Journal Issue
Share
Article

Four Decades of Fund Arrangements

Author(s):
Julio Santaella
Published Date:
July 1995
Share
  • ShareShare
Show Summary Details

I. Introduction

In the four decades following the introduction in 1952 by the International Monetary Fund of the first facility recognizable as a stand-by arrangement, more than 800 regular Fund arrangements spanning 132 member countries have been approved (Table 1). The use of Fund resources has changed substantially over the decades. What started basically as balance-of-payments support for industrial countries (the 1952 stand-by was for Belgium) gradually became destined to serve mainly developing countries. 1/ After a timid beginning in the 1950s, developing countries were receiving the majority of the increasing number of Fund arrangements in the 1960s. The collapse of the Bretton-Woods system in the 1970s interrupted momentarily the increasing trend in the use of Fund resources, but subsequent to the onset of the debt crisis in developing countries more than 250 financial arrangements were approved in the 1980s, committing in excess of SDR 60 billion of the Fund’s resources. This vigorous pace in the number of arrangements and resources committed has continued in the 1990s with the restructuring of the centrally-planned economies of Eastern Europe and the former Soviet Union into market-based systems.

Table 1.International Monetary Fund Financial Arrangements, 1952–94
195219531954195519561957195819591960196119621963196419651966196719681969197019711972
AfghanistanUCTUCTUCTUCT
Albania
Algeria
ArgentinaUCTUCTUCTUCTUCTUCTFCT
Armenia
AustraliaUCT
Bangladesh
Barbados
Belarus
BelgiumFCT
Belize
Benin
BoliviaUCTUCTUCTUCTUCTUCTUCTUCTUCTUCTUCT
BrazilUCTUCTUCTUCTFCTFCTFCTFCTFCTFCT
Bulgaria
Burkina Faso
BurundiUCTUCTUCTUCTUCTUCT
Cambodia
Cameroon
Central African Rep.
Chad
ChileUCTUCTUCTUCTUCTUCTUCTUCTUCTUCT
China, People’s Rep.
ColombiaUCTUCTUCTUCTUCTUCTUCTUCTUCTUCTUCTUCTUCTUCT
Comoros
Congo
Costa RicaUCTUCTUCTUCTUCT
Cote D’Ivoire
Croatia
CubaFCT
Cyprus
Czech Republic
Czechoslovakia
Dominica
Dominican RepublicUCTUCT
EcuadorUCTUCTUCTUCTUCTUCTUCTUCTUCT
EgyptUCTUCT
El SalvadorUCTUCTUCTUCTUCTFCTUCTUCTUCTUCTFCT
Equatorial Guinea
Estonia
Ethiopia
Fiji
FinlandFCTUCT
FranceFCTUCTUCT
Gabon
Gambia, The
Georgia
GhanaUCTUCTUCTUCT
Grenada
GuatemalaUCTUCTUCTUCTUCTUCTUCTFCT
Guinea
Guinea Bissau
GuyanaUCTFCTFCTFCTFCTFCT
HaitiUCTUCTUCTUCTUCTUCTUCTUCTUCTFCTFCTFCT
HondurasFCTUCTUCTUCTUCTUCTUCTUCTUCTUCTUCTUCT
Hungary
IcelandUCTUCTUCT
IndiaUCTUCTUCTUCT
IndonesiaFCTUCTUCTUCTUCTUCTUCT
IranUCTUCT
Israel
Italy
JamaicaUCT
JapanFCTFCT
Jordan
Kazakhstan
Kenya
KoreaUCTFCTUCTFCTUCTUCTUCTFCT
Kyrgyz Republic
Lao P.D. Republic
Latvia
Lesotho
LiberiaUCTUCTUCTUCTUCTUCTUCTUCTFCT
Lithuania
Macedonia. F.Y.R. of
Madagascar
Malawi
MaliUCTUCTUCTUCTUCT
Mauritania
Mauritius
MexicoUCTUCTFCT
Moldova
Mongolia
MoroccoUCTUCTUCTUCTUCTUCTUCT
Mozambique
MyanmarFCT
Nepal
NetherlandsFCT
New ZealandUCT
NicaraguaFCTUCTUCTUCTUCTUCTUCTUCTUCTUCT
Niger
Nigeria
PakistanFCTFCTUCTUCT
PanamaUCTFCTUCTUCTUCTFCT
Papua New Guinea
ParaguayUCTUCTUCTUCTUCTFCTFCTUCTFCT
PeruFCT, UCTUCTUCTUCTUCTUCTUCTUCTUCTUCTFCTUCTUCT
PhilippinesUCTUCTUCTUCTFCTFCTUCTUCTUCTUCT
Poland
Portugal
Romania
Russia
RwandaUCTUCTUCTUCT
Sao Tome & Principe
Senegal
Sierra LeoneUCTUCTUCT
Slovak Republic
Solomon Islands
SomaliaFCTUCTUCTUCTUCTUCTUCT
South AfricaFCTUCT
SpainUCTUCT
Sri LankaUCTUCTUCTUCTFCT
SudanUCTUCTUCTUCT
Syrian Arab Rep.FCTUCTUCT
Tanzania
Thailand
Togo
Trinidad and Tobago
TunisiaUCTUCTUCTUCTUCTUCT
TurkeyUCTUCTUCTUCTUCTUCTUCTUCTUCTUCT
UgandaFCT
Ukraine
United kingdomUCTUCTUCTUCTFCTFCTUCTUCTUCT
United StatesFCTFCT
UruguayUCTUCTUCTUCTFCTUCT
VenezuelaUCT
Viet Nam
Western Samoa
YugoslaviaUCTUCTUCTFCT, UCT
ZaireUCT
Zambia
Zimbabwe
Key: FCT: Stand-by in first credit tranche UCT: Stand-by in upper credit tranches EFF: Exended Fund FacilitySAF: Structural Adjustment Facility ESAF: Enhanced Structural Adjustment Facility STF: Systemic Transformation Facility
1973197419751976197719781979198019811982198319841985198619871988198919901991199219931994
AfghanistanUCTFCT
AlbaniaUCTESAP
AlgeriaFCTUCTUCT
ArgentinaUCTUCTUCTUCTUCTUCTUCTEFF
ArmeniaSTF
Australia
BangladeshFCTUCTUCTEFFUCTUCTSAPSAP
BarbadosUCTUCT
BarbadosSTF
Belgium
BelizeUCT
BeninSAPESAF
BoliviaUCTUCTUCT SAPESAPESAP
BrazilETFUCTUCT
BulgariaUCTUCT, STF
Burkina FasoSAPESAP
BurundiFCTUCT SAPESAP
CambodiaSTFBSAF
CameroonUCTUCTUCT
Central African Rep.UCTUCTUCTUCTUCTUCT SAPUCT
ChadSAPUCT
ChileUCTUCTUCTFCI
China, People’s Rep.FCTFCT
ColombiaFCT
ComorosSAF
CongoFCTUCTUCTUCTUCT
Costa RicaFCTUCTEFFUCTUCTUCTUCTUCTUCT
Cote D’IvoireEFFUCTUCTUCTUCTUCTUCTESAP
CroatiaUCT STF
Cuba
CyprusUCT
Czech RepublicUCT
CzechoslovakiaUCTUCT
DominicaEFFUCTSAP
Dominican RepublicBFFUCTUCTUCT
EcuadorUCTUCTUCTUCTUCTUCTUCT
EgyptUCTEFFUCTUCTEFF
El SalvadorFCTUCTUCTUCTUCT
Equatorial GuineaUCTUCTSAPESAP
EstoniaUCTUCT, STF
EthiopiaUCTSAP
FijiFCT
FinlandUCT
France
GabonUCTEFFUCTUCTUCTUCT
Gambia, TheFCTFCTUCTUCTUCT, SAPESAP
GeorgiaSTF
GhanaUCTUCTUCTUCTUTF, SAFESAF
GrenadaFCTFCTFCTUCTEFF
GuatemalaFCTUCTUCTUCT
GuineaUCTUCTUCT SAPESAP
Guinea BissauSAP
GuyanaFCTFCTFCTFCTUCTEFFEFFUCT BSAPBSAF
HaitiFCTFCTUCTUCTFCTEFFUCTUCTSAPUCT
HondurasEFFUCTUCTESAP
HungaryUCTUCTUCTUCTBPUCT
Iceland
IndiaEFFFCT UCT
IndonesiaUCT
Iran
IsraelFCTUCTUCT
ItalyUCTUCT
JamaicaUCTUCTEFFEFFEFFUCTUCTUCTUCTUCTUCTEFF
Japan
JordanUCTUCTEPF
KazakhstanSTFUCT
KenyaEFFFCTUCTUCTUCTUCTUCTUCT SAPESAPESAP
KoreaFCTFCTUCTFCTUCTUCTUCTUCT
Kyrgyz RepublicUCT, STFESAF
Lao P.D. RepublicUCTSAPESAP
LatviaUCTUCT STF
LesothoSAPESAPUCT
LiberiaFcrFCTFCTFCTUCTUCTUCTUCTUCT
LithuaniaUCTUCT STFEFF
Macedonia, F.Y.R. ofSTF
MadagascarFCTUCTUCTUCTUCTUCTUCTSAPUCTESAP
MalawiUCTUCTUCTEFFUCT ESAPUCT
MaliUCTUCTUCTUCT SAPESAP
MauritaniaFCTUCTUCTUCTUCT SAPUCTESAPBSAP
MauritiusFCTUCTUCTUCTUCTUCT
MexicoEFFEFFUCTEFF
MoldovaUCT STF
MongoliaUCTESAP
MoroccoBFFBFFUCTUCTUCTUCTUCTUCTUCT
MozambiqueSAPESAP
MyanmarFCTUCTUCTUCTUCT
NepalFCTUCTSAPBSAP
Netherlands
New Zealand
NicaraguaUCTUCTBSAP
NigerUCTUCTUCTUCT, SAPESAPUCT
NigeriaUCTUCTUCT
PakistanUCTUCTUCTBFFBFFUCT SAPUCTEFF, ESAF
PanamaFCTFCTFCTFCTUCTUCTUCTUCTUCTUCTUCT
Papua New GuineaUCTUCT
Paraguay
PeruUCTUCTUCTEFFUCTEFF
PhilippinesUCTFCTFCTEFFUCTUCTUCTUCTUCTEFFUCTEFF
PolandUCTEFFUCTUCT
PortugalFCTUCTUCT
RomaniaUCTUCTUCTUCTUCTUCT STF
RussiaFCTSTF
RwandaFCTSAP
Sao Tome & PrincipeSAP
SenegalFCTEFFUCTUCTUCTUCTUCT, SAPUCTESAPESAP
Sierra LeoneFCTUCTEFFUCTUCT SAPSAP, ESAP
Slovak RepublicUCT
Solomon IslandsUCTUCT
SomaliaUCTUCTUCTUCTUCT SAP
South AfricaFCT UCTUCT
SpainUCT
Sri LankaUCTUCTEFFUCTSAPESAP
SudanUCTUCTEFFUCTUCTUCT
Syrian Arab Rep.
TanzaniaUCTUCTUCTSAPESAP
ThailandFCTUCTUCTUCT
TogoUCTUCTUCTUCTUCTUCTUCT SAPESAPESAF
Trinidad and TobagoUCTUCT
TunisiaUCTEFF
TurkeyUCTUCTUCTUCTUCTUCT
UgandaUCTUCTUCTUCTSAPESAPESAP
UkraineSTF
United KingdomUCTUCT
United States
UruguayFCTFCTFCTFCTUCTFCTUCTUCTUCTUCT
VenezuelaEFF
Viet NamUCT STFESAP
Western SamoaFCTFCTFCTUCTUCTUCT
YugoslaviaFCTUCTUCTUCTUCTUCTUCT
ZaireFCTUCTUCTEFFUCTUCTUCTUCT SAPUCT
ZambiaFCTUCTUCTBFFUCTUCTUCT
ZimbabweUCTUCTEFF ESAP
Key: FCT: Stand-by in first credit tranche UCT: Stand-by in upper credit tranches EFF: Exended Fund FacilitySAF: Structural Adjustment Facility ESAF: Enhanced Structural Adjustment Facility STF: Systemic Transformation Facility
Key: FCT: Stand-by in first credit tranche UCT: Stand-by in upper credit tranches EFF: Exended Fund FacilitySAF: Structural Adjustment Facility ESAF: Enhanced Structural Adjustment Facility STF: Systemic Transformation Facility

