Policy Research Working Paper 9208 Moving Up the Ladder An Analysis of IDA Graduation Policy Anton Dobronogov Stephen Knack James Wilson Development Finance Vice-Presidency April 2020 Policy Research Working Paper 9208 Abstract This paper analyzes the factors that affect countries’ gradu- levels of poverty and social indicators. Through a pooled ation from International Development Association (IDA) logit estimation using panel data covering IDA-eligible assistance and develops a statistical model of graduation. countries for 1987–2016, the authors determine the fac- IDA provides concessional financing (credits, grants, tors that influenced IDA graduation decision making for and guarantees) to the world’s poorest countries to help FY1989–FY2018. They find that throughout the sample reduce poverty and improve living standards. IDA’s eli- the probability of being a graduate is positively and sta- gibility criteria include (a) absence of International Bank tistically significantly associated with income per capita, for Reconstruction and Development (IBRD) creditwor- creditworthiness, and country size. They account for the thiness and (b) gross national income (GNI) per capita shift of policy after 1998. Using an interaction dummy vari- below the IDA operational cutoff. Following several ‘reverse able to capture the pre-FY1999 period, they find that life graduations’ during the 1990s, the set of factors affecting expectancy, reduced poverty, urbanization, and institutional graduation decisions broadened to include an assessment development have been positive and significant predictors of the country’s macroeconomic prospects, risk of debt dis- of graduation status in the later period. tress, vulnerability to shocks, institutional constraints, and This paper is a product of the Development Finance Vice-Presidency. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at adobronogov@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Moving Up the Ladder: An Analysis of IDA Graduation Policy Anton Dobronogov †, Stephen Knack ‡, and James Wilson § Keywords: graduation from foreign aid, IDA, International Development Association, statistical model JEL classification codes: F3, O29 † World Bank. Corresponding author, email: adobronogov@worldbank.org ‡ World Bank. § Oxford University The authors are grateful to Stephane Guimbert and Phil Keefer for useful comments. Any remaining errors are our own. The views expressed in this paper are those of the authors, and they should not be attributed to the World Bank, its Executive Directors, or the countries they represent. 2 I Introduction This paper develops a statistical model of graduation from International Development Association (IDA) assistance and uses it to analyze the factors that affect countries’ graduation as well as the effects of the shift in graduation policy which occurred in FY1999. 1 IDA provides concessional financing (credits, grants, and guarantees) to the world’s poorest countries to help reduce poverty and improve living standards. IDA’s eligibility criteria include (a) absence of International Bank for Reconstruction and Development (IBRD) creditworthiness and (b) gross national income (GNI) per capita below the IDA operational cutoff (US$1,175 for FY2020). Before FY1999, the two major criteria for graduation from IDA were creditworthiness for IBRD for several years and a level of GNI per capita that exceeded the operational cutoff set in 1987 and adjusted since then annually based on GNI per capita calculations in accordance with the Atlas conversion factor methodology. However, the experience of eight reverse graduations which occurred in the 1990s, including that of Indonesia following the Asian financial crisis, led to a change in policy in FY1999, intent on limiting future reverse graduations. The set of factors affecting graduation decisions broadened to include other factors to account more precisely for country-specific circumstances. The only reverse graduation after this policy change was that of Syria in 2017. Similar to the investigation of IBRD graduation policy conducted by Heckelman, Knack, and Rogers (2012), this paper takes a positive approach to IDA graduation policy, analyzing factors that influenced IDA’s decisions on graduations and reverse graduations over the past three decades through a pooled logit estimation using panel data covering IDA-eligible countries for 1987–2016. We find that throughout the sample the probability of being a graduate is positively and statistically significantly associated with income per capita, creditworthiness, and country size. We account for the shift of policy in FY1999. Using an interaction dummy variable to capture the pre-FY1999 period, we find that life expectancy, reduced poverty, urbanization, and institutional development have been positive and significant predictors of graduation status in the later period. II Background IDA graduation policy aims to strike a balance between two objectives. The demand for IDA resources outstrips supply, and so there is a need to graduate countries as soon as they are ready, to focus the assistance where it is needed the most. It is also important to avoid reverse graduations—making 1 Denotes fiscal year 1999. World Bank fiscal years run from July 1 to June 30, and thus, FY2019 ended on June 30, 2019. 