70046 Yemen Energy Subsidy Reform A Report by The World Bank Visiting Mission of May 24-June 1, 2008 I. Introduction A team of World Bank staff visited Yemen during May 24 to June 1, 2008 to provide technical assistance to the Government of Yemen in addressing issues of energy subsidies. The team is grateful for very generous hospitality and effective cooperation from all the concerned entities. The team summarized the results of the mission work in a presentation to the Economic Committee. The Present members were the Deputy Prime Minister (DPM), the Minister of Finance, the Minister of Electricity, the Minister for Trade and Industry, a number of deputy ministers, and representatives from other relevant entities. This report provides a brief description of the conclusions and recommendations of the mission. II. Context Energy subsidies are known to: (a) result in inefficient use of energy resources and distortion in related technologies; (b) impose a heavy burden on the government budget and jeopardize fiscal sustainability; and (c) benefit the rich who consume the major share of the subsidized energy. The recent increase in the international energy prices has brought to the forefront the need to reduce energy subsidies through a well designed process which takes account of various consequences of price adjustment. The World Bank has reviewed the case of Yemen in a Policy Note: Options for Enabling Petroleum Products Price Reform. The Policy Note examines the impact of petroleum price adjustment on cost of living (inflationary impact); the level of poverty (distributional impact); and the government budget (fiscal impact). The paper also reviews the social protection measures that can be taken to mitigate the impact of price increase on the poor. Drawing upon the analysis of the Policy Note and the experiences of other countries, the mission has prepared a set of recommendations which are summarized in this paper. The main questions to be examined are: 1. To what extent one can reduce subsidies without increasing the price of the corresponding energy product? 2. To what extent one has to increase prices of energy products? 3. Which energy products and what speed of price increase should be adopted? 4. How to protect the vulnerable part of the population? 5. How to use the savings from energy subsidy reduction? 6. How to communicate with the public? III. Reducing the cost of energy supply First, the mission examined the possibilities of reducing energy subsidies by lowering the cost of supply. Although the mission has identified a number of efficiency improvement measures in the electricity and the petroleum refining sectors, the most immediate savings can be achieved by switching from oil to gas-based power generation. This is well recognized and incorporated in the development plan of the power sector in the form of three important gas-based power plants: Marib 1; Marib 2; and Maabar. Despite the urgency of commissioning of these plants, they are all facing significant delays. The costs of these delays are very tangible and paid from the government budget in the form of huge subsidies for diesel and fuel oil consumed in power generation. Marib 1 and Marb 2 would replace oil-based power generation in the PEC system as well as in the 187 MW rental units. The savings from commissioning these two units is estimated at $ 514 million/year (see Table 1 for details). Savings from commissioning the Maabar power plant are also quite significant. It will partly replace the existing oil-based generating capacity and partly meet the growth in electricity demand which would be oil-based in the absence of the gas-based Maabar plant. Table 1: Cost of Construction Delays of Marib 1 and Marib 2 Plant Marib 1 Marib 2 Total Capacity MW 340 400 740 Annual production GWh 2234 2628 4862 Fuel consumption, annual HFO million liters 0 525.6 526 Diesel million liters 558.45 328.5 887 Natural gas MMBTU 23231520 27331200 50562720 Gas in volume bcf 22.6 26.5 49.09 (gas in volume for 25 years) tcf 0.56 0.66 1.23 Annual subsidy HFO billion YR 0.0 1.0 1.00 Diesel billion YR 79.9 47.0 126.83 Natural gas billion YR 11.8 13.9 25.64 HFO million $ 0.0 5.0 5.02 Diesel million $ 401.3 236.1 637.36 Natural gas million $ 59.2 69.6 128.82 Net saving bilion YR 68.1 34.1 102.20 Net saving billion $ 342.1 171.4 513.55 Savings per day million $ 0.9 0.5 1.41 Annual savings 342 171 514 The subsidy picture in the electricity sector is quite complicated. There is an apparent or explicit subsidy of 5 R/kwh which is based on the difference between the average generation cost of 17 R/kwh and the average selling price of 12 R/kwh. However, the electricity sector receives full subsidy for its investment costs (estimated at 6 R/kwh) and more than 80% of its fuel cost (estimated at 13 R/kwh). Therefore the total subsidy is at least 24 R/kwh (about 12 US cents/kwh). This means that presently electricity price covers only 1/3 of the cost while the other 2/3 are subsidizes. Despite this huge subsidy, the electricity company (PEC) is not able to keep up with its ongoing financial obligations. The desperate financial picture of the electricity sector and the huge burden on the government budget are expected to continue until the new gas-based electricity generating plants come on line and enable the power sector to cut back on its use of very expensive diesel and fuel oil. The main reasons for these delays are lack of effective coordination between the gas and power sector, and the lack of a strong leadership in resolving the outstanding issues. IV. The Need for Price Increase Energy subsidies during 2007 are estimated at $ 2.4 billion. Figure 1 shows the flow of these subsidies by fuel and the consuming sectors. The subsidy associated with diesel oil is by far the largest component accounting for about 58% of total energy subsidies. Figure 1: Flow of Energy Subsidies (2007) YR ($) both in millions Domestic Total: 474 YR billion (except electricity) 223,646 ($1,124) ($2.4 billion) Diesel 86,073 ($433) Gasoline 32,939 ($166) LPG 12,306 ($62) Jet fuel 9,174 ($46) Kerosene Electricity Fuel 57,591 ($290) Subsidies HFO 52,383 ($263) Diesel Electricity Tariff 25,500 ($130) Subsidy* 6 * Rough estimate based on 2006 data The total amount of energy subsidies will be significantly higher in 2008 due to a sharp increase in the international oil price and the growth in energy consumption. As a result, energy subsidies are expected to reach R 609 billion ($3 billion), accounting for about 37% of the government current expenditures in 2008, compared to 10% in 2002. At their present levels, energy subsidies represent the largest component of the government budget exceeding social expenditures, and government wages and salaries, which