As the Fund has adapted to the world’s changing circumstances, so have the modalities of its financial assistance. After the historic introduction of the stand-by arrangement in 1952, what would later become known as the Compensatory and Contingency Financing Facility (CCFF) was established in 1963 to provide financial assistance to members experiencing exogenous shortfalls in export earnings, In 1974 the Extended Fund Facility (EFF) was introduced to allow member countries to adopt a medium-term horizon in solving their balance-of-payments adjustment problems; in 1986 the Structural Adjustment Facility (SAF) was established to provide financial assistance on concessional terms to low-income members, and one year later this concessional assistance was expanded through the Enhanced Structural Adjustment Facility (ESAF). In 1992 the Systemic Transformation Facility (STF) was introduced to assist member countries in the transition toward a market-based system. 2/

After four decades of Fund financial arrangements, a conventional wisdom has emerged in circles of academics and practitioners alike that typifies the economic situation of countries that turn to the Fund. This conventional wisdom was initially influenced by the experience and the analytical methods prevailing during the Bretton-Woods period, when a system of fixed-but-adjustable par values provided the setting for the analysis, and then it gradually incorporated new developments in both theoretical and applied economics. Under the conventional wisdom, there is the presumption of a dramatic deterioration in the economic conditions in the period prior to a Fund-supported adjustment program. Since a “balance of payments need11 is a pre-requisite for Fund financial assistance, it is not surprising that the requesting country faces critical conditions in the balance of payments. Another piece of the conventional wisdom is that the macroeconomic disequilibria prior to the approval of a financial arrangement can be characterized by high or rising inflation, real exchange rate appreciation and low or declining growth in GDP. 1/ However, in spite of four decades of Fund arrangements, the evidence gathered in this respect is scant and, at best, mixed.

In fact, the bulk of the analysis in the literature that studies Fund-supported programs has focused on the macroeconomic effects of the adjustment programs, and much less research has been done on the characteristics of countries that enter into Fund-supported programs. It would be no exaggeration to claim that despite the long experience with Fund-supported programs over the decades, much more is known about the ex-post effects of Fund-supported programs than about the factors that lead up to the adoption of programs in the first place. 2/ The sparse literature that has dealt with the period before the adoption of an adjustment program has mostly focused on the “demand for” Fund arrangements, 3/ This study attempts to bridge this gap in the literature.

This paper uses one of the most comprehensive data sets to date to analyze the initial macroeconomic conditions before Fund financial arrangements are adopted, and complements previous work in Knight and Santaella (1994). It presents evidence from 324 Fund arrangements in 78 non-oil developing countries during the period 1973-91. The paper also documents differences and similarities across decades in these stylized facts before the inception of a Fund-supported adjustment program. The choice of sample was made for both practical and conceptual reasons. On the practical side, data limitations precluded extending the sample to more countries or capturing earlier Fund arrangements.

Conceptually, the time period under consideration corresponds to the post-Bretton Woods system, a period that has not been studied sufficiently and, given the previously generated bulk of the conventional wisdom, has spurred many academic and policy-making controversies, 1/ Nevertheless, the sample used in this investigation is quite representative since it covers more than three quarters of all the financial arrangements approved in the 1973-91 period. The methodological approach followed here is very basic and relies heavily on the use of simple descriptive statistics. This approach is rather model-free in the sense of avoiding the need to impose behavioral assumptions to analyze the empirical evidence, and contrasts to the approach followed in Knight and Santaella (1994), where explicit demand and supply of Fund arrangements were postulated in order to analyze the empirical determinants of Fund arrangements.

By documenting the initial macroeconomic conditions of the Fund-supported adjustment programs, this paper can be helpful in addressing important questions. First, better knowledge of the initial conditions would enhance the design of macroeconomic policies geared to tackle the macroeconomic disequilibria of Fund member countries. Second, a good characterization of the stylized facts before adoption of a Fund-supported program would also improve the ex-post evaluation of Fund programs. Solid evidence regarding the initial macroeconomic conditions would enhance the robustness of the results of a “before-after” type of evaluation, and would also document observable differences between program and nonprogram episodes in order to undertake a “with-without” type of evaluation. 2/ Third, this paper would shed some light on the controversial area that deals with the origins of the macroeconomic imbalances that give rise to Fund-supported programs. At the risk of oversimplifying the issues, one can look at two extreme positions. One group of observers argues that macroeconomic imbalances before a Fund-supported program generally originate in the external conditions faced by the country in question, such as deterioration in the terms of trade, a slowdown of demand for the country’s exports, or adverse conditions in international financial markets. On the opposite side of the spectrum, another group of observers contends that imbalances are endogenous to the country, and in particular, brought about mostly by governments pursuing unsustainable macroeconomic policies. Reality, however, is rarely either black or white and the case studies by Killick and Malik (1992) show that in most cases a mixture of both external and internal factors have led to Fund arrangements.

The paper is organized as follows. The stylized facts that lead to a Fund-supported program are gathered in Section II and cover the evolution over time of a series of macroeconomic indicators in the three-year period preceding the approval of a Fund financial arrangement. In order to have a clear idea of the extent of economic deterioration before the inception of a Fund-supported adjustment program, a control group of nonprogram observations is included as a benchmark. The comparison with a control group will allow explicit and precise identification of how program countries differ from the behavior that is associated with normal economic conditions. Section III contains the statistical analysis and presents a series of nonparametric tests that are computed in order to study to what extent countries initiating a program are “significantly different” from the control group. This analysis is complemented with the use of discriminant analysis to ascertain which types of variables are the most useful indicators to discriminate program episodes from nonprogram observations. Section IV brings back the historical dimension of the paper by comparing the macroeconomic characteristics of the 1970s and the 1980s with respect to the period preceding a Fund-supported program. Finally, Section V concludes by summarizing the interpretation of the evidence.

II. The Three-Year Period Preceding Fund-Supported Programs

This section examines the behavior of a selected group of macroeconomic variables in the three-year period prior to the initiation of a Fund-supported adjustment program. The evidence presented below shows a particularly feeble economic situation in member countries before the inception of Fund financial arrangements: weak balance of payments, low output growth and investment rates, tough external conditions and frail public finances. This state of affairs persists during the three-year period prior to the Fund-supported program, with sharp deteriorations only evident in the overall balance of payments, stock of international reserves and external indebtedness. These stylized facts conform only partially to the conventional wisdom.