3 graduated countries eligible again for IDA would be (a) adversely affecting reputations of the countries and their governments and (b) disruptive for planning purposes. Currently, IDA employs a flexible multistage graduation process that relies on careful case-by-case analysis of specific country situations. The IDA graduation process involves multiple stages, offering countries an opportunity to gradually adjust to tighter terms of financing. A country’s transition from IDA to IBRD usually proceeds the following way: • IDA-only non-gap to IDA-only gap. Countries that have been above the IDA operational cutoff for more than two years but are not yet deemed creditworthy for IBRD financing are classified as IDA-only ‘gap’ countries. • IDA-only non-gap or IDA-only gap to blend. A positive creditworthiness assessment by IBRD leads to reclassification of a country from IDA-only non-gap or IDA-only gap status to blend status (IDA/IBRD). The assessment needs to be requested by the country. • Blend to IBRD-only. The IDA graduation process concludes 2 with a reclassification from blend status to IBRD-only borrower. The IDA decision to graduate a country to IBRD-only status is currently based on an assessment of the country’s macroeconomic prospects, risk of debt distress, vulnerability to shocks, institutional constraints, and levels of poverty and social indicators. ‘Graduation from IDA’ refers to completion of the latter stage when a country becomes ineligible for concessional 3 financing from the World Bank. Understanding how and when countries graduate from IDA borrower status is important in forecasting future demand for IDA resources and in anticipating better the risk of reverse graduation. The post-FY1999 broadening of the set of factors used to assess potential IDA graduation cases has aimed to reduce the likelihood of reverse graduations, which almost always result from crises: economic, political, or both. A wide range of findings in academic literature support the idea of such broadening. Usually reverse graduations occur when GNI per capita of a country falls below (or close to) IDA operational cutoff. Typically, this involves a major currency devaluation. Macroeconomic imbalances and risks of debt distress are the most direct and immediate causes and predictors of such devaluations in 2 During the IDA17 and IDA18 replenishments, four graduates (India, Bolivia, Sri Lanka, and Vietnam) have been receiving transitional support from IDA (credits on non-concessional IBRD terms) during the first three years after graduation. Past IDA commitments to all graduates continue to disburse as projects progress. 3 According to the World Bank methodology, financing is concessional if it includes a grant element of 35 percent or more, calculated at a 5 percent discount rate. 4 middle-income countries (Ahmed, Coulibaly, and Zlate 2017; Presbitero 2017; Reinhart, Reinhart, and Rogoff 2012). Political instability increases risks of economic crises (Bussière and Mulder 2000). Institutional quality helps avoid macroeconomic imbalances and increase resilience of the economies to shocks. This argument has been robustly supported by the literature. Acemoglu et al. (2003) find that countries that inherited more ‘extractive’ institutions from their colonial past were more likely to experience high volatility and economic crises during the postwar period. Dabla-Norris and Gündüz (2014) find institutional quality to be the strongest predictor of a growth crisis. Collier and Goderis (2012) also show that vulnerability to negative shocks is heightened in resource-rich countries with weak institutions, while Elbadawi and Soto (2016) find the vulnerability of oil-rich countries to be conditional on bad political governance. Low social indicators and high poverty rates do not help such resilience and may mean that an economic crisis of a smaller magnitude results in a reverse graduation because of its larger adverse social impact. Several studies highlight the implications for economic stability arising from households’ vulnerability to poverty as they are less able to cope with economic fluctuations due to limited access to risk-sharing mechanisms and capital markets (Calvo and Dercon 2013; Dercon et al. 2012). Briguglio (2016) stresses the importance of social indicators in economic vulnerability and economic resilience indexes. III IDA Graduation Model In this section, we develop a statistical model of IDA graduation. Sample Panel data were used for the sample period 1987–2016 to generate predicted probabilities of graduation status. Countries were included only if they had been IDA eligible for at least one year during the sample period and had a population of over 1.5 million. 4 In addition, country-year observations were included only if the GNI per capita for that year was at least 80 percent of the IDA operational cutoff value. As a consequence, this results in an unbalanced sample as not all countries exceeded the threshold for all years. The panel is further unbalanced due to missing data. Table 1 shows the 71 countries that are eligible for the study. Due to the incomplete data set, the final sample had 705 observations from 65 countries. Table 2 presents instances of graduation and reverse graduation in the sample. 