Report No. AAA33-BA Bosnia and Herzegovina Social Assistance Transfers in Bosnia and Herzegovina Moving Toward a More Sustainable and Better-Targeted Safety Net Policy Note April 30, 2009 Human Development Sector Unit Europe and Central Asia Document of the World Bank CURRENCY EQUIVALENTS (Exchange Rate Effective 04/30/2009) Currency Unit = KM KM 1.00=US$0.69 US 1.00=KM1.44 FISCAL YEAR January 1-December 31 ABBREVIATIONS AND ACRONYMS BH Bosnia and Herzegovina CEPOS Center for Policy Studies csw cvw Center for Social Work Civilian Victims of War ECA Europe and Central Asia EU EuropeanUnion FBH Federation of Bosnia and Herzegovina GDP Gross domestic product HBS Household Budget Survey HMT Hybridmeans-testing I T Informationtechnology KM Bosnianmark LCU Local currency unit LSMS Living Standards Measurement Survey MT Means-testing N W I Non-War Invalids' Benefit O&C Oversight and controls mechanisms OECD Organisation for Economic Co-operation and Development PAE Per-adult equivalent PC Per capita PMT Proxy means-testing RS Republika Srpska VAT Value-added tax VMT Verifiedincome and asset-testedprograms Vice President: Shigeo Katsu Country Director: Jane Armitage Sector Manager: Kathy A. Lindert Project Team Leader: Maniza B.Naqvi Acknowledgements Assistance, insights and information received from the authorities in Bosnia and Herzegovina during the preparationof this Policy Note is gratefully acknowledged. This note was prepared by a team comprising Messrs./Mmes. Maniza B. Naqvi, Senior Social Protection Specialist, Vedad Ramljak, Consultant, Anna Gueorguieva, Young Professional, Erwin H.R.Tiongson, Senior Country Economist, Orhan Niksic, Senior Country Economist (ECSPE), Emil Daniel Tesliuc, Senior Economist (HDNSP), Goran Tinjic, Senior Operations Officer, Michele Gragnolati, Sector Leader, (LCSHD); Mariam Khanna, Consultant (ECSPE), Sreypov Tep, Program Assistant (ECSHD), and Senad Sacic, TeamAssistant (ECCBA). The draft report was submitted to the Government in April 29, 2009. The government did not request for any changes on the final of the report. The team appreciatesthe close cooperation and collaborative process with the Government counterparts and wishes to thank the Ministries of Labor inthe Federation and Republika Srpska; PIUSESER; Federation Bureau of Employment; MinistryofHealthand Social Welfare inRepublika Srpska; the Centersfor Social Work; andthe Bosnia and Herzegovina Agency for Statistics. The team appreciatesthe comments provided by Peer Reviewers, Jeanine Braithwaite, Sr, Social Protection Economist,(HDNSP); Ivailo Izvorski, Lead Economist (EASPR); Penelope Williams, Sr. Country Officer (ECCUS); Dena Ringold, Sr. Economist (LCSHS); Tarsicio Castaneda, Safety Nets Targeting Specialist (Consultant). The team benefitedfrom management oversight and guidance provided by Messrs./Mmes. Jane Armitage, Country Director (ECCU4), Marco Mantovanelli, Country Manager (ECCBA), Tamar Manuelyan Atinc, Sector Director (ECSHD) and Kathy A. Lindert, Sector Manager (ECSHD). SOCIAL ASSISTANCE TRANSFERS INBOSNIA AND HERZEGOVINA: MOVINGTOWARD A MORESUSTAINABLE AND BETTER-TARGETED SAFETY NET CONTENTS EXECUTIVE SUMMARY............................................................................................................................ i Section 1:....................................................................................................................................................... 1 1.1. Overview of Public SpendingonNon-insuranceTransfers inBH.................................................... 1 1.2. Composition of Public Spendingon Transfers inBH........................................................................ 3 Poverty .................................................................................................................................................... 1.3.Performanceof Social SpendingTransfers: Coverage and TargetingAccuracy andtheir Impact on 10 1.4. Impact ofBenefits on Poverty.......................................................................................................... 14 1.5. Opportunity Cost of Social Spending:Crowding Out ofPublic Investments.................................. 14 1.6. Hidden Costs of UntargetedBenefits -Labor MarketImplications ................................................ 16 Section2:..................................................................................................................................................... 18 2.1. Rationalefor Targeting and Overview ofTargetingInstruments.................................................... 18 2.2. HouseholdSurveyData Usedto Develop TargetingTools ............................................................. 22 2.3. Empirical Frameworkfor the PMT and HMT Estimation............................................................... 23 2.4. Results: Proxy-means-testModel Usingthe 2007 HBS .................................................................. 26 Section3:..................................................................................................................................................... 35 Annex A: Performanceofthe Current Programs:ADePTaTables......................................................... 38 Annex B: Spendingon Non-insurance Social ProtectionCash Transfers .............................................. 62 Annex C: Details ofthe Social Pact in.................................................................................................... 63 the Federationof BosniaandHerzegovina(FBH).................................................................................. 63 Annex D:Data Description and PastProxy-means-testing(PMT) Models............................................ 64 Annex E: Analysis ofthe Underestimationof Income............................................................................ 66 Annex F: Statistical Tables ..................................................................................................................... 69 Annex G:Number of Beneficiaries and Average BenefitLevels inFBH-Administrative and HBS Data ......................................................................................................................................................... 75 References ............................................................................................................................................... 77 Figures FigureES 1: InternationalComparisons of Public Spending on Social Assistance....................................... i Figure ES 2: Upsurge inRights-based Benefits inFBH............................................................................... FigureES 3: Regressive Distribution of Social ProtectionBenefits inBH ................................................ ...ii 111 FigureES4: TargetingAccuracy of Social Assistance Benefits -InternationalComparison.................... ... 111 Figure 1.1. InternationalComparisons of Public Spendingon Social Assistance ........................................ 2 Figure 1.2: Composition of Spending onNon-insurance Cash Transfers: Average for 2005-09 (Percent of total) .............................................................................................................................................................. 4 Figure 1.3:Veteran-related Benefits in FBH................................................................................................ 5 Figure 1.4: Veteran-related Benefits inRS................................................................................................... 5 Figure 1.5: Composition of CivilianBenefits............................................................................................... 