The Impact of the Public Distribution System on Living Standards Some Microsimulations 5

Transcription

1 The Impact of the Public Distribution System on Living Standards Some Microsimulations 5 1. Introduction According to the World Bank (2007), the Iraqi Public Distribution System (PDS) is probably the largest food-based transfer program operating anywhere in the world. The PDS was started when sanctions were imposed in 1990, providing domestically produced food to the entire population. In 1996, the United Nations agreed to allow food imports under the Oil for Food Programme, thereby permitting the size of the rations to more than double. The entire Iraqi population is eligible to receive the ration freely, although until recently a small nominal fee equivalent to US$0.16 per person per month was needed. 6 At present, the PDS provides 85 percent of caloric needs. 7 The size of the PDS is so large that it has major effects on the Iraqi economy. Advantages are clear: Food rations keep people out of. Disadvantages include the following. First, the fiscal cost of the program is large. The Iraqi Ministry of Finance estimates that the PDS will absorb 5.7 percent of the general budget in the year 2009, corresponding to ID 4,200 billion (US$3.5 billion). Second, the PDS has distortionary effects on the structure of domestic food market prices and hence on the allocation of resources within the country. Third, although the PDS is an important safety net, its role is accomplished inefficiently, mainly because of the absence of targeting. Given its crucial role, PDS reform has been hotly debated in Iraq in past years, and the issue is currently under scrutiny as part of the government s effort to design a broader national development strategy. A variety of proposals have been put forth previously, but their assessment is made difficult by a lack of data that prevents the use of adequate methods of analysis. For instance, to estimate how the Iraqi economy would react if the PDS were changed/eliminated one could rely on computable general equilibrium (CGE) models. These are ubiquitous in policy analysis. However, CGE analysis is currently not an option for Iraq because of lack of suitable data. In this document we present the results of static microsimulation exercises, a strategy made possible by the availability of the Iraq Household Socio-Economic Survey (IHSES), a large-scale, nationally representative household budget survey fielded from October 2006 through November Microsimulations represent a first, admittedly rudimentary, attempt to estimate the impact of different reform plans of the PDS on the living standards of the Iraqi population. What microsimulations can help shed light on is the first stage of the transition process from a universally subsidized economic system (the PDS) to a market-based one. In our microsimulations we analyze what would happen if, after the elimination of the PDS, the prices of the items included in the rations increased. We consider a range of price increases (from 25 percent to 200 percent). The largest increase can be interpreted as an approximation of what could happen in the short run after the elimination of the PDS. In contrast, if one assumes that in the medium/long run inflation will be kept under control, the zero price change situation 5 Document prepared by Nicola Amendola (Univ. of Rome Tor Vergata ), Susan Razzaz (The World Bank), Marina Sorrentino (Italian Statistical Institute), and Giovanni Vecchi (Univ. of Rome Tor Vergata ). Maria Edo provided excellent research assistance. 6 See World Bank (2007: 12). There also are retail costs that differ across governorates. 7 The energy caloric requirement for Iraq has been estimated to equal 2,337 kilocalories/person/day. Food rations distributed by the PDS provide 1,998 kilocalories/person/day. 562

2 could offer some insight into the final social costs that would be borne. However, no attempt is made to estimate the final outcome of the process or the transitional dynamics toward the final outcome (the steady state equilibrium outcome in the economist s parlance) this goes beyond the scope of our analysis. The paper is organized as follows. Section 2 describes the IHSES survey. Section 3 supplies the motivation for the microsimulations on PDS reforms discussed in the subsequent sections. Section 4 examines what would happen if the PDS were eliminated without any compensating measures. In Section 5 we simulate substitution of the PDS with two baseline cash transfer schemes, namely universal cash transfers (no targeting) and a second simulation whereby the poor are the only beneficiaries of an equalsize cash transfer (perfect targeting). In Section 6 we simulate the effects of geographical targeting on indicators, while in Section 7 we analyze a proxy means-testing targeting algorithm. Section 8 discusses, in light of the obtained results, the dilemma facing policy makers confronting the issue of PDS reform. 2. Iraq Household Socio-Economic Survey Data The aim of this section is to provide the reader with essential information about data availability and constraints. The Iraq Household Socio-Economic Survey (IHSES) was fielded during October 2006 November The Central Organization for Statistics and Information Technology (COSIT) and the Kurdistan Region Statistics Office (KRSO) were the main organizations responsible for survey design and implementation. Almost 18,000 households were interviewed over the course of a year, thus collecting data for more than 127,000 individuals. 8 The questionnaire is organized into five sections. The first contains information on household composition, the rations received through the PDS, housing, health, activities, entertainment and hobbies, labor market participation, and job-seeking activities. The second section collects information on all commodities and services purchased and consumed during the 30 and 90 days prior to the interview. The third section collects data on food and recurring nonfood expenditures. Information is also provided on each source of employment during the previous year, all nonwage earnings, and income from sources such as property, remittances from family members abroad, traditional Islamic charity (zakat) and government transfers (different for food rations). Moreover, it contains information on durable goods owned by the household, lending and credit, and recent experiences of shocks and the responses to them. The fourth section consists of a daily ledger in which each sample household can record all consumptions and expenditures. Finally, the fifth section of the questionnaire records the activities during a 24-hour period of every member of the household older than 10 years. The fifth section was applied to a subsample of one-third of the IHSES households. 3. Microsimulations Theory offers limited guidance on the best method to predict what would happen if PDS were to be eliminated. A number of factors, including data availability and political feasibility, severely limit the set of implementable methods. 8 A comprehensive description of the survey is provided in Tabulation Report Iraq Household Socio-Economic Survey 2007 (IHSES 2008). In particular, see Volume I for objectives and methodologies, and Volume III for the questionnaire. The full tabulation report (898 pages) can be downloaded in English and Arabic from or 563