account for 25% and 30% of the budget, respectively. Thus the fiscal impact of energy subsidies is clearly unsustainable. The distributional impact of energy subsidies is also of significant concern because these subsidies are sharply regressive benefiting the rich. The 20 % richest receive 43% of these subsidies while the 20% poorest receive only 8%. However, sudden removal of subsidies without compensation increases the poverty rate by 9.2 percentage points (7.6 points for urban / 9 points for rural), resulting in 1.5 million Yemenis falling into poverty. While the need for reducing energy subsidies is well established, the government should design measures to minimize the hardship on the poor and the lower middle class before embarking upon any subsidy reform. V. Social Protection The experience in other countries indicates that the government should launch a program of transitional relief for the lower and middle income population at the same time that it reduces energy subsidies. This relief can take a variety of forms including unconditional and conditional cash transfers, as well as targeted socio-economic activities. The Government of Yemen has recognized this requirement, and has identified the Social Welfare Fund (SWF) as a possible instrument for transitional relief in the form of unconditional cash relief. The mission’s review of the potential options confirms the suitability of SFW with three major caveats. First, aside from energy subsidy, SWF plays a fundamental role in poverty reduction. For that role to be effective, SWF needed to improve: (i) its method of screening (targeting); (ii) its application pool and processes; and (iii) its cash delivery mechanism. The existing targeting of SWF is categorical (orphans, widows, disables, etc.) and needs to be adjusted to a more explicit measure of poverty. An EU supported TA has proposed a scoring method which would use a number of household characteristics to proxy the income status of potential applicants. The World Bank has conducted a “proxy means testing� to examine how closely those household characteristics are correlated with household welfare (consumption expenditure). These exercises have been now completed and a formula has been adopted. There are also some on going efforts to improve the application pool. A survey has been piloted to identify potential beneficiaries of SWF. Awareness initiatives have been taken to publicize the SWF facility even in remote rural areas. Finally, the cash delivery mechanism is under improvement through some innovative arrangements. While improving various aspects of SWF is a continuous process, the instrument is now considered acceptable and operational. However, SWF needs significant further support to strengthen its institutional capacity. Second, the use of SWF in relation to energy subsidy adjustment should be structured separately from SWF’s baseline operation. The transitional relief has to be: (a) explicitly designated as compensation for energy price increase; (b) well publicized and communicated with the public; (c) temporary within an explicit time-frame; and (d) covering a larger portion of the population than those falling below the poverty line. The present coverage of SWF is limited to about 1 million cases. The new application pool of SWF is expected to reach 1.5 to 2 million cases, which would correspond with a population of 7 to 8 million. The mission recommended the use of this wider application pool for transitional relief associated with energy subsidy reduction. The mission also recommended including public service employees (about 900,000 employees representing about a 5 million population) in the pool for transitional relief. The sum of the two populations, after adjusting for the overlap of recipients would represent 10 million which is likely to cover the most vulnerable group affected by energy subsidy reduction. Third, in addition to the unconditional cash transfer, the government should use part of savings from subsidy reform to support high priority socio-economic activities. This is an integral component of the plan for using the relevant savings (discussed in the next section). VI. How to use the Savings from Energy Subsidy Reform The international experience indicates that the savings from subsidy reform can go to waste if they are not explicitly and effectively managed. Normally these savings are allocated to two distinct purposes. First, part of the savings, e.g., 30%, could be used to compensate the poor and near poor for the direct and indirect impacts of energy price hikes. Second, the remainder of these savings would go to finance the budget deficit and to increase expenditure on high priority projects/activities. There are two significant risks in the process: (a) that the compensation is untargeted; and (b) that the rest of the savings are channeled to existing mediocre programs and projects. To mitigate against the first risk, the cash (and non-cash) transfers have to be well targeted, implemented, and monitored. The discussion in the previous section is aimed at clarifying the improvements that need to be made to ensure more effective cash transfer. To mitigate against the second risk, the government would need to provide an explicit and transparent account of how it plans to use the remainder of savings from subsidy reform. Projects which target the poor like education (e.g., operational aids to schools), health (e.g., health care and health insurance for the poor), rural infrastructure (e.g., village roads, water, irrigation) are economically, socially and politically attractive. The allocation process should favor projects which are well focused and visible. More importantly, the allocation process should take a systematic and transparent form which can be easily demonstrated to the public. In other words, the subsidy reduction should be taken as an opportunity to promote a strategic approach to allocate government expenditures; and to improve pro-poor spending generally. As an illustrative example, the proposed subsidy reduction schedule (Section VII) results in an overall savings of about $ 6 billion over the course of 2.5 years. If the government allocates 30% (about $ 1 billion) 1of these savings to transitional cash transfer, still there will be some $ 2 billion in savings which should be allocated to funding the budget deficit and to financing high priority socio-economic projects. Normally the savings from subsidy reduction will have to be used within a medium -term macroeconomic framework and public investment plan (PIP). However, the existing budgetary process may not provide sufficient transparency and accountability. To provide a clear and presentable picture, the government would need to: (1) re-work