) Indonesia

1991 0.038 4.24 0.083 0 1

1992 0.039 �6.42 0.085 0 1

1993 0.039 �9.14 0.086 0 1

1994 0.041 �4.18 0.090 0 1

1995 0.072 6.88 0.179 0 1

1996 0.060 �5.28 0.148 1 1

1997 0.361 37.27 0.266 1 1

1998 0.791 35.05 0.556 0 0

1999 0.600 �21.24 0.599 1 1

2000 0.458 �7.95 0.384 0 0

(i) Malaysia

1991 0.054 1.31 0.114 0 1

1992 0.052 �12.80 0.126 0 1

1993 0.037 �12.82 0.085 0 1

1994 0.054 �7.54 0.116 0 1

(Continues)


DOI: 10.1002/jid


Table 3. (Continued)


Marketpressureindex





1995 0.069 4.86 0.136 0 1

1996 0.063 �0.49 0.137 0 1

1997 0.130 32.38 0.232 0 1

1998 0.210 �5.31 0.418 0 1

1999 0.049 �17.90 0.173 0 1

2000 n/a �2.69 n/a 0 1

(j) China

1991 0.019 �8.45 0.037 0 1

1992 0.032 16.24 0.072 0 1

1993 0.026 9.58 0.066 0 1

1994 0.017 5.36 0.035 0 1

1995 0.028 �7.03 0.074 0 1

1996 0.029 �17.15 0.078 0 1

1997 0.030 �12.96 0.080 0 1

1998 0.028 �12.23 0.073 0 1

1999 0.033 �15.64 0.080 0 1

2000 0.041 �1.13 0.099 0 1

(k) Russia (probabilities compared against full-sample average only)

1991 n/a n/a. n/a 1 1

1992 n/a n/a n/a 0 1

1993 n/a n/a n/a 0 1

1994 0.662z n/a 0.861z 1 1

1995 0.565z n/a 0.761z 1 1

1996 0.734 �7.32 0.742 0 0

1997 0.405 �42.52 0.328 0 0

1998 0.859 59.25 0.822 1 1

1999 0.514 �0.76 0.420 0 0

2000 0.200 �56.71 0.205 0 1

Notes: Numbers in bold are above the regional sample average for middle income signing countries. The predictedprobabilities in column 1 are derived from the full model used for middle income countries. The MPI is positivewhere the index shows a weighted combination of exchange rate depreciation, reserve loss and rising interest rates.Observations marked with zwere not available in the estimation with the market pressure index (MPI) and are thusdrawn from an estimation without MPI. In this case, the general pattern of probabilities was examined to ensurethat it was reasonably comparable to that from the estimation with MPI.


Table 3(a–k) shows the evolution of the probabilities of an IMF agreement for a number

of years (column 1) around the ‘crises’ experienced either by the country itself or its

regional neighbours. The second column of Table 3 shows the predicted probabilities of

countries signing an agreement in the following year based on data from the current year,

for the 10-year period from 1991 to 2000. The third column identifies the market pressure

index as a proxy for basic economic pressures on the country’s exchange rate and thus its

external economic position; a higher positive number indicates greater pressure. The fourth

column shows the expected probability of an agreement in the following year based on the

model excluding the effect of previous IMF programs. The probabilities in this column,

when compared to those in column 2, help to identify the effect of past IMF agreements in


DOI: 10.1002/jid


affecting the predicted probability of a current program. Thus, we would anticipate that for

a country that has not had (has had) a recent IMF program, the predicted probability of

there being one would increase (decrease) when the IMF indicator is excluded from the

estimation. Column 5 indicates whether in fact an agreement was signed in the following

year (indicated by the entry ‘1’) or not (identified by a ‘0’). In the sixth column, an entry of

‘1’ means that a country was ‘eligible’ to sign an agreement in the sense that it was not

already operating with a Fund arrangement. Thus, for example, on the basis of the

characteristics identified in our best-fitting basic model, Mexico would have been expected

to sign an agreement in 1995 with a probability of 0.518 (recorded in the table for the year

before, i.e. 1994), and it did indeed do so. The equivalent probability based on 1995

characteristics was 0.628, but Mexico did not sign an agreement in 1996 because it was

already operating with the agreement from the previous year.

