Using a cross-country sample of 169 IMF-supported programs and detailed studies of 15 programs, this evaluation report examines various aspects of fiscal adjustment in IMF-supported programs. It presents evidence that does not support some critics’ view that IMF-supported programs typically adopt a one-size-fits-all approach to fiscal adjustment, nor the perception that programs always involve austerity by targeting reductions in public spending. The report also proposes a number of recommendations for IMF surveillance and program design in the future.

Appendix 1

Table A1.1.

Determinants of the Envisaged and Actual Fiscal Adjustment (T−1 to T+1) in IMF-Supported Programs

article image
Note: Equation estimated through ordinary least squares with White-corrected (heteroskedasticity-consistent) standard errors. *, **, and *** denote significance at the 90 percennt, 95 percent, and 99 percent confidence levels, respectively.
Table A1.2.

Determinants of the Differences Between Envisaged and Actual Fiscal Adjustment

article image
Notes: Equation estimated through ordinary least squares with White-corrected (heteroskedasticity-consistent) standard errors. * = significant at the 99 percent confidence level.Definition of variables∆GBALA – ∆GBALE : Difference between actual and envisaged changes in the fiscal balance from T–1 to T+1.GrowthAT+ 1 – GrowthET+ 1: Differences between actual and envisaged real GDP growth at year T+1.GBALAT– 1 – GBALET– 1: Difference in the fiscal balance between the WEO (actual) and MONA (envisaged) databases.Transition: Dummy for transition countries.

Appendix 2: Code Book for Assessing the Need for Fiscal Adjustment

Do program documents clearly explain the source of the existing or potential balance of payments problem motivating the program?1

  • “Unsatisfactory”: The program document provides no explicit reference to any existing or impending external imbalance either from a flow or stock type that the program aims to correct or prevent.

  • “Marginally satisfactory”: The program document makes some quick reference to an existing or possible external imbalance, but does not provide any detailed discussion of the problem. The reader is therefore unclear about whether there is a balance of payments problem, what the nature of the problem is, and how the program is expected to correct it.

  • “Satisfactory”: The program document identifies, discusses, and critically analyzes the sources of the balance of payments problem the IMF-supported program is trying to correct. The document clearly explains the nature of the balance of payments problem calling for IMF involvement and the strategy that the program will follow to tackle it.

  • “Highly satisfactory”: In addition to the characteristics under “satisfactory,” the program document would clearly identify whether the external financing gap calling for IMF involvement resulted from a current or capital account deficit and whether it stemmed from the public or private sector.

In light of the above, do documents explain the country-specific mechanism by which the fiscal adjustment will help improve the balance of payments problem (or more generally the problem that called for the Fund’s involvement)?

  • “Unsatisfactory”: The program document makes no reference to the country-specific mechanism through which the envisaged fiscal adjustment will assist in solving or preventing the problems associated with the external imbalance.

  • “Marginally satisfactory”: The program documents refer to a possible link between fiscal adjustment and the external problems and imbalances mentioned above but provide virtually no discussion of how the mechanism that links the two will operate.

  • “Satisfactory”: The program document clearly describes and explains the mechanism through which the envisaged fiscal adjustment is going to contribute to solve or prevent the existing or possible balance of payments problem.

  • “Highly satisfactory”: Same as in previous category, but the program either provides a comprehensive analysis of these questions or includes a medium-term assessment of the relationship between these two variables.

Do documents explain the factors determining the pace and magnitude of the fiscal deficit adjustment, in particular its magnitude relative to the envisaged current account adjustment (e.g., fiscal adjustment as a fraction of the total adjustment)?

  • “Unsatisfactory”: Program documents do not compare the direction and size of the change in the fiscal and current account balances over the life of the program.

  • “Marginally satisfactory”: Program documents make some connection between how the magnitude of the envisaged fiscal adjustment is related to the magnitude of the envisaged current account adjustment, but provide practically no explanation or analysis of the envisaged joint evolution of these variables. Alternatively, a program document that makes no verbal connection between these two indicators but provides a table with information on the evolution of saving and investment balances of both the public and private sector has also been classified here.

  • “Satisfactory”: The program document provides a clear sense of the pace of “burden sharing” between adjustment in the private and public sector.

  • “Highly satisfactory”: Same as “satisfactory,” but the document also provides an analysis of the factors affecting the likely evolution of the current account, fiscal deficit, and private savings-investment balance, including a medium-term table with disaggregated data on savings and investment of the public and private sector.

If there are other factors influencing the envisaged fiscal deficit adjustment (other than balance of payments considerations), do documents explain clearly how they influence that adjustment?

  • “Unsatisfactory”: The program documents do not point out which macroeconomic imbalances or problems, if any, the envisaged fiscal adjustment is expected to correct, or why a reduction of the fiscal deficit under the program is the appropriate economic policy to follow.

  • “Marginally satisfactory”: The program documents give some general reasons why the fiscal adjustment might be necessary (high inflation, debt sustainability, and financing problem) but the language is vague and does not analyze the problem with sufficient detail.

  • “Satisfactory”: The program documents provide a clear explanation of the objectives of the fiscal adjustment in terms of some well-defined macroeconomic objective (free resources for the private sector, reduce inflation, and bring the public debt to a sustainable path) and the reader is given a good and unequivocal sense of why the fiscal adjustment is necessary.

  • “Highly satisfactory”: The document not only provides a good analysis of why the fiscal adjustment is necessary but also a clear explanation of why the precise magnitude of the envisaged adjustment being proposed (and not some other magnitude) is necessary.

Do documents explain the rationale for the composition of the fiscal deficit adjustment? In other words, is there a good explanation of why the adjustment has to be done through revenues or expenditures or a combination of the two?