7 Figure 1.6: Changing Composition of CivilianBenefits in FBH.................................................................. 7 Figure 1.7: Spendingon Benefits is Fairly Steady inRS.............................................................................. 8 Figure 1.8: UpsurgeinRights-based Benefits inFBH .................................................................................. 8 Figure 1.9: The Rise inUnemployedDemobilized Soldiers Followingthe Introductiono fBenefits........10 Figure 1.10: Coverage of Social Protection and Social Assistance Benefits inBH, H B S 2007 .................11 Figure 1.11: Regressive Distribution of Social ProtectionBenefits in BH................................................. 12 Figure 1.12: TargetingAccuracy of Social Assistance Benefits InternationalComparison - .................... 12 Figure 1.13: Weak TargetingAccuracy o f Specific Social Benefits Programs: FBHand RS .................... 13 Figure 1.14: InternationalComparison of TargetingAccuracy ................................................................. 13 Figure 1.15: Public Investments as a Percentageof Total Public Expenditures (2006 ............................... 15 Figure 1.16: Veterans Beneficiaries, by Disability Category...................................................................... 17 Distribution ................................................................................................................................................. Figure2.1:HouseholdMonthly Per Capita Income and Consumptionover Quantiles o f Income 25 Figure 2.2: Composition of HouseholdIncome Over the Distribution of Welfare..................................... 26 Figure2.3: Baseline PMT Model-Actual and Predicted Values ............................................................. 28 Figure 2.4: Incidence and Distribution of Inclusion Errors of EachModel................................................ 31 Figure2.5: Share of Beneficiaries inthe Bottom Quintile -InternationalComparisons .......................... 32 FigureAA 1: Lorenz Curves and Venn Diagrams . ..................................................................................... 57 Tables Table 1.1: Expenditureon Non-insurance Social Protection Cash Transfers in BH,.................................... 2 Table 1.2: Spending on Non-Contributory Social Assistance CashTransfers inBHas Percent of Budgets14 Table 1.3: Demobilized Soldiers as Percent of Total Number of Unemployed.......................................... 16 Table 2.1: A Spectrum of Targeting Instruments Based on Individual Assessment................................... 21 Table 2.2: PMT Errors of Inclusion and Exclusion .................................................................................... 30 Population, by Decile.................................................................................................................................. Table 2.3: Distribution o f Beneficiaries of the 2007 and 2004 PMT Models Covering 20 Percent of the 34 Table AA. 1:PopulationDemographics..................................................................................................... 38 Table AA.2: AverageTransfer Value, Per Capita ..................................................................................... 39 Table AA.3: Average Transfer Value. Per Capita. BeneficiaryHouseholdsof IndicatedTransfer Only .40 Table AA.4: Coverage ............................................................................................................................... 41 Table AA.5: Coverage inRepublikaSrpska.............................................................................................. 42 Table AA.6: CoverageinFederationofBH.............................................................................................. 43 Table AA.7: DistributionofBeneficiaries................................................................................................. 44 Table AA. 8: Distributionof Beneficiaries inRepublikaSrpska................................................................ 45 Table AA. 9: Distributionof Beneficiaries inFederationofBH................................................................ 46 Table AA. 10: DistributionofBenefits (TargetingAccuracy) ................................................................... 47 Table AA. 11:DistributionofBenefits (TargetingAccuracy) inRepublikaSrpska.................................. 48 Table AA. 12: DistributionofBenefits (TargetingAccuracy) inFederationofBosnia andHerzegovina.48 Table AA. 13:Generosity ........................................................................................................................... 49 Table AA. 14:UndercoverageandLeakage............................................................................................... 50 Table AA. 15: ImpactofProgramson Povertyand InequalityMeasures-Simulatingthe Absence ofthe Program....................................................................................................................................................... 51 Table AA. 16: Coady-Grosh-Hoddinott(CGH) Indicator.......................................................................... 52 Table AA. 17:Coady-Grosh-Hoddinott(CGH) Indicator.Benefits'Incidence........................................ -53 Table AA. 18:Transfer DuplicationinEachPopulationGroup (YO).......................................................... 54 Table AA. 19:Social ProgramOverlap(%)............................................................................................... 55 56 Table AA. 21: Coverage (Based onpre-transferwelfare) ......................................................................... Table AA.20: SocialProgramOverlap. by IncomeQuintile(%).............................................................. 58 Table AA.22: DistributionofBeneficiaries(Basedonpre-transferwelfare) ............................................ 59 Table AA.24: DistributionofBenefits(TargetingAccuracy) ................................................................... TableAA.23: Distributionof Benefits(TargetingAccuracy) (Based on pre-transferwelfare).................60 61 Table AF. 1:BaselineModel...................................................................................................................... 69 Table AF 2: Entity-levelModels............................................................................................................... . 71 Table AF.3: PMTRegressionby UrbanandRuralAreas .......................................................................... 73 Table AG. 1: Comparisonof H B S Estimates andAdministrative.............................................................. 75 Table AG. 2: ComparisonofH B S QuestionnaireDefinitionsandthe Actual BenefitTitles.................... 