the macro framework, under the assumption of phased subsidy removal as recommended in the next section. This will help staying the course on fiscal sustainability and macro economic management. (2) identify projects under PIP that could be presented as low-hanging 1 The proportion of savings allocated to transitional relief is normally in the range of 30% to 40%. It is often set at a magnitude that would relief the consumers who would consume certain amount of fuel (e.g., 30 liters of diesel per month). The transitional relief should be at the first stage calculated based on the initial price increase in order to avoid over-commitment by the Government. fruits in education, health and rural infrastructure that have a great development outcome, and have wider shared benefits with short gestation lags; and (3) examine the possibilities of ring-fencing select investment programs from being distorted. VII. Schedule of Price Adjustment The Bank’s Policy Note has developed a number of scenarios for petroleum price adjustments. The mission used those scenarios to examine the priority and extent of price adjustments. The most important considerations in a schedule of price adjustment are: (a) the share of subsidy to each product; (b) the risk of (distortive or inefficient) substitution among the fuels; and (c) the social, political and practical challenges associated with the price increase of each energy product. As mentioned earlier, in the case of Yemen the amount of subsidy on diesel oil accounts for close to 60% of the total energy subsidies. Gasoline is the next big ticket item. Thus, diesel oil and gasoline should be the main focus of price adjustment at an early stage. However, kerosene presents some serious risk of substitution for diesel oil and therefore should be priced to avoid such substitution. Therefore we recommend that the government focuses on three products: diesel oil, gasoline, and kerosene at the initial stage of subsidy reduction. The speed of price adjustment should also consider the present level of per unit subsidy. Table 2 shows the prices of petroleum products compared with the cost of supply. Again, diesel oil and kerosene are most subsidized products while the current price is about 26% of the cost of supply if we take as the benchmark the average international price of the last quarter of 2007; the ratio drops to 24% if we use the average international price of the first quarter 2008. The gasoline price covers a significantly higher portion of the cost of supply. Table 2: Extent of Fuel Subsidies Reference cost (YR/liter) Price paid Share of cost paid (%) (YR/liter) 4th qtr. 2007 1st qtr. 2008 4th qtr. 2007 1st qtr. 2008 Diesel for transport 136 148 35 26 24 and other uses Diesel for electricity 136 148 17 13 11 HFO for electricity 99 101 25 25 25 Gasoline 125 133 60 48 45 Kerosene 135 144 35 26 24 Jet fuel 135 144 36 27 25 LPG 136 158 24 18 15 Source: International Prices – World Bank. Based on the above considerations we recommend a price adjustment package including (a) initial doubling with steady semester increase of diesel and kerosene prices; and (b) a steady semester increase of gasoline prices. The price of LPG would not be increased at an early stage because of the potentially high political and social cost associated with the LPG price increase. The proposed price adjustments are shown in Figures 2 and 3 below. Figure 2: Proposed Scenario for Adjusting Diesel and Kerosene Prices Figure 3: Proposed Scenario for Adjusting Gasoline Prices The mission estimated the potential savings and the inflationary impacts of the recommended package with the assumption that the first price increase would take place in May 2009 (Table 3). At an early stage the government should only consider the savings associated with the first price adjustment while allocating resources to transitional relief. However, the planning process should take account of total savings and arrive at a list of high priority projects and activities that can be funded with such savings. This list is not only essential for encouraging pro-poor projects and ensuring a systematic targeting, but also required for communication with the public. Table 3: Potential Savings and Inflationary Impact of the Proposed Schedule 2nd Semester 2009 2010 2011 Cumulative Potential savings ($ Million) $427 $2,016 $3,671 $6,113 Impact on cost of living (%) increase 7.3% 5.8% 3.5% 17.5% The government would need to set a date for the first price adjustment. For the sake of the discussion in this note we have assumed a date of May 2009 for the first price adjustment. To adhere to this date the government would need to initiate certain preparation activities as soon as possible. VIII. Communication Strategy The government recognizes the importance of a well formulated communication strategy that needs to be launched in a very calculated and systematic manner. The experience in other countries has shown that effective communication is a make-or-break aspect of subsidy adjustment. The communication strategy should start at least 6 months prior to the first price hike. At its initial stage it should concentrate on various well focused messages to explain the existing situation, e.g., the international price trends, the amount of subsidies, the distribution of subsidies, the deprived potential projects, etc. The next phase, still prior to the first price hike, should focus on how the government plans to protect the vulnerable groups, and how to use the savings from subsidy reduction to fund highest priority projects. Finally the communication strategy should continue through the implementation of the subsidy reform explaining challenges, practical problems, and modifications. The government would need substantial support in preparation of a well designed communication strategy. During the mission’s discussion with the donors, DFID expressed their potential willingness to provide support after the government has made a decision about the date and the schedule of the price adjustment. The World Bank should also provide substantive support (cross country experience and professional expertise) while DFID would help the Government prepare and implemnet the communication strategy. IX. Conclusions and Next Steps The first priority in addressing energy subsidies is to reduce the cost of energy supply in any possible way. In this regard the government needs to pursue various (recommended) measures to increase electricity savings (e.g., CFLs). However, the most urgent matter relates to decisions on construction/completion of Marib 1, Marib 2, and Maabar projects. Lack of coordination jeopardizes the timely implementation of these projects resulting in a huge cost to the government and the economy. The