4 We chose this threshold because countries with population of 1.5 million or less are eligible for IDA credits on most concessional terms defined in IDA’s Small Islands Exception policy. Because of their heightened vulnerability to external shocks and limited economies of scale and scope, small states tend to need concessional financing for longer, so their graduations are usually delayed. 5 Table 1. Countries in Sample with IDA Eligibility, FY1989–FY2018a Afghanistan Ghana Nigeria Albania Guinea North Macedonia Angola Guinea-Bissau Pakistan Armenia Haiti Papua New Guinea Azerbaijan Honduras Philippines Bangladesh India Rwanda Benin Indonesia Senegal Bolivia Kenya Serbia Bosnia and Herzegovina Kosovo Sierra Leone Burkina Faso Kyrgyz Republic Somalia Burundi Lao PDR South Sudan Cambodia Lesotho Sri Lanka Cameroon Liberia Sudan Central African Republic Madagascar Syria Chad Malawi Tajikistan China Mali Tanzania Congo, Democratic Republic of Mauritania Togo Congo, Republic of Moldova Uganda Côte d'Ivoire Mongolia Uzbekistan Egypt Mozambique Vietnam Eritrea Myanmar Yemen Ethiopia Nepal Zambia Gambia, The Nicaragua Zimbabwe Georgia Niger Note: a. The following IDA-eligible or formerly IDA-eligible countries are excluded from the sample due to their population size (less than 1.5 million) and/or lack of data availability: Bhutan, Cabo Verde, Comoros, Djibouti, Dominica, Equatorial Guinea, Grenada, Guyana, Kiribati, Maldives, the Marshall Islands, the Federated States of Micronesia, Montenegro, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Samoa, São Tomé and Principe, the Solomon Islands, Timor-Leste, Tonga, Tuvalu, and Vanuatu. 6 Table 2. Instances of Graduation and Reverse Graduation in the Sample Fiscal year of last IDA ‘Reverse graduates’ - fiscal year Country credit on initial reentered graduation Albania FY2008 Angola FY2014 Armenia FY2014 Azerbaijan FY2011 Bolivia FY2017 Bosnia and Herzegovina FY2014 Cameroon FY1981 FY1994 China FY1999 Congo, Republic of FY1982 FY1994 Côte d'Ivoire FY1973 FY1992 Egypt FY1981, FY1999 FY1991 Georgia FY2014 Honduras FY1980 FY1991 India FY2014 Indonesia FY1980, FY2008 FY1999 Nicaragua FY1981 FY1991 Nigeria FY1965 FY1989 North Macedonia FY2002 Papua New Guinea FY1983 FY2003 Philippines FY1979, FY1993 FY1991 Serbia FY2008 Sri Lanka FY2017 Syria FY1974 FY2017 Vietnam FY2017 Zimbabwe FY1983 FY1992 To be candidates for graduation under the policy, countries should be over the GNI per capita threshold and creditworthy. China is an exception, graduating at the end of FY1999 because it was considered creditworthy, despite its per capita GNI being only about 90 percent of the IDA threshold at the time. GNI per capita is sometimes adjusted retrospectively, and the currently available (‘ex post GNI’) data may not accurately reflect the income data that were available when graduation decisions were made (‘contemporaneous GNI’). This may be true particularly for the earlier part of the sample period. By including all country years for which GNI per capita (measured ex post) exceeds 80 percent of the operational cutoff, we err on the side of including too many instead of too few graduation candidates. If we included only observations in which ex post GNI per capita exceeds the operational cutoff, we would very likely be dropping some cases where contemporaneous GNI exceeded the cutoff. Note that this 80 percent rule also ensures that China’s graduation is not excluded from the sample. Estimation method The estimation method follows Heckelman, Knack, and Rogers (2012). The dependent variable is graduation status, taking on the value of 1 for country-year observations corresponding to having 7 graduated from IDA borrowing and 0 for eligible borrowers. We therefore used a logit estimation model. We run pooled logit regressions and correct standard errors for clustering by country. We also test the model with a probit estimator and find similar results. All data come from the World Development Indicators (WDI) database, unless stated otherwise. As the graduation status of a country for the coming year cannot be determined by data for that same year, the IDA decisions are based on historical data. This means that the model has to reflect a two-year lag between the IDA graduate status in a certain fiscal year and the datapoints. For example, it is assumed that the FY2018 graduation decisions are based on calendar year 2016 data, as they are the most recent data available at the time the determination is made. This analysis treats graduation and reverse graduation symmetrically. In practice, graduation and reverse graduation are not fully symmetric. Reverse graduations usually occur because of a crisis resulting in a major fall in income per capita and/or loss of IBRD creditworthiness, whereas decisions on graduation are affected by a broader set of factors. Notwithstanding this limitation, our analysis helps reveal whether IDA graduates differ systematically from IDA borrowers in ways consistent with the IDA graduation policy and, as demonstrated in the following paragraphs, explains graduation decisions reasonably well. Explanatory variables In this model we are using the data that proxy for the issues analyzed when graduation decisions are made. In some instances, the data used in the model were not available when the decision was made. There are also