51485 IDA15 MID-TERM REVIEW IDA's Performance Based Allocation and Development Results: An Update International Development Association IDA Resource Mobilization Department (CFPIR) October 2009 Abbreviations and Acronyms AIDS Acquired Immuno Deficiency Syndrome CPIA Country Policy and Institutional Assessment CPIA (A-C) Country Policy and Institutional Assessment, Clusters A-C CPIA (D) Country Policy and Institutional Assessment, Cluster D CPR Country Performance Rating CRS Creditor Reporting System DPT3 Diphtheria, Pertussis and Tetanus Vaccine EVI Economic Vulnerability Index HIV Human Immuno-deficiency Virus GDP Gross Domestic Product GNI Gross National Income HAI Human Assets Index HDI Human Development Index IBRD International Bank of Reconstruction and Development IDA International Development Association LDC Least Developed Countries ODA Official Development Assistance OECD-DAC Organisation for Economic Cooperation in Development, Development Assistance Committee PBA Performance Based Allocation PORT IDA Portfolio Rating RBA Results Based Allocation SDR Special Drawing Rights UN United Nations UNDP United Nations Development Programme WDI World Development Indicators Table of Contents EXECUTIVE SUMMARY ............................................................................................... I I. INTRODUCTION ......................................................................................................... 1 II. IDA'S PERFORMANCE BASED ALLOCATION AND DEVELOPMENT RESULTS: AN UPDATE ................................................................................................. 2 A. A Model Linking Policies and Institutions to Outcomes .............................................. 2 B. Results with Updated Data (2004-06) ............................................................................ 5 C. Are Outcomes Correlated with IDA Allocation or Commitments? ............................... 6 D. Is IDA More or Less Selective Compared to Other Donors? ........................................ 9 III. IDA'S PERFORMANCE BASED ALLOCATION: THE BALANCE BETWEEN PERFORMANCE AND NEEDS .............................................................. 10 A. What are the Various Measures of Needs in IDA's PBA System? ............................. 10 B. How Does the Weight on GNI per capita Change as the Various Measures of Needs are Introduced?............................................................................................................ 12 C. To What Extent Do Capping and Exceptional Allocations Shift Resources to Countries with Greater Development Needs? ............................................................................. 15 IV. TRADE-OFFS BETWEEN PERFORMANCE BASED ALLOCATION AND RESULTS BASED ALLOCATION SYSTEMS .......................................................... 17 A. A Results Based Allocation (RBA) Model ................................................................. 18 B. Comparison between PBA and RBA: Trade-offs and Distributional Implications ..... 18 V. CONCLUSIONS ........................................................................................................ 23 ANNEX 1. REGRESSION RESULTS: NON-HDI INDICATORS ........................... 26 Boxes Box 1. Data Sources ............................................................................................................3 Box 2. Endogeneity of the CPIA/Allocation/Commitments: How big a problem is it? ......9 Table of Contents (cont'd) Tables: Table 1. Human Development Index and policy composite, 1995-2006 ........................... 5 Table 2. Human Development Index and IDA allocation, 1995-2006 ............................... 7 Table 3. Human Development Index and IDA commitments, 1995-2006 ......................... 7 Table 4. Selectivity of IDA vis-à-vis other donors, 1995-2006 ........................................ 10 Table 5. How the Coefficient on GNI per capita shifts with various measures of needs (Dependent variable is IDA allocation)............................................................... 13 Figures: Figure 1. Distribution of IDA resources when using GNI per capita versus under-5 mortality rate in the PBA, 2008 data .............................................................. 14 Figure 2. Regional distribution of IDA resources when using GNI per capita versus Under-5 mortality rates in the PBA, 2008 data............................................... 14 Figure 3. Impact of capping and exceptional allocations in shifting IDA resources to LDCs , 2009 data ............................................................................................ 15 Figure 4. Impact of capping and exceptional allocations on redistribution of resources to poorest countries, 2009 data ........................................................................... 16 Figure 5. Impact of capping and exceptional allocations on redistribution of resources across regions, 2009 data ................................................................................ 17 Figure 6. Distributional Implications of PBA and RBA, by country classification ......... 19 Figure 7. Distributional implications of PBA and RBA, by region .................................. 19 Figure 8. Distributional implications of PBA and RBA, by country classification .......... 20 Figure 9. Distributional implications of PBA and RBA, by region .................................. 21 Figure 10. Correlation between IDA14 allocations and governance (Cluster D of CPIA ratings) score under the PBA and RBA models.............................................. 22 EXECUTIVE SUMMARY i. As part of the IDA15 replenishment discussions, IDA undertook a study entitled "Selectivity and Performance: IDA's Country Assessment and Development Effectiveness" (henceforth referred to simply as the Selectivity paper). The paper analyzed the links between IDA's country policy and institutional assessment (CPIA) and development results during the 1985-2004 period. ii. This paper responds to the request by IDA Deputies to provide an update to this study at Mid-term Review of IDA15, focusing on the link between IDA allocations and development results. The paper updates the SP paper using additional new data covering the period 2004-06. Furthermore, the paper extends the analysis of the SP in four main respects: (i) examines whether there is a significant direct correlation between development outcomes and IDA allocations and commitments; (ii) examines whether or not other donors are more or less selective in their aid allocations vis-à-vis IDA; (iii) explores the relationship between needs and performance in IDA's current Performance Based Allocation (PBA) system and development results; and, (iv) analyzes the trade-offs between the current PBA system and a system that would allocate resources based on development results, also referred to as a Results-Based Allocation System, or RBA. iii. Both the update and the extended analysis fundamentally retain the original model in the Selectivity paper. The model included both the average level of the Country Policy and Institutional Assessment (CPIA) and its change over the decade as right-hand-side variables. In addition, three control variables were included: (i) an Africa dummy, to allow for possible factors that may not be captured in the CPIA ratings but are likely to affect both growth and other development outcomes (e.g., being landlocked, disease burden, etc); (ii) an indicator of the prevalence of HIV/AIDS; and, (iii) the initial value of the development indicator to allow for possible convergence effects. Development outcomes (dependent variable) are proxied by the following indicators: changes in the Human Development Index (HDI), changes in the under-5 mortality rate, changes in the immunization rate, and real growth of GDP per head. iv. The updated analysis confirms the key findings of the Selectivity paper that countries with high policy and institutional performance rating, as measured by CPIA and CPR, averaged over decades, have better improvements in human development and growth outcomes than countries with a low rating. Both the average CPIA and its change are (as in the Selectivity paper) significant correlates of the HDI, a measure of development outcome, at the 99 percent level and estimated coefficients are now slightly higher suggesting a stronger relationship between the right hand side variables used in the model and the HDI. Results for two other development indicators - the immunization rate change and GDP per capita growth rate also corroborate the finding between HDI and CPIA. ii v. The key findings of the extended analysis are summarized below. vi. Correlations between IDA allocation/commitments and development outcomes. During the IDA15 replenishment discussions, IDA Deputies requested that Management further examine the relationship between development outcomes and the allocation of IDA resources across countries, while recognizing that the outcomes cannot be attributed to IDA alone. More specifically, instead of examining the effectiveness of the PBA system indirectly through the relationship between CPIA/CPR and outcomes, it was requested that the direct relationship between IDA allocations and outcomes be examined. The analysis of the relationship between average IDA allocations per capita/ average IDA commitments per capita and changes in HDI shows that countries with higher per capita IDA allocation have on average experienced greater improvements in HDI. This is true for both