The sample of Fund financial arrangements was obtained from a total population of 91 non-oil developing countries during the 1973-93 period. From this population 324 financial arrangements were approved by the Fund in support of adjustment programs (Table 2). These program episodes span 78 countries and include annual data for the three-year period preceding four different types of Fund financial arrangements that contain conditionality elements: stand-by arrangements in the upper credit tranches, EFF, SAF and ESAF.

Table 2.Program Episodes and Control Group Observations
1973197419751976197719781979198019811982198319841985198619871988198919901991
AfghanistanUCT
AlgeriaUCT
ArgentinaUCTUCTUCTUCTUCTUCTUCT
Bahrain
BangladeshUCTUCTEFPUCTUCTSAFESAF
BarbadosUCT
BeninSAF
BoliviaUCTUCTUCT.WFESAF
Botswana
BrazilEFFUCT
Burkina FasoSAF
BurundiUCTJAFB5AF
CameroonUCTUCT
Central African Rep.UCTUCTUCTUCTUCTUCT5AF
ChadSAF
ChileUCTUCTUCTEFF
China People’s Rep.
Colombia
CongoUCTUCTUCT
Costa RicaUCTGFFUCTUCTUCTUCTUCT
Cote D’IvoireEFFUCTUCTUCTUCTUCTUCT
CyprusUCT
Dominican RepublicEPFUCTUCT
EcuadorUCTUCTUCTUCTUCTUCT
EgyptUCTEFPUCTUCT
El SalvadorUCTUCT
EthiopiaUCT
Fiji
GabonUCTEFPUCTUCTUCT
Gambia, TheUCTUCTUCTSAFESAP
GhanaUCTUCTUCTUCTBFP5APESAF
GuatemalaUCTUCT
GuineaUCTUCTUCT JAPBSAF
GuyanaUCTBFFBPFUCT^SAI
HaitiUCTUCTEFPUCTUCTSAPUCT
HondurasBPFUCTUCT
IndiaEFPUCT
IndonesiaUCT
IsraelUCTUCT
JamaicaUCTUCTEFPBPFBFFUCTUCTUCTUCTUCTUCT
JordanUCT
KenyaEFPUCTUCTUCTUCTUCTUCTJAPBSAF
KoreaUCTUCTUCTUCTUCT
Lebanon
LesothoSAFESAP
LiberiaUCTUCTUCTUCTUCT
MadagascarUCTUCTUCTUCTUCTUCTSAPUCTESAF
MalawiUCTUCTUCTBPPUCT^SAF
Malaysia
MaliUCTUCTUCTUCT3AP
Malta
MauritaniaUCTUCTUCTUCTJAPUCTeup
MauritiusUCTUCTUCTUCTUCT
MexicoBFFEFFUCTEFP
MoroccoEFFBFFUCTUCTUCTUCTUCTUCT
MyanmarUCTUCTUCTUCT
NepalUCTSAF
NicaraguaUCTUCT
NigerUCTUCTUCTUCT4AFESAF
NigeriaUCTUCTUCT
PakistanUCTUCTUCTEFFEFFUCT, SAF
PanamaUCTUCTUCTUCTUCTUCT
Paraguay
PeruUCTUCTUCTEFFUCT
PhilippinesUCTEFFUCTUCTUCTUCTUCTBFFUCT
RomaniaUCTUCTUCTUCT
RwandaSAF
SenegalEFFUCTUCTUCTUCTUCT.SAFUCTESAf
Sierra LeoneUCTEFFUCTUCT^AF
Singapore
SomaliaUCTUCTUCTUCTUCT.SAP
South AfricaUCTUCT
Sri LankaUCTUCTEFPUCTSAPESAF
SudanUCTUCTEFFUCTUCTUCT
Swaziland
Syrian Arab Rep.
TanzaniaUCTUCTUCTSAFESAP
ThailandUCTDCTUCT
TogoUCTUCTUCTUCTUCTUCTUCT.SAF6SAF
Trinidad and TobagoUCTUCT
TunisiaUCT6FF
TurkeyUCTUCTUCTUCTUCT
UgandaUCTUCTUCTUCTSAFESAF
UruguayUCTUCTUCTUCT
Venezuela6FF
Western SamoaUCTUCTUCT
Yemen Arab Rep.
Yemen, P.D. Rep.
YugoslaviaUCTUCTUCTUCTUCTUCT
ZaireUCTUCTEFFUCTUCTUCTUCT, SAFUCT
ZambiaUCTUCTEPFUCTUCTUCT
Note: Shaded cells indicate control group observations.Key to Program Episodes:UCT: Stand - by in upper credit tranchesEFF: Extended Fund FacilitySAF: Structural Adjustment FacilityESAF: Enhanced Structural Adjustment Facility.
Note: Shaded cells indicate control group observations.Key to Program Episodes:UCT: Stand - by in upper credit tranchesEFF: Extended Fund FacilitySAF: Structural Adjustment FacilityESAF: Enhanced Structural Adjustment Facility.

In order to assess the distinctive features of the program episodes, the behavior of the main economic variables is compared with a control group of observations. In studies of the ex-post effects of Fund-supported programs, the typical objective for using a control group is to have a counterfactual; namely, under the assumption that the group of nonprogram countries shares the same exogenous environment as the program episodes, the control group provides an approximation of what would have happened in the absence of a program. 1/ The objective of a control group in this study is rather different. The idea of the control group is to capture the macroeconomic performance that predominates under “normal circumstances” in countries that do not enter into a Fund-supported program during an extended period of time. The control group thus includes countries that do not enter into Fund arrangements either because their economic conditions do not warrant one or because, in spite of having the need, they do not feel inclined to enter into one (Knight and Santaella, 1994). Given the clear difficulty of disentangling these two types in the control group, no attempt will be made here to separate them. However, countries that do not enter into an arrangement because they do not need one are likely to dominate the characteristics of the whole group, since it would be hard for a country in need of an arrangement to avoid a program indefinitely. This feature of the control group would make it a legitimate indicator of macroeconomic performance under “normal” or “noncritical” circumstances.

Specifically, in this study the control group observations were also selected from the same 91 country population during 1973-91. The control group observations were defined as those country-year observations that did not have a Fund-supported program in place for at least the next three consecutive years. Three years without a Fund financial arrangement is an arbitrary but reasonable selection criterion that yielded a sample made up of 693 observations spanning a total of 84 different countries (see Table 2). In sum, out of 1729 potential observations in the population, 40 percent were classified as control group observations and 19 percent as program episodes; each of the latter comprises a period of three years prior to the adoption of a Fund financial arrangement.

In comparing the evolution over time during the period preceding the adjustment program with the control group, a distinction is made between the behavior of macroeconomic targets or outcome variables, external sector indicators and domestic policy variables. This distinction is useful in dealing with the question of the origins of the macroeconomic disequilibria that lead to an adjustment program. As a first step, the characteristics of the two sample groups are described making use of some basic descriptive statistics. As a second step, a formal statistical analysis is performed in Section III in order to test for the apparent differences in the macroeconomic indicators between the two groups.

1. Macroeconomic outcomes

Table 3 shows the evolution of the sample distribution of some macroeconomic indicators--the overall balance of payments, the current account, the stock of international reserves, the inflation rate, the growth of GDP per capita, the domestic investment rate and the real effective exchange rate--over the three-year period prior to the arrangement’s approval. The table compares the distributions of these variables in the third, second and first year before the initiation of the Fund-supported program--labeled “t-3”, “t-2” and Mt-1”--with the control group’s sample distribution. Figure 1 displays this evolution over time for the median program episodes against the first, second and third quartiles of the control group. 1/

Table 3.Hacroeconomic Outcomes in Program and Control Groups: Distribution of Samples
Program episodesControl
t-3t-2t-1group
Balance of payments (percentage of GDP)
First quartile-4.9-5.4-5.9-1.6
Median-1.6-2.3-2.70.6
Third quartile0.2-0.1-0.43.2
Mean-2.8-3.3-3.80.5
Current account (percentage of GDP)
First quartile-9.0-8.6-8.6-7.3
Median-5.0-5.2-5.2-3.2
Third quartile-1.9-2.2-2.80.5
Mean-6.5-6.3-6.4-3.9
International reserves (months of imports)
First quartile1.00.90.82.0
Median1.91.81.64.1
Third quartile3.63.43.06.8
Mean2.62.42.35.2
Consumer prices (percentage change)
First quartile7.57.67.76.3
Median’11.913.312.711.4
Third quartile24.626.227.417.5
Mean105.254.798.620.1
GDP per capita (percentage change)
First quartile-2.9-2.9-3.1-1.1
Median0.80.50.02.2
Third quartile3.53.02.85.4
Mean0.40.1-0.11.9
Investment (percentage of GDP)
First quartile15.414.714.617.4
Median21.220.920.023.0
Third quartile27.226.526.430.2
Mean22.021.221.124.6
Real effective exchange rate (percentage change)
First quartile-6.0-6.3-7.9-4.5
Median-1.1-0.5-1.50.6
Third quartile5.05.44.26.1
Mean-0.20.60.02.6

Figure 1ECONOMIC OUTCOMES

The stylized facts that emerge from this evidence are suggestive. As expected, the most consistent pattern that can be inferred, is that the inception of a Fund-supported program is preceded by a sharp deterioration in the external accounts of the member country. Three years before the program, the median overall balance of payments is already as low as -1.6 percent of GDP, the same as the deficit of only the first quartile of the control group. From this deficit, the balance of payments worsens steadily to reach -2.7 percent of GDP in the year prior to the program. This behavior is almost mirrored by the evolution of the stock of international reserves: three years before the program the median stock of international reserves is only 1.9 months of imports, below the first quartile of the control group, and it falls to 1.6 months of imports by the year prior to the adjustment program. The external current account also starts from a level that is substantially weaker than the one observed in the control group: the median current account deficit in the program countries is 5.0 percent of GDP three years before the arrangement compared with the 3.2 percent for the control group. The median current account deficit reaches 5.2 percent of GDP two years before the program and seems to stabilize at that level for the next year. Another important feature of the program episodes is that the dispersion of the stock of international reserves is substantially smaller (and centered around a lower level) for the program episodes than for the control group. 2/ It is clear that program episodes start from a very weak external position that deteriorates markedly up to the point of falling to almost unsustainable levels of foreign exchange reserves before the arrangement’s approval.