variables that have been used for recent graduation decisions but are not available for most years used in this study and are therefore excluded. Regressors in the model include two income-related indicators: the ratio of GNI per capita to the operational cutoff and (the log of 1 plus) the number of consecutive years in which GNI per capita has exceeded the operational cutoff. If income falls below the cutoff in any given year, this indicator resets at 0. The number of consecutive years is capped at 9, and a separate variable captures (the log of 1 plus) the number of years in excess of 9 that a country has been above the cutoff but not graduated. This captures the notion that a country that has been above the cutoff for so long may become less likely to graduate due to other factors. For example, there may be a negative element not included in the model that explains why a country has not graduated despite its income status. We proxy for IBRD creditworthiness with the Institutional Investors Creditworthiness Index. It is based on surveys of international bankers, with scores potentially ranging from 0 (highest risk of default on sovereign debt) to 100 (lowest risk). Creditworthiness and income per capita have been historically the 8 main indicators of readiness for IDA graduation. As noted previously, the approach to graduation has become more circumspect and rounded, to limit the risk of future reverse graduations. This encourages the inclusion of development indicators. Poverty headcount is potentially an important determinant of graduation, as poverty reduction is a major objective of IDA lending. The data are taken from the World Bank’s PovcalNet database. Coverage is not universal and as such this reduces the sample size, with some IDA-eligible countries having no data at all, while missing country-year observations are accounted for through linear interpolation, 5 but only for at most two consecutive missing years. We proxy for political and institutional development using the International Country Risk Guide (ICRG) Quality of Government Index produced by the PRS Group. This index combines scores for ‘corruption’, ‘law and order’, and ‘bureaucracy quality’ to give a value from 0 to 1, where higher values indicate higher quality of government. 6 To capture the role of vulnerability to shocks, country size (measured by log of population) is included as, other things being equal, larger countries are likely to have more diversified economies less vulnerable to weather-related disasters or adverse commodity-specific or sector-specific shocks. Larger countries may also benefit from economies of scale in administrative capacity and other aspects of institutional development (regulatory systems, courts, independent audit, other accountability institutions, and so on). In addition, there is a cap on allocations to large blend countries that was established to limit the share of IDA resources allocated to populations with large countries and existing access to IBRD. Historically this has covered China, India, Indonesia, and Pakistan, with only the latter still being IDA eligible. The cap meant that Pakistan’s IDA18 allocation was capped at 7 percent of IDA’s total country-allocable envelope. For some of the countries subject to the cap, the IDA financing as a share of gross domestic product (GDP) can become very low, 7 and as a result, they may graduate from IDA sooner than they would have otherwise. A second measure of vulnerability is the degree of dependence on resource revenues. As such, we include natural resource rents as a share of GDP. Because prices of oil, metals, and other commodities are 5 The PovcalNet methodology is described as “interpolating and extrapolating for economies in which survey data are not available in the reference year but are available either before, after, or both. The reference years range from 1981 to 2015. The more survey data are available (that is, the more data for different years), the more accurate the interpolation.” http://iresearch.worldbank.org/PovcalNet/methodology.aspx 6 For countries in which ICRG data were not available, we used the Worldwide Governance Indicators to estimate the ICRG value to maintain a larger sample size. This was done through a linear regression of ICRG rating on these indicators, with the predicted ICRG strongly correlated with actual ICRG. 7 For India, average IDA commitments were just 0.1 percent of GDP in the years of their last replenishment before graduation. 9 volatile, dependence of the economies on resource revenues is likely to be associated with large fluctuations in per capita income and creditworthiness. This consideration may induce greater caution regarding the advisability of graduation for resource-dependent countries. Some others aspects of vulnerability are likely captured by other variables in the model, such as GNI per capita and creditworthiness. This is supported by the literature, for example, Briguglio et al. (2009) and Carment et al. (2008) find per capita income levels to be the main factor influencing vulnerability. There is also the possibility that institutionally weak but resource-rich countries are graduated earlier, reflecting their low demand for IDA resources or a view by some donor constituencies that such countries are not the most appropriate beneficiaries of their aid. This is difficult to capture with a single institutional indicator or an interaction term, because there are many resource-rich countries with weak governance that have remained on IDA support, whose governments are more cooperative. Angola and Equatorial Guinea are the best examples of countries that may have graduated earlier than would normally have been the case and to account for this in the model, a dummy variable is included for these two countries. 