government needs to establish a high level task force to monitor and solve coordination problems. It should also resolve issues in gas pricing and gas allocation. The next immediate priority is to agree on the schedule of adjusting the prices of petroleum products. The recommended schedule for various fuels is as follows: (a) diesel oil: initial doubling with steady semester increase; (b) kerosene: initial doubling with steady semester increase; and (c) gasoline: steady semester increase. Associated with this schedule we recommend a transitional relief package with the following characteristics: 1. Amount: 30% of savings, though one could consider higher ratios 2. Population covered: 12 million though there will be a 2-3 million overlap 3. Channel of distribution: SWF and public service employees 4. Frequency of distribution: quarterly for 2 years though one could consider a longer period. The preparatory process can progress further only when the government sets a date for the first price adjustment. The preparatory process would include formulation of a well designed communication strategy. This process should start some 9 to 10 months prior to the first price adjustment. It is therefore already a tight schedule if the government plans to introduce the first price adjustment in May 2009. DFID should be requested to initiate the assistance in this regard. The preparation process should also include identifying pro- poor projects which are well focused and visible. These projects should be introduced in the communication strategy to ensure the public that the savings from subsidy reform are used in the most effective manner. The World Bank team plans to help in identification of these projects. Finally, implementation of the subsidy reform would require the designation of a leader (champion) who can bring together the political, social and economic aspects of the matter, and to steer the communication with the public. Annex: The Results the Proxy Means Test (PMT) This annex briefly describes the development of a proxy means test (PMT) that can be used to determine the population eligible to receive benefits from the Social Welfare Fund (SWF) cash transfer program. Simulated poverty outcomes are generated from applying the PMT and other targeting approaches to the Yemeni population using the Household Budget Survey 2005-2006. The first section reviews the background and the current targeting procedure being considered. The second section discusses the proxy means test, and the third section compares the PMT to several other targeting methods including the score-based approach and the concluding section describes potential administrative costs associated with the targeting methods. I. Background The SWF was developed in 1996 to provide cash transfers to the most needy and vulnerable Yemenis. The program has gradually expanded from 102,000 to nearly 945,000 cases by 2006 with a budget of about YR 16 million. Program targeting was based on categories of eligibility assumed to comprise especially vulnerable groups, including single mothers, orphans and the elderly. However, the poverty of applicants was not rigorously assessed. In 2006, only 13 percent of the poor actually received transfers from SWF, while nearly 45 percent of those who received benefits were not poor.2 With the ongoing assistance of the EU and more recently the World Bank, the SWF is in the process of fundamental reform, including changing the legislation establishing eligibility, the targeting methods and the administration of the program. It is expected that the SWF will play a more prominent role in the social safety net system. A key reform is to require that individuals should be considered to be poor to be eligible to receive SWF cash benefits. The proposed legislation will make poverty a precondition for the receipt of SWF benefits, making the assessment of individual poverty status by SWF staff of critical importance. The remainder of the annex examines which targeting methods perform well in terms of identifying the poor, focusing attention on two proposed variations of a proxy means test (PMT). A full PMT, the currently proposed scoring approach, universal targeting and basic geographic targeting are considered. II. Scoring Provided there are no information or cost constraints, the ideal poverty targeting mechanism is means testing. Benefits are given to households or individuals with incomes below a pre-established threshold, where incomes are determined through tax, workplace and other administrative records. Means testing is used with social programs in many OECD and industrialized countries. However, in developing countries such as 2 Government of Yemen, World Bank, UNDP. November 2007. Yemen Poverty Assessment Volume 1:Main Report. Yemen, collecting household income information can be costly and records are inadequate. Households may underestimate income, especially from self-employment or the informal sector constituting a majority of workers in Yemen. The poor therefore must be targeted using other methods. It has been proposed to move from a categorical targeting to a scoring approach similar to a proxy means test. A score is calculated for each applicant to the SWF by assigning points or weights to certain observable characteristics that are expected to be associated with poverty and vulnerability. Those applicants with a score above an established threshold are accepted into the program. The characteristics are obtained from a survey or application form filled out by the applicant and verified by cross checking available administrative records and through a followup home visit by social workers. This is in fact a version of a proxy means test. This scoring approach has been piloted by the SWF with EU assistance in several governorates, and the points assigned to each characteristic have been revised based on pilot data. The characteristics include the application category (disabled, elderly, orphan, single woman without caretaker, unemployed), family size, educational status of applicant, type of house, ownership of various assets However, the points or weights assigned to each characteristic are not based on any clear association with poverty. II. Proxy Means Test An alternative approach that may improve targeting outcomes is a full proxy means test (PMT). The advantage of the PMT is that it provides fairly good individual-level targeting of program benefits based on poverty status using a relatively small amount of information, without having to collect information on incomes or expenditures that are unreliable. Development of a PMT requires nationally representative household data that has information on incomes, expenditures, and a variety of household and socioeconomic characteristics. The 2005-2006 Household Budget Survey (HBS) provides sufficient information on Yemeni households on which to base a PMT. The objective is to predict actual consumption using variables that can be easily collected from applicants and used to determine eligibility for the program.3 The exercise involves two steps. First, a regression is estimated using the 2005-2006 HBS data that predicts household consumption expenditures. Per capita expenditures are regressed on a large vector of explanatory variables from demographic characteristics to geographic and residential features to income and employment aspects. Econometric methods are used to determine which set of characteristics is most effective in predicting consumption, and derives the points or weights assigned to each characteristic. The PMT differs from the scoring approach primarily in the application of econometric techniques to determine the characteristics and weights. 