The comparative behavior of the other macroeconomic outcome variables is less dissimilar between program and non-program episodes. Although the median inflation rate starts higher in the program episodes than in the control group, the difference appears fairly small (11.9 percent per annum three years before the program versus 11.4 percent in the control group). Interestingly enough, the median inflation rate in the program episodes does not accelerate as the approval time gets closer: it reaches only 13.3 percent two years before the program, and then declines to 12.7 in the year prior to it. However, the dispersion of inflation is substantially higher in the program episodes than in the control group observations. In fact, high inflation program observations--such as that corresponding to the arrangement in the aftermath of the Bolivian hyperinflation--drive the average inflation rate well above the median inflation for this group.

The behavior of the real effective exchange rate seems to be at variance with the conventional wisdom of a real appreciation. In fact, the real exchange rate of program episodes depreciates continuously as the program gets closer: the median cumulative depreciation is almost 6 percent in the three years prior to the program. It turns out then, that contrary to a conventional wisdom mostly forged during fixed exchange rate regimes, real exchange rates in program episodes start to react to a worsening external situation by depreciating well before the inception of a Fund-supported adjustment program.

More in line with the conventional wisdom, the median rate of growth of GDP per capita is substantially lower in the period prior to the programs than in the control group, and it deteriorates steadily from 0.8 percent in t-3 to about zero in t-1. Another interesting piece of evidence relates to domestic investment. Program episodes exhibit a fairly constant median rate of investment of around 20 percent of GDP during the three year period, much lower than the 23 percent of GDP for the control group.

2. External indicators

This section focuses on the following variables for the external sector: the terms of trade, growth of export markets and external debt indicators. Overall, the evidence pertaining to these variables is quite interesting. As can be seen from Table 4 and Figure 2, the median terms of trade deteriorates slightly over time for the program group, falling a cumulative 6 percent over the three years prior to the arrangement, while for the control group the terms of trade experience a modest improvement (0.6 percent per annum). Interestingly enough, both samples exhibit roughly the same degree of terms of trade variability according to the interquartile range. The median growth rate of export markets is almost identical for the control group and for the program episodes three years before the program (3.5 percent per annum), but it deteriorates slightly over time for the program countries (3.2 percent one year before the program).

Table 4.External Indicators in Program and Control Groups: Distribution of Samples
Program episodesControl
t-3t-2t-1Group
Terms of trade (percentage change)
First quartile-10.4-9.1-9.0-6.9
Median-2.0-1.5-2.50.6
Third quartile6.05.64.09.3
Mean-0.4-1.1-0.83.4
Export markets (percentage change)
First quartile2.12.22.22.3
Median3.53.33.23.5
Third quartile4.34.24.14.7
Mean3.23.23.03.5
External debt service
(percentage of exports)
First quartile11.112.313.04.5
Median20.621.522.110.1
Third quartile32.034.734.419.0
Mean24.025.625.214.1
External debt (percentage of GDP)
First quartile26.629.932.812.4
Median42.547.350.624.8
Third quartile65.671.677.041.4
Mean52.758.163.833.6

Figure 2EXTERNAL ENVIRONMENT

Finally, both the external debt service and the total external debt ratios do show sharp differences between the program episodes and the control group. Although these two variables are not entirely exogenous because they reflect to a large extent policy decisions undertaken in previous periods, the external debt service will have an important exogenous component: the interest rate prevailing in international capital markets. The median external debt service ratio for program countries is already well above the third quartile of the control group three years before the program (20.6 percent of exports). This ratio grows steadily over time, reaching 22.1 percent of exports in the year prior to the arrangement. Similarly, total external debt is substantially higher for program countries than for the control group. The median debt to GDP ratio starts at 42.5 percent, also above the third quartile of the control group, and rises over time to reach 50.6 percent of GDP one year before the program. Also evident from the data is the higher dispersion of these two variables for the program episodes one year before the program.

3. Policy variables

The next piece of evidence deals with the behavior of economic policy variables--monetary, fiscal and nominal exchange rate indicators. The overall picture that emerges is that program countries differ substantially from the control group with respect to the stance of their fiscal and exchange rate policies, but not so much with respect to their monetary and overall credit policy.

The three fiscal policy indicators (the fiscal balance, the flow of net government borrowing from the banking system, and the rate of expansion of domestic credit to the government) show that the fiscal performance of program countries is substantially weaker than that prevailing in the control group. For example, the median fiscal deficit in program countries three years before the program is 5.6 percent of GDP compared to 3.4 percent in the control group, while the median rate of growth of domestic credit to the public sector in program countries in the same period is 25.6 percent per annum compared to 15.9 percent in the control group (Table 5 and Figure 3). Perhaps the most interesting feature revealed by the data is that the stance of fiscal policy is roughly constant during the three-year period prior to the approval of the financial arrangement; i.e. public finances do not deteriorate markedly before the inception of the adjustment program.

Table 5.Macroeconocnic Policy in Program and Control Groups: Distribution of Samples
Program EpisodesControl
t-3t-2t-1Group
Flow of government borrowing
(percentage of GDP)
First quartile1.10.71.2-0.6
Median3.43.23.21.2
Third quartile6.66.56.74.2
Mean4.34.54.13.0
Fiscal balance
(percentage of GDP)
First quartile-10.1-9.6-9.9-8.6
Median-5.6-5.4-5.8-3.4
Third quartile-2.9-2.8-3.1-0.8
Mean-7.0-6.8-6.9-5.9
Domestic credit
(percentage change)
First quartile12.410.810.110.1
Median22.021.620.820.4
Third quartile35.738.536.938.2
Mean125.653.2160.833.4
Broad money supply
(percentage change)
First quartile10.610.611.412.2
Median20.218.919.119.7
Third quartile31.728.232.829.9
Mean92.047.589.627.2
Domestic credit to the government
(percentage change)
First quartile9.07.26.0-11.6
Median25.622.723.215.9
Third quartile59.252.260.447.9
Mean-25.064.2685.6-19.6
Domestic credit to the government
(percentage of total)
First quartile13.015.417.0-0.2
Median32.833.134.914.4
Third quartile51.151.652.741.6
Mean32.533.034.558.1
Nominal effective exchange rate
(percentage change)
First quartile-11.5-12.3-17.6-4.8
Median-1.8-2.5-3.0-0.1
Third quartile1.42.02.33.8
Mean-8.2-8.7-10.5-2.1

Figure 3ECONOMIC POLICY

Figure 4Histograms by Group

The relative similarity of monetary policy in the program episodes and the control group is rather surprising. Although the growth rate of total domestic credit is initially slightly higher in program episodes three years before the program than in the control group--22 percent per annum versus 20.4 percent--it falls continuously over time, converging to the control group one year before the program. A similar picture is depicted by the growth rate of broad money supply, which remains roughly constant during the three year period prior to the adjustment program at about the median growth rate of the control group (19.7 percent per annum). The apparent similarity of the growth rates of total domestic credit taken together with the larger expansion of credit to the public sector in program countries, implies that the governments’s share of credit is growing over time. The median share of credit to the government starts at about 33 percent of total credit three years before the program and it ends close to 35 percent the year prior to the arrangement. These figures are substantially higher than the government’s share in the control group: only 14.4 percent of total credit.

Finally, program episodes differ from the control group in the pattern displayed by exchange rate policy. Program countries start depreciating more than the control group; the median annual percentage change in the nominal effective exchange rate is -1.8 percent three years before the program, against -0.1 percent for the control group. As time goes by, program countries accelerate the rate of depreciation, and by the year prior to the arrangement the nominal depreciation rate is -3.0 percent per annum. Not only do exchange rates depreciate more during program episodes, but they also exhibit a larger dispersion.

III. Statistical Analysis

The previous section has provided a broad overview of a set of selected macroeconomic indicators distinguishing between program episodes and a control group. In this section a formal statistical analysis is carried out with a twofold purpose. First, a series of nonparametric tests are performed In order to determine whether the apparent differences between program episodes and control group observations are statistically significant. Second, a discriminant analysis is presented to ascertain which variables are the most useful indicators In distinguishing program and nonprogram observations.