the full sample, which includes post conflict and capped countries, as well as a sample excluding the latter two categories of countries. The correlation becomes stronger as we move from the full sample of IDA countries to the sample without post-conflict and capped countries, as doing so would effectively correct for "distortions" caused by exceptions to the pure PBA system. vii. Selectivity of IDA vis-à-vis other donors. The paper uses ODA commitments, excluding debt relief and emergency, and the same control variables used in the above model to assess: (i) whether non-IDA donors channel more resources to countries with better improvements in development outcomes; and (ii) how this selectivity compares to that of IDA. The analysis shows that the relationship between HDI changes and commitments per capita is not statistically significant for any of the non-IDA donor groupings (multilateral, bilateral, all). viii. Balance between needs and performance in IDA's PBA. While performance is the main determinant within the formula for IDA allocations under the current PBA system, IDA also takes needs into account in various ways--the GNI per capita component of the allocation formula, population, minimum base allocation, capping of allocations to blend countries with large populations, and through exceptional allocations to post conflict and re-engaging countries with special needs. The paper explores two issues: (i) how the weight on GNI per capita changes as a consequence of introducing the various measures of needs in the PBA formula; (ii) to what extent the capping and exceptional post conflict and reengagement allocations shift the volume of resources to countries with more development needs. We proxy "development needs" by two measures: first, the share of IDA resources received by Least Developed Countries (LDC) as defined by the United Nations; and second, the share of Africa (and to a lesser extent of South Asia) following the World Bank's regional groupings. Using the IDA allocation as a dependent variable and regressing it on logs (so that each estimated coefficient is a measure of elasticity) of CPR, population, and GNI per capita the paper finds that as the various measures of needs are introduced the coefficient on GNI per capita increases. The elasticity of allocation with respect to GNI per capita moves from -0.124 under the pure PBA, to -0.131 with the introduction of minimum allocations; to -0.133 when capping is introduced; and to a further -0.142 when post conflict and reengagement allocations are iii introduced, i.e., a one percent increase in GNI per capita leads to a 0.14 decline in allocations (as opposed to a 0.12 decline under the pure PBA). We can therefore conclude that, as minimum allocations and exceptions are introduced, not only are more funds allocated to poorer countries as a group, but also that within this group, more resources flow to relatively poorer countries as measured by their GNI per capita. The paper also explores to what extent capping shifts resources to countries with greater development needs. Using the UN's definition of LDCs as a proxy for countries with greater "development needs", the analysis finds that capping India and Pakistan increases LDCs' access to IDA resources from 21 percent to 48 percent of total IDA resources--more than doubling the resources of LDCs. Grouping countries by income per capita quintiles and using that as a proxy for needs, the paper finds that resources are redistributed primarily to the bottom three quintiles by the amount of SDR 3.7 billion, about 44 percent of total IDA resources. Using the World Bank's regional grouping as a benchmark, capping of India and Pakistan more than doubles IDA resources available to Africa from 20 percent to 49 percent. It is also the case that post conflict and reengaging exceptional allocations redistribute resources from the third and second quintiles to the first (poorest) quintile though the amounts are not large, in the order of SDR 250 million. ix. Trade-off between PBA and RBA. Finally, the paper examines how allocations would change if development outcomes (results) were used instead of performance (CPR) to determine allocations. The results show that a (static) RBA which predicates IDA allocation on HDI levels shifts resources from LDCs to the relatively better off low income countries, with the share of LDCs dropping from 48 percent under the current PBA system to 29 percent if an RBA were used. Similarly, by region, IDA resources flowing to Africa would decline by half if a static RBA were used while resources going to Europe and Central Asia, Latin America and Caribbean, and Middle East and North Africa would nearly triple given their higher developmental outcome indicators. However, if a "dynamic" RBA which predicates IDA allocation on changes in HDI (instead of levels) were to be used, the share of IDA resources flowing to LDCs increases from 49 percent to 63 percent, with the corresponding share of resources flowing to the remaining IDA countries shrinking from 51 to 37 percent. Similarly, Africa's share would increase from 48 percent under the PBA to 52 percent under this method and South Asia's from 33 to 36 percent, while the share of other regions drops. x. However, a "dynamic" RBA model raises serious concerns in terms of aid volatility, policy incentives, and the time lag between outcome data and allocations. First, volatility in IDA allocations will increase--by more than a five-fold--relative to the PBA system. Second, low income countries with a mid-to-high range of development outcomes will find it difficult to increase their IDA allocation simply because their outcomes may not move rapidly. Third, making allocations conditional on changes in outcomes means that the time lag between allocation and outcome indicators increases even further to three years or more, making the process very much backward looking. Finally, an RBA system may penalize countries for the effects of exogenous shocks; it tends to reduce allocations for fragile states with weak development outcomes; and it iv requires reliable statistical indicators on outcomes ­ which are lacking in many IDA- eligible countries. xi. In conclusion, the analyses undertaken in this paper shows that IDA allocations, which are based on the PBA system, are positively linked to development results. In addition, while IDA's PBA system relies primarily on performance, it also includes a strong "needs" component (through capping and exceptional assistance to post-conflict and re-engaging countries) which ensures that resources are directed to the poorest countries. Finally, while reliance on a "dynamic" RBA system would increase allocations to LDCs, it will significantly raise volatility relative to the PBA system and penalize countries in the mid to high range of development outcomes as well as fragile states and countries where outcomes are affected by exogenous shocks. I. INTRODUCTION 1. As part of the IDA15 replenishment discussions, IDA undertook a study entitled "Selectivity and Performance: IDA's Country Assessment and Development Effectiveness"1 (henceforth referred to simply as the Selectivity paper). The paper analyzed the link between IDA's country policy and institutional assessment (CPIA) and development results. 2. IDA Deputies requested that this study be updated at the Mid-term Review of IDA15, focusing on the links between IDA allocations and development results. This paper responds to that request. The paper updates the SP paper using additional new data covering the period 2004-06. Furthermore, the paper extends the analysis of the SP in four main respects: (i) examines whether there is a significant direct correlation between development outcomes and IDA allocations and commitments; (ii) examines whether or not other donors are more or less selective in their aid allocations vis-à-vis IDA; (iii) explores the relationship between needs and performance in IDA's current Performance Based Allocation (PBA) system and development results; and, (iv) analyzes the trade-offs between the current PBA system and a system that would allocate resources based on development results, also referred to as a Results-Based Allocation System, or RBA. 3. The paper is organized as follows. Section II updates as well as extends the Selectivity paper. It updates the results of that study for new development outcomes and performance data covering the period 2004-06. It extends the analysis of the Selectivity paper in two dimensions: first by replacing CPIA and CPR with IDA country allocation and commitment data to examine whether the results hold under the latter; and, secondly, by examining whether other donors are more or less selective in their aid allocation decisions vis-à-vis IDA. Section III examines the balance between performance and needs in IDA's resource allocation system. More specifically, it examines: (i) how needs are captured in the present allocation system and how well the GNI per capita indicator captures development needs; and, (ii) to what extent capping of blend countries with large populations and special allocations to post conflict and reengaging countries shift resources to countries with greater development needs. Section IV analyzes the trade-offs between the current performance based allocation and a hypothetical results-based allocation system, where development outcomes instead of country performance are used in determining country allocations. Based on such simulations, the section will assess the limitations of moving away from the current system to a RBA system. Section V offers summary and conclusions. 