Revisiting IMF expenditure conditionality

¹ IEO (Citation2003) estimates that the average targeted fiscal balance improved by about 1.7% of GDP over two years in the IMF-supported programmes during the 1993–2001 period.

² Please see the Data and Stylized Facts section for definitions of different types of IMF conditionality. See also https://www.imf.org/en/About/Factsheets/Sheets/2016/08/02/21/28/IMF-Conditionality.

³ The focus on poverty reduction in LIDCs partly explains the higher number of conditions in LIDC programmes (IMF Citation2017a).

⁴ See Appendix B for detailed statistics on the application and conformity to expenditure conditionality.

⁵ Examples of exploring long-run effects of non-structural conditionality are: effect of conditionality on health spending on increasing health expenditures, and impact of conditionality on wage spending on containing the wage bill.

⁶ As stated by Stubbs et al. (Citation2020), the literature and their study enrich our understanding of the consequences of conditionality, they tell us less about the effect of IMF technical assistance.

⁷ Conditionality classified under Public Financial Management. Details are available in Appendix A.

⁸ Dabla-Norris, Ho, and Kyobe (Citation2016) find that productivity and economic growth in emerging markets could benefit substantially from tailored institutional and structural reforms.

⁹ For our purposes, it does not matter if IMF programmes are consecutive or not as the study’s focus is on exploring the long-run impact of conditionality.

¹⁰ The data are unreliable prior to 1992, which then determined the study’s starting point.

¹¹ Our data suggests that expenditure conditionality is met at a higher rate compared with other types of conditionality.

¹² To be consistent across different estimations, we set the number of lags to 2. Incorporating more lags would not be possible for the CS-ARDL approach as the time series dimension of our data is not large enough.

¹³ In fact, it allows for causality to run both ways (Chudik and others Citation2018).

¹⁴ Regarding the quantitative spending categories, we explore only those spending conditions that could have potential impact on the spending category of interest. For instance, for wage outlays as dependent variable, we have explored the impact of conditionality on wage expenditures, or for health and education spending, we have investigated the impact of spending floor on social spending.

¹⁵ The sample of countries varies from one regression to another due to two factors. First, the sample for each regression depends not only on the availability of the dependent and explanatory variables, but also whether or not a country had the conditionality of interest during its programme. If a country did not have specific condition during IMF programme, the coefficient of interest would not be identifiable for the regression for that country. Second, for each dependent variable, the data are winsorized and outliers are dropped. Therefore, depending on the dependent variable and whether it is scaled by GDP or government expenditure, the regressions may include few more or less countries.

¹⁶ Estimations are available upon request.

¹⁷ As explained subsequently, one should note that reported social expenditures are not necessarily uniformly defined across countries. In addition, other expenditures may have been classified as social expenditures because of the IMF programme conditionality.

¹⁸ We also re-ran our regressions for the sub-sample of countries classified as fragile. Given the small sample size, only few results are found to be statistically robust. They show that general PFM conditionality (such as developing a PFM strategy, monitoring operations and financial operations of public enterprises) has a positive impact on enhancing public investment; a reduction in expenditure arrears creates fiscal space for productive spending in the long term; and improved budget preparation systems help increase the share of education sector in total budget outlays. This shows that a long-term institutional development helps improve expenditure outcomes in fragile states.

¹⁹ We repeat the regressions for wage spending in real per capita terms and find that a minimum floor on social spending, especially in low-income countries, lowers them in the long-run, suggesting that social spending floors lead governments to have a better mix of wage and non-wage costs in social sectors (results not reported in the paper and are available upon request). One should note that the definition of social spending varies across countries and may include spending on social safety nets and health and education sectors. The precise coverage of social sector spending depends on the agreement between the authorities and IMF staff for each programme. A thorough investigation of the evolution of each component is necessary to understand how floors on social spending have impacted other budget components such as overall wage outlays.

²⁰ The range is obtained based on the size of the statistically significant coefficients.

²¹ As of October 2016 IMF World Economic Outlook.

²² All quantitative measures are ceilings except for social protection (social spending) and payment of arrears.