3.2 Interpretation

How should we interpret the findings reported in Table 3(a–k)? First, our ability to explain

IMF programs remains limited. There are many potential economic and political factors

that may idiosyncratically help determine the incidence of programs and no one standard

model is likely to capture all the nuances that will vary from case to case. Expectations

should not be set too high. Case studies may be required to fully capture the complexities of

individual IMF programs. However, our results do nevertheless reveal discernible

differences both between and within sub-groups of countries.

Second, and despite the previously mentioned difficulty, our findings do allow us to

make interesting observations about individual countries and their engagement (or lack of

engagement) with the IMF over the period studied. For example, China would not have

warranted a program on the basis of the factors captured by our analysis. And no program

was signed. At the other extreme, and relative to the mean level at which programs are put

in place, Russia would have been expected to have a program in each and every year except

in 2001. In fact, Russia did negotiate an Extended Fund Facility program with the IMF in

March 1996 and operated under an IMF agreement for every year in the sample for which

data are available, except for 2001. Our empirical investigation also identifies the

importance of market pressures in Russia at the time of the 1998 crisis.

From amongst the CAC countries as identified by the IMF, there is a somewhat different

picture for those in Latin America than for those in Asia. As noted earlier, in the case of

Mexico a program became increasingly likely over the period from 1991 until its peak in

1995. Both the market pressure index and the predicted probabilities emerging from the

models including and excluding IMF past program indicators are well above the group

average for 1994 and 1995, corresponding to the signing of an agreement in 1995 and its

continuation into 1996. The subsequent 1999 agreement, however, was hinted at only

weakly by market pressures, and occurred in the year least expected by our model. Finally,

the elevated (though still below average) signing probabilities generated by the model for

1996 and 1997 appear to be the result of the lagged program indicator, as the model

excluding this variable gave much lower predicted probabilities (comparing columns

2 and 4).

For Brazil our model indicates a high likelihood of signing an agreement 1994, a year of

heightened market pressure as well. However, an agreement did not occur until 1998; an

event perhaps anticipated by rising market pressure in 1997, ahead of the sharp increase in


DOI: 10.1002/jid


the probability (in 1998) of a program in the following year as predicted by the model. To

some extent, this reflects the impact of the 1997 program; the version of the model

excluding the indicator of previous IMF programs did generate a rising trend in the

predicted probabilities after 1995.

The model generated fairly high predicted probabilities of signing for Argentina from

about 1993 onwards, with peaks in 1995 and 1999 corresponding to two of the four

agreements signed over the 10 years reported in the table. In both instances, however, the

predicted probability was above average for the year preceding the actual agreement as

well, and fails to predict Argentina’s precautionary EFF in 1998. In only one of these cases,

the 1996 agreement, did the market pressure index even weakly suggest the possibility of

an arrangement with the Fund. In the model that excluded the IMF program indicator, the

above average predicted probabilities corresponded well to the actual signing events, but

missed the 1992 and 1998 agreements. It is also interesting to note that Argentina was one

of the few IMF members to consider signing precautionary programs, which are not

anticipated to be drawn on and which may therefore be adopted under quite different

circumstances than agreements where there is an associated immediate resource need.

The two other Latin American countries reported in Table 3 tell contrasting stories. On

the one hand, Chile is not predicted to have a program on the basis of our model, and it did

not, despite higher market pressure in 1998. Uruguay, on the other hand, entered into

programs even though these were usually not predicted by the model. The 1992 agreement

did, however, coincidewith market pressure in 1991, and this was also slightly elevated just

prior to the 1996 agreement. While the model’s predicted probabilities were rising from

1997 to 1999, reaching a peak that corresponded to the 2000 agreement, most of the

frequent arrangements entered into by Uruguay have only a weak correspondence with the

model. Uruguay has a history of adopting precautionary IMF programs and its government

may have been looking at them to transmit a positive signal about the commitment to

economic reform. Alternatively, it might have been that governments in Uruguay involved

the Fund in order to tip the domestic political balance in favour of reform or to be able to

use the Fund as a scapegoat that could be blamed for unpopular reform measures. But as

noted earlier, our model does not seek to capture such political motivations.