76 EXECUTIVESUMMARY Bosnia and Herzegovina (BH) spends 4 percent of its gross domestic product (GDP) on non- insurance social protection cash transfers.' With such a significant share of the country's GDP going to these transfers, BHis one of the highest spenders inthe Europe and Central Asia (ECA) region (Figure ES.l). By comparison, the regional countries' expenditure average i s 1.6 percent of GDP, and the Organisation for Economic Co-operation and Development (OECD) nations' average i s 2.5 percent. This level of spending on non-insurance social protection cashtransfers i s fiscally unsustainable, particularly given the impacts of the global financial and economic crisis onpublic revenues. Figure ES 1: InternationalComparisons of Public Spending on SocialAssistance PublicSpendingon Social Assistance,% of GDP Croatia - 1 Bosnia-Herzegovina Hungary Ukraine OECD Uzbekistan Macedonia Russia Moldova Armenia Latvia Estonia Belarus Lithuania Bulgaria Kyrgyzstan Kosovo Serbia Romania Kazakhstan Albania Poland Georgia Turkey Azerbaijan Tajikistan 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00 Source: Lindertand others (2008). For historical reasons, social benefits in BH have been heavily dominated by "rights-based" programs designed to protect war veterans or their surviving dependents ("veteran-related 1Health, pension, and unemployment insurance programs are examples o f insurance-based schemes. A significant proportion o f the ostensibly insurance-based benefits (for example, special pensions) are also financed through the general government budget. i benefits"). Veteran-related benefits absorb about three-quarters o f total spending on non- insurance social protection cash transfers. The share i s slightly lower inthe Federationo f Bosnia and Herzegovina (FBH) than in Republika Srpska (RS)-the two Entities that make up Bosnia and Herzegovina. Both Entities also operate a number o f civilian benefits that account for about one-quarter of total spending on non-insurance social protection cash transfers. These include means-tested programs such as Social Assistance Benefits and the Child Protection Allowance. In addition, FBHhas two rights-based disability benefit programs that have increased significantly over time ~~ - (FigureES.2). Figure ES 2: Upsurgein Rights-basedBenefitsin FBH Veterans Benefits (Rights Based) Civilian Disability Benefits (Rights Based: NWI+CVW) Civilian MT Benefits (Means-Tested: - 1 SA+CA) Source: Authors' calculations usingHBS2007 data. Despite significant fiscal outlays on non-insurance social protection cash transfers, their coverage o f the poor is low. Moreover, when viewed in the aggregate, non-insurance social protection cash transfers are regressive innature; that is, a higher share o f overall expenditure on these benefits i s going to people in the richer quintiles o f BH's population. In contrast, those in the poorest quintile receive only 18 percent o f overall non-insurance social protection cash transfers-a smaller proportion than their share o f the total population o f BH (each quintile represents 20 percent o f the population rankedby consumption) (Figures ES.3 and ES.4). ii Figure ES 3: Regressive Distributionof Social Protection BenefitsinBH BiH:TargetingAccuracy of Social Protection Benefits (Distributional Incidence; 2007 HBS) .-5 +a 25.0 ?L20.0 -All Socia11 Protection Y .e a 2 15.0 > .xi u s -All .-wE Social Insurance (Pensions) 2u 5 - 2=w 10.0 -- All SocialAssistance EW 5.0 (Veterans+Civilian) i I .. I 1 I r x0 01 42 Q3 Q4 45 Source; Authors' calculations using HBS 2007 data. Figure ES 4: TargetingAccuracy of Social AssistanceBenefits- InternationalComparison r WeakTargetingAccuracy of Social Assistance Benefits: BiH, FBH, RSwith International Comparison 70% 60% 50% 40% 30% 20% 10% 0% I Source: Nguyen and Lindert (2009) and authors' calculations using HBS 2007 data (for BH). Within the sphere of non-insurance social protection cash transfers, veteran-related benefits are the most regressive, with 27 percent of veteran-related benefits going to people in the richest quintile of the population, while those in the poorest quintile receive less than 15 percent of veteran-related benefits. Civilian benefits are somewhat better targeted. Twenty-six percent of the expenditure on the Child ProtectionAllowance (which is means-tested) and 30 percent of the expenditure on the Social Assistance Benefit (also means-tested), Non-War Invalids' Benefit, and Civilian Victims' o f War Benefit reach the poorest quintiles o f the BH population. Nevertheless, these outcomes are not very good when compared with other countries. iii Means-tested benefits are better targeted in RS, where those in the poorest quintile receive 48 percent o f Social Assistance benefits and 35 percent o f Child Protection Allowances. This performance is reasonable by international standards for poverty-focused programs, although there is certainly room for improvement since some programs in new European Union Member States in the ECA attain targeting accuracy outcomes o f 70 to 80 percent (on a par with means- tested programs inthe United States and Brazil). The poverty-reduction impact o f non-insurance social protection cash transfers i s quite limited and falls short o f regional and international norms. As mentioned, BHspends, on average, about 4 percent o f GDP on these benefits. However, coverage of the poor i s low (about 15 percent o f those in the bottom quintile report receiving veteran or civilian benefits) and benefits are generally regressive (those inthe poorest quintile receive 18 percent o f total non-insurance social protection cash transfers in BH). Given those patterns, it i s not surprisingthat poverty-reduction impact is negligible. Indeed, the poverty headcount rate i s estimated in the 2007 Household Budget Survey (HBS) at about 18 percent o f the population with the transfers counted in total consumption (incomes). Without the transfers, the poverty headcount would increase only slightly to 19.2 percent o f the population (so transfers reduced poverty by only 1.2 percentage points, or 6 percent). By way o f contrast, the poverty impact o f social insurance benefits (pensions) i s much higher-without these transfers poverty would increase to 25.8 percent o f the population. The opportunity costs o f public spending on generally regressive transfers are also high. Public expenditures on non-insurance social protection cash transfers absorb a huge share o f the Entities' respective budgets. This level of spending requires buoyant public revenues. However, public revenues will be under continuing pressure in view o f the impending economic crisis. Moreover, devoting a large proportion o f public funds to social transfers has the effect o f crowding out resources that could be devoted to public investments-which will be increasingly needed to stimulate growth as the economy begins to sag under the impact o f the world economic crisis. In addition, there is evidence that some rights-based programs create disincentives for employment. This situation is fiscally unsustainable, economically inefficient, and socially inequitable. BH needs to completely overhaul its non-insurance social protection cash transfer programs. There are many ways in which BH could reform these programs and put in place measures aimed at developing a social safety net that is: (a) less o f a burden on public resources, (b) more efficient, and (c) better targeted to the poor. Specifically, it is recommended that the governments in BH consider a three-pronged approach with measures to: 0 Improve and introduce targeting mechanisms to better channel resources to the poor; 0 Strengthen benefits administration and beneficiary registry systems; and, 0 Rationalize disability-related benefit schemes. An increasingly widespread recognition o f the need for rationalization of the non-insurance social protection cash benefits is discernible inboth the decision-making circles and inthe public discourse in BH. This i s a new development. In FBH, in particular, it i s increasingly clear to decisionmakers that the unsustainably large portion o f these social transfers inrelation