The evidence presented below supports some of the patterns reported earlier in this paper and provides new insights. It shows that program episodes and the control group are statistically different in all their macroeconomic characteristics, with the exception of the expansion of a broad money aggregate and of total credit. Moreover, the discriminant analyses suggest that some indicators--such as the stock of international reserves and the external debt variables--appear to be quite powerful In detecting the Incidence of Fund-supported programs.

1. Nonparametric tests

Unlvariate nonparametric tests indicate in what precise sense program episodes are different from the control group observations. The nonparametric statistics used test the equality of medians (Wilcoxon rank-sum test), the equality of distributions (Kolmogorov-Smirnov test) and, the equality of populations (Kruskal-Wallis test) of the two samples. 1/ It must be recalled that nonparametric statistical tests require fewer and weaker assumptions than parametric tests. In particular, nonparametric tests are better equipped to deal with samples made up of observations from different populations. Moreover, they are also less distorted by problems of measurement error. The traditional disadvantage of nonparametric tests--i.e., the “waste” of information reflected in a low power-efficiency ratio--can be overcome with sufficiently large samples, such as those used In the present study.

The results of the nonparametric tests for the comparison of “t-1” (the year immediately preceding a Fund-supported program) and the control group are presented in Table 6. 1/ With respect to the macroeconomic outcome variables, the evidence is overwhelming: all the differences described In the previous section are statistically significant. In the case of the overall balance of payments, the balance in the external current account, the stock of international reserves, the growth rate of per capita income, the investment rate and the real effective depreciation rate, the Wilcoxon rank-sum test rejects the null hypothesis that the program episode median is equal to the control group median In favor of the alternative of a smaller median for the program episodes; in the case of the Inflation rate the favored alternative hypothesis is that the median is higher for the program episodes. Not only the medians are different but the Kolmogorov-Smirnov test also rejects the null of equality of sample distribution functions for all the macroeconomic outcomes. Finally, the Kruskal-Wallis test indicates that in fact the program episodes and the control group come from different populations. An additional fact--but not reported due to space limitations--is that basically all the nonparametric statistics increase over time, thus indicating deepening differences in macroeconomic outcomes between the program episodes ai\d the control group as the adoption of the program gets closer.

Table 6.Macroeconomic Differences Between Program Episodes in (t-1) and the Control Group: Nonparametric Statistical Analysis
Wilcoxon rank-sum:Kolmogorov-Smirnov:Kruskal-Wallis:
equality of mediansequality of distributionequality of populations
Zsignif.DSignif.HSignif.
I. Macroeconomic Outcomes
Balance of payments
(percentage of GDP)-12.490.0000.3980.000155.900.000
Current account
(percentage of GDP)-6.200.0000.2360.00038.490.000
International reserves
(months of imports)-11.740.0000.3940.000137.790.000
Consumer prices
(percentage change)3.660.0000.1540.00013.380.000
GDP per capita
(percentage change)-6.160.0000.2270.00037.990.000
Investment
(percentage of GDP)-5.170.0000.1610.00026.710.000
Real effective exchange rate
(percentage change)-4.080.0000.1410.00016.660.000
II. External Indicators
Terms of trade
(percentage change)-3.540.0000.1350.00112.570.000
Export markets
(percentage change)-3.780.0000.1440.00014.320.000
External debt service
(percentage of exports)11.610.0000.3540.000134.890.000
External debt
(percentage of GDP)13.000.0000.4040.000168.930.000
III. Macroeconomic Policy
Flow of government borrowing
(percentage of GDP)6.670.0000.2730.00044.530.000
Fiscal balance
(percentage of GDP)-5.660.0000.2340.00032.080.000
Domestic credit
(percentage change)0.330.7410.0520.5920.110.741
Broad money supply
(percentage change)-0.270.7840.0600.4060.080.784
Domestic credit to the government
(percentage change)4.040.0000.1540.00016.290.000
Nominal effective exchange rate
(percentage change)-5.530.0000.2120.00030.600.000

The nonparametric tests also corroborate the stylised facts detected for the external sector indicators (terms of trade, export markets and external debt variables). The null hypotheses of equality of medians, equality of distributions and equality of populations are all strongly rejected. In the case of this set of variables, differences between the program episodes and the control group also tend to increase as the program gets closer, differences being substantially more pronounced for the external debt indicators.

Results that emerge from the nonparametric tests performed on policy variables are even more Interesting. They confirm the observed differences between program and nonprogram observations in terms of fiscal and exchange rate policies, as well as the similarities with respect to monetary and credit policy. The null hypotheses of equality of medians, equality of distributions and equality of populations between the program episodes and the control group are strongly rejected for the flow of net government borrowing, the fiscal balance, the growth of domestic credit to the government and for the rate of nominal effective depreciation. However, the story is different for the selected monetary and credit indicators. In the case of the growth of total domestic credit the same three null hypotheses are weakly rejected (with confidence levels ranging between 10 and 20 percent) for three and two years prior to the inception of the program. But one year before the program, the control group and the program episodes seem to share the median domestic credit expansion, its distribution and even to be drawn from the same population. In the case of the growth of the money supply, it is clear that the groups are almost indistinguishable all through the three-year period before a program: none of the null hypotheses is rejected at conventional significance levels.

In sum, the nonparametric tests verify that the macroeconomic initial conditions of program episodes are significantly different in almost every respect from the empirical regularities observed in the control group. In particular, program episodes exhibit weaker balance of payments, growth, investment, external conditions and fiscal policy than the control group; they are also characterized by a higher degree of external indebtedness and inflation, while their exchange rates are more depreciated in both nominal and real terms than those of the control group. The program and control groups are different three years before the inception of a Fund-supported program, and these differences are accentuated in most cases as the program approval approximates. According to the sample of this study, the only possible exceptions to this pattern of statistical differences correspond to the growth rates of broad money and overall credit, where the two groups appear statistically similar.

2. Discriminant analyses

Having observed that program episodes and the control group differ basically in all their macroeconomic characteristics, an interesting question is to determine which of these indicators are able to discriminate better between the two groups. Such evidence will provide further insights on the controversial issue of the origins of macroeconomic imbalances that lead to a Fund-supported adjustment program, that is, whether macroeconomic outcome characteristics, external environments, or domestic macroeconomic policies are important in differentiating program episodes. In fact, the discriminant analysis can also be viewed as a first-pass signal detection exercise in the identification of Fund-supported programs: discriminant functions yield Illustrative “rules-of-thumb” that pick programs when a given variable exceeds an estimated threshold.

Before describing the results some preliminaries are necessary. In the discriminant analysis the two samples P^ and P2 from the program and nonprogram observations are merged. The space defined by the vector of macroeconomic indicators x Is then bisected into regions R^ and R2. These two regions are defined by the following conditions:

R^: Xj_ such that f(xj_) ≥ c*

R2: xi such that f(xj_) < c*

The classification rule is as follows: observation i is allocated to P]_ if x^ falls in R^ or to p£ if it falls in R2. For simplicity just a linear function is used so the discriminant function f(x^) becomes d’x^ and, under standard assumptions, the cutoff c* is 4 d’(X^+x^), where the vector of parameters d is obtained by maximizing the between-group sample variance relative to the within-group sample variance, and x^ is the sample mean of Pi. 1/ Observations are partitioned for each of the indicators discussed above in a series of univariate discriminant analyses.

In evaluating the discriminating power of each variable, three statistics commonly used in the signal detection literature are considered: the model sensitivity, specificity, and the area under the receiver operating characteristic (ROC) curve. 2/ Sensitivity is defined as the fraction of program episodes that are correctly classified by the classification rule (true positives). Similarly, specificity is defined as the fraction of nonprogram observations that are correctly classified (true negatives). A model thus discriminates perfectly the two groups in the sample if it yields unitary sensitivity and specificity rates. Except for the case of a perfectly discriminating model, a trade-off exists between sensitivity and specificity: varying the cutoff point improves the classification in one criterion but worsens it in the other. This trade-off is similar to the one encountered in statistical inference between Type I and Type II errors. The ROC curve graphs sensitivity versus (1-specificity) as the cutoff point is varied in a logit model--with the cutoff of the univariate linear discriminant function being just a special case--and thus captures the trade-off between the two criteria. 3/ The area below the ROC curve provides a summary indicator of the discriminating power of a model, being equal to unity for a perfectly discriminating model and 0.5 for a model with no discriminating power, When an economic indicator yields a higher area under the ROC locus, it implies that higher sensitivity can be achieved for the same specificity, which can be interpreted as a stronger signal and thus, as an economic Indicator that is more powerful in the detection of programs.

The results of the discriminant analysis for macroeconomic outcomes, external indicators and policy variables are presented in Table 7. Due to space limitations and similarity of results, only results pertaining to one year before the program vis-à-vis the control group are presented. Given the evidence presented earlier, the results are not that surprising. With respect to the macroeconomic outcomes indicators, the stock of international reserves turned out to be the most powerful variable in sorting out program and nonprogram observations. The linear discriminant model implied a cutoff of international reserves equal to 3.75 months of imports, below which observations were classified as program episodes and above which as nonprogram observations. Bisecting the sample in that way, the univariate model based on international reserves has a remarkable sensitivity, picking up almost 80 percent of the program episodes. Despite its rather low specificity, the stock of international reserves has a strong discriminating power as shown by the large area under the ROC curve (0.73).