8 Finally, we include social and human development indicators. In the model, we proxy these with (log of) life expectancy at birth and the percentage of the population living in urban areas (where access to improved water, sanitation, education, and health services tends to be higher). To control for the change in approach to graduation before and after FY1999, a pre-FY1999 dummy variable is introduced to capture pre-FY1999 effects. This is interacted with the additional development indicators that have become a more significant part of the graduation appraisals since FY1999. The expectation would be that the interacted term would capture the relative neglect of these factors in the earlier part of the sample. Also included as a control is the level of GNI per capita for the Organisation for Economic Co-operation and Development (OECD) countries, relative to the operational cutoff. The expected sign here is ambiguous. If OECD income is high, then it may signal that more aid is available and so countries are less likely to be graduated, but it may also reflect stronger global economic conditions that should support the transition out of IDA status, encouraging graduation. Lastly, a dummy is included to capture the countries who reverse graduated in the pre-FY1999 period. Several countries— such as Cameroon, the Republic of Congo, Nigeria—began the sample period as graduates, despite scoring poorly in several factors that hold a greater weight in the post-FY1999 decision-making process. Therefore, they do not align well with the model’s objectives and so the dummy helps steer the predictive model toward today’s policy regime. 8 However, Equatorial Guinea is excluded from the final regression due to its population falling under 1.5 million. 10 The model faced some limitations as it was not possible to capture certain potential predictors of graduation. For example, there was no satisfactory way of including political instability in the model. This will be a factor when considering a country for graduation, but no variable was found to adequately capture this effect. Moreover, if the model were to be used for projecting future IDA graduations, it would be virtually impossible to know whether countries will remain in conflict, and for how long, or whether conflicts will emerge in the forecast period. As a result of this exclusion, the model may produce overly optimistic probabilities for conflict-afflicted countries, although the institutional development indicator may help limit this effect to an extent. IV Empirical Results Table 3 presents the logit and probit marginal effects, which are similar. The pre-FY1999 interactions are not presented in this table (but are shown in the Appendix) as the primary focus is on the ongoing effects and the predictive power of the model for projection of future IDA-eligible population. The results are in line with expectations. The probability of being a graduate is increasing in the size and the wealth of the country and in its creditworthiness. The social and human development indicators—life expectancy and urban population share—are both positive and significantly correlate with the probability of being a graduate. We tested other measures of education and health—including primary completion, youth dependency ratio, and infant and under-five mortality rates—that turned out to be less significant predictors of being a graduate. The model picks up the nonlinearity of the time that a country has spent with its income per capita above the operational threshold. As hypothesized, the chances of being a graduate are positively associated with time spent above the cutoff, but this is only the case up to a certain point. After a long period above the operational cutoff, it appears that the likelihood of being a graduate begins to reverse. This effect may capture unobserved factors, such as country-specific effects or reflect the omission of a measure of political instability. The model picks up economic vulnerability through population size, as small states are seen to be more vulnerable due to a lack of economies and scope, all other things being equal (Briguglio et al. 2006). The output is consistent with this, with the coefficient positive and significant. 