3 The first question that needs to be addressed when considering the issue of reforming PDS is: what would happen if PDS were eliminated without any compensating measures? While the simulated situation is probably unlikely, the analysis supplies a useful starting point, to be interpreted as a baseline providing the order of magnitude of the social costs under the worst case scenario. Section 4 will describe the simulation results. Even before seeing those results, it is easy to imagine that the elimination of a program as large as the PDS could substantially worsen Iraqis social conditions. Hence it is reasonable to think that compensating measures could be introduced if the PDS were removed. In particular, cash transfers could be implemented. As a first step in assessing the impact of cash transfers in lieu of the PDS, we have addressed a second question: what would happen if the universal rations were substituted for by universal cash transfers? That is, we have studied the consequences of removing the distortionary effects of the in-kind PDS transfers and substituting for them with uniformly distributed cash transfers (see Section 5.1). Transfers of equal size for all beneficiaries are considered in all simulated cash transfer schemes. Uniform distribution, however, would dampen the scheme s effectiveness in reduction. To assess the full potential for this aim of cash transfers, we have considered the effects of a system in which all poor Iraqis are perfectly identified (that is, the per capita expenditure of each Iraqi is observed and all those for whom this indicator is below the line threshold are identified) and equal size cash transfers go to all of them and only to them (see Section 5.2). The assumption of perfect targeting of all poor Iraqis cannot be met in practice because we do not actually observe the per capita expenditure (PCE) of all Iraqis. This would require the collection of detailed information on all citizens consumption expenditures and its regular update. Hence we study two popular targeting schemes: geographical targeting, in which transfer beneficiaries are identified by where they live (Section 6); and proxy means-testing targeting, in which indicators of well-being are constructed and used to rank applicants and assess whether they are eligible for receiving the transfer (see Section 7). 4. What Would Happen if the PDS Were Eliminated? To answer this question, we make three key assumptions. First, we assume that the PDS is permanently and abruptly removed with no compensating measure. Second, households are assumed to experience whether they are poor or not a reduction in their living standards exactly equal to the market value of the rations received. This rules out both access to credit which would allow households to mitigate the impact of rations removal on their living standards and use of personal savings. Households are therefore assumed to lack any means to cope with negative shocks, regardless of their status. Third, we allow relative prices to change as a consequence of increases in the prices of the items that were included in rations. In particular we model equal percentage price increases for all included items. This implies that the consumer price index also rises. 9 This is the expected scenario in the immediate aftermath of eliminating the rationing system, as a consequence of 15 years of crowding out by the PDS, artificially suppressed retail prices, lack of investment, lack of access to credit, and a largely noncompetitive business environment that imply limited capacity in parts of the food supply chain, which will take time and effort to ameliorate (World Bank, 2007). Furthermore, because one of the most important components of PDS rations is cereals, we take into account the effect of price increases on the income of cereal-producing farmers. What would happen then under this scenario to household living standards? 9 See Appendix of this annex for details. 564

4 Table What Would Happen if the Public Distribution System Were Eliminated? price increase by headcount gap gap squared Gini index among the poor ,151 56, ,129 55, ,275 55, ,573 54, ,011 53, ,576 53, ,258 52, ,048 52, ,938 51,498 actual ,944 61,624 Source: Authors estimates using 2007 IHSES data Note: The line is ID 76,896/person/month. A convenient way to read Table is to separate the first row from all the others. The first row shows what happens if prices do not change. If we compare it with the actual situation (last row), the main results can be summarized as follows. First, the incidence of increases from 22.9 percent to 34.4 percent. Second, the depth of doubles: the gap index rises from 4.5 percent to 9.1 percent. Third, the severity of increases dramatically. The gap squared index increases by a factor of 2.5 (from 1.4 percent to 3.5 percent). This latter result shows that if we attribute a higher weight/importance to the poorest among the poor, the elimination of the PDS increases the measure by 2.5 times (or equivalently, increases by 350 percent)! Fourth, inequality increases. The Gini index increases from 28.5 percent to 31.6 percent. All and inequality indicators worsen dramatically overall. Row 1, however, by not allowing any price to increase is quite unrealistic in describing the situation in the immediate aftermath of the PDS removal. In particular, it fails to account for the reduction in living standards due to the fact that prices of the items included in the rations are likely to increase. Subsequent rows in Table assume that the prices of rations increase by 25 percent increments to reach 200 percent. Note that 100 percent implies that the prices double, while 200 percent means the prices triple. The last row of the table shows the actually observed and inequality rates. Table shows that the impact on and inequality measures crucially depends on the severity of the price increase. A number of findings are worthy of comment. First, the removal of the PDS has a very negative impact on the incidence of (measured by the headcount index): the number of individuals classified as poor increases from the current 6.9 million to 11 million under the best-case scenario (which assumes authorities successfully avoid excessive increases in the relative price of rations, containing them at the 25 percent level), and up to 14.7 million should ration prices triple (that is, a relative price increase of 200 percent). 565