1 This earlier study was prepared by the Office of the Chief Economist (DECVP), Development Economics and Research Group Vice Presidency, February 2007. 2 II. IDA'S PERFORMANCE BASED ALLOCATION AND DEVELOPMENT RESULTS: AN UPDATE 4. This section updates as well as extends the analysis of the Selectivity paper. It updates the results of the paper for new data covering the period 2004-06 -- data that has become available since the original study was conducted. It extends the analysis of the Selectivity paper in two dimensions: first we replace CPIA and CPR with IDA country allocation and commitment data to examine whether the results hold under the latter; and, secondly, we examine whether or not other donors are more or less selective in their aid allocation decisions vis-à-vis IDA. A. A Model Linking Policies and Institutions to Outcomes 5. Both the update and the extension of the analysis fundamentally retain the original model developed in the Selectivity Paper to assess the statistical relationship between country ratings and development outcomes. As argued in the Selectivity Paper, the issues discussed should be considered within the development assistance model often referred to as the "New Partnership Model," which emphasizes country-driven development supported by donors, and places special emphasis on development effectiveness. In order to maximize the effectiveness, particular attention is placed on the quality of country systems, and a focus on "selectivity" more compatible with country ownership that may guarantee the best outcomes given the resources employed. "Selectivity, in turn raises the need for some indicator of development effectiveness that is not overly determined by a rigid set of policies, and its use to guide the allocation of financial support2." 6. The CPIA, as the central pillar of IDA's PBA system, is the core of the Country Performance Rating on the basis of which allocations are made to IDA- eligible countries. The CPIA is a broad indicator of the quality of development policies and institutions, as assessed by World Bank staff. It focuses on key elements within a country's direct control, rather than outcomes such as GDP growth rates, which can be influenced by a number of external factors outside the control of policymakers. At the same time, it measures actual progress rather than relying on a simple projection of good intentions. Finally, CPIA assessments in their final form are the result of a careful and extensive review across the World Bank. 7. CPIA ratings tend to move slowly over time. As mentioned in the Selectivity paper, barring countries at the extremes, CPIA changes are very limited in scope. However, substantial changes over a longer time frame of 10 years or more are not unusual: a typical strong performer can transit from an average rating of 3 to a robust rating of 4 over a decade. CPIA changes show a negative correlation with initial values: when a consistently strong performer experiences a change, it is likely for the worse. Conversely, a poor performer will likely improve its rating upwards. For these reasons, we make use of long-term averages to take into account both measurement errors arising 2 IDA. 2007. "Selectivity and Performance: IDA's Country Assessment and Development Effectiveness", Development Economics, Office of the Chief Economist (DECVP). February. 3 from the subjective nature of some CPIA components, and forward and backward lags between policy indicators and outcomes. However, using long-term averages can conceal very different development trajectories. An average CPIA rating of 3.2 can reflect either a country whose policies have been rated as stable around an average score, or a country whose initial strong ratings have fallen abruptly, or still a country that has experienced consistent improvement from a low base. We therefore include the change in CPIA (in addition to average level) in order to capture these trend dynamics. Box 1. Data Sources As in the original study, the HDI data are obtained from the UNDP. The HDI is an average of three indexes: life expectancy; real GDP per capita (in PPP dollar); and educational attainment, with educational attainment being a combination of literacy and enrolment data.3 Since the UNDP regularly revises the HDI data, we use the latest HDI data series as published on the United Nation's Human Development Report statistics website.4 The available data covers the years 1980, 1985, 1990, 1995, 2000, 2003, 2004, 2005, and 2006. However, because data are not consistently available for all countries for all years, in some cases we extrapolate missing data by finding the best fit and applying the resulting coefficients to calculate the missing data point. We do this in order to keep as many countries as possible in the analysis. Under-5 mortality per 1000 live births, GDP growth rates in constant 2000 US Dollars, and HIV prevalence as percentage of population ages 15-49 are derived from the World Bank's World Development Indicators (WDI) 2009 dataset. Immunization rates, measured by Diphtheria, Pertussis and Tetanus (DTP3) vaccination, are derived from the World Health Organization's World Health Statistics 2009 dataset. The full sample consists of 45 IDA countries and, as in the "Selectivity and Performance" paper, excludes countries in the ECA region and small states. In contrast to the Selectivity Paper, however, we do not include China or Egypt5 in our analysis as both countries graduated from IDA in FY99. Source: World Bank staff and http://hdr.undp.org/en/statistics/ 8. The original model included both the average level of the CPIA/CPR and its change over the decade as right-hand-side variables. In addition, three control variables were added: (i) an Africa dummy, to allow for possible factors (such as the burden of diseases such as malaria, high dependency ratios, the greater prevalence of landlocked countries, economic sparseness and fragmentation, adverse neighborhood effects and other geographic features that contribute to high transport costs) that may not be captured in the CPIA ratings but are likely to affect both growth and other development outcomes; (ii) an indicator of the prevalence of HIV/AIDS; and, (iii) the initial value of the development indicator to allow for possible convergence effects. Development results are proxied by the following indicators: changes in Human 3 In the HDI calculation, minima and maxima are defined for each index: 25 years to 85 years for life expectancy, $100 and $40,000 for real GDP per capita, and 0-100% for the four categories combined in the educational attainment index (Educational Attainment = 2*(adult literacy) + (combined primary, secondary, and tertiary enrollment ratios). The index is defined as (actual ­ minimum) / (maximum ­ minimum). The overall HDI is then a simple average of the Life Expectancy, Real GDP, and Educational Attainment Indexes. 4 See http://hdr.undp.org/en/statistics/ 5 China and Egypt were included in the Selectivity Paper to reduce adverse selection effects when running the model for the period covering 1985-1995. 4 Development Index (HDI), changes in the under-5 mortality rate, changes in the immunization rate, and real growth of GDP per head. 9. The model used in the Selectivity paper and in the current analysis is: (Development outcome) = a1*(CPIA average) + a2*(CPIA or CPR) + a3*(Africa dummy) + a4*(HIV/AIDS incidence) + a5*(initial development outcome) where is the change over the decade. The expected sign pattern for the model is: a1 and a2 positive; a3, a4 and a5 negative. 10. The update seeks to confirm whether the key conclusions of the Selectivity Paper hold with the extended data. The key conclusions of the Selectivity Paper were two-fold. First, countries with high policy and institutional performance rating, as measured by CPIA and CPR, averaged over decades, have higher improvements in human development and growth outcomes than countries with a low rating. Secondly, for the same level of policy and institutional performance, it is much more difficult to achieve development outcomes in Africa. While the PBA system has de facto provided for this difficulty by capping IDA allocations to large creditworthy blends, this capping, however, has reduced the level of development results achieved through the PBA. 11. The extension of the analysis addresses two key related questions. First, during the IDA15 replenishment discussions, IDA Deputies requested that Management further examine the relationship between development outcomes and the allocation of IDA resources across countries, while recognizing that the outcomes cannot be attributed to IDA alone.6 More specifically, instead of examining the effectiveness of the PBA system indirectly through the relationship between CPIA and outcomes, it was requested that the direct relationship between IDA allocations and outcomes be examined. The first part of the extension therefore analyzes the relationship between average IDA allocations per capita/ average IDA commitments per capita and Human Development Index (HDI) changes. Second, while the principle of selectivity (deploying resources where they can have the greatest positive impact) is relevant to IDA's and indeed all donors' efforts to improve development effectiveness, its practical application may vary from one donor to another. Therefore we analyze whether and to what extent other non-IDA donors exercise selectivity in their resource allocation. Three groups of non-IDA donors are considered for this analysis: (i) non-IDA multilateral ODA, (ii) bilateral ODA, and (iii) non-IDA ODA inclusive of both bilateral and multilateral ODA. To the extent that wider selectivity (beyond IDA) may be needed to leverage development outcomes through resource allocation, this analysis will provide useful insights on the broader challenges faced by the donor community as a whole. 