  • “Unsatisfactory”: The program documents provide a list of expenditure and revenue measures associated with the fiscal deficit reduction, but do not explain why the burden of adjustment has to fall on revenues and expenditures; or how the specific share of adjustment revenue and expenditures has been designed.

  • “Marginally satisfactory”: The program documents refer to how the adjustment will be effected (including a sense of the envisaged revenue and expenditure changes), but do not provide a clear rationale of why this specific composition between revenue and expenditures is optimal or necessary.

  • “Satisfactory”: The program documents provide a clear sense of why the specific composition of the adjustment (between revenue and expenditures) is the appropriate one. It includes indicators of what percentage of GDP specific revenue and expenditure measures are going to yield.

  • “Highly satisfactory”: In addition to providing a good explanation of the envisaged composition of the adjustment, the documents provide some analysis of the structure of revenue and expenditure (aimed at identifying major weaknesses in the structure of public finance) and a relatively detailed analysis of how intra-revenue or intra-expenditure changes are going to contribute to the adjustment.


In the less likely case that the IMF-supported program did not respond to a balance of payments difficulty, the same criteria would apply but with regard to the specific reasons that motivated the program.

Appendix 3

Table A3.1.

Levels of Grants in a Sample of Sub-Saharan African Countries

article image
Source: Program documents.
Table A3.2.

Changes in Levels of Grants

article image
Source: Program documents.
Table A3.3.

Aid Flows Under IMF-Supported Programs, 1995–2001

article image
Sources: Program documents, and IEO staff estimates.

Appendix 4: Explanatory Variables and Methodological Issues in the Analysis of Social Spending in IMF-Supported Programs

In order to appropriately assess the impact of the IMF on social spending using a multivariate regression framework, we need to take into account at least three methodological problems: (1) missing variable bias, (2) serial correlation and nonstationarity, and (3) the endogeneity of IMF-supported programs (for a more extensive discussion of these methodological issues, including an analysis of alternative estimating techniques such as the Generalized Evaluation Estimator, see Martin and Segura-Ubiergo (forthcoming).

To avoid a missing variable bias, the following control variables were defined using data from the World Bank’s World Development Indicators and the IMF’s World Economic Outlook (see Table A4.4 of this appendix for the summary statistics, including means for the “with IMF” and “without IMF” groups). Two other control variables (health_priv and ca_y) had insignificant coefficients and were excluded from the final regressions.

article image

The above control variables explain some of the differences in spending between countries, but there may be residual country differences in spending not captured by them. To account for that, the empirical model was also estimated with country dummies (fixed effects), that is, which allowed for a different level of average spending for each country.

To address the problem of serial correlation and nonstationarity we used a dynamic model that clearly separates short- and medium-term effects. Although there are different models that can serve this purpose, we decided to use an Autoregressive Moving Average process (ARIMA), which seemed to fit the data rather well. A first-order process on the dependent and independent variables was enough to obtain residuals without further detectable serial correlation or unit roots. The following equation was used:


where Sit denotes social spending in country “i” and period “t”, Xit is the vector of exogenous variables defined above, and IMFit measures the presence of an IMF-supported program as proxied by the instruments defined below. L is the lag operator (i.e., LZZt1 , for any variable Z), D is the first-difference operator (DZtZtZt1 ) and uit are the residuals.

An alternative and equivalent way of writing (1) is:


where (1γ)·α2=α1and(1γ)·β2=β1. In this specification, changes in the dependent variables, D Sit, can be seen as the result of two effects: contemporaneous change in the explanatory variables (with an impact determined by the coefficients α1andβ1) ; and gradual adjustment to an “equilibrium” level of spending, determined by the coefficients α2andβ2 . Transitory changes in the independent variables do not change the long-run “equilibrium” level, so that the effect decays geometrically at the rate (1γ) after the second period.

To address the endogeneity issue, the following instruments were used to “predict” the presence of an IMF-supported program:

  • current account deficit as fraction of GDP in the previous year (as proxy of external crisis);

  • growth in the previous year (proxy of unsustainable expansion);

  • income per capita (IMF-supported programs less likely in high-income countries);

  • presence of an IMF-supported program in the previous year;

  • government balance as share of GDP in the previous year; and

  • democracy index (as in the control variables).

To explore the robustness of the result we compared the results with those obtained with alternative estimation methods and with different subsamples of countries (see Table A4.2 and Table A4.3).

Table A4.1.

ARIMA Model with Control Variables and Endogenous IMF-Supported Programs

article image
Note: See the text for variable definitions. An initial L indicates a lagged value and D the first difference. IMF(predicted) is the estimated value of the IMF variable with the following instruments: lagged values of IMF, growth, CA/GDP, government balance/GDP, democracy index, and GDP per capita in U.S. dollars. The actual estimated equation is IMF(predicted)=0.148+0.696IMF(-1)-0.003growth(-1)+0.001ca-y(-1)+0.001cgbal(-1)-0.043democracy-0.011.gdpusdpc(41.94***)(-2.58***)(-0.69)(0.60)(-3.26***)(-4.85***)



Table A4.2.

Summary of Robustness Analysis

article image
Table A4.3.

Summary of Regression Results

article image
article image
article image
Note: IMF variable measured as proportion of the years under an IMF-supported program. The number of asterisks indicates the significance level for the test that the coefficient is different from zero: *** for 99 percent, ** for 95 percent, and * for 90 percent.

Estimate of serial correlation of the regression.

CONC = Stand-By or EFF programs; NONCONC = SAF, ESAF, or PRGF programs.

Table A4.4.

Control Variables for Social Spending

article image

Statistically significant differences in means are indicated by *** (99 percent confidence level) or **(95 percent).

Table A4.5.

List of Countries and Subsamples

article image
article image
Note: S1 = one to five years; S2 = one to ten years; and S3 = five or more years.