5 Moreover the removal of the PDS has a large impact on the depth of. Average living standards of the poor would decrease by 9 16 percent (see the last column of Table 9.1-1), depending on the magnitude of the price increase. This is clearly captured by the gap index, which shows that on average the expenditure shortfall of poor Iraqis with respect to the line increases from 4.5 percent to percent. Note that the gap index provides an estimate of the per capita cost of eliminating. According to the estimates in Table (third column), while PDS is in place this result could be reached if each nonpoor Iraqi citizen were to pay ID 4,500 per month toward a transfer to the poor. The total amount required to eradicate in the presence of the PDS would be of the order of ID 1,261 billion per year (corresponding to US$1.0 billion). 10 Our simulations show that if we remove PDS the bill for eradicating increases: with a 25 percent price increase, it would be equal to ID 2,805 billion per year (US$2.2 billion); with a 200 percent price increase, it would rise to ID 4,610 billion (US$3.5 billion). 11 Figure provides a visual illustration. Figure Actual versus Simulated PCE Distribution for PDS Elimination with No Compensating Measures Source: Authors estimates using 2007 IHSES data In sum, the results presented show that the elimination of the PDS without any compensating measures would be socially devastating, even under the very optimistic and unrealistic scenario that prices do not increase (here, with reference to the time horizon, prices are more likely to increase by large percentages in the short run, while it is more likely that inflation could be contained in the long run). Furthermore, inflation is key in determining the magnitude of the social costs. While here we are considering estimates at the national level, Appendix shows the results of the simulation carried out separately for urban and rural areas. 10 Assuming an exchange rate of US$1 = ID 1, These calculations underestimate what might be required in reality because they are based on the unrealistic assumption of perfect targeting. By perfect targeting we mean giving the poor just enough to bring them up to the line, with no leakages to the nonpoor. Perfect targeting is of course an ideal unlikely to be met in practice. 566

6 5. Monetization of the PDS: Baseline Schemes In this section we consider the introduction of cash transfers as compensating measures when the PDS is eliminated. We examine two cash transfer schemes that can be considered as references for the more realistic ones discussed later in the annex (namely, geographic and proxy means-testing targeting). In the first subsection (5.1) we consider a universal cash transfer scheme in which a budget equal to the market value of the currently distributed rations is divided into equal transfers for supply to all Iraqis. Hence the PDS would be replaced by a system that would preserve the current universality but eliminate the inefficiencies associated with the in-kind nature of existing transfers. The total budget used for transfers in all simulations is ID 4,259 billion (approximately US$3.3 billion, assuming an exchange rate of $US1 = ID 1,300). This IHSES-based figure is remarkably close to the figure estimated by the Iraqi Ministry of Finance, according to which the total financial cost of the food system for 2007 amounted to ID 4,055 billion (approximately $US3.1 billion assuming an exchange rate of $US1 = ID 1,300). 12 With the budget considered here, a universal cash transfer system would provide each Iraqi with ID 11,793/month. 13 If we were to consider the Ministry of Finance figure and we assumed that 10 percent of the total financial cost was absorbed by administrative costs, 14 the total budget would equal ID 3,650 billion, which would correspond to ID 10,105/person/month. Thus the universal cash transfer program would cover 86 percent of the value of PDS. In the second subsection (5.2) we present the results of a system for which the same total budget is used, but equal size cash transfers are received only by people classified as poor using the line estimated at ID 76,986/person/month. Following this definition, the scheme can be referred to as one of perfect targeting. Why should targeting be considered? When aiming to reduce with a given budget, the efficiency of the alleviation measures can be maximized if the cash transfers are distributed only to poor individuals and the amount for each recipient equals the difference between a measure of welfare and a line. This would be an ideal targeted scheme. In real life, the design of transfer schemes is affected by a series of costs. Examples of such costs include those for collecting information on who is and is not poor; those incurred to acquire transfers, such as indirect costs related to the possibility that eligibility criteria induce households to change behaviors; the social stigma that may be associated with being identified as a recipient of transfers; and political costs. 15 A number of targeting mechanism types exist: ranging from those based on an assessment of the individual/household means (including all forms of means testing, from those requiring verified information to those using proxies), to group targeting in which all individuals in a group become recipient, to self-selection (see, for example, Coady, Grosh, and Hoddinott, 2004). After a comprehensive review of 122 interventions in 48 countries, Coady, Grosh, and Hoddinott (2004) conclude, generalizing the results by Grosh (1994), that targeting can work but that it does not always work everywhere. More specifically, they state that moderately progressive targeting (that is, a distribution of transfers that favors poor people rather than dispensing a random allocation of resources) is possible in many places, 12 See World Bank (2009: 44, table 4.1). 13 Population size in 2007, as estimated on the basis of IHSES, is 30,097, This is in line with the experience in other countries. 15 Analyses of the benefits and costs of targeting in developing countries are discussed by Atkinson (1995); Besley and Kanbur (1993); Devereux (1999); Grosh (1994); van de Walle (1998); and Coady, Grosh and Hoddinott (2004). 567