Table 7.Macroeconomic Characteristics of Program Episodes in (t-1) and The Control Group: Univariate Discriminant Analysis
LinearModelModelArea
VariableCutoff 1/Sensitivity 2/Specificity 3/Under ROC 4/
I. Macroeconomic Outcomes
Balance of payments
(percentage of GDP)-1.6759.5775.320.7427
Current account
(percentage of GDP)-5.1650.6264.650.6206
International reserves
(months of imports)3.7579.9454.110.7282
Consumer prices
(percentage change)59.3311.1195.820.5711
GDP per capita
(percentage change)0.8756.4860.320.619&
Investment
(percentage of GDP)22.8262.3852.280.6033
Real effective exchange rate
(percentage change)1.3161.5645.010.5801
II. External Indicators
Terms of trade
(percentage change)1.3064.5146.900.5689
Export markets
(percentage change)3.2952.8057.470.5745
External debt service
(percentage of exports)19.6356.8375.610.7262
External debt
(percentage of GDP)48.7251.7181.100.7534
III. Macroeconomic Policy
Flow of government borrowing
(percentage of GDP)3.5546.1372.550.6421
Fiscal balance
(percentage of GDP)-6.4245.3768.250.6101
Domestic credit
(percentage change)97.087.4193.270.5064
Broad money supply
(percentage change)58.3910.8094.080.4947
Domestic credit to the government
(percentage change)332.993.7396.880.5790
Nominal effective exchange rate
(percentage change)-6.2841.5678.840.6085

Threshold derived from the linear discriminant function that bisects sample into the two different groups.

Percentage of program observations classified as such by the linear discriminant function (true positives).

Percentage of control group observations classified as such by the linear discriminant function (true negatives).

Area under the receiver operating characteristic (ROC) curve--the graph defined by varying the cutoff point in a logistic regression in the sensitivity versus (1-specificity) plane.

Threshold derived from the linear discriminant function that bisects sample into the two different groups.

Percentage of program observations classified as such by the linear discriminant function (true positives).

Percentage of control group observations classified as such by the linear discriminant function (true negatives).

Area under the receiver operating characteristic (ROC) curve--the graph defined by varying the cutoff point in a logistic regression in the sensitivity versus (1-specificity) plane.

Other macroeconomic outcomes do not fare as well as the stock of international reserves in sorting out the two groups from the sample. All the rest of the linear discriminant models have lower ability to recognize program episodes (i.e. they all have lower sensitivity), and only models based on the balance of payments, the current account, the inflation rate and the growth of per capita GDP pick better nonprogram observations (i.e. higher specificity) than the model based on international reserves. The particularly low sensitivity of the inflation rate--matched by a very high specificity–reflects the fact that the linear discriminant function is heavily influenced by outliers, as evidenced by the high cutoff inflation rate. A summary assessment of the discriminating power of these macroeconomic indicators based on the area under the ROC curve shows that possibly only the overall deficit in the balance of payments is as useful as the stock of international reserves in sorting out the two groups. In other words, except for international reserves and the balance-of-payments, no substantial discriminating gains are obtained by varying the cutoff point of other macroeconomic outcome variables: sensitivity losses are almost matched by specificity gains.

With respect to the external indicators, a linear discriminant function based on the terms of trade yields both the most sensitive and the least specific classification, the growth of export markets fares poorly on both counts, and the external debt variables are mildly sensitive but fare better on the specificity count. It is interesting, however, to note that the external debt variables are quite promising discriminators when varying the cutoff point In a logistic univariate regression, as illustrated by the high areas under their respective ROC curves. For example, if one reduces the cutoff of the external debt to GDP ratio by 5 percentage points in order to capture more programs, then the sensitivity rate improves by 7.5 percent with an associated sacrifice of k.2 percentage points in the specificity count.

Finally, considering the models based on the policy variables, the evidence confirms some of the previous results. Although neither of the linear discriminant functions seem to yield classifications that are very sensitive, models based on fiscal and exchange rate policy indicators perform better than those based on monetary and credit indicators. The latter are highly specific models and, as was the case of the inflation rate, seem to be affected by the influence of outlier observations. Considering the summary indicator of the discriminating abilities, the area under the ROC curves suggests again that fiscal and exchange rate policies are more capable of sorting out the two groups in the sample than monetary and credit policy indicators.

Some of the cutoff points estimated with the linear discriminant function have some interest in their own right. They provide a ready estimate of the threshold value for different macroeconomic indicators that separates program episodes from the control group. As noted earlier, observations with a stock of international reserves of less than 3.75 months of imports are classified by the univariate linear discriminant function as program episodes. Also illustrative are the results that when a deficit in excess of 1.67 percent of GDP in the balance of payments, or 5.16 percent of GDP in the current account is recorded, the observations are classified as program episodes; an external debt service exceeding 19.63 percent of exports, or an external debt above 48.72 percent of GDP, or fiscal deficits in excess of 6.42 percent of GDP also correspond to program episodes according to the univariate linear discriminant functions.

Overall, this exploratory investigation based on a univariate discriminant analysis suggests that the stock of international reserves, the overall balance of payments and the external debt variables are the most promising indicators in determining whether an observation belongs to either the program or the nonprogram groups. In principle, it should be possible to improve the discriminating power of the univariate methods presented here by using more general multivariate models, such as a broader discriminant function, multivariate logistic or probit regressions as is done in Knight and Santaella (1994). However, application of such methods lies beyond the basic stylized facts sought in this investigation.

The results of this statistical analysis can provide some insights on the issue of the origins of macroeconomic disequilibria before Fund-supported adjustment programs. As is natural to expect, the results seem supportive of a middle-ground view; disequilibria observed prior to programs have their origins both to domestic and external factors rather than one or the other. The fact that external debt variables are quite important in differentiating program episodes from the control group points to this conclusion, since these variables are the result of policy decisions in previous periods (as reflected in the accumulation of debt) as well as the result of movement in important exogenous variables such as the interest rate prevailing in international capital markets. This conclusion is reinforced by the documented behavior in the terms of trade, the fiscal stance, and in nominal exchange rate policies.

IV. Programs in the 1970s versus Programs in the 1980s

Perhaps one of the most interesting issues dealing with the empirical regularities before the inception of Fund-supported programs, is the question of whether the stylized facts described in the preceding sections have been relatively stable over time. The evidence presented in this section indicates that in fact some particularly stark features have differentiated initial conditions of Fund-supported programs in the 1970s from programs in the 1980s. This section compares macroeconomic outcomes, external indicators and macroeconomic policy variables prior to 91 Fund-arrangements during 1973-80 to these prior to the remaining 233 arrangements in our sample during 1981-91 (Tables 8-10).

Table 8.Macroeconomic Outcomes in Program Episodes in the 1970s and 1980s: Distribution of Samples
1973-801981-91
t-3t-2t-1t-3t-2t-1
Balance of payments
(percentage of GDP)
First quartile-2.1-2.9-3.9-6.0-6.4-6.3
Median-0.2-0.5-1.3-2.6-3.0-3.2
Third quartile1.6100.5-0.5-0.8-0.8
Mean-0.7-1.1-2.2-3.6-4.2-4.4
Current account
(percentage of GDP)
First quartile-7.5-8.2-8.9-9.6-8.8-8.5
Median-3.5-4.7-5.4-5.4-5.3-5.2
Third quartile-0.8-2.1-2.0-2.2-2.2-3.1
Mean-5.4-5.9-6.9-6.9-6.4-6.2
International reserves
(months of imports)
First quartile1.71.61.10.80.70.7
Median2.82.31.91.71.51.5
Third quartile4.03.63.43.43.32.9
Mean3.12.72.42.42.32.2
Consumer prices
(percentage change)
First quartile6.17.38.07.87.97.6
Median10.610.713.313.014.212.5
Third quartile17.421.423.028.028.329.5
Mean15.421.532.9140.367.6124.2
GDP per capita
(percentage of GDP)
First quartile-1.4-1.1-1.4-3.5-3.4-3.6
Median1.61.91.70.5-0.2-0.3
Third quartile5.44.24.53.02.51.9
Mean1.81.51.3-0.2-0.5-0.7
Investment
(Percentage of GDP)
First quartile15.615.515.515.114.414.4
Median21.823.724.721.119.318.9
Third quartile31.329.830.926.624.824.2
Mean23.423.624.021.320.320.0
Real effective exchange rate
(percentage change)
First quartile-4.5-4.5-7.8-6.7-7.4-8.0
Median1.21.20.2-1.6-1.1-1.8
Third quartile5.95.95.44.25.13.5
Mean2.12.1-0.1-1.1
Table 9.External Indicators in Program Episodes in the 1970s and 1980s: Distribution of Samples
1973-801981-91
t-3t-2t-1t-3t-2t-1
Terms of Trade
(percentage change)
First quartile-7.3-10.5-9.9-11.0-9.0-8.3
Median-0.3-1.8-2.2-2.8-1.4-2.6
Third quartile15.A7.96.8A.AA.63.6
Mean3.0-0.5-0.9-1.7-1.4-0.7
Export markets
(percentage change)
First quartile3.43.43.31.91.91.8
Median4.44.44.13.23.12.8
Third quartile5.25.15.03.93.73.6
Mean4.14.14.02.92.82.7
External debt service
(percentage of exports)
First quartile8.711.312.112.013.113.1
Median16.518.220.623.123.222.7
Third quartile24.431.030.834.437.235.6
Mean21.523.424.325.026.525.5
External debt
(percentage of GDP)
First quartile17.018.522.532.235.238.1
Median29.432.437.048.152.459.1
Third quartile42.548.554.472.579.287.5
Mean33.337.540.660.166.072.5
Table 10.Macroeconomic Policy in Program Episodes in the 1970s and 1980s: Distribution of Samples
1973-801981-91
t-3t-2t-1t-3t-2t-1
Flow of government borrowing
(percentage of GDP)
First quartile0.71.41.81.20.70.9
Median3.74.13.73.43.03.1
Third quartile6.46.88.26.76.36.2
Mean4.44.55.24.34.53.7
Fiscal balance
(percentage of GDP)
First quartile-10.8-9.9-12.5-9.9-9.5-9.2
Median-5.4-5.6-6.9-5.9-5.4-5.4
Third quartile-2.9-2.5-3.4-3.0-2.9-3.0
Mean-7.6-7.1-7.9-6.8-6.6-6.5
Domestic credit
(percentage change)
First quartile17.818.618.311.19.68.3
Median24.227.428.620.819.817.7
Third quartile35.941.641.634.836.133.4
Mean29.339.047.9162.458.6204.9
Broad money supply
(percentage change)
First quartile14.514.813.810.19.310.6
Median22.219.820.718.618.317.9
Third quartile31.630.232.231.727.633.6
Mean37.829.236.5113.254.6110.3
Domestic credit to the government
(percentage change)
First quartile8.414.613.49.43.95.1
Median25.732.539.325.519.819.7
Third quartile61.362.782.359.043.445.5
Mean73.776.0199.8-62.759.7871.2
Nominal effective exchange rate
(percentage change)
First quartile-8.0-8.0-10.8-15.4-16.5-19.5
Median-1.6-1.6-1.9-2.2—31-4.3
Third quartile1.31.31.31.42.42.9
Mean-3.9-3.9-8.0-9.9-10.6-11.5