to the rest o f the FBH budget is counterproductive to the objective o f adequately protecting those in acute iv need, especially in view o f the meager coverage o f the poor by the current programs. This realization has provided a new impetus for reform and given a new context for debate on this issue. Inturn, this has led to a policy dialogue. Inthe past, this dialogue was almost completely absent or, at best, quite weak. Inearlier years, there was no political or institutional will for such dialogue, and public debate remainedclouded inthe politically and emotionally charged debates. The Government o f FBH has recently taken some important steps toward rationalizing its existing non-insurance social protection cash transfer programs, with the announcement o f the "Social Pact." The Pact essentially sets out a commitment by key stakeholders (including trade union representatives and employers) to push for reforms to rationalize the safety net inFBH on a number o f different fronts, including improved targeting, fiscal restraints, improved benefit registries, oversight and controls, improvements to the Medical Examinations Institute (responsible for overseeing certification for disability benefits), and so forth. Quite unexpectedly, and under mounting fiscal pressure, the Parliament o f FBH has, at the government's initiative, adopted amendments that significantly reform the area o f the Non-War Invalids' Benefits program. Thus, the non-war invalids' program will dispense benefits only to invalids who are certified as being 90 percent and 100percent disabled. This i s incontrast to the previous practice whereby invalids whose disability ranged from 60 percent to 100 percent were all eligible for benefits. Moreover, iffurther amendments are adopted, the program would also cover congenital invalids o f a lower percentage o f disability until age 18 or, if infull-time education, untilage 27. While the reform was heralded as quite substantial, it is estimated that it will be fiscally neutral in2009 and, owing to the arrears that have been accumulated, the full effects will probably not be felt until 2010 or even 2011. For the purposes o f this paper (and the proposed Bank project) the reforms in question are not material because: (a) the effects will be rather limited in the medium term, and (b) they do not represent a fundamental change inthe official thinking but are merely a temporizing measure designed to reduce the government deficit for the time being. Given this new "window o f opportunity" for reforms-and increasingpressure to rationalize the safety net-the government has requested assistance in developing updated tools that could be used to improve the targeting of non-insurance social benefits. An improved targeting mechanism will help raise the issues above the politically expedient rights-based approach to a more neutral needs-based approach that enhances benefits for those who are socially and economically vulnerable. Using proxies to estimate the welfare o f a household, such as the Proxy Means Testing (PMT) and Hybrid Means Testing (HMT) methods, i s one way to improve the targeting mechanism. These methods have been proven to work particularly well in countries with a highlevel o finformality andwhere personaland householdincome is difficult to verify. To contribute to an ongoing debate in this area, this Policy Note simulates a PMT model using the latest microdata for BH-the 2007 Household Budget Survey ( H B S j a n d finds that the potential improvement over the current means-testing (MT) programs i s substantial. This Note updates the scoring formula based on HBS 2004 done in June 2008. The PMT model predicted distribution with the HBS 2007 data remains as strong as was estimated with the 2004 data. The P M T model i s comparable to, and by some measures an improvement over, previous BH-related PMT models. It also appears to be comparable to the performance of PMT models in other countries. Should the PMT scheme be implementedwell, the empirical predictions suggest that a substantial improvement over the results achieved by means-tested programs in BH during 2001-07 i s to be expected. Currently, the targeting accuracy, as measured by funds disbursed to the poorest 20 percent of the population, of the BH MT programs such as the Child Protection Allowance and benefits awarded via the Centers for Social Work i s around 25 percent, while the forecasted targeting efficiency-should PMT be used-is above 55 percent. In other words, implementing PMT for the two abovementioned sets o f benefit schemes would double the efficiency with which the authorities can target the poor who are the most in need o f these benefits. Nevertheless, despite relatively good P M T simulation results, empirical simulations, similar to any analytic forecasts, have certain limitations as predictors o f actual success o f proposed reforms inthis sphere. Above all, the biggest limitation i s the fact that a great deal o f success o f any PMT program depends on how well the reforms are implemented on the ground (see Castaneda and Lindert 2003). On the technical estimation side, the limitations o f the predictions include: (a) inadequate income data from the HBS, (b) lack o f disaggregated social protection transfer information, and (c) lack o f administratively feasible proxies in the HBS data. Furthermore, a comparison between MT, PMT, and HMT methods i s not possible due to underreported income-that is, because we do not have correct information on income, we cannot say how well income does as a predictor o f welfare. Reforming safety nets is usually an iterative and ongoing process that takes place over a significant period o f time. Initial measures could involve the development o f technical tools (for example, targeting mechanisms) and legislative reforms to pave the way for implementing improved benefittargeting and benefit administration and management. While there are ample technical opportunities for strengthening and reforming the safety net in BH (many o f which are discussed inthis report), the Entity Governments will need to strike a careful balance between fiscal pressures for reform (which are increasing under the global crisis) and political support for such measures. Such a balance will needto come into play indecisions about (a) which programs to target, (b) how narrow to target them (setting levels o f thresholds that focus narrowly on the poor or more broadly on "lower-income groups," for example), and (c) how fast to proceed (bold, broad-sweeping, and fast reforms versus a more gradual approach). A strategy for continued consultations and clear communication o f the rationale and need for reforms will also needto accompany any technical strategy for improving the system to balance political support for reforms with the fiscal, efficiency, and equity objectives for overhaulingthe system. vi SECTION 1: ANASSESSMENT EXISTINGNON-INSURANCE PROTECTION TRANSFERS OF SOCIAL IN BOSNIAAND HERZEGOVINA This section assesses existing non-insurance social protection cash transfers in Bosnia and Herzegovina (BH) with an overview o f public spending on such transfers (very high), the composition o f spending (biased toward rights-based transfers to the detriment o f needs-based benefits), coverage and targeting accuracy (weak), andpoverty impacts (negligible). It also notes that these high levels of spending on untargeted transfers likely crowd out resources for public investments-which will be increasingly needed to help stimulate the economy inthe face o f the