3 A more detailed discussion of PMT techniques can be found in Grosh and Baker (1995). The final regression results with the estimated coefficients are shown in Table A1. The best predictions of per capita expenditure were generated by using individual regressions for each governorate. There are therefore 21 separate equations shown in Table A1. The R-squared coefficients, or fit of the regressions, are within the ranges typical for PMTs used in other countries. On average, about half of the variation in per capita consumption observed in the sample is explained by the variables. However, the equations have better explanatory power in some governorates than others. The variables account for more than 65 percent of the variation in Al-Maharah and 63 percent in Al-Baida and Sanaa governorates for example, while 43 percent of the variation is explained in Hadramout, the region with the lowest R-squared coefficient. The coefficients corresponding to the explanatory variables have the expected signs. For example, households that have many people tend to have lower per capita consumption than others, as do those that use wood or manure for daily cooking, as suggested by the negative coefficients. Conversely, households living in urban areas or possessing assets such as a car or refrigerator have higher per capita consumption, other things being equal. The second step is to use the predicted consumption levels from the regressions to establish the eligibility for program benefits. In the simulation exercise, the population in the HBS is used to estimate how many households and individuals would be eligible. In practice, the applicants surveyed by SWF or those who apply at SWF offices would have their characteristics inserted into the appropriate regression to establish eligibility. A simple approach is to decide on a cutoff level of per capita consumption, and all households with predicted consumption below the cutoff would be eligible to receive benefits. All others would be excluded from participation. Different cutoff levels can be selected depending on the available budget for the program and the proportion of the population that will be covered. Because the predicted consumption levels do not perfectly correspond to actual consumption, there will inevitably be cases in which households or individuals are not found to be eligible, but they should be eligible if actual consumption were used instead. There will also be those found to be eligible that would not be eligible using actual consumption cutoff levels. These targeting errors are known as “errors of exclusion� (or undercoverage) and “error of inclusion� (leakage), respectively. All targeting methods have undercoverage and leakage to differing extents and no targeting method can be said to be perfect. Table A1: Proxy Means Test Regressions by Governorate Dependent variable: Log of Per Capita Household Consumption Ibb Abyan Sana’a city Al-baida Taiz Al-jawf Hajja Al-hodeida Hadramout Log of household size -0.59 -0.55 -0.64 -0.55 -0.66 -0.54 -0.61 -0.67 -0.46 [7.55]*** [4.05]*** [11.08]*** [5.85]*** [8.70]*** [3.69]*** [6.66]*** [11.27]*** [7.28]*** Number that never attended school 0.06 0.06 -0.02 0.01 0.05 -0.04 -0.01 0 0 [3.51]*** [2.34]** [1.08] [0.33] [1.82]* [1.59] [0.62] [0.24] [0.07] Number attending school 0.08 0.09 0.01 0.01 0.08 -0.01 0.02 0.04 0.01 [4.12]*** [3.35]*** [0.37] [0.53] [3.09]*** [0.35] [0.96] [2.46]** [0.64] Number that attended school 0.09 0.07 0.03 0.04 0.08 0.05 0.01 0.05 0.05 [4.93]*** [2.63]** [1.88]* [2.00]* [2.80]*** [1.45] [0.66] [2.61]** [2.90]*** Household has employed people 0.02 0.03 0.04 0.04 0.03 0.04 0.05 0.05 -0.01 [0.96] [1.53] [3.65]*** [2.11]** [1.58] [2.82]*** [3.57]*** [3.61]*** [0.56] Number of members that are single -0.07 -0.1 -0.02 -0.03 -0.08 -0.01 -0.03 -0.05 -0.01 [4.26]*** [4.20]*** [1.13] [1.57] [3.60]*** [0.60] [1.81]* [2.88]*** [0.43] Number of pre-school children 0.03 0.03 -0.02 0.01 0.07 0 0.01 0.03 -0.03 [1.70]* [1.36] [1.68]* [0.41] [3.56]*** [0.02] [0.50] [1.76]* [2.38]** Household has electricity 0.05 -0.09 -0.18 0 0.06 0.09 0.15 0.01 0 [0.58] [0.76] [1.02] [0.04] [0.89] [1.50] [1.96]* [0.13] [0.04] Type of household: house -0.03 -0.07 -0.04 -0.14 -0.04 0.21 -0.01 -0.04 -0.02 [0.54] [1.09] [1.55] [2.73]*** [0.90] [3.51]*** [0.29] [0.88] [0.27] Type of floor: ceramic 0.23 0.11 0.11 0.19 0.11 0.62 -0.04 0.18 0.21 [4.53]*** [1.86]* [3.87]*** [3.94]*** [2.24]** [5.50]*** [0.54] [3.76]*** [3.67]*** Energy to cook: wood/manure 0.01 0.37 -0.46 -0.14 -0.28 0.59 -0.1 0.04 0.07 [0.06] [2.45]** [3.39]*** [0.31] [2.11]** [6.41]*** [1.15] [0.60] [0.70] Energy to cook: gas 0.14 0.51 -0.26 -0.01 -0.24 0.59 0.18 0.04 -0.05 [0.86] [3.34]*** [2.80]*** [0.02] [2.15]** [6.58]*** [2.33]** [0.93] [0.59] Has water connection 0.02 -0.04 0.12 0.14 0.22 0.17 -0.16 0.18 -0.12 [0.41] [0.58] [3.44]*** [2.36]** [4.33]*** [3.18]*** [2.44]** [3.16]*** [2.15]** Has car/minibus 0.16 0.28 0.3 0.25 0.31 0.2 0.22 0.27 0.1 [3.48]*** [4.94]*** [11.62]*** [4.89]*** [5.52]*** [4.92]*** [4.07]*** [4.26]*** [2.89]*** Has motorcycle -0.09 0.27 -0.19 0.01 0.05 0.13 0.14 0.14 0.02 [0.66] [0.85] [1.99]** [0.09] [0.43] [2.11]** [1.75]* [3.25]*** [0.61] Has refrigerator 0.06 0.24 0.12 0.03 0.1 0.13 0.2 0.18 -0.04 [1.16] [3.18]*** [3.38]*** [0.61] [1.95]* [2.37]** [2.96]*** [3.63]*** [0.55] Has TV 0.09 0.09 0.2 0.1 0.11 0.1 0.14 0.1 0.03 [2.11]** [1.27] [4.91]*** [2.14]** [2.19]** [2.64]** [2.40]** [2.26]** [0.63] Has washing machine 0.18 0.1 0.14 0.2 0.12 0.19 0.12 0.12 0.12 [3.90]*** [2.29]** [4.06]*** [3.76]*** [2.74]*** [3.46]*** [1.70]* [2.68]*** [2.80]*** Has sewing machine -0.03 0.08 0.05 0.05 0.07 0.02 -0.11 0.01 -0.03 [0.57] [1.85]* [1.77]* [0.83] [1.38] [0.44] [2.17]** [0.19] [0.62] Has mobile telephone 0.2 0.18 0.25 0.16 0.27 0.12 0.25 0.29 0.16 [4.88]*** [4.01]*** [8.72]*** [3.31]*** [6.50]*** [2.64]** [6.77]*** [9.01]*** [3.95]*** Urban area 0.06 0.06 -0.01 0.38 -0.06 0.03 0.04 0.03 0.02 [0.96] [0.74] [0.24] [4.24]*** [0.67] [0.54] [0.47] [0.40] [0.27] Own or rent a plot for agriculture 0.1 0.21 0.16 0.04 0.12 0 0.12 0.13 0.06 [2.14]** [3.28]*** [4.00]*** [0.57] [2.11]** [0.00] [2.85]*** [3.26]*** [1.01] Has livestock -0.04 -0.09 0.03 -0.01 0.05 -0.01 0.11 0.03 -0.03 [0.74] [2.04]** [0.52] [0.20] [0.85] [0.15] [2.32]** [0.70] [0.68] Number of disabled/chronically ill -0.03 0 0 -0.01 -0.03 0.02 -0.06 -0.03 0 [1.36] [0.06] [0.02] [0.82] [1.36] [0.43] [3.35]*** [2.05]** [0.03] Number of males aged 60 or more -0.02 0.1 0.01 0.03 0.02 -0.01 -0.02 0.04 0.02 [0.47] [1.70]* [0.29] [0.71] [0.49] [0.22] [0.50] [1.18] [0.50] Number of females aged 55 or more -0.05 -0.08 0.02 -0.08 -0.04 0.02 0 -0.07 -0.01 [0.98] [1.37] [0.63] [1.68]* [1.00] [0.33] [0.00] [1.93]* [0.38] Constant 11.88 11.46 12.67 11.81 12.19 11.23 12.11 11.96 12.27 [59.41]*** [57.30]*** [63.50]*** [25.69]*** [90.83]*** [45.67]*** [85.90]*** [119.29]*** [93.16]*** Table A1: Proxy Means Test Regressions by Governorate Dependent variable: Log of Per Capita Household Consumption Dhamar Shabwah Sa’adah Sana’a region Aden Laheg M areb Al-mahweet Al-maharh Log of household size -0.55 -0.48 -0.64 -0.4 -0.8 -0.69 -0.77 -0.56 -0.56 [7.30]*** [4.28]*** [6.95]*** [3.36]*** [12.84]*** [8.04]*** [6.80]*** [7.60]*** [4.25]*** Number that never attended school 0.02 -0.01 0.03 0.01 0.02 0.02 -0.04 0.01 -0.11 [0.79] [0.57] [1.67] [0.68] [1.24] [0.76] [1.50] [0.61] [3.43]*** Number attending school 0.03 0.05 0.03 0 0.09 0.02 0.03 0.06 0 [1.27] [1.86]* [1.72]* [0.22] [4.42]*** [0.50] [1.43] [2.31]** [0.15] Number that attended school 0.06 0.04 0.03 0.07 0.08 0.04 0.04 0.08 -0.06 [2.53]** [1.73]* [1.37] [2.09]** [4.12]*** [1.23] [1.43] [3.97]*** [1.96]* Household has employed people 0.05 0.03 0.08 -0.03 0.05 0.05 0.02 0.02 0.08 [2.73]*** [1.13] [4.31]*** [1.21] [3.17]*** [1.90]* [0.72] [1.40] [3.24]*** Number of members that are single -0.06 -0.04 -0.05 -0.02 -0.07 -0.03 -0.01 -0.07 0 [3.02]*** [1.94]* [2.24]** [1.14] [3.58]*** [1.10] [0.32] [3.42]*** [0.08] Number of pre-school children 0.02 0.02 0.05 -0.03 0.06 0.03 0.07 0.04 0.04 [1.03] [0.92] [2.65]** [1.50] [3.30]*** [1.28] [2.78]*** [2.13]** [1.57] Household has electricity 0.09 0.27 0 0.25 0.33 0.15 0.09 0.14 -0.24 [1.19] [1.33] [0.03] [2.87]*** [1.77]* [1.08] [0.64] [2.57]** [1.29] Ty pe of household: house 0.01 -0.09 -0.09 -0.32 -0.01 0.04 0.02 0 0.31 [0.18] [0.85] [0.82] [3.16]*** [0.18] [0.28] [0.27] [0.01] [2.74]** Ty pe of floor: ceramic 0.12 0.37 0.21 0.08 0.15 0.22 0.38 0.03 -0.41 [1.43] [4.64]*** [2.11]** [0.94] [4.15]*** [2.33]** [3.53]*** [0.60] [3.30]*** Energy to cook: wood/manure 0.2 -0.06 -0.34 -0.14 0.17 0.12 -0.17 -0.15 -0.11 [2.39]** [0.37] [2.13]** [2.63]** [1.17] [0.63] [1.59] [0.95] [1.87]* Energy to cook: gas 0.23 -0.13 -0.35 0 0.08 0.13 -0.01 -0.07 0.34 [3.77]*** [0.80] [2.21]** [.] [0.74] [0.63] [0.15] [0.51] [1.73]* Has water connection -0.01 -0.01 -0.02 0.16 -0.12 -0.12 -0.13 0.03 -0.21 [0.11] [0.09] [0.30] [2.53]** [0.33] [1.16] [1.63] [0.73] [1.50] Has car/minibus 0.25 0.25 0.17 0.23 0.31 0.19 0.33 0.08 0.33 [4.59]*** [4.76]*** [3.47]*** [7.27]*** [7.78]*** [2.50]** [7.01]*** [1.65] [5.81]*** Has motorcycle 0.15 -0.43 0.36 0.24 -0.14 0.12 -0.05 0.3 0.06 [2.47]** [2.32]** [4.08]*** [3.17]*** [1.71]* [1.49] [0.16] [1.94]* [0.61] Has refrigerator 0.01 0.02 0.1 0.01 -0.09 0 0.06 0.06 0.09 [0.19] [0.23] [1.53] [0.15] [1.48] [0.04] [0.64] [1.08] [1.11] Has TV 0.17 0.07 0.18 0.01 0.16 0.12 0.08 0.09 0.14 [3.42]*** [1.14] [3.55]*** [0.13] [4.39]*** [1.78]* [1.10] [2.26]** [2.00]* Has washing machine 0.1 0.02 0.15 -0.09 0.18 0.1 0 0.15 0.3 [1.89]* [0.30] [1.93]* [0.91] [4.02]*** [1.47] [0.04] [2.75]*** [4.37]*** Has sewing machine 0.02 0.02 0.05 0.07 0 0.23 -0.11 -0.01 -0.11 [0.32] [0.28] [0.94] [0.90] [0.09] [2.64]** [1.57] [0.17] [0.96] Has mobile telephone 0.27 0.3 0.13 0.18 0.21 0.19 0.25 0.17 0.12 [5.62]*** [3.43]*** [2.86]*** [2.64]** [5.60]*** [3.45]*** [4.08]*** [3.89]*** [2.32]** Urban area -0.08 0.01 0.03 0 0 0.36 -0.01 -0.09 -0.03 [0.94] [0.11] [0.30] [.] [.] [4.03]*** [0.11] [1.31] [0.24] Own or rent a plot for agriculture 0.15 -0.02 0.16 0.1 -0.1 0 -0.06 0.12 0.11 [3.12]*** [0.21] [2.81]*** [1.37] [0.51] [0.02] [0.82] [2.38]** [1.01] Has livestock 0.04 -0.04 0.06 0.04 -0.04 0.12 0.11 0 -0.06 [0.87] [0.56] [1.00] [0.59] [0.59] [2.06]** [1.61] [0.09] [0.75] Number of disabled/chronically ill -0.01 0.02 0.02 -0.03 0 -0.03 -0.01 -0.03 0.01 [0.38] [0.63] [1.22] [0.98] [0.07] [0.80] [0.31] [0.84] [0.18] Number of males aged 60 or more 0.01 0.03 -0.11 0.12 0.05 0.05 -0.07 0.01 0.01 [0.28] [0.39] [3.13]*** [2.38]** [1.27] [0.59] [0.68] [0.30] [0.15] Number of females aged 55 or more -0.16 0.01 0 -0.06 0.02 -0.07 0.09 -0.12 0.03 [3.82]*** [0.07] [0.06] [1.06] [0.56] [1.16] [0.76] [2.91]*** [0.38] Constant 11.87 11.84 12.39 12.25 12.2 11.64 12.5 12.06 12.38 [77.93]*** [64.13]*** [69.47]*** [58.12]*** [39.20]*** [48.77]*** [80.62]*** [98.64]*** [53.87]*** Observations 676 356 542 267 707 513 377 537 277 R-squared 0.45 0.5 0.41 0.63 0.6 0.45 0.59 0.48 0.65 Table A1: Proxy Means Test Regressions by Governorate Dependent Variable: Log of Per Capita Household Consumption Amran Al-dhale Remah Log of household size -0.56 -0.54 -0.67 [5.98]*** [5.27]*** [7.20]*** Number that never attended school 0.02 -0.02 0.02 [0.94] [0.54] [0.56] Number attending school 0.05 0.02 0.04 [1.90]* [0.91] [1.18] Number that attended school 0.07 0 0.05 [2.98]*** [0.15] [1.41] Household has employed people 0.02 0.06 0.06 [0.88] [2.20]** [2.68]** Number of members that are single -0.03 -0.03 -0.05 [1.14] [1.35] [1.90]* Number of pre-school children 0.02 0.05 0.06 [0.97] [2.51]** [2.38]** Household has electricity -0.01 0.09 0.11 [0.11] [0.91] [1.09] Ty pe of household: house -0.04 -0.07 -0.05 [0.73] [0.72] [0.63] Ty pe of floor: ceramic 0.13 0.23 0.44 [1.73]* [3.29]*** [2.44]** Energy to cook: wood/manure -0.28 -0.63 -0.34 [0.89] [3.85]*** [2.67]** Energy to cook: gas -0.18 -0.56 -0.04 [0.57] [3.56]*** [0.23] Has water connection -0.11 0.08 0.02 [1.42] [0.94] [0.09] Has car/minibus 0.21 0.28 0.36 [4.39]*** [4.90]*** [2.11]** Has motorcycle 0.2 -0.03 0 [1.67] [0.31] [.] Has refrigerator 0.2 0.05 -0.02 [2.52]** [0.86] [0.05] Has TV 0.17 0.08 -0.1 [3.13]*** [1.44] [0.75] Has washing machine 0.06 0.13 -0.06 [1.01] [1.89]* [0.29] Has sewing machine 0.03 0.08 -0.06 [0.44] [0.89] [0.40] Has mobile telephone 0.18 0.21 0.25 [3.65]*** [3.68]*** [3.66]*** Urban area 0.1 -0.03 0.28 [0.84] [0.25] [2.77]** Own or rent a plot for agriculture 0.04 0.05 0.05 [0.59] [0.95] [0.82] Has livestock 0.01 -0.01 0.1 [0.11] [0.15] [1.17] Number