7 so that should be the expectation when implementing a targeting scheme. They find a weak ranking of outcomes achieved by different mechanisms, but stress that implementation matters tremendously in determining the results of a program. Universal Cash Transfers In this section (5.1) we discuss what would happen if the PDS were replaced by cash transfers, equal in nominal amount and distributed to all Iraqis. We also model increases in the prices of the items included in rations and their effects on the income of cereal farmers. Table Results of PDS Replacement by Universal Cash Transfers price increase by headcount Source: Authors estimates using 2007 IHSES data Given the considered design of the transfer scheme, if prices do not increase, the monetization of the PDS does not produce statistically significant changes in and inequality measures (compare the first and last rows of Table 9.1-2). However, price increases lead to a worsening of and inequality indicators, which can be substantial for the larger price increases considered here. Perfectly Targeted Cash Transfers gap gap squared Gini index among the poor ,396 61, ,000 61, ,797 60, ,769 59, ,900 59, ,176 58, ,585 57, ,118 57, ,764 56,401 actual ,944 61,624 In this section (5.2) we maximize the effectiveness of transfers in terms of reduction by considering a system in which all poor Iraqis are perfectly identified (via their observed PCE) and are given a transfer. Its nominal amount is fixed for all poor people and thus does not depend on the distance of the recipient s PCE from the line. The results of this simulation, as summarized in Table 9.1-3, show that the potential for reduction with targeting is very large. In fact, if perfect targeting of all poor Iraqis were feasible, very substantial decreases in all and inequality measures could be achieved. Moreover, the improvement in social conditions would happen not only if prices could be prevented from rising, but also with considerable price increases. For example, all and inequality indicators would be significantly better than the actual ones even with a doubling (plus 100 percent) of the prices of items currently included in rations. 568

8 Table Results of PDS Replacement by Perfectly Targeted Cash Transfers price increase by headcount gap gap squared Gini index among the poor real cash transfer (ID/benefi ciary/mont h) ,787 68,729 35, ,228 70,267 35, ,824 70,218 34, ,563 69,701 33, ,432 69,009 33, ,415 68,295 32, ,503 67,630 32, ,690 66,977 31, ,964 66,159 30,810 actual ,944 61,624 Source: Authors estimates using 2007 IHSES data Figure supplies a visual illustration of these results. Figure Actual versus Simulated PCE Distribution with Perfectly Targeted Cash Transfers Source: Authors estimates using 2007 IHSES data 569

9 6. Replacement of PDS by Geographically Targeted Cash Transfers In this section we consider transfer programs that allocate an equal-size cash transfer to everyone who lives in selected geographical areas. Targeting based, at least partly, on residential location is very common (see, for example, Coady, Grosh, and Hoddinott, 2004). To identify the areas to be targeted, we first consider a partition made of 54 geographical areas (18 governorates divided into rural, governorate centers, and other urban centers). Then we rank the 54 areas on the basis of their headcount index, and we define cutoff lines so as to cover the areas where 33 percent, 50 percent, 67 percent, and 75 percent of all poor Iraqis live. This means targeting, respectively, 9, 15, 22, and 30 areas out of the total 54. This procedure and its implications are described in more detail in Appendix The effects of geographical targeting on indicators are illustrated in Figure A first clear-cut result is that the share of poor people can be reduced with respect to the observed 22.9 percent figure, as long as the number of targeted areas is large enough (9 areas are too few) and increases in ration prices are not too large (below 50 percent). The improvement with respect to the current situation is observed irrespective of the chosen measure. However, on the basis of the gap and the -gap squared indices, the improvement is achievable, at least in the presence of some price increase, only with a larger number of targeted areas (at least 22). Moreover, if we aim at reducing the incidence of, the optimal number of targeted areas depends on the range of the increases in ration prices. If they are below 100 percent, it is optimal to target 22 areas (the red broken line being the lowest for this range of inflation values). If they are between 100 and 175 percent, it is optimal to target 15 areas. And if prices triple, the lowest number of poor people is achieved when 9 areas are targeted. Figure Replacement of the Public Distribution System by Geographical Targeting Using a Headcount Index Source: Authors estimates using 2007 IHSES data 570