The first interesting difference between the initial conditions of Fund-supported programs is that, despite the similarity of the deficits in the external current account, programs during 1981-91 registered substantially larger deficits in the overall balance of payments before the inception of the financial arrangements. In fact, balance of payments deficits twice as large during 1981-91 than during 1973-80 were reflected in lower stocks of international reserves.

This weakening of the member countries’ external position during the 1980s appears to be closely related to the external debt crisis and, to a lesser extent, to the lower rate of growth of the world economy. The differences in the external debt figures before the adoption of Fund-supported adjustment programs in 1981-91 vis-à-vis those in the earlier period are glaring. The median external debt In the immediate year preceding a Fund financial arrangement climbed from 37 percent of GDP in 1973-80 to a staggering 59 percent of GDP in 1981-91. With respect to the external debt service, the actual increase from one decade to the other is not as marked--perhaps due to the accumulation of arrears--and was basically concentrated in the higher end of the distribution.

It is interesting to note that developing countries undertaking adjustment programs in 1981-91 faced lower growth in export markets than they did in 1973-80 (the median growth rate of export markets declined from 4.1 percent to 2.8 percent). On the other hand, there are no appreciable differences between the two sample periods in the evolution of the terms of trade before the Inception of Fund-supported adjustment programs.

Macroeconomic adjustment to these adverse conditions on the external front Is evident from, on the one hand, the evolution of output, investment, the real exchange rate and to a lesser extent the inflation rate, and on the other hand, to the stance of macroeconomic policies. Dramatic differences across time are observed in the growth of GDP per capita and the investment rate. The median growth rate of GDP per capita before a Fund-supported program declined from 1.7 percent In 1973-80 to -0.3 percent in 1981-91, while the fall in the median investment to GDP ratio was from 25 percent to 19 percent.

The changing external conditions, chief among which was the debt crisis afflicting developing countries in the 1980s, also had its impact on real exchange rates. While the median real effective exchange rate was slightly appreciating before a Fund program in 1973-80, it depreciated during the later period. This finding is quite interesting, since it reconciles the conventional wisdom’s perception of a real appreciation before a Fund program with the behavior of real exchange rates prior to the debt crisis and the associated large increase in use of Fund resources. Of course, this result also shows that the earlier results for the whole sample period 1973-91 are driven by the characteristics of the substantially higher number of programs in the 1981-91 subperiod.

Curiously enough, programs in the lower half of the distribution of the inflation rate appear to exhibit fairly similar levels of price increases across the two decades. However, programs in the higher end of the distribution did experience higher inflation rates before programs during 1981-91 than during 1973-80. This is evidenced in the increase of the mean inflation rate before a program from 33 percent in 1973-80 to 124 percent in 1981-91.

The overall impression from the evolution of macroeconomic policies before a Fund financial arrangement, is that member countries followed somewhat tighter financial policies in 1981-91 than they did in the previous years. With respect to fiscal and budgetary policies, the behavior of the flow of government borrowing, the fiscal balance and the expansion of domestic credit to the government indicate a tighter fiscal stance in the 1980s than in the 1970s. For instance, the median fiscal balance before a Fund-supported program declined from almost 7 percent of GDP in 1973-80 to 5.4 percent of GDP in 1981-91. A similar conclusion can be reached with respect to monetary and credit policies. The median rates of growth of domestic credit and broad money before a financial arrangement declined from 29 percent and 21 percent in 1973-80 respectively, to 18 percent in 1981-91 for both monetary aggregates. However, it is noteworthy that some program episodes at the higher end of the distribution--as was the case with the inflation rate--did experience higher expansion of monetary aggregates in 1981-91 than before, as shown by the evolution of the mean rates of growth.

Finally program episodes during 1981-91 were preceded by higher rates of nominal effective depreciation than in the period 1973-80. The median nominal effective exchange rate fell almost 2 percent in 1973-80, but more than 4 percent in 1981-91.

V. Summary and Conclusions

This paper has examined the basic empirical regularities that distinguish macroeconomic performance before the approval of a Fund financial arrangement from the performance of countries that do not enter into a financial arrangement and are therefore assumed to have “normal” or “sustainable” macroeconomic conditions. This issue is important because it permits one to document the stylized facts that precede the adoption of a macroeconomic adjustment program, stylized facts that the conventional wisdom has many times inferred mostly from casual observation without any rigorous statistical analysis. Determining the characteristics of program episodes before the approval of a Fund arrangement would also provide some answers about the sources of the macroeconomic disequilibria that trigger the need for a Fund financial arrangement and the adjustment program that it supports.

Using a large sample of 324 Fund-supported programs during the 1973-91 period, evidence indicates that there are important differences in the macroeconomic characteristics between program episodes and the control group. The starting macroeconomic conditions prevailing in program episiodes are significantly different from the ones observed in the control group. Program episodes exhibit weaker balance of payments, output growth, investment, external conditions and fiscal policy than the control group; they are also characterized by a higher degree of external indebtedness and inflation, and their exchange rates are more depreciated in both nominal and real terms than those of the control group. Only in the case of the growth rates of broad money and overall credit do the two groups appear to be statistically similar. The discriminant analyses suggest that the stock of international reserves( the overall balance of payments and the external debt variables are some of the most promising indicators in detecting whether an observation belongs to either the program or the control group.

On the basis of the evidence presented here, one can put together a story about the average situation of countries that enter into Fund-supported programs. Countries that enter into a Fund-supported arrangement seem to follow more expansionary fiscal policies characterized by relatively large fiscal deficits. According to the sample in this study, the median country copes with such a seemingly unsustainable fiscal policy by increasing the public sector liabilities both with domestic and foreign creditors, trying to keep a “tight” monetary policy and by depreciating the nominal effective exchange rate. This policy stance seems to avoid an explosive rise in inflation in the short-run, but it does so at the expense of crowding-out the private sector in the credit market, depressing the rate of investment and slowing down the growth of output. On top of these conditions in the domestic economy, the median program episode in the sample is also affected more severely by external shocks than the control group: terms of trade, growth of export markets, but especially by conditions in international capital markets, behave relatively more adversely for program episodes. Any negative impact coming from this external environment accelerates the unravelling of the precarious conditions originating from domestic developments, and precipitates a balance of payments crisis. It must be recognized, however, that the “story” is just that; in reality, actual developments leading to the adoption of a Fund-supported program would vary significantly from case to case as shown by the high variability displayed by program episodes.

The stylized facts have changed over time as evidenced by the documented differences across decades. Comparing the period before Fund programs in 1973-80 with programs in 1981-91, it turns out that initial positions in the latter period were characterized by tougher external conditions, namely the developing countries’ debt crisis and the sluggish expansion of export markets. This situation was reflected in weaker balance of payments, lower output growth and investment, more depreciated real exchange rates and somewhat tighter macroeconomic policies before the adoption of Fund arrangements in the 1981-91 period than in the 1973-80 period.

Characterization of these macroeconomic stylized facts in the period prior to the inception of Fund financial arrangements should have two important ramifications. First, it should improve the design of macroeconomic policy oriented towards tackling economic disequilibria that give rise to the need for Fund arrangements. Second, by providing additional benchmarks, these stylized facts should also improve the ex-post evaluation of Fund programs. It should be clear from the results presented here that in order to evaluate Fund-supported programs using a “with-without” approach, considerable care should be taken to control for the large differences between program and nonprogram countries.