11 Table 3. Results of Logit and Probit Regressions Estimates for graduate (1 = graduate, 0 = non-graduate) Logit Probit GNI per capita/operational cutoff 0.091*** 0.088*** (4.94) (5.14) Log of years over cutoff - capped at 9 0.122*** 0.108*** (5.10) (5.58) Log of years above cutoff in excess of 9 −0.033** −0.035 (−2.45) (−2.44) Creditworthiness 0.006*** 0.006*** (4.65) (5.14) Poverty headcount −0.332 −0.383** (−1.54) (−1.97) Log of population 0.057*** 0.051*** (4.91) (5.13) Log of life expectancy 0.694*** 0.682*** (3.09) (2.96) Urban population share 0.422** 0.413** (2.49) (2.54) Institutional development 0.465* 0.516** (1.80) (2.31) Resource rents/GDP −0.017 −0.024 (−0.08) (−0.11) Log of OECD GNI per capita/cutoff 1.383*** 1.338*** (7.32) (7.36) Angola dummy 0.126 0.149 (1.19) (1.56) Number of observations 705 705 Number of countries 65 65 pseudo-R2 0.78 0.78 Mean dependent variable 0.27 0.27 Correctly predicted (%) 94.9 94.5 Note: Summary statistics are displayed in Appendix in Table 4, pre-1999 interactions in Table 5. t-statistics (in parentheses, beneath marginal effects) are adjusted for clustering by country; * indicates significance at the 10 percent level, ** at the 5 percent level, and *** at the 1 percent level for two-tailed tests. Predictive power The pseudo-R2 (or proportionate reduction in the log likelihood) for the logit models used to generate predicted probabilities is 0.78. In ‘predicting’ graduation status of the sample observations, the model correctly classifies 95 percent of them. By contrast, a ‘null model’ would classify only 73 percent correctly (by predicting, in the absence of any information other than the proportion of observations that are graduates, that none of them were). 12 Specifically, we examine how the model has predicted historical changes in graduation status by identifying type 1 and type 2 errors, respectively, when the model falsely predicts a graduation or alternatively does not predict an actual IDA graduation. There are 16 country graduations in the sample for which there are sufficient data, and in all cases, the model predicted graduation. The model produced three type 1 errors, but in all instances—China, Bosnia and Herzegovina, and Mongolia —the prediction was early, with these countries having later graduated (or in Mongolia’s case, scheduled to graduate). We also analyzed the predictions of reverse graduations. There are 11 of these in the sample for which data were available,9 of which the model falsely predicted a reverse graduation just once, for Papua New Guinea, but it did subsequently reverse graduate four years later. There were no cases of reverse graduation that the model did not predict. The model’s explanatory power in sample overstates its ability to predict accurately out of sample. The model is likely ‘overfitted’ by, for example, including a human development variable (life expectancy) that was significantly related to graduation but excluding others (for example, infant mortality) that were not. To the extent that health-related outcomes influence graduation decisions, it is unlikely that life expectancy is in fact the only relevant measure. Alternate regressions Table 6 in the Appendix shows some alternate regressions that were run. It is shown that exports as a share of GDP did not appear to have any meaningful association with being a graduate, while the two proxies for political instability—a ‘battle dummy’ equal to 1 if there were greater than 500 battle deaths in a given year and a dummy if there was ‘conflict ongoing’—also failed to exhibit a statistically significant relationship. From the same database as the headcount measure, poverty gap is used as an alternative poverty indicator and does not materially alter the output. The final column in Table 6 shows the output for the main logit regression based only on the post- FY1999 period. This reduces the sample size, though not substantially, given that the data availability is much better in the latter part of the original sample. The model fit is even greater, with a pseudo-R2 of 0.85 and over 96 percent of outcomes correctly predicted. Table 7 shows the analogous probit estimates. Comparison to IBRD findings Heckelman, Knack, and Rogers (2012) carried out an analysis of IBRD graduation policy using similar methods. We briefly compare our results with this study, noting that countries graduating from IBRD generally have different characteristics to countries looking to progress from IDA. For instance, the 9 Data for Syria are not available for several years preceding its reverse graduation in FY2017. 13 FY2020 graduation discussion income threshold for IBRD is US$6,975 compared to US$1,175 for IDA. As a result, the assessment criteria for graduation are different. For IBRD graduation, there is less focus on issues such as poverty and human development indicators than for IDA, as it is assumed that by the time a country is looking to graduate from IBRD, these are no longer major concerns. They find that graduation from IBRD is also positively associated with income per capita, institutional development, and creditworthiness. Additionally, they find statistically significant positive associations with export concentration, tax revenue as a share of GDP, banking sector depth (proxied by M2 as a share of GDP), and a negative association with bank crises. We found no relationship with export concentration and IDA graduation, while for the other variables there were no adequate data available for the IDA countries to carry out the analysis. The covariates that are significant predictors in the IBRD analysis are typically factors that are more critical for middle-income countries, for example, the level of financial sector depth and stability. By contrast, for IDA graduations, the key factors are more closely related to poverty and human development. V Conclusions This paper analyzes the factors that