global economic crisis. There is some evidence suggesting that they also dampen incentives for adult employment. The section concludes with recommendations for overhauling the system o f social transfer benefits inBH. 1.1.OVERVIEWOFPUBLICSPENDINGONNON-INSURANCE TRANSFERSBH IN Bosnia and Herzegovina spends a significant share of its gross domestic product (GDP) on non-insurance social protection transfers. Public spending on transfers has averaged about 4 percent o f GDP (Table 1.1). The BH economy grew by about 40 percent during 2002-08, as did expenditures on non-insurance social protection cash transfers. Since the overall spending on non-insurance social protection cash transfers has continuously grown during this period, it i s reasonable to suppose (given its size) that this form o f expenditure has tended to crowd out other forms o f expenditure inBH. O n the whole, while this form o f public spending may have boosted somewhat the aggregate consumption o f consumer goods and services over the period in question, it has done little to enhance the country's growth prospects, or, indeed, to reach many o f the poor. The opportunity cost of spending a large proportion o f GDP today on social transfers also implies foregoing greater competitiveness o f the BH economy and higher-yielding opportunities for this capital4pportunities that could increase the overall wealth and welfare o f BHcitizens in years to come. Hence, spending 4 percent of today's GDP diverts the resources from investmentsthat could produce greater GDP growth inthe future. By devoting a significant proportion o f its resources to consumption, BH society is, implicitly, putting a very low price on the wealth and welfare o f the coming generations. 1 Table 1.1: Expenditure on Non-insurance Social Protection Cash Transfers inBH, as Percent of GDP 2002 2003 2004 2005 2006 2007 2008 CivilianBenefits 0.9 1.o 1.o 0.8 1.5 1.4 1.4 Veterans' Benefits 3.1 3.0 2.8 2.6 2.6 2.6 2.5 Total as YOof GDP 4.0 4.0 3.8 3.4 4.1 3.9 3.9 Note:Nominal BHGDP 13,736 14,505 15,786 16,928 19,106 21,641 22,831 inmillionKM Source: Central Bank of BHand World Bank staffestimates. In fact, BH is among the highest spenders in the broader region (Figure 1.1). On average, countries inthe Eastern Europe and Central Asia (ECA) region spend about 1.6 percent of GDP on non-contributory social protection transfers while, on average, Organisation for Economic Co-operation and Development (OECD) member states spend 2.5 percent of their respective GDP. With the exception of Croatia, Bosnia and Herzegovina outspends all countries in the region. Figure 1.1. International Comparisons of Public Spending on Social Assistance PublicSpendingon Social Assistance,% of GDP Croatia Bosnia-Herzegovina Hungary Ukraine OECD Uzbekistan Macedonia Russia Moldova Armenia Latvia Estonia Belarus Lithuania Bulgaria Kyrgyzstan KOSOVO Serbia Romania Kazakhstan Albania Poland Georgia Turkey Azerbaijan Tajikistan 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00 Source: Lindert and others (2008). BH's high level of spending on non-insurance social protection transfers is fiscally unsustainable, particularly given the impending impacts of the global financial and economic crisis on public revenues. As a result of the global crisis, the country is likely to experience a contraction in remittances, exports, employment, and the availability of credit. 2 During 2008, the effects o f the global financial and economic crisis were not immediately apparent in BH. However, it i s now clear that the BHeconomy will feel the impact o f the global crisis, the effects o f which are likely to be transmitted through four main mechanisms: (a) a credit squeeze, (b) a fall in demand for BH exports and/or a fall intheir prices, (c) a drop inthe size o f remittances from the BH diaspora, and (d) a reduction in bilateral donor assistance. The credit crunch and a substantial reduction in the growth o f exports are already evident from the official statistics for the fourth quarter o f 2008. While the impact o f reduced remittances has not yet emerged, the size o f remittances i s likely to fall during2009 and insubsequent years. Like all other countries in Europe, BHis almost certain to experience a reduction in its rate of growth and, consequently, a fiscal contraction. In 2009, the rate o f real GDP growth i s likely be halved and the recovery is expected to be slow even after the eventual trough. On the one hand, unemployment i s expected to rise as the private sector sheds jobs in the coming months. On the other hand, poverty i s likely to be exacerbated further as job losses in other countries translate into reduced remittances for households in BH. Moreover, the authorities are likely to face a significant fall the size o f their revenues as the expected slowdown of economic activity negatively affects tax collection, especially since the bulk o f the BH public revenues come from indirect taxes, which are particularly vulnerable to falls inconsumption. Furthermore, the continued liberalization o f trade with the European Union (EU) will also put a dent in tax revenues. The projected fall in revenues will render high spending non-insurance social protection transfers even more untenable. High levels o f spending on transfers (4 percent o f GDP) are unlikely to be sustainable and will continue to crowd out spending on investments (as discussed in more detail below), which are crucially needed to help stimulate the economy. At the same time, improving the efficiency and targeting o f social benefits to those in real need would be a much more effective way to protect the vulnerable (and would create fiscal space for public investments) than scaling-up the existing programs. As such, the crisis will further create pressures for a rationalization o fthe social safety net (as discussed inmore detail below). 1.2. COMPOSITION OF PUBLICSPENDINGONTRANSFERSBH IN F o r historical reasons, non-insurance social protection transfers in BH have been heavily dominated by measures designed to protect veterans and/or their surviving dependents. Veteran-related benefits in Republika Srpska (RS) absorb about three-quarters o f total spending on non-contributory cash transfers (Figure 1.2), and this share has been fairly steady during 2005-09. Inthe same period, veteran-related benefits in FBHabsorbed an average o f 66 percent o f spending on the abovementioned transfers (Figure 1.2), though this share has declined from 82 percent o f the total in 2005 to 57 percent by 2009. Since 2006, FBH's overall spending on non- insurance social protection cash transfers was largely driven by the allocations made for demobilized soldiers and non-war invalids' benefits. The introduction o f the benefits for demobilized soldiers and non-war invalids (in 2006) has produced exponential growth in both types o f benefits with spendingon civilian benefits rising almost threefold during2006-08. 