TECHNICAL APPENDIX

With two subsamples P1 and P2 of sizes n1 and n2, let T be the sum of the ranks when a combined sample of n = n1 + n2 observations is ordered in ascending order of magnitude. T is called the Wilcoxon rank-sum statistic, which can be approximated by a standard normal distribution:

where the expected value of T is

and the standard deviation is

For the Kolmogorov-Smirnov test, let F(x) and G(x) be the empirical distributions for the sample to compare. Define the following statistics

then the combined statistic is

which can be shown to have the following asymptotic P-value

For the Kruskal-Wallis test with two samples, let and T^ and be the sum of the ranks for each of the subsamples. Define the following statistic

which has a sampling distribution that is approximately x^ with one degree of freedom.

In the discriminant analysis a linear function d’x is found such that it provides the best discrimination between two groups P^ and P2. Define

The vector of parameters d is selected by maximizing the variance between groups relative to the variance within groups

which gives

The means of the discriminant function in the two samples are:

Given an observation x^, one computes the discriminant function evaluated at that point,

and assigns it to P1 if yi is closer to, than to ȳ1, than to ȳ2; that is, if yi>½(ȳ12) then yi is assigned to P1, otherwise yi is assigned to P2.

Define the sensitivity rate as the fraction of predicted program observations given that they were actually program observations, P(program|program); and the specificity rate as the fraction of predicted control group observations conditional on belonging to the control group, P(control|control). Then the area under the ROC curve is:

References

    AmemiyaTakeshi (1985) Advanced EconometricsCambridge: Harvard University Press.

    BagciPinar and WilliamPerraudin (1995) “Do IMF Programmes Work?”miraeoUniversity of Cambridge.

    BirdGraham (1990) “Evaluating the Effects of IMF-Supported Adjustment Programmes: An Analytical Commentary on Interpreting the Empirical Evidence,” in International Finance and the Less Developed Countriesed. byKatePhylaktis and MahmoodPradhanNew York: St. Martin’s Press.

    • Search Google Scholar
    • Export Citation

    BirdGraham (1995) IMF Lending to Developing Countries: Issues and EvidenceLondon: Routledge.

    BirdGraham and TimothyOrroe (1981) “An Analysis of Drawings on the International Monetary Fund by Developing Countries,”World DevelopmentVol. 9 (June) pp. 56368.

    • Crossref
    • Search Google Scholar
    • Export Citation

    BuiraAriel (1983) “IMF Financial Programs and Conditionality,”Journal of Development EconomicsVol. 12 (February-April) pp. 11136.

    • Crossref
    • Search Google Scholar
    • Export Citation

    ConwayPatrick (1994) “IMF Lending Programs: Participation and Impact,”Journal of Development EconomicsVol. 45 (December) pp. 36591.

    • Crossref
    • Search Google Scholar
    • Export Citation

    CorneliusPeter (1987) “The Demand for IMF Credits by Sub-Saharan African Countries,”Economic LettersVol. 23No. 1 pp. 99102.

    CoxD.R. and E.J.Sneli (1989) Analysis of Binary DataNew York: Chapman and Hall.

    de VriesMargaret G. (1976) The International Monetary Fund 1966-71: The System Under Stress Vol. I: NarrativeWashington, DC.

    de VriesMargaret G. (1985) The International Monetary Fund 1972-78: Cooperation on Trial Vols. I and II: Narrative and AnalysisWashington, DC.

    • Search Google Scholar
    • Export Citation

    de VriesMargaret G. (1987) Balance of Payments Adjustment 1945 to 1986: The IMF ExperienceWashington, DC.

    EdwardsSebastian and Julio A.Santaella (1993) “Devaluation Controversies in the Developing Countries: Lessons from the Bretton Woods Era,” in A Retrospective on the Bretton Woods Systemed. byMichael D.Bordo and BarryEichengreenChicago: University of Chicago Press.

    • Search Google Scholar
    • Export Citation

    EganJames (1975) Signal Detection Theory and ROC AnalysisNew York: Academic Press.

    GoldsteinMorris and PeterMontiel (1986) “Evaluating Fund Stabilization Programs with Multicountry Data: Some Methodological Pitfalls”Staff PapersVol. 33No. 2 (June) pp. 30444.

    • Crossref
    • Search Google Scholar
    • Export Citation

    GreenD.M. and J.A.Swets (1974) Signal Detection Theory and FsychophysicsHuntington, NY: Krieger.

    GylfasonThorvaldur (1987) “Credit Policy and Economic Activity in Developing Countries with IMF Stabilization Programs”Princeton Studies in International FinanceNo. 60Princeton: Princeton University.

    • Search Google Scholar
    • Export Citation

    HorsefieldJ. Keith (1969) The International Monetary Fund 1945-65 Vol. I: ChronicleWashington, DC.

    HorsefieldJ. Keith and Margaret G.de Vries (1969) eds. The International Monetary Fund 1945-65 Vol. II: AnalysisWashington, DC.

    JoyceJoseph (1992) “The Economic Characteristics df IMF Program Countries,”Economic LettersVol. 38 (February) pp. 23742.

    KhanMohsin (1990) “The Macroeconomic Effects of Fund-Supported Adjustment Programs”Staff PapersVol. 37No. 2 (June) pp. 195231.

    KillickTony and MoazzamMalik (1992) “Country Experiences with IMF Programmes in the 1980s”The World EconomyVol. 15 (September) pp. 599632.

    • Crossref
    • Search Google Scholar
    • Export Citation

    KillickTony and MarcusManuel (1992) “What Can We Know About the Effects of IMF Programmes?”The World EconomyVol. 15 (September) pp. 57597.

    • Crossref
    • Search Google Scholar
    • Export Citation

    KnightMalcolm and Julio A.Santaella (1994) “Economic Determinants of Fund Financial Arrangements,”IMF Working Paper 94/36 (March).

    • Search Google Scholar
    • Export Citation

    MaddalaG.S. (1983) Limited-Dependent and Qualitative Variables in EconometricsCambridge: Cambridge University Press.

    McDonaldJudith (1986) “Factors Influencing Countries’ Decision to Use Credit From the International Monetary Fund,”Ph.D. dissertationPrinceton University.

    • Search Google Scholar
    • Export Citation

    SachsJeffrey (1989) “Conditionality, Debt Relief and Developing Country Debt Crisis,” in Developing Country Debt and Economic Performanceed. byJeffreySachsChicago: University of Chicago Press.

    • Crossref
    • Search Google Scholar
    • Export Citation
1/Comments from Patrick Conway, Nadeem Haque, Mobsin Khan, Don Mathieson, Miguel Savastano and Peter Wickham are gratefully appreciated, Ravina Malkani, Brooks Calvo and Kote Nikoi provided helpful research assistance. The views expressed here do not necessarily represent those of the International Monetary Fund.
1/In fact, the last time major industrial countries received financial assistance was in 1977 when the stand-by arrangements for Italy and the United Kingdom were approved.
2/Other special facilities that have been used by member countries include the buffer stock financing facility and the temporary oil facilities in the mid-1970s. For a historical analysis of the use of Fund resources and its experience with balance-of-payments adjustment, see Horsefield (1969), Horsefield and de Vries (1969) and de Vries (1976, 1985, 1987).
1/For instance, Buira (1983) and Bird (1990) describe the situation before the adoption of an adjustment program as characterized by imbalances in an economy that are usually reflected in the prevalence of inflation and balance of payments deficits. See also Sachs (1989) and Edwards and Santaella (1993) for more elaborate depictions of the conventional wisdom.
2/Recent papers on the macroeconomic effects of adjustment programs supported by Fund arrangements are Khan (1990), Killick, Malik and Manuel (1992), Killick and Malik (1992), Conway (1994), and Bagci and Perraudin (1995).
1/For a study of Fund-supported adjustment programs in the context of 48 devaluation episodes during the Bretton-Woods period, see Edwards and Santaella (1993).
2/On the evaluation of macroeconomic programs, see Goldstein and Montiel (1986) and Khan (1990).
1/For instance Gylfason (1987) attempted to select a control group that confronted balance of payments difficulties similar to the ones faced by Fund-supported programs. For a methodological discussion, see Goldstein and Montiel (1986) and Khan (1990).
1/Histograms for the program episodes in “t-1” and for the control group are presented in Figure 4 in the Appendix.
2/See also the histograms in the Appendix, where it is apparent that the program episodes’ distributions of the balance of payments, stock of international reserves and current account lie to the left of the distributions of the control group--i.e. technically they are first order stochastically dominated.
1/Definition of these tests can also be found in the Appendix.
1/Nonparametric tests for “t-2H and “t-3”--available upon request--are very similar to the tests for “t-1”.
1/See Maddala (1983), Amemiya (1985) and the Appendix.
3/In the logit model the predicted probability of a positive outcome for observation I is:
An observation i is classified as positive if p^≥ p, and otherwise is classified as negative. The usual assumption is that px = 4, but in the ROC curve the cutoff p* is varied to yield different classifications. For the relation between discriminant and logit analysis, see Cox and Snell (1989).

Other Resources Citing This Publication