6 Chairman's summary: IDA15 meeting, Paris, March 2007. 5 B. Results with Updated Data (2004-06) 12. The relationship between changes in HDI and CPIA is presented in Table 1. Both the average CPIA and its change are, as in the Selectivity paper, significant correlates of the increase in the HDI at the 99 percent level of confidence. The estimated coefficients are now slightly higher, suggesting a possibly stronger relationship between the independent variables described in the model and the HDI.7 Table 1. Human Development Index and policy composite, 1995-2006 (Dependent Variable is Change in HDI) HDI change 1995-2006 HDI change 1995-2004 Updated "Selectivity Paper" CPIA average 0.027 *** 0.026 *** CPIA change 0.032 *** 0.018 *** Africa dummy -0.036 *** -0.028 ** HIV prevalence rate -0.005 *** -0.0042 *** Initial HDI -0.207 *** -0.12 *** Constant 0.104 *** 0.041 R-squared 0.67 0.67 Observations 45 48 Source: World Bank staff calculations. Note: ***, **, and * indicate 99, 95 and 90 per cent confidence levels respectively 13. The full results of the regression analysis for the other indicators of development outcomes--immunization rate, GDP per capita growth rate and under-5 mortality rate-- are presented in Annex 1. The key findings are as follows: Immunization rate change and CPIA. As in the Selectivity paper, immunization rate is positively correlated with country performance--i.e. countries with higher performance (as measured by CPIA average) tend to show faster improvements in the immunization rate.8 GDP per capita growth rate change and CPIA. Using growth of GDP per capita as a proxy for development results, we find that the coefficients on performance (CPIA average and CPIA change) are significant but only after oil-exporting countries are excluded from the sample. Given that the two most recent years included in the analysis (2005 and 2006) have been marked by significant increases in commodity prices, particularly oil prices, this result may in turn be not surprising. 7 Detailed results are presented in the Annex. 8 This correlation, however, appears to be weaker both in terms of magnitude and statistical significance. 6 14. The results of the updated analysis thus broadly confirm the findings in the Selectivity paper: 9 First, countries with high policy and institutional performance (as measured by CPIA and change in CPIA) have seen larger improvements in their HDI, higher immunization rate, and higher growth of GDP per capita. Second, for the same level of policy and institutional performance, it is much more difficult to achieve development outcomes in Africa. The Africa effect highlights the debate on "needs versus performance" and the complexity involved in moving to "results-based" aid--an issue discussed at length in Sections III and IV below. C. Are Outcomes Correlated with IDA Allocation or Commitments? 15. IDA Deputies have expressed interest in examining in more detail the relationship between development outcomes and the allocation of IDA resources across countries.10 This section therefore explores the link between development outcomes as measured by the HDI and IDA allocations and commitments. The purpose here is not to test for any causal links between allocations/commitments and outcomes. Rather, the regressions aim to confirm the existence of a correlation between allocations/commitments and development outcomes in the long run. In other words, they aim to verify whether IDA resources flow to countries that are experiencing improved development outcomes in the long run. 16. We replace CPIA by actual IDA allocation per capita and IDA commitment per capita11 and examine whether the results above hold. Instead of examining the effectiveness of IDA's resource allocation system indirectly through the relationship between CPIA and development outcomes, we examine the direct relationship between IDA allocation per capita and development outcome as measured by HDI. Because post- conflict and capped countries are allocated resources outside of the PBA, and their data points could otherwise distort the "normal" relationship between performance and IDA allocations/commitments, we do this analysis for a full sample of IDA countries as well as a sample excluding these countries. The results are presented in Table 2 and Table 3. 17. The results in Table 2 show that countries with higher per capita IDA allocation have on average experienced greater improvements in HDI. This is true for both for the full sample, which includes post-conflict and capped countries, as well as the sample excluding the latter two categories of countries. The correlation becomes stronger as we move from the full sample to the sample without post-conflict and capped countries. This is as expected, since by excluding capped and post-conflict countries, we are correcting for "distortions" caused by exceptions to the pure PBA system. However, 9 The only exception is the relationship between the changes in under-5 mortality rate and average CPIA. Whereas the original paper found a significant relationship between the two, the updated analysis finds that the signs are as expected but the coefficients are insignificant at 90 percent level. Data revision may have caused the observed changes in coefficients as well as their statistical significance. 10 Chairman's summary; IDA15 meeting, Paris, March 2007. http://siteresources.worldbank.org/IDA/Resources/Seminar%20PDFs/73449- 1172525976405/3492866-1172527584498/SummaryParis2007.pdf 11 IDA allocation data are in Special Drawing Rights (SDR). 7 there appears to be no relationship between changes in HDI and changes in allocations per capita. Table 2. Human Development Index and IDA allocation, 1995-2006 (Dependent Variable is Change in HDI) Data Set Full sample Excluding capped and post conflict countries Average allocations per capita 95-06 0.006 *** 0.010 *** Allocations per capita change 95-06 0.001 0.000 Africa dummy -0.050 *** -0.051 *** HIV prevalence rate -0.004 *** -0.005*** Initial HDI -0.239 *** -0.300 *** Constant 0.188 *** 0.196 *** R-squared 0.61 0.75 Observations 44 35 Source: World Bank staff calculations. 18. Similarly, we find a positive and statistically significant (at the 99 percent level) correlation between HDI changes and average IDA commitments per capita (Table 3). Countries with higher IDA commitments per capita, on average, showed greater improvements in HDI. This is true both for the full sample and for the sample excluding post conflict and capped countries. As with allocations, excluding post-conflict and capped countries increases both the statistical significance and the size of the coefficients. Finally, as with allocations, there is no significant relationship between HDI changes and changes in commitments per capita. Table 3. Human Development Index and IDA commitments, 1995-2006 (Dependent Variable is Change in HDI) Data Set Full sample Excluding capped and post conflict countries Commitments per capita average 95-06 0.004 *** 0.005 *** Commitments per capita change 95-06 -0.000 -0.000 Africa dummy -0.051 *** -0.051 *** HIV prevalence rate -0.004*** -0.004 *** Initial HDI -0.246 *** -0.295 *** Constant 0.206 *** 0.224 *** R-squared 0.66 0.76 Observations 43 34 Source: World Bank staff calculations. 19. The coefficients on commitment per capita are smaller than that on allocations per capita (0.01 vs. 0.005). This may be explained by three factors. First, countries may frontload allocations from future years (within a replenishment cycle) in order to accommodate operational needs, such as large and lumpy infrastructure 8 operations.12 Second, countries may receive exceptional allocations for a variety of reasons including natural disasters, re-engagement with IDA after a prolonged period or following conflict, arrears clearance operations. These exceptions may in turn lead to higher commitments than what the country would have received using the PBA system. The additional flexibility granted under these specific circumstances (i.e. agreed allocations) can therefore cause commitments to diverge from performance-driven allocations. These factors may also explain why the coefficients on commitment per capita are slightly smaller than that on allocations per capita. Third, whereas CPIA ratings are independent of each other, allocations and commitments are instead the result of competition among countries for limited IDA resources. Although the benchmarking phase of the CPIA rating attempts to ensure that ratings are set at the right level and are consistent across countries and regions,13 the country specific ratings are largely independently determined based on an assessment of the actual policy and institutional progress made. On the allocation (and commitment) side, however, because IDA allocation is a zero-sum game and countries compete for a limited amount of IDA resources, what ultimately matters is their relative performance in terms of CPIA ratings.14 Therefore, whereas a country's CPIA rating may increase or decrease in line with its actual policy and institutional progress, its allocations and commitments may follow a somewhat different path, dependent also on other IDA countries' performance. 