3 Figure 1.2: Compositionof SpendingonNon-insurance CashTransfers: Average for 2005-09 (Percent of total) 100% 80% 60% 40% 20% 0% FBH RS Veterans' Benefits Civilian Benefits Source: Entitybudgetsand World Bank staff estimates. In both Entities, veterans and their survivors benefit from a plethora of legislatively mandated benefits (Figures 1.3 and 1.4), including: 0 Military Invalids ' Benefit (Veterans); Legislation gives the right to benefits on the basis o f individuals' physical disability, regardless o f their financial means and/or employment status. These benefits are rights-based (rather than needs-based). Survivor Dependents' Benefit (Survivors); Legislation gives the right to benefits on the basis o f individuals' relationship status to the deceased person, who could be classified as a fallen soldier or deceased military invalid (veteran), regardless o f their financial means and/or employment status. These benefits are rights-based (rather than needs-based). Demobilized Soldiers' Allowance; FBH legislation mandates the right to benefits on the basis o f individuals' wartime involvement, regardless of their financial means. These benefits are rights-based, though dependent on formal unemployment status. This benefit i s particular to FBH. Medal Holders' Allowance: Legislation mandates the right to benefits on the basis of individuals' receipt o fthe highest military decorations. These benefits are rights-based. A detailed overview ofthe benefits and spending thereon is given inAnnex B. 4 Figure 1.3: Veteran-related Benefits in FBH Figure 1.4: Veteran-related Benefits in RS - - - - I _ FBH:CompositionofVeterans-Related Benefits RS: CompositionofVeterans-Related Benefits (% of spendingon Vets Benefits, Periodaverage 2005-09) ("hof spending onVets Benefits, Periodaverage, 2005-09) 0% Bi Mil InvalidsAllowance Mil Invalids &Survivor ISurvivorDependentBenefits Benefits PX SpecialSupplement DemobilizedSoldiers Allowance MedalHolders Allowance Medal Holders Allowance Other A Other Source: World Bank Benefil; Survey. (BetweenJanuary and April 2008, the World Bank mounted a survey of all public authorities' benefit programs in BH [Entity and sub-Entity], which yielded significant results. These results have been summarizedina table in Annex B. Updatesto the originaldata were made in late 2008 andearly 2009. WorldBankBenefit Survey thus refers to this exercise.) Both Entities operate a number of civilian benefits, though these receive relatively less funding than veteran-related benefits. Civilian benefits account for about one-third of total spending on non-insurance transfers inFBH and about one-quarter o f such outlays inRS (Figure 1.2). Some civilian benefits-such as the Social Assistance Benefit Program and the Child Protection Allowance program, are means-tested. FBH also operates two types o f disability benefits for non-war invalids and civilian victims o f war, which are rights-based. Civilian benefits include: o Social Assistance: According to FBH legislation, it is up to Cantonal social protection laws to set the amounts and criteria for regular social assistance, while in RS, the legislation determines the amount based on family size and income. Eligibility for a permanent cash benefit may be awarded to a personwith no other source o f income, no family support network, and no ability to work. One-off social assistance i s also provided on an as-needed basis to persons in temporary difficulty. Receipt o f this benefit does not constitute an entitlement to regular benefits. These benefits are means-tested. o Child Protection Allowance: Legislation prescribes means-tested benefits in cash andin-kindfor mother and child. o Non-War Invalids 'BeneJit (NWI) -Disability BeneJits: FBHlegislation gives the right to benefits on the basis of individuals' physical disability, regardless o f their means and/or employment status. These benefits are rights-based and are particular to FBH. o Civilian Victims ofwar (CW): FBHlegislation gives the right to benefits on the basis o f an individual's physical disability (or relationship status to the deceased person who could be classified as a CVW), regardless o f their financial means 5 and/or employment status. These benefits are rights-based and are particular to FBH. The current composition o f civilian beneficiaries is illustrated inFigure 1.5. The compositionof civilian benefitshas evolved in favor of rights-based disability benefits in FBH. With the introduction of benefits for non-war invalids and civilian victims of war, the composition o f civilian benefits has changed significantly over the past five years (Figure 1.6). Whereas in 2005, means-tested social assistance and child protection benefits dominated, by 2009, these accounted for only 15 percent and 17 percent o f spending on civilian transfers. Disability benefits for non-war invalids increased substantially, accounting for over half o f spending on civilian-related non-contributory transfers by 2009. Owing to mounting fiscal pressure, the FBH Parliament has, at the government's initiative, adopted amendments that significantly reform the area o f Non-War Invalids' Benefits program. Thus, the benefits program in question will cover only the 100 percent and 90 percent non-war invalids, as opposed to the 60 percent to 100 percent invalids, as previously. Moreover, if further amendments (discussed at the time) are adopted inthe near future, the program would also cover congenital invalids o f a lower percentage o f disability until age 18 or, if in full-time education, until age 27. While the reform was heralded as quite substantial, it is estimated that it will be fiscally neutral in2009 and, owing to the arrears that have been accumulated, the full effects will probably not be felt until 2010 or even 2011. Nevertheless, while they appear substantial on paper, the reforms in question are not material because: (a) the effects thereof will be rather limited in the medium term, and (b) they do not represent a parametric change in the official thinking but are merely a temporizing measure designed to reduce the government deficit for the time being. While this step is a far cry from the sequenced and measured approach advocated in this Note, the sheer fiscal necessity posed bythe worsening fiscal situation inFBHhas ledto this dramatic action. Therefore, it i s important that the authorities reform other benefit programs on time and before circumstances force them to adopt crude and socially unpopular reform measures that may still leave behind the most vulnerable members o f society. Therefore, substantively, the reforms in question have altered neither the fiscal outlook nor the policy landscape ina meaningful manner. 6 Figure 1.5: Composition of CivilianBenefits Figure 1.6: Changing Composition of Civilian BenefitsinFBH Other Other Civilianvictims of war Civilian victims ofwar Non-war invaIids Non-war invalids Child benefits(MT) @Childbenefits (MT) Social ksistance (Mi) Social Assistance(MT) FBH RS 2005 2006 2007 2008 2009 Source: World BankBenefit Survey. This shift in composition of benefits in FBH has also been accompanied by an expansion in the overall envelope for both veteran-related and civilian benefits. Broad trends discernible in Figures 1.7 and 1.8 indicate that, over the years, expenditures on non-contributory cash transfers grew at a faster rate inFBHthan inRS. Overall, there has been an increase in spending on rights-based benefits in FBH (Figure 1.8). Since 2006, FBH's overall spending on non-insurance social protection cash transfers was largely driven by the allocations made for demobilized soldiers and non-war invalids' benefits. The introduction o f the benefits for demobilized soldiers and non-war invalids (in 2006) has produced exponential growth in both types o f benefits with spending on civilian benefits rising almost threefold during 2006-08 and spending on veteran-related benefits nearly doubling inthe same period. So far, RS has largely resisted pressure to introduce similar benefits, even though 2007 saw the introduction o f a token annual benefit for demobilized soldiers whose cost at KM12 million per year (first budgeted in 2008) is manageable, even though policy implications thereof are a cause for potential future concern. Meanwhile, the amounts devoted to the military invalids' benefits, although they remain very high, have been stagnating somewhat in both FBH and RS. 7 Figure 1.7: Spendingon Benefits is Fairly Steady inRS 450 - ; 400 -- .c 0 =0 E: -300 .