contribute toward the decision of whether or not to graduate a country from IDA eligibility. Through statistical analysis of 65 countries that have been IDA eligible at some point over 1987–2016, we find that the World Bank’s approach to graduation is borne out by the empirical evidence. We find that countries are more likely to graduate if they are wealthier, are highly populated, and have greater creditworthiness. In addition, for the period after the Asian financial crisis, when the graduation policy became more circumspect, we find evidence that human development indicators—life expectancy and urban population share—as well as poverty and institutional development became robust predictors of likelihood of being a graduate. The predictive power of our model is high, correctly predicting 95 percent of graduation statuses. 14 Appendix. Data Description and Full Results of the Regressions, Including Alternative Specifications Data description All data come from the World Bank WDI database, unless stated otherwise. GNI per capita: Gross national income divided by midyear population, Atlas method (current US$) IDA operational cutoff: Threshold which GNI per capita should be above to be considered a candidate for graduation. Source: IDA Creditworthiness: Institutional Investors Creditworthiness Index, ranging from 0 (least creditworthy) to 100. Poverty headcount: The proportion of a population that lives below the international poverty line (US$1.90 a day in 2011 PPP). Source: PovcalNet database Institutional development: The ICRG Quality of Government index produced by the PRS Group, with scores ranging from 0 to 1 (highest quality of government) Urban population share: Percentage of population living in urban areas as defined by the country Resource rents/GDP: Total natural resources rents are defined as the sum of oil rents, natural gas rents, coal rents (hard and soft), mineral rents, and forest rents Worldwide Governance Indicators: ‘Government effectiveness’, ‘regulatory quality,’ and ‘rule of law’ used to proxy ICRG, where data are not available. Table 4. Countries with GNI Per Capita over 80 Percent of IDA Operational Cutoff: Summary Statistics N Mean Min Max Standard deviation GNI per capita 764 1,893 510 8,690 1,354 GNI per capita/operational cutoff 764 1.75 0.80 7.59 1.09 Years over cutoff 764 5.86 0 26 5.73 Creditworthiness 742 29.0 3.6 80.2 14.3 Poverty headcount 880 0.21 0.00 0.96 0.22 Log(population) 1001 16.3 13.8 21.0 1.5 Life expectancy 999 64.2 41.4 78.5 8.7 Urban population share 970 0.42 0.11 0.69 0.14 ICRG rating 690 0.40 0.04 0.69 0.13 Resource rents/GDP 830 0.09 0.00 0.60 0.11 OECD real GNI per capita 1,010 33,412 24,991 39,416 4,526 15 Table 5. Full Results Estimates for graduate (1 = graduate, 0 = non-graduate) Logit Probit GNI per capita/operational cutoff 0.091*** 0.088*** (4.94) (5.14) Log of years over cutoff - capped at 9 0.122*** 0.108*** (5.10) (5.58) Log of years above cutoff in excess of 9 −0.033** −0.035 (−2.45) (−2.44) Creditworthiness 0.006*** 0.006*** (4.65) (5.14) Poverty headcount −0.332 −0.383** (−1.54) (−1.97) Log of population 0.057*** 0.051*** (4.91) (5.13) Log of life expectancy 0.694*** 0.682*** (3.09) (2.96) Urban population share 0.422** 0.413** (2.49) (2.54) Institutional development 0.465* 0.516** (1.80) (2.31) Resource rents/GDP −0.017 −0.024 (−0.08) (−0.11) Log of OECD GNI per capita/cutoff 1.383*** 1.338*** (7.32) (7.36) Poverty headcount × pre-1999 0.44 0.543 (1.57) (2.08) Log of life expectancy × pre-1999 −0.022 −0.026 (−0.41) (−0.52) Urban population share × pre-1999 0.281 0.28 (0.81) (0.88) Institutional development × pre-1999 −0.038 −0.049 (−0.15) (−0.21) Resource rents/GDP × pre-1999 0.924** 0.835** (2.06) 2.49 Angola dummy 0.126 0.149 (1.19) (1.56) Number of observations 705 705 Number of countries 65 65 pseudo-R 2 0.78 0.78 Mean dependent variable 0.27 0.27 Correctly predicted (%) 94.9 94.5 Note: t-statistics (in parentheses, beneath marginal effects) are adjusted for clustering by country; * indicates significance at the 10 percent level, ** at the 5 percent level, and *** at the 1 percent level for two-tailed tests. 16 Table 6. Alternative Specifications (logit) Estimates for graduate (1 = graduate, 0 = non-graduate) FY89– FY89– FY89– FY89– FY00– FY18 FY18 FY18 FY18 FY18 GNI per capita/operational cutoff 0.089*** 0.091*** 0.095*** 0.093*** 0.086*** (4.67) (4.89) (5.15) (5.30) (4.34) Log of years over cutoff - capped at 9 0.132*** 0.121*** 0.122*** 0.121*** 0.061*** (5.37) (5.07) (5.53) (5.20) (2.15) Log of years above cutoff in excess of −0.033** −0.033** −0.033** −0.031** −0.027** 9 (−2.25) (−2.47) (−2.39) (−2.36) (−2.04) Creditworthiness 0.007*** 0.006*** 0.005*** 0.006*** 0.002** (4.91) (4.70) (4.75) (4.79) (2.32) Poverty headcount −0.361 −0.332 −0.294 −0.547** (−1.60) (−1.55) (−1.56) (−2.11) Poverty gap −0.894 (−1.44) Log of population 0.054*** 0.057*** 0.056*** 0.055*** 0.102*** (3.70) (4.90) (4.80) (4.51) (5.07) Log of life expectancy 0.780*** 0.685*** 0.665*** 0.704*** 1.617*** (3.06) (3.11) (2.92) (3.07) (2.66) Urban population share 0.429** 0.420** 0.404** 0.430** 0.929** (2.36) (2.48) (2.58) (2.53) (3.15) Institutional development 0.571** 0.466* 0.470* 0.429* 0.583*** (1.97) (1.81) (1.89) (1.67) (3.04) Resource rents/GDP 0.025 −0.022 0.000 0.005 −0.268* (−0.10) (−0.10) (−0.00) (−0.02) (−1.73) Log of OECD GNI per capita/cutoff 1.447*** 1.375*** 1.414*** 1.395*** 1.065*** (7.27) (7.30) (8.37) (7.56) (5.84) Angola dummy 0.173 0.125 0.11 0.109 0.333** (1.51) (1.19) (1.13) (1.11) (2.03) Exports/GDP −0.100 (−0.90) Battle dummy −0.010 (−0.72) Conflict ongoing 0.11 (0.36) Number of observations 674 705 705 705 581 Number of countries 64 65 65 65 65 pseudo-R2 0.78 0.78 0.78 0.79 0.85 Mean dependent variable 0.28 0.27 0.27 0.27 0.23 Correctly predicted (%) 94.5 94.5 94.8 94.6 96.4 Note: t-statistics (in parentheses, beneath marginal effects) are adjusted for clustering by country; * indicates significance at the 10 percent level, ** at the 5 percent level, and *** at the 1 percent level for two-tailed tests. 