10 Table Replacement of the Public Distribution System by Geographically Targeted Cash Transfers in 22 Locations cash transfer among the (ID/benefi price gap poor ciary/mont increase headcount squared Gini index h) by gap ,940 62,152 32, ,521 61,525 31, ,305 60,914 30, ,271 60,196 29, ,404 59,862 28, ,689 59,110 27, ,112 58,600 26, ,662 58,119 26, ,330 57,447 25,496 actual ,944 61,624 Source: Authors estimates using 2007 IHSES data Table presents the results for 22 targeted areas (those for the other sets of targeted areas and for lower total transfer budgets are presented in Appendix of this annex). Finally, we need to remember that a targeting based on where people live necessarily implies a substantial geographical redistribution of poor and nonpoor people. Summing up the above evidence, we can draw a number of useful conclusions. First, the optimal number of targeted areas depends on the rate of inflation. This emphasizes, once again, the crucial role played by the inflation induced by the elimination of the PDS. The error of exclusion (that is, the share of poor people who do not receive benefits) and of inclusion (the share of beneficiaries who are not poor) in geographical targeting are indicators of undercoverage and leakage, commonly used to assess efficiency in tackling and the cost effectiveness of targeting, respectively (see, for example, Coady, Grosh, and Hoddinott, 2004). For 22 targeted areas, around onethird of the poor people do not receive transfers (corresponding to 35 percent of poor households), and around 60 percent of beneficiaries are not poor (corresponding to 69 percent of all households receiving transfers). Furthermore, even in the scenario where 30 areas are targeted, a few governorates have no targeted areas within their borders. Because the absence of targeted areas implies no program resources transferred to the population of that governorate, the actual implementation of this geographical targeting could be politically difficult. The results illustrated in this section suggest that geographical targeting has some important advantages, including the fact that once targeted areas are identified, so are all the beneficiaries. Seemingly this makes the scheme relatively easy to implement. However, in practice, implementation is extremely complicated because the optimal number of targeted areas depends on (and turns out to be very sensitive to) the inflation rate. 571

11 7. Proxy Means-Tested Transfers While geographical targeting is based on residential location, other targeting mechanisms start from measures of the resources available to each potential beneficiary (individual or household). In particular, means testing, another quite widely employed method, requires the collection of information on the total income or consumption of the potential beneficiaries. This can happen either via a visit to the house of the potential beneficiary by an officer working on the program, or via a visit to a program office by the claimant individual/household. The information thus collected can either be taken at face value or verified (by documentation supplied by the potential beneficiary or by using other sources, for example fiscal databases). Benefits are then granted if the claimant s resources are found to be lower than a threshold or cutoff. The characteristics of this targeting mechanism make it more suitable for countries where economic transactions are usually well documented (see Coady, Grosh, and Hoddinott, 2004). Yet even in those, it implies incurring substantial costs for the collection and, possibly, verification of a large number of variables for each potential beneficiary. A variant of means testing that is more appropriate for settings where informal economic transactions are widespread and that is based on a smaller information set is proxy means testing (PMT). This mechanism is more recent, coming into use in 1980 in Chile and in other Latin American countries, and later applied in areas as disparate as Armenia, Indonesia, and Turkey (Coady, Grosh, and Hoddinott, 2004). PMT requires the assessment of a small set of household characteristics that, as in means testing, can be observed either via a visit by a program officer to the claimant s house or by the claimant s visit to a program office. These characteristics should be related to a measure of welfare such as consumption or income, should be easy to observe and verify, should not be easily manipulated by the claimant, and should not change rapidly with time. This last characteristic is essential because the reassessment of a claimant s situation is usually infrequent, often a few years after the initial approval. Hence, this mechanism is appropriate to tackle chronic or extreme (see, for example, Grosh et al., 2008) rather than related to shocks. The value of each of the observed characteristics for a given claimant is used, together with previously estimated weights, to calculate a score. As in means testing, this is compared with a threshold to assess whether the claimant actually qualifies for program benefits. The performances of PMT targeting systems have been considered comparatively good for the share of benefits that go to the poorest people (see, for example, Grosh, 1994; and Castaneda and Lindert, 2005). However, PMT may risk failing to identify the poorest of the poor due to the difficulties in accurately estimating very low consumption expenditures (see Grosh and Baker, 1995; and Hou 2008). To overcome this problem, it could be assessed whether it is feasible and it would be suitable to complement PMT with other targeting methods, such as geographical targeting, or with efforts at the community level to involve the poorest people in the program (see Coady, Grosh, and Hoddinott, 2004). In the following, we present the simulation results for implementation of PMT-based cash transfers as substitutes for the PDS. The PMT algorithm has been built via the following steps: Estimate a regression model for the PCE on the basis of a set of observable and verifiable household characteristics. 16 Define the PMT formula via the estimated regression coefficients. This formula is then used, together with the observed values of the considered characteristics, to calculate a score for each household. Identify the threshold to be used to select beneficiaries by minimizing the headcount index that PMT-based transfers would produce. 16 The estimations were run by OLS, as it usually done due to the convenience and ease of interpretation of the results. See Appendix to this annex for details. 572