- .-tn350; 2 250200 c E 2 G= 150 -- a c 100 -- d 50 2 -- aa > 0 - I , , I ,1 -Veteran-related Benefits RS -Civilian BenefflsRS Source: World BankBenefitSurvey. Figure 1.8: Upsurge in Rights-based Benefits in FBH FBH:TrendsinSpendingonVeterans & CivilianBenefits (millionKM) -Veteran s Benefits (Rights Based) 250 -Civilian Disability 200 Benefits (Rights Based: 150 NWI+CVW) 100 -- Civilian MT Benefits 50 (Means-Tested: 0 SA+CA) 2005 2006 2007 2008 2009 Source: World Bank Benefit Survey. 8 Over time, the' proliferation of right-based benefits has led to increased beneficiary numbers at the expense of means-based programs,especially in FBH. During2002-06, the number of "persons not having sufficient income to support themselves" in BH has risen from 168,890 to 174,330 while the number o f recipients o f permanent social assistance has declined by 30 percent, from 19,779 to 13,819 (BHAS 2008). On the other hand, the same period has witnessed a near doubling in the number o f "one-off' social assistance payments awarded. Bearing inmindthat the maximums for most "one-off' social assistance payments are KM200 to KM250 and that these allowances cannot be claimed more than three times in a single calendar year, this i s a very inefficient way o f alleviating poverty, especially since "one-off' allowances are mostly dependent on the discretion of executive officeholders (for example, mayors) rather thanan explicit, formal test ofeligibility. Right-basedbenefits have driven the overall increases in non-insurancesocial spending, especially in FBH. Thus, a notable trend has been the expansion o f beneficiary numbers and resources devoted to the (right-based) non-war invalids' benefit in FBH. The open-ended nature o f this benefit, its wide eligibility criteria, and the opportunities for double-dipping that it affords have resulted in the fivefold increase in the resource envelope devoted to this benefit-from KM30 millionthat was paid out in2006 to nearly KM157 million that was budgeted for 2008. Since the introductionof the NWI benefit in 2006, the number of beneficiarieshas tripled. The authorities are struggling to process a backlog o f some 100,000 applications, the slow processing of which i s as much the result o f their lack of administrative capacity as it i s an attempt at fiscal restraint by indirect means. This would appear to indicate that, generally, the benefits inFBHhave been a result o f an organic growth process and based on the notions of right and reward rather than a demonstrable material need. As such, this has produced a vicious cycle that has resulted in either increases of the existing benefits or the introduction of new benefits- without any regard to the potential poverty alleviation impact o f such schemes or even the overall fiscal balance. Another example of how expansion in beneficiary numbers plays out in practice is the impact of the demobilized soldiers' benefit in FBH, introduced in October 2006. During 2004-06, prior to the introduction o f this benefit, the number o f unemployed demobilized soldiers registeredwith the EmploymentBureaus had beenaround 60,000. As Figure 1.9 shows, the number o f demobilized soldiers who are officially registered as unemployed has risen in the run-upto the passage of the relevant legislation and thereafter. Consequently, the Employment Bureaus have recorded a 46 percent increase inthe number of people registered as unemployed demobilized soldiers between September 2006 and December 2007. 9 Figure 1.9: The Rise in Unemployed Demobilized SoldiersFollowingthe Introductionof Benefits : 1 m - 70000 6oooO 5 m - 4 m - 30000 - 20000 - 1 m - 0 1 I I I I I I I I I I I I I Source: FBHBureauo f Employment. Interestingly, the same law provided for social assistance benefits for those demobilized soldiers who were not eligible for a pension but were old enough (65 ,years o f age) to be taken off the Employment Bureau rolls. The cash payment for this subcategory of the demobilized soldiers' benefitwas subject to a means test. Information gatheredthrough interviews with officials from the relevant authorities indicates that the take-up has been extremely poor, with fewer than 200 beneficiaries. Tentatively, this would suggest that even a simple means-test targeting applied to all potential claimants would have radically influenced the number o f people eligible to claim this sort of benefit. As it is, the number o f registered beneficiaries appears to be growing on a monthly basis. 1.3. PERFORMANCEOF SOCIAL SPENDINGTRANSFERS: COVERAGEAND TARGETING ACCURACY AND THEIR IMPACT ON POVERTY Household survey data allow for an independent analysis of patterns in the distribution of non-contributory transfers. The 2007 Household Budget Survey (HBS) provides a snapshot o f the characteristics o f the population through a representative sample at the country level (BH), as well as for each Entity. Data collection was conducted across the year, with 7,468 households divided into samples and interviewed at monthly intervals. Survey modules cover consumption, income, sociodemographic characteristics, and so forth. Fortunately, the 2007 HBS also included a fairly detailed module on receipt of benefits from social protection programs.2 This allows for an independent analysis of the coverage, targeting accuracy and impacts of these programs. Typically, household survey data do a betterjob capturing the distribution o f benefits across the population (quintiles) because such surveys present a representative sample according to key socioeconomic strata. They perform less well at capturing coverage o f specific programs because *The module covers most programs, including a variety of contributory social insurance programs (various pensions) and a range of civilian and veterans non-contributory transfers (though two civilian benefits in the FBH were lumpedtogether into a single category:NWIand CVW). 10 they are not typically designed to be a representative sample o f beneficiaries o f specific programs. Despite significant fiscal outlays (4 percent of GDP), coverage of non-contributory transfers is low. Overall, only 12.4 percent o f the population reports receiving benefits from non-contributory social assistance transfers (civilian or veteran-related) in BH as a whole. The share reporting coverage o f such benefits i s slightly higher among the poorest quintile (15.1 percent) than the richest (9.7 percent). A much larger share o f the population reports receiving social insurance benefits (40 percent), and about halfthe population reports receiving any type of benefits (contribution-based social insurance and/or non-contributory social transfers), as shown inFigure 1.10. As expected, coverage of veteran-related benefits is higher thancivilian benefits, and coverage o f veteran-related benefits i s highest among the middle and upper quintiles than those inthe poorest quintile. Figure 1.10: Coverage of Social Protection and Social Assistance Benefits inBH,H B S 2007 A. Social ProtectionBenefits B.SocialAssistance Benefits - 70 1 16 z .= 60 -AllSocial c 14 \ n n --AllNon-Insurance a2 .-g8 n Protection 50 --AllSocial Insurance .->12 \ (Pensions) ?: 10 \\ + . -. .... f 40 -e ....(VeteranstCiviiian) Veterans Benefits .x,xxuII,-* c All Non-Insurance -.