Relative to what might have been expected, given the overall difficulty in predicting IMF

programs, our model therefore performs reasonably well in explaining those in Latin

America during the 1990s. While the correspondence between the predicted probabilities

emerging from the estimating model and what happened is certainly not perfect, the model

still manages to anticipate several of them and often captures the increasing probability of

programs in advance of them occurring.

This is less the case when we turn to the Asian countries in our sample around the time of

the 1997/1998 crisis. For Thailand, relative to the full-sample mean for Asia, our model

predicted higher likelihoods of a program in 1998, 1999 and 2001 after the program had

been signed. There were no signings in the first two of these years because the crisis had hit

in 1997 and the agreement signed in August of that year extended over the subsequent

2 years. The market pressure index for 1997 was high, but there was no foreshadowing in

1996 of the crisis and IMF agreements as indicated by our model. The Thai agreement, and

the crisis that provoked it, are not readily apparent until they actually occur.

For Korea, the negotiation of an agreement with the Fund in December 1997 is equally

surprising given the predicted probabilities of our model. Again, the predicted probability

of an agreement in the following year and the market pressure index do not increase

substantially until 1997, so the signals are again rather delayed. The same pattern again


DOI: 10.1002/jid


occurs for Indonesia, with most signals failing to anticipate either the crisis or the IMF

programs. In Indonesia, there was elevated market pressure in 1995, 2 years before the

actual crisis, but the pressure was not sustained.

The Asian country that perhaps conforms best to our model is Malaysia, which, like

Chile, avoided the IMF altogether despite the market pressure exerted in 1997. In no year

was the predicted probability of an agreement very high. In the estimating model without

IMF indicators, there is an elevated predicted probability of one in 1999, 2 years after the

crisis. China also conforms to expectations based on our model in that it stayed away from

the Fund. Indeed, our model is effectively better at predicting the absence of programs in

the Asian sample than their incidence.

Our results show that the Asian financial crisis was largely unpredicted by conventional

models of IMF usage. This may provide some vindication of the Fund’s apparent

shortcomings in failing to anticipate it. While all of the Asian countries examined here

faced high market pressure in 1997, our model generates only a modest increase in the

predicted probability of agreements in that year.

Our results also lend some support to the claim that Malaysia opted for the alternative of

not entering into an IMF program at a time when economic factors suggested that a

program was probable. Only some support, however, since, as noted earlier, the model

predicts that a program would have been relatively unlikely from 1996 to 1998. It is only in

1999 that the variables in our model would have predicted a program. Within our Asian

sample we therefore have cases where a program was predicted but did not occur

(Malaysia) and cases where a program was not predicted but did occur, or occurred earlier

than anticipated (Thailand and Korea).

While broadly consistent with the claim that some countries in Asia were different

inasmuch as they negotiated IMF arrangements when theywould not have been expected to

do so on the basis of conventional determinants, our results are inconsistent with the

suggestion that there is a larger distinct group of CAC countries incorporating not only

Asian economies but also Latin American ones, where standard explanations of

arrangements perform very badly.

Three additional points of interest may be identified from our results. First, it may be the

case that annual data are just too crude to capture the dynamics of an unfolding crisis. We

do, however, have measures of percentage change over the course of a year for such

variables as reserves-to-imports and real exchange rate movements which ought to reflect

these events, if only imperfectly. Second, the fact that IMF arrangements are not

foreshadowed by the data on key economic variables provides some support for the

argument that such events are inherently difficult to predict. Third, the economic variables

in our model continued to worsen in many of the countries after they had signed an IMF

agreement. This casts some doubt on the ability of the Fund to quickly restore economic

stability in client countries, as reflected by the increasing probability of IMF programs

(even excluding the effects of contemporary programs on the probability of future ones),

and this observation may contain an implicit message about the effectiveness of

conditionality.