12 Calculate the transfer per beneficiary per month (as a ratio between the total budget and the number of beneficiaries). 17 After experimenting with different specifications, we have chosen one in which we try to strike a balance between observability, verifiability, nonmanipulability, and stability of the household characteristics included as regressors, and statistical explanatory power. 18 The preferred specification includes sociodemographic characteristics of the household (household size, household head s gender and age, a dependency ratio, educational attainment of the household head), housing characteristics (whether the house is rented, crowding, ceiling and floor materials, quality of the sewage system and of the road leading to the house), geographical location, and labor market status of the household head and whether he/she works in the public sector. It also includes whether a household member suffers from a disability. As documented in Appendix 9.1-4, the econometric results are remarkably satisfactory considering the special Iraqi conditions during the data collection period. In particular, the R-squared of the chosen specification is in line with those in the literature (see, for example, Sharif, 2009; Hou, 2008; Narayan, Viswanath, and Yoshida, 2005; Ahmed and Bouis, 2002; Glinskaya and Grosh, 1997; and Grosh and Baker, 1995). Furthermore, regressors generally are significant and their estimated coefficients have the expected signs. The headcount index minimization criterion leads to a threshold equal to the 36 th percentile of the simulated PCE distribution (see Appendix 9.1-4). Table Replacement of PDS by Proxy Means-Testing Targeted Cash Transfers Using Poverty and Inequality Indicators cash transfer among the (ID/benefi price gap poor ciary/mont increase headcount squared Gini index h) by gap ,822 64,058 33, ,328 63,444 33, ,686 62,585 33, ,385 60,511 33, ,982 57,867 33, ,030 53,901 33, ,027 49,190 33, ,378 43,861 33, ,377 38,513 33,545 actual ,944 61,624 Source: Authors estimates using 2007 IHSES data 17 A more complicated program design could allow the size of the transfer to vary with the distance between the estimates score and the threshold. For illustrative purposes, we have maintained a simple program design. 18 See the column final in Table 9.1-A4.1 in Appendix of this annex. 573

13 Figure Actual versus Simulated PCE Distribution Using Proxy Means-Testing Targeted Cash Transfers Source: Authors estimates using 2007 IHSES data Table shows the results of the simulations, and Figure illustrates their effects. As usual, we can interpret the first row of Table as our best approximation to what might happen in the medium/long run. We can observe that all and inequality indicators significantly improve with respect to the actual figures. This improvement remains also for relatively contained ration price increases (up to 50 percent). However, if the table s larger inflation values (which we interpret as our best approximation of short-run eventualities) should be incurred, this transfer scheme would result in a very severe worsening of the current situation and extremely high social costs. Table presents the measurement of coverage errors for both individuals and households. In particular, we can observe that 22.7 percent of poor Iraqis, corresponding to 29.1 percent of poor households, fail to obtain transfers. This is the error of exclusion or undercoverage. On the other hand, 49.7 percent of beneficiary individuals, corresponding to 51.7 percent of beneficiary families, are not poor. This is the error of inclusion or leakage. 574

14 Table Replacement of PDS by Proxy Means-Testing Targeted Cash Transfers Using Undercoverage and Leakage Indicators individuals (absolute numbers) target group non-target group total beneficiary 5,321,059 5,259,714 10,580,773 non-beneficiary 1,562,588 17,953,639 19,516,227 total 6,883,647 23,213,353 30,097,000 individuals target group non-target group beneficiary non-beneficiary households (absolute numbers) target group non-target group total beneficiary 513, ,513 1,062,956 non-beneficiary 210,736 3,111,356 3,322,092 total 724,179 3,660,869 4,385,048 households target group non-target group beneficiary non-beneficiary Source: Authors estimates using 2007 IHSES data It is worth remembering that a trade-off exists between leakage and undercoverage. To reduce leakage, the assessment mechanism to identify beneficiaries must become stricter. In turn, this is likely to increase the number of poor people who fail to obtain transfers. Moreover, the relative relevance of leakage and undercoverage depends on policy makers preferences and purposes. If the aim is minimizing, undercoverage should be the worry. If the aim is efficient use of the budget allocated to transfers, the greater concern would be for leakage. Note that given the availability of the PCE in IHSES data and the line we have calculated undercoverage and leakage for poor and nonpoor people as defined by whether their observed PCE is below or above the line. In the literature, these indicators often are calculated without reference to a line but to the first x percentiles of the PCE distribution. Because the line approximately corresponds to the 23 rd percentile of the PCE, we can compare our results with those in the literature for similar percentiles of the PCE distribution. Usually in the literature the cutoff level and the target population are the same percentile of the PCE distribution. In our case, the target population corresponds approximately to the 23 rd percentile of the actual PCE distribution, while the cutoff line is the 36 th percentile of the simulated PCE distribution. The literature contains many examples of these indicators for PMT programs: Sharif (2009), for Bangladesh for the 20 th percentile of the actual PCE, undercoverage equals 57 percent and leakage equals 40 percent; for the 25 th percentile of the actual PCE, undercoverage equals 49 percent, leakage equals 3 3 percent. 575