of disabled/chronically ill -0.04 0.04 -0.04 [1.51] [0.86] [1.42] Number of males aged 60 or more 0.01 0.07 0.05 [0.24] [1.14] [0.76] Number of females aged 55 or more -0.02 -0.08 -0.04 [0.34] [1.22] [0.83] Constant 11.96 12.59 12.27 [34.96]*** [80.90]*** [84.57]*** Observations 535 385 282 R-squared 0.45 0.53 0.55 Table A2 displays the inclusion and exclusion errors from the PMT models along with those of the scoring method assuming a cutoff score for eligibility at 25 or higher. The last row of the table indicates that the PMT overall has an exclusion error of 35 percent, implying that 35 percent of those who are actually poor (according to the poverty line established for the HBS) would be found ineligible using the PMT. Therefore 65 percent of the poor would be correctly identified and could receive benefits. The corresponding inclusion error is 15 percent, implying that 15 percent of those who are found to eligible under the PMT would be nonpoor. This targeting outcome is superior to the scoring method, which has an error of exclusion rate of 51 percent and an inclusion error of 24 percent. The PMT does a much better job of correctly identifying the poor households and individuals. This is consistent with the fact that the PMT is specifically designed to identify the poorest, while the scoring method is an ad-hoc association of characteristics not necessarily related to poverty status. Table A2: Exclusion and Inclusion Errors Proxy Means Test Scoring (+ 25) Poverty Exclusion Inclusion Number of Exclusion Inclusion Number of Governorate Level error error beneficiaries error error beneficiaries Ibb 29.5 53 13 522,012 52 25 711,414 Abyan 49.6 24 34 235,732 65 19 115,591 Sana’a city 13.0 73 2 85,453 88 1 44,154 Al-baida 53.6 17 37 342,308 68 6 109,477 Taiz 40.1 35 22 932,259 60 25 735,866 Al-jawf 44.0 22 19 188,183 48 14 129,201 Hajja 52.3 24 22 748,810 23 41 884,670 Al-hodeida 39.0 38 19 802,991 29 45 1,256,659 Hadramout 37.5 45 11 267,917 90 7 80,530 Dhamar 26.1 55 13 312,140 45 31 543,285 Shabwah 58.1 22 24 272,551 74 4 80,095 Sa’adah 17.4 77 3 46,839 59 43 315,578 Sana’a region 24.6 44 7 191,105 24 25 368,814 Aden 16.7 58 4 54,292 94 1 12,410 Laheg 44.8 28 33 352,799 54 22 230,615 Mareb 49.2 16 17 110,719 19 25 116,307 Al-mahweet 32.3 35 12 128,505 23 30 201,373 Al-maharh 7.4 75 4 4,271 91 8 6,429 Amran 74.4 11 34 657,349 67 13 248,038 Al-dhale 34.5 45 16 144,924 66 21 129,125 Remah 43.3 24 20 183,760 22 50 260,381 Yemen 36.5 35 15 6,584,919 51 24 6,580,011 The PMT coverage compares quite well with international experience. Figure A1 shows the coverage rates among the poor for several safety nets programs that use PMT. The simulated PMT for Yemen would cover a larger share of the poor than any of the active programs in the figure. However, it is difficult to compare a simulated model to active programs, since actual coverage will be affected by the available budget and the implementation of the program (applicant intake, processing, verification and followup, etc.) and is not simply a matter of the targeting formula. Nonetheless, the PMT model is promising. Figure A1: Coverage of the Poor Using the Proxy Means Test 70 65 60 60 Percentage coverage 50 39 40 34 34 30 20 10 0 s T IR F la de PM SU co SH da Es le en ni a hi bi sa rtu m C um Ye ol po B ol O C il az o ic Br ex M Coverage defined as the percent of eligible poor households or the percent of those in the bottom consumption quintile who receive benefits. Source: Author’s calculations, Castaneda and Lindert et. al. 2005. III. A Comparison of Simulated Poverty Outcomes Improving the targeting of the SWF by including more individuals who are poor and excluding the better off will allow the same resources to help more needy people. Further, an expansion of the budget would provide better coverage and improve the adequacy of assistance. This section investigates the likely effects on poverty outcomes that would obtain by using different targeting approaches. Two cases are examined, reflecting the basic policy choices that must be made by government and SWF: • Case 1: Keep the number of beneficiaries constant, but increase the benefits to each as the budget increases. 20 • Case 2: Determine a fixed benefit formula, but increase the coverage of the eligible population as the budget increases A mix between these two extremes is also likely, whereby there is some expansion of beneficiaries along with a moderate increase in benefits. This has occurred with the recent policy of doubling SWF benefits to a maximum of YR 4,000 and expanding the coverage of the SWF. However, the hypothetical cases above will serve to illustrate the properties of the targeting approaches. Case 1: Increasing Benefits Figure A2 shows the reduction in headcount poverty that would result if an increasing level of benefits is transferred to individuals identified under the various targeting approaches. With no budget, no benefits would be transferred, so all methods yield the current poverty rate of 36.5 percent. At a budget of about YR 20 billion (similar to the SWF budget for 2006), the PMT test yields the lowest poverty rate of all methods. Two versions of the PMT are shown, one using only the variables available in the survey form that has been piloted recently by the SWF (labeled PMT-APP in Figure A2) and the other using several variables available in the HBS but not reflected in the SWF questionnaire (labeled PMT-New). Both versions first allocate budget geographically based on the governorate-level poverty rate and then apply the PMT to determine eligibility and benefits. Both versions yield a poverty rate of about 35 percent with a budget of YR 20 billion. Figure A2 Case 1: Poverty Levels Using Different Targeting Methods Increasing Benefits, Constant Number of Beneficiaries 37.00 36.50 36.00 35.50 Poverty Level 35.00 34.50 34.00 33.50 33.00 0 5 10 15 20 25 30 35 40 Cost (in Billions YR) PMT - APP PMT - New UBT GEO + UBT Scoring 18+ Score 25 21 The other targeting methods produce higher poverty rates. With a YR 20 billion budget, the scoring approach results in a poverty rate of about 35.3 percent, depending on which score is taken as the threshold. The universal transfer (UBT) results in the smallest drop in poverty, since no criteria are used to distinguish the eligible population. While everyone would receive a benefit under UBT, it is so small that there are very small effects on poverty. Geographic targeting improves on UBT, but it is still inferior to the scoring and PMT approaches. As the budget expands and benefits are increased, the reduction in poverty increases across all methods. However, the decline is greatest for the PMT versions. If the SWF increased to YR 40 billion, roughly double the current size, poverty