20. Overall, the conclusion is that the results hold when we substitute the CPIA by IDA allocation and IDA commitment per capita. Allocations/commitments perform in much the same way as the CPIA in the regression analysis. There is a statistically significant relationship between average IDA allocation/ commitment per capita and changes in HDI over 1995-06. In both cases, we find a statistically significant correlation between allocation/commitments and HDI--countries with higher average allocations/commitments per capita make more progress in increasing their human development outcomes. As expected, the correlations become stronger when capped and post conflict countries (which can be viewed as exceptions to the PBA system) are excluded from the sample. 21. These results, however, should be taken with some caution. First, although we find a positive correlation between development outcomes and the allocation of IDA resources across countries, it must be emphasized that we are not claiming that IDA funds cause development outcomes. The analysis here does not address the attribution issue. Second, IDA allocations and commitments can be considered as partly endogenous to development results over a long period such as a decade, a potential problem that is not explicitly addressed through a rigorous testing for endogeneity in this paper (Box 2). 12 Small countries - with correspondingly small performance-based allocations ­ can, for example, use their entire replenishment allocations in one year. 13 For further information, please see: http://web.worldbank.org/WBSITE/EXTERNAL/EXTABOUTUS/IDA/0,,contentMDK:20941073~pa gePK:51236175~piPK:437394~theSitePK:73154,00.html 14 For example, if country X maintains a constant CPIA rating of three over two years but all other countries improve their ratings, country X will ­ everything else constant ­ receive fewer funds than in the previous year. 9 What the results show is that, ex post, IDA puts its money in countries that are achieving better improvements in development outcomes. Box 2. Endogeneity of the CPIA/Allocation/Commitments: How big a problem is it? The Selectivity Paper acknowledged that the CPIA was unlikely to be exogenous to outcomes as analysts will use whatever information is available, including on outcomes, to set ratings. Indeed, one would expect CPIA-type assessments to be influenced by outcomes, and with averaging, the problem of feedback is not possible to avoid. However, the paper tried to test for "naive endogeneity"--where the latter is defined as analysts responding in a mechanistic way to outcomes, including those impacted by external factors or exogenous shocks.15 Using data from 1985 to 2005, the paper examined the responsiveness of CPIA ratings to short-run GDP growth rates (which are not only highly variable but also driven by weather, terms of trade or other shocks in poor countries). It found that the most recent annual GDP growth rate available at the time of a CPIA rating has low predictive power for the change in the CPIA. On that basis, it concluded that there was little evidence that the CPIA ratings have responded in a "naïve" way to shocks that influence development outcomes. If the CPIA is partly endogenous to outcomes, it is reasonable to assume that IDA allocations and commitments are as well. Indeed, allocations/commitments can be seen as slow-responding variables to development results, much like the CPIA. While the naïve endogeneity tests suggest the problem may not be serious, a further in-depth examination of the degree of endogeneity of allocations/commitments to development outcomes is nevertheless needed to establish the scope of the problem and the implications for the correlation results in this paper. Source: IDA. 2007. "Selectivity and Performance: IDA's Country Assessment and Development Effectiveness", Office of the Chief Economist, Development Economics and Research Group Vice Presidency, February. Pp 16-17. D. Is IDA More or Less Selective Compared to Other Donors? 22. We now turn to how IDA's selectivity in allocating resources compares with that of non-IDA donors. We will use ODA flows, excluding debt relief and emergency aid,16 to assess: (i) whether the non-IDA donors channel more resources to countries with better development outcomes; and (ii) how this selectivity compares to that of IDA. We use aid commitment data publicly available through OECD-DAC's Creditor Reporting System (CRS). 23. We run the same regression model using non-IDA commitments per capita by different donor groupings for the same set of IDA countries examined in the previous sections. The results are presented in Table 4. 15 For more discussion on this, please see, IDA, 2007, "Selectivity and Performance: IDA's Country Assessment and Development Effectiveness", Office of the Chief Economist (DECVP), Development Economics and Research Group Vice Presidency, February. Pp 16-17. 16 This is in order to focus on aid destined to development programs rather than relief or emergency aid. 10 Table 4. Selectivity of IDA vis-à-vis other donors, 1995-2006 (Dependent Variable is Change in HDI) Data Set IDA Non-IDA Bilateral Non-IDA ODA Multilateral ODA ODA Post Conflict and Caps included? No No No No Commitments per capita average 95-06 0.005 *** 0.001 0.0007 0.0004 Commitments per capita change 95-06 -0.000 0.0003 -0.0004 -0.00008 Africa dummy -0.051 *** -0.031 * -0.027 -0.027 HIV prevalence rate -0.004 *** -0.005 *** -0.005 *** -0.005 *** Initial HDI -0.295 *** -0.245 *** -0.258 *** -0.247 *** Constant 0.224 *** 0.205 *** 0.209 *** 0.202 *** R-squared 0.76 0.59 0.59 0.58 Observations 34 35 35 35 Source: World Bank staff calculations. 24. We find that the relationship between HDI changes and commitments per capita is not statistically significant (at the 90 percent level) for any of the non-IDA donor groupings. The results suggest that commitments by non-IDA donors do not appear to be linked to improvements in development outcomes, as measured by changes in the HDI. 25. Finally, it is worth noting that the prevalence of HIV/AIDS has a statistically significant negative effect on the changes in HDI for all donor groupings (with similar coefficients), whereas the initial level of HDI's negative coefficient is the consequence of conditional convergence. Both are important variables influencing the level of development, as measured by the HDI. III. IDA'S PERFORMANCE BASED ALLOCATION: THE BALANCE BETWEEN PERFORMANCE AND NEEDS A. What are the Various Measures of Needs in IDA's PBA System? 26. While country performance is the main determinant of country allocations, IDA also considers country needs in its resource allocation system. Country needs are taken into account through a variety of ways: through the IDA GNI per capita cutoff, the GNI per capita component in the allocation formula, population, minimum base allocation, capping of allocations to blend countries with large populations, and through exceptional allocations to post-conflict and re-engaging countries with special needs. The IDA cutoff grants access to IDA concessional resources to countries whose GNI per capita per annum in FY10 is below US$ 1,135 (excluding small island exceptions.) The cutoff level is such that 60 percent of IDA countries (68 percent 11 of IDA countries excluding small island exceptions) are considered low-income countries. Conversely, all countries classified as low income are recipients of IDA resources. Population directly enters the PBA formula with a log-linear weight of one. This guarantees that allocations are granted proportionately to a country's population for all levels of poverty. Each IDA country receives a minimum base allocation of SDR1.5 million per annum. In terms of per capita allocations, this benefits small states, which have a unique set of needs, not captured by the GNI per capita indicator, such as greater vulnerability to shocks and periodic recurrence of natural disasters.17 GNI per capita directly enters the allocation formula with a log-linear weight of -0.125, where a country with a higher GNI per capita receives less allocation than a country with a lower GNI per capita, all else being equal. Capping: Historically IDA has set limits on its allocations to countries with large populations whose per capita GNI is below the IDA operational cut off but which have access to IBRD and other market resources because they are creditworthy. Without a cap on their allocations, these countries would receive the bulk of IDA resources because of their large populations. IDA currently caps the performance- based three-year allocations of two countries-- India and Pakistan--at 11 percent and 7 percent, respectively, of total IDA's commitment authority for a replenishment period.18 The capping of India and Pakistan redistributes resources from South Asia to other regions, in particular to Africa, thereby allowing to meet the differential resource needs of IDA countries more effectively. IDA's exceptional allocations to post-conflict and re-engaging countries similarly drives more resources towards those countries whose development needs may be greater either because they have been in conflict or have disengaged from IDA for a prolonged period of time. 27. In the remainder of this section we explore two issues: (i) how the weight on GNI per capita changes as a consequence of introducing the various measures of needs in the PBA formula; and, (ii) to what extent the capping and exceptional post conflict and reengagement allocations shift the volume of resources to countries with more development needs. While development needs are hard to measure or capture, we proxy "development needs" by two measures: first, the share of IDA resources received by the Least Developed Countries (LDCs) as defined by the United Nations; and second, the share of Africa (and to a lesser extent of South Asia) following the World Bank's 17 For further details, see "IDA's Performance-Based Allocation System: Simplification of the Formula and Other Outstanding Issues", Washington, DC, September 2007. http://siteresources.worldbank.org/IDA/Resources/Seminar%20PDFs/73449- 1172525976405/3492866-1175095887430/PBA_Sept.2007.pdf 18 The fixed percentages are after deductions of donor arrears from total commitment authority. Indonesia, which set to graduate from IDA by end-FY08, has been capped at 3 percent until now. 12 regional groupings. The United Nations country classification of LDCs is based on three criteria--income, human development, and economic vulnerability criteria--and as such can be viewed as identifying countries with greatest development needs from the perspective of poverty. B. How Does the Weight on GNI per capita Change as the Various Measures of Needs are Introduced? 