17 Table 7. Alternative Specifications (probit) Estimates for graduate (1 = graduate, 0 = non-graduate) FY89– FY89– FY89– FY89– FY00– FY18 FY18 FY18 FY18 FY18 GNI per capita/operational cutoff 0.087*** 0.088*** 0.094*** 0.090*** 0.088*** (4.94) (5.12) (5.43) (5.48) (4.75) Log of years over cutoff - capped at 9 0.115*** 0.106*** 0.111*** 0.107*** 0.061*** (5.50) (5.56) (6.59) (5.60) (2.42) Log of years above cutoff in excess of −0.035** −0.035** −0.034** −0.033** −0.026** 9 (−2.32) (−2.47) (−2.37) (−2.32) (−2.19) Creditworthiness 0.007*** 0.006*** 0.006*** 0.006*** 0.002** (5.31) (5.23) (5.41) (5.31) (2.58) Poverty headcount −0.418** −0.379** −0.325* −0.501** (−2.03) (−1.97) (−1.96) (−2.29) Poverty gap −1.101* (−1.85) Log of population 0.049*** 0.051*** 0.051*** 0.049*** 0.096*** (3.88) (5.11) (5.27) (4.79) (6.35) Log of life expectancy 0.749*** 0.666*** 0.665*** 0.695*** 1.458*** (2.90) (2.92) (3.00) (3.04) (2.76) Urban population share 0.419** 0.409** 0.392** 0.431*** 0.881*** (2.44) (2.51) (2.56) (2.63) (4.20) Institutional development 0.598** 0.516** 0.511** 0.476** 0.569*** (2.43) (2.32) (2.40) (2.11) (3.58) Resource rents/GDP 0.001 −0.031 0.013 −0.008 −0.238 (−0.00) (−0.14) (−0.06) (−0.04) (−1.60) Log of OECD GNI per capita/cutoff 1394*** 1.326*** 1.397*** 1.350*** 1.088*** (7.49) (7.36) (8.53) (7.61) (6.20) Angola dummy 0.188* 0.146 0.124 0.138 0.292** (1.83) (1.54) (1.45) (1.51) (2.13) Exports/GDP −0.070 (−0.63) Battle dummy −0.018 (−1.34) Conflict ongoing 0.017 (0.54) Number of observations 674 705 705 705 581 Number of countries 64 65 65 65 65 pseudo-R2 0.78 0.78 0.79 0.78 0.85 Mean dependent variable 0.28 0.27 0.27 0.27 0.23 Correctly predicted (%) 94.4 94.5 94.5 94.6 96.0 Note: t-statistics (in parentheses, beneath marginal effects) are adjusted for clustering by country; * indicates significance at the 10 percent level, ** at the 5 percent level, and *** at the 1 percent level for two-tailed tests. 18 References Acemoglu, D., S. Johnson, J. Robinson, and Y. Thaicharoen. 2003. “Institutional Causes, Macroeconomic Symptoms: Volatility, Crises and Growth.” Journal of Monetary Economics 50 (1): 49–123. Ahmed, S., B. Coulibaly, and A. Zlate. 2017. “International Financial Spillovers to Emerging Market Economies: How Important Are Economic Fundamentals?” Journal of International Money and Finance 76: 133–152. Briguglio, L. 2016. “Exposure to External Shocks and Economic Resilience of Countries: Evidence from Global Indicators.” Journal of Economic Studies 43 (6): 1057–1078. Briguglio, L., G. Cordina, and N. Farrugia, and S. Vella. 2009. “Economic Vulnerability and Resilience: Concepts and Measurements.” Oxford Development Studies 37 (3): 229–247 ———. 2006. “Conceptualizing and Measuring Economic Resilience.” In Building the Economic Resilience of Small States, edited by L. Briguglio, C. Cordina and E. J. Kisanga. Malta: Islands and Small States Institute and London: Commonwealth Secretariat. Bussière, M., and C. Mulder. 2000. “Political Instability and Economic Vulnerability.” International Journal of Finance & Economics 5 (4): 309−30. Calvo, C., and S. Dercon. 2013. “Vulnerability to Individual and Aggregate Poverty.” Social Choice and Welfare 41 (4): 721–740. Carment, D., Y. Samy, and S. Prest. 2008. “State Fragility and Implications for Aid Allocation: An Empirical Analysis.” Conflict Management and Peace Science 25 (4): 349–373. Collier, P., and B. Goderis. 2012. “Commodity Prices and Growth: An Empirical Investigation.” European Economic Review 56 (6): 1241 –1260. Dabla-Norris, E., and Y. Gündüz. 2014. “Exogenous Shocks and Growth Crises in Low-Income Countries: A Vulnerability Index.” World Development 59: 360–378. Dercon, S., J. Hoddinott, and T. Woldehanna. 2012. “Growth and Chronic Poverty: Evidence from Rural Communities in Ethiopia.” Journal of Development Studies 48 (2): 238–253. Elbadawi, I., and R. Soto. 2016. “Resource Rents, Political Institutions and Economic Growth.” In Understanding and Avoiding the Oil Curse in Resource-rich Arab Economies, edited by I. Elbadawi and H. Selim, 187–224. Cambridge: Cambridge University Press. Heckelman, J., S. Knack, and H. Rogers. 2012. “Crossing the Threshold: A Positive Analysis of IBRD Graduation Policy.” Review of International Organizations 7 (2): 145–176. Presbitero, A. F. 2012. “Total Public Debt and Growth in Developing Countries.” European Journal of Development Research 24 (4): 606–626. Reinhart, C. M., V. R. Reinhart, and K. S. Rogoff. 2012. “Public Debt Overhangs: Advanced-Economy Episodes since 1800.” Journal of Economic Perspectives 26 (3): 69–86.