15 Narayan and Yoshida (2005), for Sri Lanka for a cutoff level at the 30 th percentile of the predicted PCE distribution, undercoverage is 37 percent and leakage is 37 percent (the two indicators have equal value). Hou (2008), for Pakistan for a cutoff level at the 30 th percentile of the actual PCE distribution, undercoverage is 48 percent and leakage is 35 percent. Grosh and Baker (1995), for Jamaica for a cutoff level at the 30 th percentile of the actual PCE distribution, undercoverage equals 69 percent while leakage equals 44 percent. Castenada and Lindert (2005) for a cutoff level at the 30 th percentile of the actual PCE distribution for urban Bolivia: undercoverage equals 39 percent and leakage equals 24 percent; for urban Peru (same cutoff level): undercoverage equals 54 percent while leakage is 35 percent. The comparatively high leakage and low undercoverage obtained here with respect to the just mentioned papers can probably be explained with the choice of a cutoff threshold higher than that defining the target group. Furthermore the PMT algorithm was built using IHSES data collected in The robustness of the results should be verified on the data of a new survey wave, when they become available. To conclude, we can argue that a PMT targeting mechanism seems worth exploring in the Iraqi case given the encouraging results obtained here with an algorithm that is not fine-tuned, but only aimed at assessing the potential of the targeting system. However, the simulation shows that the improvement in social conditions is attainable only if inflation in the prices of the items so far included in the rations is kept under control. If this cannot be guaranteed, the risks of very high social costs cannot be disregarded. 8. A Dilemma for Policy Makers In this paper we have explored a wide range of options for PDS elimination. The main aim has been to gain insight into what the implications for each type of strategy could be. For this reason, the results presented should be considered as a starting point toward what could be obtained if each strategy were carefully fine-tuned. In this sense, we interpret our results as useful in that they provide guidelines for further research. That said, a number of relevant and clear-cut findings have emerged. The main finding of our analysis is that the choice between geographical targeting and PMT is not straightforward since it crucially depends on the expected inflation rates. We find that if the reform of the PDS scheme succeeds in keeping ration price increases under control, then PMT performs better than geographical targeting in many respects. All indicators improve more and there is less concern for politically sensitive issues. On the other hand, geographical targeting performs better than PMT under scenarios with larger expected inflation rates. In particular, it is worth noting that to the extent to which the authorities are not confident of being able to prevent inflation rates similar to the largest ones considered here, the outcome of the PMT could become socially devastating. Moreover, the outcome of geographical targeting depends on the number of targeted areas. But the optimal value for this parameter changes with the inflation in the prices of the items thus far included in the rations, a variable that can be difficult to predict and can be subject to large variations over time. This feature may make the best possible outcome of geographical targeting hard to reach in practice. 576

16 REFERENCES Ahmed, A., and H. Bouis Weighing What s Practical: Proxy Means Testing for Targeting Food Subsidies in Egypt. Food Policy 27: Atkinson, A On Targeting Social Security: Theory and Western Experience with Family Benefits. In D. van de Walle and K. Nead (eds.), Public Spending and the Poor. Baltimore: John Hopkins University Press. Besley, T., and R. Kanbur Principles of Targeting. In M. Lipton and J. van der Gaag (eds.), Including the Poor. Washington, D.C.: World Bank. Bigman and Fofack Geographical Targeting for Poverty Alleviation. World Bank Economic Review 14 (1): Castaneda, T., and K. Lindert Designing and Implementing Household Targeting Systems: Lessons from Latin America and the United States. Social Protection Discussion Paper Series No. 0526, World Bank, Washington, D.C. Coady, D., M. Grosh, and J. Hoddinott Targeting of Transfers in Developing Countries: Review of Lessons and Experience. Washington, D.C.: The International Bank for Reconstruction and Development. Devereux, S Targeting Transfers. IDS Bulletin 30 (2): Glinskaya, E., and M. Grosh Proxy Means Testing and Social Assistance in Armenia. Mimeographed document, Development Economics Research Group, World Bank, Washington, D.C. Grosh, M Administering Targeted Social Programs in Latin America: From Platitudes to Practice. Regional and Sectoral Studies, World Bank, Washington, D.C. Grosh, M., and J. Baker Proxy Means Tests for Targeting Social Programs: Simulations and Speculation. Working Paper No. 118, Living Standards Measurement Study, World Bank, Washington, D.C. Grosh, M., C. Del Ninno, E. Tesliuc, and A. Ouerghi For Protection and Promotion: The Design and Implementation of Effective Safety Nets. World Bank, Washington, D.C. Hou, X Challenges for Targeting the Bottom Ten Percent: Evidence from Pakistan. Mimeographed document, World Bank, Washington, D.C. Narayan, A., T. Viswanath, and N. Yoshida Sri Lankan Welfare Reform. Poverty and Social Impact Analysis, World Bank, Washington, D.C. Narayan, A., and N. Yoshida Proxy Means Test for Targeting Welfare Benefits in Sri Lanka. Report No. SASPR-7. World Bank, Washington, D.C. Sharif, I. A Building a Targeting System for Bangladesh Based on Proxy Means Testing. Discussion Paper No. 0914, Social Protection and Labor, World Bank, Washington, D.C. van de Walle, D Protecting the Poor in Vietnam s Emerging Market Economy. Policy Research Working Paper No. 1969, World Bank, Washington, D.C. World Bank Considering the Future of the Iraqi Public Distribution System. World Bank, Washington, D.C World Bank Iraq Poverty Assessment Report. Draft, (June 8). 577