28. We start with a logarithmic transformation of the pure PBA model19 where: log(Allocations) = f [log(cpr), log(pop), log(gnipc)] The coefficients on the variables would then show the elasticity of allocations with respect to the variable in question. Under the pure model, the elasticities of allocation with respect to the country performance rating, population, and GNI per capita are, respectively, 5, 1 and ­0.125. 29. We then run regressions to examine how the coefficients on GNI per capita change when the following needs features of the PBA are introduced:20 (i) minimum base allocation; (ii) capping of blend countries with large populations ­ India and Pakistan; and (iii) exceptional allocations to post-conflict and reengagement countries. The regressions are done for regular IDA countries (excluding small countries with population of less than one million, capped countries, and post-conflict and re-engaging countries as these countries constitute exceptions to the PBA formula). 30. The results are presented in Table 5. The first column shows the elasticities under the pure PBA model. The rest of the columns show how the coefficients on the PBA model shift as minimum allocations, capping of allocations, and post conflict and reengagement allocations are introduced. 19 IDA. 2007. "IDA's Performance Based Allocation System: Simplification of the Formula and Other Outstanding Issues", Washington, DC, September. 20 Moreover, the model uses allocations as computed prior to grant discounts and MDRI being applied. The latter are modifications to the final allocations to countries which take into account the greater concessionality of grants (compared to loans) and the foregone repayment on cancelled debt to IDA: two elements which cannot be modeled in the regression. 13 Table 5. How the Coefficient on GNI per capita shifts with various measures of needs (Dependent variable is IDA allocation) Pure PBA- Pure PBA-based PBA with PBA with capping, based formula formula with capping 1/ and post conflict, and minimum re-engaging allocations allocations 2/ Log of Country Performance Rating 4.996 *** 4.553 *** 4.780 *** 4.768 *** Log of Population 1.000 *** 0.941 *** 0.961 *** 0.962 *** Log of GNI per capita -0.124 *** -0.131 *** -0.133 *** -0.142 *** R-squared 1.00 0.99 0.99 0.99 Observations 72 56 54 44 Source: World Bank staff calculations. Note: 1/Small countries and capped countries are excluded. 2/Small, capped, post conflict and reengaging countries are excluded. 31. The results show that, as the various measures of needs are introduced, the (absolute) coefficient on GNI per capita increases steadily. The elasticity of allocation with respect to GNI per capita moves from -0.124 under the pure PBA to -0.131 with the introduction of minimum allocations; to -0.133 when capping is introduced; and to a further -0.142 when post conflict and reengagement allocations are introduced. In other words, a one percent increase in GNI per capita leads to a 0.14 decline in allocations when exceptions are introduced (as opposed to a 0.12 decline under the pure PBA).We can therefore conclude that, as minimum allocations and exceptions are introduced, not only are more funds allocated to poorer countries as a group, but also that within this group more resources flow to relatively poorer countries as measured by their GNI per capita. 32. To what extent does GNI per capita capture development needs? No single poverty indicator can capture all aspects of development needs and GNI per capita is no exception. Nevertheless, we examine the extent to which GNI captures development needs in two ways. First, we look at the simple correlations between GNI per capita and development outcomes such as HDI and under-5 mortality rates. Second, we compare allocations per capita by quintiles under two cases: when countries are allocated resources on the basis of the PBA using GNI per capita (the current formula) versus allocations based on a PBA formula substituting GNI per capita by under-5 mortality quintiles. Simulations are based on the 2008 allocation data to take into account the latest under-5 mortality rates available. 33. The results suggest that the GNI per capita captures development needs fairly well. First, the correlations between GNI and development results are fairly strong ­ about 51 percent between GNI per capita and HDI and about 64 percent between GNI per capita and under-5 mortality rate. Second, keeping everything else constant but replacing the GNI per capita by under-5 mortality rates in the formula, the resulting allocation distribution is not significantly different ­ i.e. the percent of allocations flowing to IDA countries does not change significantly when using GNI per capita or 14 under-5 mortality rates (Figure 1). The results hold at an aggregated regional level, as shown in Figure 2. Figure 1. Distribution of IDA resources when using GNI per capita versus under-5 mortality rate in the PBA, 2008 data Source: World Bank staff calculations based on WDI data. Figure 2. Regional distribution of IDA resources when using GNI per capita versus Under-5 mortality rates in the PBA, 2008 data Source: World Bank staff calculations based on WDI data. 15 C. To What Extent Do Capping and Exceptional Allocations Shift Resources to Countries with Greater Development Needs? 34. The capping of IDA countries with large populations that have access to market- based financing redistributes resources to countries with greater development needs, particularly those in Africa. We assess the impact of this measure in terms of shifting resources. 35. First, we define countries with greatest development needs as being the Least Development Countries (LDCs), as defined by the UN. The United Nations' Committee for Development Policy in its 2006 triennial review of the list of Least Developed Countries defined them on the basis of the following three criteria: (i) a low-income criterion, based on a three-year average estimate of GNI per capita (under US$745 for inclusion, above US$900 for graduation); (ii) a human capital status criterion, involving a composite Human Assets Index (HAI); and (iii) an economic vulnerability criterion, involving a composite Economic Vulnerability Index (EVI). To be added to the list, a country must satisfy all the three criteria. This definition or country classification therefore takes into account a more comprehensive framework of needs from the perspective of poverty.21 Figure 3. Impact of capping and exceptional allocations in shifting IDA resources to LDCs , 2009 data Source: World Bank staff calculations. 36. Using this definition, capping India and Pakistan increases LDC's access to IDA resources from 21 percent to 48 percent of total IDA resources (Figure 3). Granting further special allocations through the post conflict and reengagement windows 21 For further information, please refer to http://www.unohrlls.org/en/ldc/related/59/ 16 further increases this share to 49 percent. Through capping and exceptional post conflict allocations, LDCs are allocated more than double the amounts they would receive if the system were to be based exclusively on performance. The apparently smaller impact of the post-conflict and re-engaging exceptions is explained by the fact that these set of countries are all within the LDCs group. Granting them exceptional allocations therefore causes an intra-group redistribution of resources towards countries with special reconstruction and recovery needs. 