would be reduced to 33.5 percent using PMT. It would drop to 34.2 percent using the scoring approach, and would stay at 35.2 percent if universal transfers were adopted. An important conclusion from this analysis is that poverty rates will be little affected regardless of the targeting approach used unless benefits can be increased. At YR 40 billion, approximately 660,000 people would be lifted out of poverty using the PMT compared to the baseline zero budget, while 506,000 would be above the poverty line using the scoring approach. A difference of 154,000 individuals is not a huge effect on poverty. Figure A3 Case 1: Poverty Gap Using Different Targeting Methods Increasing Benefits, Constant Beneficiaries 12.00 11.50 11.00 10.50 Poverty GAP 10.00 9.50 9.00 8.50 8.00 0 5 10 15 20 25 30 35 40 Cost (in Billions YR) PMT - APP PMT - New UBT GEO + UBT Scoring 25+ Scoring 18+ 22 However, given budget limitations and the extent and severity of poverty in Yemen, the goal of the SWF should be to ensure that the available resources are concentrated on the neediest individuals. Fully half of the poor are clustered 75 percent below the poverty line, according to the Yemen poverty assessment. Therefore, providing relatively small benefits to the poorest of the poor will not raise them above poverty, but should reduce the severity of poverty. This effect is illustrated in Figure A3. The chart shows the reduction in the poverty gap for each targeting method as benefits increase. The poverty gap is the additional consumption (or income) needed to raise the average person out of poverty, expressed as a share of the poverty line.4 With a budget of YR 20 billion, the poverty gap drops to about 9.4 under the scoring approach while it falls to 9.2 using the PMT. The gap is 2 percent lower with PMT at YR 20 billion, while at YR 40 billion the gap becomes 4.5 percent lower with the PMT (to 8..4). The severity of poverty is less with the PMT than the scoring approach and improves as the budget increases. Figure A4 Case 2: Poverty Levels Using PMT and Scoring Methods Fixed Benefit Formula, Increasing Number of Beneficiaries 37.00 16,000,000 36.50 14,000,000 36.00 12,000,000 Number of beneficiaries 35.50 10,000,000 Poverty Level 35.00 8,000,000 34.50 6,000,000 34.00 4,000,000 33.50 2,000,000 33.00 0 0 10 20 30 40 50 60 70 80 90 100 Cost (in Billions YR) PMT - APP PMT - New Scoring Beneficiaries - PMT Beneficiairies - Scoring 4 Household-specific poverty lines are used in the analysis, so the poverty gap can be interpreted as the share of the relevant poverty line consumption level needed to raise each person to the poverty line, averaged across the population. 23 Case 2: Increasing Beneficiaries The second exercise involves fixing the benefit formula and increasing the eligible beneficiaries as the budget increases. Benefits are assumed to follow the new formula established for the SWF: beneficiaries are provided YR 2,000 per month plus YR 200 for each household member up to a maximum of YR 4,000. Beneficiaries are gradually expanded by taking new cases in the order of eligibility established by the targeting approach. For the PMT, individuals with the lowest predicted consumption are taken first, moving up the list as the budget allows. A similar protocol is followed for the scoring approach. The universal transfer and pure geographic approaches are not included in the exercise as there is no meaningful way to select beneficiaries. Figure A4 shows the reduction in headcount poverty for the PMT and scoring methods and budgets assuming the fixed benefit formula and increasing beneficiaries. The left vertical axis shows the poverty rate, while the right axis shows the increase in the number of beneficiaries. With a budget of YR 20 billion, approximately 800,000 cases would receive benefits, and the poverty rates would be roughly equal under either scoring or PMT. While the PMT identifies more truly poor individuals than through scoring, these people are very far from the poverty line, so the limited fixed benefits are not enough to affect overall poverty. If the budget were to double to YR 40 billion, the PMT is seen to have a larger effect on poverty than the scoring approach, maintaining an essentially constant difference up to YR 100 million. Figure A5 Case 2: Poverty Gap Using PMT and Scoring Methods Fixed Benefit Formula, Increasing Number of Beneficiaries 10.50 16,000,000 14,000,000 10.00 12,000,000 Number of beneficiaries 10,000,000 9.50 Poverty GAP 8,000,000 9.00 6,000,000 4,000,000 8.50 2,000,000 8.00 0 0 10 20 30 40 50 60 70 80 90 100 Cost (in Billions YR) PMT - APP PMT - New Scoring Beneficiaries - PMT Beneficiairies - Scoring 24 The changes in the poverty gap assuming a fixed benefit formula and gradually increasing beneficiaries is displayed in Figure A5. The poverty gap is lower with PMT for every budget level. The PMT using additional variables (PMT New) produces a slightly larger decline in the poverty gap than the PMT based on the current SWF survey. III. Conclusions Five conclusions can be drawn from the foregoing exercise. • The proxy means test (PMT) in combination with geographic targeting exhibits superior performance to the other approaches. It includes more poor individuals (65 percent of the poor are identified by the PMT versus 51 percent under the scoring approach) and has lower exclusion errors (15 percent for the PMT, 24 percent for scoring). • The scoring approach is comparable to the PMT. Scoring is an improvement over the earlier categorical targeting practices, and compares well with the full PMT especially at lower budget levels. • The gains from using the PMT increase at higher budget levels. At the very low levels recently available to SWF (between YR 17-20 billion), there is little observable difference between the poverty reduction performance of the PMT and scoring approaches. Small benefits, even correctly targeted, will not change the headcount poverty rate since many of the poor are very far below the poverty line. As budgets increase, there is a larger improvement evident using the PMT. • Because PMT targets more benefits to the very poor, it will reduce the depth of poverty more quickly than other methods, including scoring. The reduction in severity is greater the larger the benefits, but at current budget levels the effects will be limited. • The PMT can be implemented without delaying the SWF reform process. A national survey to identify the poor will be administered soon by SWF. One version of the PMT was constructed using variables from the questionnaire, exhibiting good targeting properties. Therefore the survey does not need to be altered or delayed, and the PMT can be implemented quickly as part of the development of the database. 25