4 CONCLUDING REMARKS

This study examines whether the determinants of IMF arrangements vary across and within

country groupings. Do low income countries come to the IMF in a different set of


DOI: 10.1002/jid


circumstances than those surrounding middle income countries? Furthermore, do all the

countries classified by the IMF as CAC countries share the same characteristics when they

turn to the Fund for financial assistance?More particularly, during the 1990s, was there was

a specific and unpredictable Asian effect on IMF lending? Were the Asian programs

predictable on the basis of what we know about the determinants of IMF arrangements?

The paper examines these issues by constructing an econometric model of IMF

arrangements and by testing it against sub-samples of countries. This exercise suggests that

there are important differences between low income and middle income countries. In

neither case does our model provide a completely satisfactory guide, but it is generally

rather better at explaining the incidence of programs in middle income countries where

pressures in the foreign exchangemarket, weak economic growth and external debt seem to

exert a significant impact. Across both country groupings recent engagement with the IMF

appears to provide a reasonably reliable guide to the probability of there being

contemporary programs. It is noteworthy that external indebtedness is a characteristic

highlighted by the IMF in their ‘classic’ case scenario.

While no single model is particularly successful at explaining IMF arrangements, we test

whether our best-fit model performs less well in explaining arrangements in the CAC

countries identified by the IMF in order to see whether these countries form a discernible

and distinct group.

Our findings suggest that they do not. Overall, the model is rather better at explaining

arrangements in Latin American CACs than in Asian CACs. But there are also variations

within these regional sub-groups. For Thailand, Indonesia and Korea our analysis suggests

that there were factors that contributed to the arrangements with the Fund that are poorly

captured by our model, in spite of its hinting at the increasing probability of programs.

Although our results are not inconsistent with other research that has claimed that there was

a significant Asian effect, our methodology allows us to identify important intra regional

differences.

If there is a subset of countries that are different from conventional users of IMF

resources in the sense that they turn to the Fund earlier than standard economic

determinants would suggest or in an unpredictable way, this raises the question of the

appropriate policy response. Unpredictability makes it difficult to anticipate future calls on

IMF resources and therefore to analyse the required lending capacity of the Fund. Perhaps

greater flexibility is required in responding quickly to circumstances as they arise. The

IMF’s Contingent Credit Lines schemewas an attempt to provide such support but it was ill

designed and abandoned having never been used. Yet if the IMF cannot devise a suitable

facility for rapidly responding to somewhat unanticipated requests for help, it may be

unsurprising to find that countries opt to provide self-insurance by building up their own

reserves or by negotiating reserve pooling schemes outside the IMF.

REFERENCES

Anderson T, Harr T, Tarp F. 2006. On US politics and IMF lending. European Economic Review in

press.

Bird G. 1996. Borrowing from the IMF: the policy implications of recent empirical research. World

Development 9(6): 563–568.

Bird G, Hussain M, Joyce JP. 2004. Many happy returns? Recidivism and the IMF. Journal of

International Money and Finance 23: 231–271.


DOI: 10.1002/jid


Bird G, Orme T. 1981. An analysis of drawings on the International Monetary Fund by developing

countries. World Development 9(6): 563–568.

Bird G, Rowlands D. 2001. IMF lending: how is it affected by economic, political and institutional

factors? Journal of Policy Reform 4(3): 243–270.

Bird G, Rowlands D. 2002. IMF lending: an analysis of prediction failures. Journal of Policy Reform

5(3): 173–186.

Cornelius P. 1987. The demand for IMF credits by sub-Saharan countries. Economics Letters 23: 99–

102.

Conway P. 1994. IMF lending programs: participation and impact. Journal of Development

Economics 45(2): 355–393.

International Monetary Fund. 2004. Fund-Supported Programs: Objectives and Outcomes, SM/04/

404. IMF: Washington, DC.

Joyce JP. 1992. The economic characteristics of IMF program countries. Economics Letters 38(2):

237–242.

Knight M, Santaella JA. 1997. Economic determinants of IMF financial arrangements. Journal of

Development Economics 54(2): 405–436.