37. Second, we look at the impact of capping and special allocations in shifting resources to the poorest countries. We do so by presenting the percent of resources allocated to IDA countries divided into poverty quintiles, and in each case (pure PBA formula, PBA formula with capping, and PBA formula with capping and post conflict and re-engaging exceptions). IDA countries are ranked by GNI per capita from lowest to highest and divided in quintiles, each one with an equal number of countries. We present both the percent of resources allocated to each quintile, and the average GNI per capita of each quintile on the legend on the right hand-side. Figure 4. Impact of capping and exceptional allocations on redistribution of resources to poorest countries, 2009 data Source: World Bank staff calculations based on WDI data. 38. Figure 4 represents shares of IDA allocations to IDA countries by poverty quintiles. Capping releases resources that are distributed, in accordance to the PBA formula, across all countries. The final effect is analogous to an increase in overall IDA funds for distribution. Resources are redistributed primarily to the bottom three quintiles, which experience an increase of resources available in the order of SDR 3.7 billion for one fiscal year. Furthermore, post-conflict and re-engaging exceptional allocations redistribute resources in the order of SDR 250 million from the third and second quintile to the first (the poorest). By region, capping more than doubles IDA resources available to Africa from 20 percent to 49 percent; and that of East Asia from 5 to 11 percent (Figure 5). 17 Figure 5. Impact of capping and exceptional allocations on redistribution of resources across regions, 2009 data Source: World Bank staff calculations based on WDI data. IV. TRADE-OFFS BETWEEN PERFORMANCE BASED ALLOCATION AND RESULTS BASED ALLOCATION SYSTEMS 39. In recent years, there has been much debate about whether performance- based or results-based allocation is more appropriate to increase development effectiveness. Kanbur (2005) has, for example, suggested to reform IDA's performance based allocation formula so that the formula explicitly includes development outcomes.22 In this section, we analyze the trade-offs between a results-based aid allocation system and the current performance based allocation system, namely: (i) how allocations would change if development outcomes (e.g., HDI) were used instead of CPR to determine allocations; and, (ii) what are the limitations of allocating resources based on results. We do so by modifying the Performance Based Allocation (PBA) formula to simulate an allocation system based purely on development outcomes. 22 See, for example, Kanbur, Ravi, 2005, "Reforming the Formula: A Modest Proposal for Introducing Development Outcomes in IDA Allocation Procedures". 18 A. A Results Based Allocation (RBA) Model 40. The PBA formula, as modified during the IDA15 Replenishment negotiations,23 is given by: (define A-C, D and PORT) Country Performance Rating = (0.24 * CPIA A-C + 0.68 * CPIA D + 0.08 * PORT) Where CPIA A-C refers to Clusters A to C of the Country Policy and Institutional Assessment (CPIA); CPIA D refers to Cluster D of the CPIA; and, PORT stands for the IDA portfolio rating. IDA country allocation = f (Country performance rating5.0, Population1.0, GNI/capita-0.125) In order to generate a results-based allocation model (RBA), we substitute HDI for CPIA, after having rebased it to the same scale as the CPIA (1-6), and calculate the country performance ratings as: Country Performance Rating = (0.92 * HDI + 0.08 * PORT). 41. We then run the modified results based allocation model (RBA) with population, GNI per capita and all other features of the PBA model (e.g., operational cutoff, base allocation, etc) unchanged. The model is run for all IDA countries, maintaining the present capped, post conflict, and re-engaging countries classifications. Capped countries continue therefore to receive the same allocations, while the phasing down of the special allocations to the post conflict and re-engaging countries would be not to their PBA levels, but to what is implied by the RBA model. Finally, due to the time lag in outcome data and limited availability of the updated HDI data, we use 2008 data. B. Comparison between PBA and RBA: Trade-offs and Distributional Implications 42. As anticipated, a purely outcome-based allocation method shifts resources from LDCs to the relatively better off low income countries. The share of IDA resources going to LDCs drops from 48 percent under the current performance based formula, to 29 percent (Figure 6). 23 Reference IDA15 Replenishment paper: "IDA's Performance-Based Allocation System: Options for Simplifying the Formula and Reducing Volatility", February 2007 19 Figure 6. Distributional Implications of PBA and RBA, by country classification 100% Percent of Total IDA Resources Non-LDC IDA countries 80% 52% 71% 60% 40% 48% 20% 29% Least Developed Countries (LDC) 0% Performance-based Results-based allocations, 2008 allocations, 2008 Source: World Bank staff calculations based on WDI and UN data. 43. Additionally, we compare the resulting allocation distributional implications by region (Figure 7). Resources flowing to Africa would halve from 50 percent to 25 percent; resources to East Asia and Pacific would nearly double from 12 percent to 22 percent; and resources allocated to Europe and Central Asia, Latin America and Caribbean, and Middle East and North Africa would nearly treble, in accordance with the higher development outcomes experienced by these regions. Figure 7. Distributional implications of PBA and RBA, by region Source: World Bank staff calculations based on WDI data. 20 44. Another way of applying a results-based allocation model is to allocate resources to countries based on changes in development outcomes as opposed to levels of outcomes, also referred to as a "dynamic" RBA system. This "dynamic" interpretation hinges on the actual increase in outcomes as compared to taking into account a static snapshot of outcomes, where allocation depends on levels, not changes. In this way, countries with low development outcomes would not be penalized with lower resources at a time when their needs could be greatest, provided they are making improvements. 45. Under this "dynamic" RBA model, more resources will get allocated to the LDCs (Figure 8). We run allocations for the aggregated IDA14 period based on changes in HDI. The results show that the share of IDA resources flowing to LDCs increases from 49 percent to 63 percent, with the corresponding share of resources flowing to remaining IDA countries shrinking from 51 to 37 percent. One reason for the increase in resources flowing to LDCs is the conditional convergence effect: changes in absolute levels of HDI are easier and more frequent when the initial level of development outcome is low. Figure 8. Distributional implications of PBA and RBA, by country classification 21 Figure 9. Distributional implications of PBA and RBA, by region Source: World Bank staff calculations based on WDI data. 46. Similarly, the regional distributions shift toward Africa compared to the static RBA model (Figure 9). Africa's share increase by four percent (from 48 percent to 52 percent) and South Asia's by three percent, while East Asia's drops by 5 percent and the remaining regions experience a collective two percent drop. Resources shift by a maximum of five percent among regional groups. 47. However, a "dynamic" RBA model raises serious concerns in terms of aid volatility, policy incentives, and the time lag between outcome data and allocations. Aid allocations become much more volatile under an RBA model that relies on changes in development outcomes. The standard deviation of aid allocated over the IDA14 period increases from an average of 10.5 under the current PBA model to 55.3 under the RBA-- a more than five-fold increase. Whereas the median percentage increase and decrease in allocations over the same period is between +12 and -11 percent under the PBA, with the "dynamic" RBA model it increases to +48 and -40 percent. Aid volatility is often singled out by recipient countries and in development literature as a serious concern because it generates uncertainty in terms of allowing the recipient countries in designing and implementing effective multi-year development programs. 22 Figure 10. Correlation between IDA14 allocations and governance (Cluster D of CPIA ratings) score under the PBA and RBA models PBA Model "Dynamic" RBA Model 25.0 25.0 Average S DR per cap ita p er ann um, IDA14 Averag e SDR p er cap ita per annum, IDA14 20.0 20.0 15.0 15.0 10.0 10.0 5.0 5.0 0.0 1.50 2.00 2.50 3.00 3.50 4.00 4.50 0.0 -5.0 1.50 2.00 2.50 3.00 3.50 4.00 4.50 Averag e IDA14 cluster D rating Averag e IDA14 cluster D rating Source: World Bank staff calculations based on WDI data. 48. Moreover, policy incentives are changed under the "dynamic" RBA. First, countries with medium to high levels of HDI may find it harder to see their allocations increase because their outcomes do not move rapidly. The conditional convergence effect benefits LDCs, and other low income countries with a mid to high-range of HDI making consistent efforts may end up not being rewarded through higher allocations. Secondly, relative to the PBA model, there will be a much weaker correlation between allocation and governance--as measured by the CPIA cluster D of governance--under the "dynamic" RBA. Figures 10 offers a comparison of the scatterplots of IDA14 average allocations against the average governance ratings (as measured by the CPIA cluster D) under the PBA and dynamic RBA model simulations. Under the dynamic RBA model, the link becomes significantly weaker, as the scatterplots show. 49. With the "dynamic" RBA, the time-lag between the allocation exercise and the underlying data's measurement further increases. Whereas CPIA scores bear a one year lag between their measurement year and the year for which they are used to determine allocation, HDI data typically has a two year lag. Making use of changes in HDI means that this time lag increases even further to three years more. The process therefore becomes very much backward looking (even more backward looking than the static RBA), and rewarding recent policy and institutional efforts becomes difficult. 