17 Appendix How Ration Price Increases Affect the General Consumer Price Index The Laspeyres index is defined as follows: [9.1-A1.1] where r is for ration. Elimination of the PDS implies that ration prices increase. Assume that. [9.1-A1.2] This means that ration prices increase by new price index becomes percent when the Public Distribution System is removed. The [9.1-A1.3] Equation can be simplified for computational purposes as follows: [9.1-A1.4] where is the average budget share that households spend on ration items, measured on the quantities at time zero. An analogous procedure can be applied to the Paasche index. By so doing we obtain P ' P w [9.1-A1-5] t t t r t where w r is the average budget share that households spend on ration items, measured on the quantities at time t. The Fisher index can be therefore expressed as follows:. [9.1-A1.6] The index above was used to simulate the price increases due to the elimination of the PDS. 578

18 Appendix What Would Happen in Urban and Rural Areas and to Cereal-Producing Farmers if the PDS Were Eliminated? In Tables 9.1-A2.1 through 9.1-A2.3, we can see the effects on the urban and rural population and on cereal-farming households of the elimination of PDS accompanied by increases in prices when the impact of these price increases on cereal-farming households are considered. For the urban population (Table 9.1-A2.1), the simulations show large increases in both the headcount index and the gap indices. The headcount reaches 40.2 percent and the gap index 12.1 percent if prices increase by 200 percent (that is, if they triple). Comparing these figures with the observed 16.1 percent for the headcount index and 2.7 percent for the gap index reveals that the worsening of conditions for the urban population is striking. Even if the shares of poor by urban/rural area change in favor of the rural population, individuals there also suffer dramatically from PDS elimination. From Table 9.1-A2.2, we can see that a 200 percent price increase after the elimination of PDS would make 60.1 percent of the rural population poor and generate a gap index of 22.6 percent (this compares with the observed headcount index equaling 39.3 percent and the gap equaling 9.0 percent). Table 9.1-A2.1. PDS Is Eliminated, Prices Increase, and Their Impacts on Cereal Farmer Incomes Are Taken into Account Effects on the Urban Population Price increase by Headcount Poverty gap Poverty gap squared Gini index Mean PCE Mean PCE among the poor ,433 59, ,699 58, ,128 57, ,708 57, ,425 56, ,270 55, ,232 54, ,303 54, ,475 53,650 actual ,944 61,624 Source: Authors estimates using 2007 IHSES data 579

19 Table 9.1-A2.2. PDS Is Eliminated, Prices Increase, and Their Impacts on Cereal Farmer Incomes Are Taken into Account Effects on the Rural Population price increase by headcount gap gap squared Gini index among the poor ,213 52, ,927 52, ,819 51, ,869 50, ,062 50, ,383 49, ,821 49, ,365 48, ,006 47,990 actual ,944 61,624 Source: Authors estimates using 2007 IHSES data For cereal-farming households (Table 9.1-A2.3), the rates are estimated to be by and large similar to those observed for rural households (headcount index equals 39.3 percent; gap index equals 9.0 percent; gap squared index equals 3.1 percent). Table 9.1-A2.3. PDS Is Eliminated, Prices Increase, and Their Impacts on Cereal Farmer Incomes Are Taken into Account Effects on Cereal-Producing Households price increase by headcount gap gap squared Gini index among the poor ,213 52, ,096 52, ,385 52, ,833 52, ,425 52, ,146 51, ,985 51, ,930 51, ,973 51,791 actual ,944 61,624 Source: Authors estimates using 2007 IHSES data 580

20 Once the PDS is eliminated and prices have increased, cereal producing farmers become poorer than they are with PDS. True, the pattern of simulated rates shows that the higher the price increase the smaller the increase in indicators, but cereal producing farmers are definitely worse off than in the presence of the PDS. Even with large price increases (that is, large positive income effects), the headcount index would be slightly smaller than in the presence of PDS, but the gap and the gap squared indices would be substantially larger. Thus, for cereal-producing households the positive income effect fails to compensate for the negative price effect (loss of purchasing power on all nonproduced goods). All this suggests that one must be careful in using the terms winners and losers. Cereal producing farmers are winners relative to all other Iraqis but not in absolute terms (while all the other Iraqis are losers both in relative and absolute terms). In other words, PDS removal is not a zero or positive sum game. The analysis suggests that all Iraqis are likely to suffer, with some suffering more than others. 581