Rowlands D. 1995. Political and Economic Determinants of IMF Conditional Credit Allocations,

1973-1989, Development Working Paper, Norman Paterson School of International Affairs,

Ottawa.

Reinhart C, Rogoff K. 2004. The modern history of exchange rate arrangements: a reinterpretation.

Quarterly Journal of Economics 119(1): 1–48.

Thacker SC. 1999. The high politics of IMF lending. World Politics 52(1): 38–75.

APPENDIX: DATA DEFINITIONS AND SOURCES

Signing of an IMF agreement in the following year: A binary variable indicating that a high

conditionality IMF agreement (Stand-by, EFF, SAF/ESAF/PRGF) is signed in the

following calendar year, given that a country was eligible to sign one. Source: IMF, Annual

report, various years.

GNP per capita: GNI per capita in thousands of $U.S., Atlas method (World Bank,

World Development Indicator) deflated by U.S. consumer price index (IMF: IMF Financial

Statistics).

GDP growth: Percentage change in GDP from the previous year (annual per cent).

Source: World Bank, World Development Indicators NY.GDP.MKTP.KD.ZG.

Reserves-to-imports: Total foreign reserves divided by total imports of goods and

services (both in current $U.S.). Source: World Bank, Global Development Indicators.

Per cent change in reserves-to-imports: The percentage change in the reserves-to-import

ratio from the previous year to the current year, as a proportion of the previous year.

Current account balance/GDP: The current account balance divided by total GDP (both

in current $U.S.). Source: World Bank, Global Development Indicators.

Per cent change in the current account: The percentage change in the current account

balance from the previous year to the current year, expressed as a percentage of the

previous year.

Source: World Bank, Global Development Indicators.

Real exchange rate depreciation: The official number of domestic currency units per

$U.S. multiplied by the ratio of the U.S. consumer price index to the country’s consumer

price index. This number is calculated for the current year and for 3 years previously


DOI: 10.1002/jid


(adjusting for changes in base years) and the difference between the two is expressed as a

proportion of the value from 3 years before. Source: World Bank, World Development

Indicators.

Debt–service ratio: Total long-term debt service payments divided by total exports of

goods and services (all in U.S. dollars). Source: World Bank, World Development

Indicators.

Per cent change in the debt–service ratio: The percentage change in the total debt service

payments-to-exports ratio from the previous year to the current year, expressed as a

percentage of the previous year.

Public external debt-to-GDP ratio: The ratio of public and publicly guaranteed long-

term debt expressed as a ratio of total GDP. Source: World Bank, World Development

Indicators.

Current rescheduling: A binary indicator of whether or not the country had to reschedule

some portion of its debt (principal or interest, official or private) in the current year, which

requires by convention an IMF agreement to be in place. Source: World Bank, Global

Development Finance.

Reschedulings in past years: The number of years out of the previous 2 years in which a

country rescheduled some portion of its official or private interest or principal repayments.

Source: World Bank, Global Development Finance.

Fixed exchange rate regime: A binary variable indicating that the country operated under

a de facto fixed exchange rate regime. Source: Reinhart and Rogoff (2004).

Flexible exchange rate regime: A binary variable indicating that the country operated

under a de facto flexible exchange rate regime. Source: Reinhart and Rogoff (2004).

Inflation: Percentage annual increase in the consumer price index. Source: World Bank,

World Development Indicators.

Market pressure index: A variable measuring changes in interest rates, exchange rate

values and foreign reserve levels as an indicator of the pressure on exchange rates

experienced by a county. Source: Alex Mandilaras.

Current IMF programme in effect: A binary variable indicating that the country operated

for at least part of the current year under an existing IMF agreement. Source: IMFAnnual

Report (various years).

IMF programme in effect in the prior year: A binary variable indicating that the country

operated for at least part of the previous year under an existing IMF agreement. Source:

IMF Annual Report (various years).


DOI: 10.1002/jid

A disaggregated empirical analysis of the determinants of IMF arrangements: Does one model fit all?

Documents

Transcript of A disaggregated empirical analysis of the determinants of IMF arrangements: Does one model fit all?