50. Finally, in addition to the re-distributional consequences, implementing a results-based allocation system entails several other challenges. Among additional challenges faced in implementing a results based allocation system are (i) a results based system would penalize countries for the effect of exogenous variables (e.g., terms of trade shocks such as those arising out of the current global financial crisis; natural disasters; 23 climate related problems e.g. droughts and floods; pandemics e.g., HIV/AIDS and avian flu); (ii) for aid agencies operating in several low income countries, results based allocations require data which is comparable across both time and across countries, and given the weak and unreliable statistical and monitoring systems in many developing countries that lack frequent surveys, such data may be very difficult to obtain; and (iii) countries with the highest needs may be those least likely to be making progress on development results (e.g., post-conflict and fragile states) and would be penalized by focusing only on outcome based aid allocation decisions. V. CONCLUSIONS 51. The updated analysis confirms the early finding of the Selective Paper that countries with high policy and institutional performance rating, as measured by CPIA, averaged over decades, have better improvements in human development and growth outcomes than countries with lower ratings. Both the average CPIA and its change are significant correlates of the Human Development Index (HDI), a measure of development outcome, at the 99 percent level and estimated coefficients are now slightly higher suggesting a stronger relationship between the independent variables used in the model and the HDI. Analysis for two other development indicators, the immunization rate change and GDP per capita growth, also corroborates the same finding. 52. The extended analysis offers five main conclusions. First, the analysis of the relationship between average IDA allocations/commitments and changes in Human Development Index (HDI) shows that countries with higher per capita IDA allocation have on average experienced greater improvements in HDI. This is true for both the full sample, which includes post conflict and capped countries, as well as a sample excluding the latter two categories of countries. The correlation becomes stronger as one moves from the full sample of IDA countries to the sample without post-conflict and capped countries since by excluding capped and post conflict countries, we are correcting for "distortions" caused by exceptions to the pure PBA system. 53. Second, the paper applies the same model but using non-IDA ODA commitments, excluding debt relief and emergency, for different non-IDA donor groupings to examine whether the donor community as a whole is channeling more resources to countries with better improvements in development outcomes, and how this selectivity compares to that of IDA. Our analysis shows that the relationship between HDI changes and commitments per capita is not statistically significant for any of the non-IDA donor groupings (multilateral, bilateral, all). 54. Third, the paper revisits the debate on performance versus needs in IDA's PBA and examines how the weight on GNI per capita changes as a consequence of introducing the various measures of needs in the PBA formula as well as to what extent the capping and exceptional post conflict and reengagement allocations shift the volume of resources to countries with more development needs. The paper finds that the (absolute) coefficient on GNI per capita moves steadily from -0.124 under the pure PBA to -0.131 with the introduction of minimum allocations; to -0.133 when capping is introduced; and to a 24 further -0.142 when post conflict and reengagement allocations are introduced, suggesting that not only are more funds allocated to poorer countries as a group as minimum allocations and exceptions are introduced, but also that within this group, more resources flow to relatively poorer countries as measured by their GNI per capita. 55. Fourth, still on the issue of performance versus needs, the paper explores to what extent capping of IBRD creditworthy countries with large population (i.e. India and Pakistan) shifts resources to countries with greater development needs. Using the UN's definition of Least Developed Countries (LDCs) as a proxy measure for "development needs", the analysis concludes that capping India and Pakistan increases LDCs access to IDA resources from 21 percent to 48 percent of total IDA resources--more than doubling the resources of LDCs. Special allocations through the post conflict and reengagement windows further increase this share to 49 percent. By region, capping more than doubles IDA resources available to Africa from 20 percent to 49 percent. 56. Finally, the paper examines how allocations would change if development outcomes were used instead of PBA to determine allocations. The results show that a (static) RBA which predicates IDA allocation on HDI levels shifts resources from LDCs to the relatively better off low income countries, with the share of LDCs dropping from 48 percent under the current PBA system to 29 percent if an RBA were used. Similarly, by region, IDA resources flowing to Africa would decline by half if a static RBA were used while resources going to Europe and Central Asia, Latin America and Caribbean, and Middle East and North Africa would nearly triple given their higher developmental outcome indicators. However, when a "dynamic" RBA which predicates IDA allocation on changes in HDI (instead of levels) is used, the share of IDA resources flowing to LDCs increases from 49 percent to 63 percent, with the corresponding share of resources flowing to the remaining IDA countries shrinking from 51 to 37 percent. Similarly, Africa's share would increase from 48 percent under the PBA to 52 percent under this method and South Asia's from 33 to 36 percent, while the share of other regions drops. 57. However, a "dynamic" RBA model raises serious concerns in terms of aid volatility, policy incentives, and the time lag between outcome data and allocations. First, volatility in IDA allocations will increase ­ by more than a five-fold--relative to the PB system. Second, low income countries with a mid to high-range of development outcomes will find it difficult to increase their IDA allocation simply because their outcomes may not move. Third, making allocations conditional on changes in outcomes means that the time lag between allocation and outcome indicators increases even further from two to three years or more, making the process very much backward looking. Finally, an RBA system may penalize countries for the effects of exogenous shocks; it tends to reduce allocations for fragile states with weak development outcomes; and it requires reliable statistical indicators on outcomes ­ which are lacking in many IDA- eligible countries. 25 REFERENCES IDA. 2007a. "Selectivity and Performance: IDA's Country Assessment and Development Effectiveness", Development Economics, Office of the Chief Economist (DECVP), Washington, DC February. IDA. 2007b. "IDA's Performance-Based Allocation System: Options for Simplifying the Formula and Reducing Volatility", February. IDA. 2007c. "IDA's Performance-Based Allocation System: Simplification of the Formula and Other Outstanding Issues", Washington, DC, September. IDA. 2007d. Chairman's Summary: IDA15 Replenishment Meeting, Paris, March. http://siteresources.worldbank.org/IDA/Resources/Seminar%20PDFs/73449- 1172525976405/3492866-1172527584498/SummaryParis2007.pdf Kanbur, Ravi, 2005. "Reforming the Formula: A Modest Proposal for Introducing Development Outcomes in IDA Allocation Procedures". 26 Annex 1. Regression Results: Non-HDI Indicators Table 1.1. Under-5 Mortality Rate and CPIA Under-5 mortality rate change Under-5 mortality rate change 1995-2006 1995-2004 Data Set Updated "Selectivity Paper" CPIA average -6.243 -14.11 ** CPIA change -6.145 -4.8 Africa dummy 28.811 *** 18.60 ** HIV prevalence rate 0.641 0.63 Initial mortality rate -0.202 *** -0.17 *** Constant 4.447 39.66* R-squared 0.28 0.31 Observations 48 49 Table 1.2. Immunization Rate and CPIA Immunization rate change 1995- Immunization rate change 1995-2004 2006 Data Set Updated "Selectivity Paper" CPIA average 6.269 * 10.45 ** CPIA change 4.219 3.78 Africa dummy 1.163 1.75 HIV prevalence rate 0.307 -0.35 Initial immunization rate -0.828 *** -0.60 *** Constant 52.23 ** 15.87 R-squared 0.803 0.53 Observations 48 47 Source: World Bank staff calculations Table 1.3. Average Real GDP per capita Growth Rate and CPIA Average Real GDP per capita Average Real GDP per capita growth 1995-2006 growth 1995-2004 Data Set Updated Updated "Selectivity "Selectivity Paper" Paper" Oil countries Including "oil Excluding "oil Including Excluding "oil countries" countries" "oil countries" countries" CPIA average 0.45 2.53*** 1.50 ** 3.70 *** CPIA change 1.48** 1.45** 0.95 * 0.64 Africa dummy -1.77** -2.17*** -1.55 * -2.97 *** HIV prevalence rate 0.008 0.026 -0.016 0.049 Initial GDP per capita -0.0017 -0.0031** -0.00052 -0.0054 *** (1995) Constant 1.95 -4.06 -1.6 -7.0 *** R-squared 0.19 0.41 0.24 0.58 Observations 47 43 51 44 Source: World Bank staff calculations * significant at 10%; ** significant at 5%; *** significant at 1%