- 30 \ (VeteranrtCivilian) 2 8 -2 6 h T - . + ' ' --"-Civilian Child .; I 20 - 'a Protection \ Y s .-cs4 :10 --" Civilian Other (SA, rill- s NWI,CW) r = 2 c0 y o 1 Y O QIQ2 Q3 Q4 Q5 Q1 Q2 Q3 Q4 95 Source: Authors' calculations using HBS 2007 data. Targeting accuracy is fairly weak overall, with a higher share of benefits going to those in richer quintiles than poorer. Overall, the distribution o f social protection benefits is regressive in BH. Those in the poorest quintile (representing 20 percent o f the population) receive 16.9 percent o f total social protection benefits (similar for social insurance and total social assistance benefits), as shown in Figure 1.11. The distribution o f overall social assistance benefits is slightly progressive inRS, where those inthe poorest quintile receive about 25.7 percent o f non- contributory social benefits, compared to 14.1 percent for those in the poorest quintile in FBH. However, even this slightly progressive outcome i s relatively weak compared to outcomes in manycountries inthe ECA region (Figure 1.12). 11 Figure 1.11: RegressiveDistributionof SocialProtectionBenefitsinBH I BiH:TargetingAccuracyof SocialProtection Benefits (Distributional Incidence; 2007 HBS) -All Social Protection -All Social Insurance -All (Pensions) Social Assistance - (Veterans+Civilian) ae 0 01 0 2 0 3 Q4 Q5 - - ~. Source: Authors' calculations using HBS2007 data. Figure 1.12: TargetingAccuracy of SocialAssistanceBenefits InternationalComparison - WeakTargetingAccuracy of Social Assistance Benefits: BiH, FBH, RSwith International Comparison 2 60% 70% , . .E 50% c, 40% X 30% g8 20% 10% 0 0% Source: Nguyenand Lindert (2009) and authors' calculations using HBS 2007 data (for BH). Within the sphere of non-contributory social benefits, veteran-related benefits are the most regressive, with 26.7 percent of veteran-related benefits reaching those in the richest quintile of the population, while those inthe poorest quintile receive less than 15 percent of these benefits. Civilian child protection allowance (which i s means-tested) and other benefits (SA+NWI+CVW) are somewhat better targeted overall, with 25 to 30 percent of such benefits going to the poorest quintile, respectively, though these outcomes are not very good compared with those in other countries (Figure 1.12). 12 Means-tested benefits are better targeted in the Republika Srpska, where those in the poorest quintile receive 47.7 percent o f CSW benefits and 35.4 percent o f child protection allowances (Figure 1.13). This performance i s reasonable by international standards for poverty- focused programs (Figure 1.14), though there is certainly room for improvement (some programs in ECA attain targeting accuracy outcomes of 70 to 80 percent-on a par with means-tested programs inthe United States andBrazil). Figure 1.13: Weak TargetingAccuracy of Specific SocialBenefitsPrograms:FBHand RS -0, JC 90.0 a '- 80.0 S 70.0 8 60.0 50.0 $ 40.0 'El '8% 30.0 20.0 5d - 10.0 ul s Civilian Civilian Civilian Veterans Civilian Veterans w- Child Benefits - 80 Social Child Other (SA, Benefits - Assistance Protection NWI, CVW) RS Protection FBH - RS - RS - FBH - FBH Source: Authors' calculations using H B S 2007 data. - Figure 1.14: InternationalComparison of TargetingAccuracy _I _ _ __ TargetingAccuracy of Specific Poverty-Focused AssistanceBenefits 90% ........... .................... . __ ..,"~--.,._I.". I ............ 80% 70% ................ .... 60% . . . . . . . . . 50% 40% 30% 20% 10% 0% -_- - _ I_ " ~ - -- __ Source: Lindert and others (2006); Nguyenand Lindert (2009). 13 1.4. IMPACT OF BENEFITSONPOVERTY With low coverage and weak targeting accuracy, it is not surprising that the poverty impacts of non-contributory social benefits are negligible. As discussed, the whole country spends, on average, about 4 percent of its GDP on non-contributory social benefits. However, coverage of the poor is low (about 15 percent o f those in the bottom quintile report receiving veteran-related or civilian benefits) and benefits are generally regressive (those in the poorest quintile receive 18 percent of total non-contributory benefits in BH overall). Given those patterns, it is not surprising that poverty impacts are negligible. Indeed, HBS 2007 estimates the poverty headcount rate at about 18 percent of the population with the transfers counted intotal consumption (incomes). Without the transfers, the poverty headcount would increase only slightly to 19.2 percent o f the population. Therefore, transfers reduced poverty by only 1.2 percentage points, or 6 percent. When this poverty-related impact of the non-insurance cash transfers i s compared to the poverty-related impact o f insurance-based benefits (pensions) there i s a stark contrast: without pensions, poverty would increase to 25.8 percent o fthe population. 1.5. OPPORTUNITY COST OF SOCIAL SPENDING: CROWDING OUT OF PUBLICINVESTMENTS Public expenditures on non-insurancesocial assistance benefitsabsorb a large share of the Entities' respective annual budgets. When the public expenditures on non-insurance social protection cash transfers are expressed as a proportion of the Entities' respective annual budgets, the staggering proportion o f resources that authorities inFBHdevote to these transfers i s revealed (see Table 1.2). Such transfers absorb 41 percent o f total budgetary spending in the FBH, compared with 14 percent in RS. Hence, it would appear that non- insurance social protection cash transfers currently account for some 19 percent of FBH's consolidated expenditures and 7 percent o f RS's. Table 1.2: Spendingon Non-Contributory Social Assistance Cash Transfers in B Has Percentof Budgets 2002 2003 2004 2005 2006 2007 Federation ofBH 35.1 35.3 36.6 42.9 39.0 41.0 I Republika Srpska 13.6 13.7 14.7 15.8 13.9 13.9 Source: Officialbudgetsand WorldBank staffestimates. This high level of spending requires a high level of revenue collection. On the revenue- gathering side, the authorities are burdening the private sector and individuals with hightaxes thereby reducing the scope for private investment andjob creation. Consequently, tax burdens (particularly inFBH)have remained stubbornly highinrelation to OECD/EU-8 average^.^ For instance, during 2004-07, the size o f overall BH public revenues collected on the basis o f direct taxes and social contributions rose by 0.1 percent and 0.3 percent of GDP, respectively, WorldBank's BHFiscalUpdate. 14 while the introduction o f the value-added tax (VAT) in 2006 resulted in a 4.2 percent year- over-year, rise.4 High spending on benefits has the effect of crowdingout resources that could be devoted to public investments. On the expenditure side, public investment represents only 6 percent o f the country's GDP, which means that the prospects for economic growth and job creation are limited. Public investment as a percentage o f GDP i s even lower in FBH-far lower than other countries in the region (Figure 1.15). Though it is difficult to put this expenditure in a contrasting counterfactual scenario, a greater proportion o f public expenditures devoted to investments and capital goods would have had the potential to stimulate the country's economic growth. Figure 1.15: PublicInvestments as a Percentageof TotalPublic Expenditures (2006 P u b l i c I n v e s t m e n t a s a p e r c e n t a g e o f t o t a l p u b l i c e x p e n d i t u r e s ( 2 0 0 6 ) Source: World Bank BHFiscalUpdate 2008. The combination of high public outlays on regressive social transfers with low public investments puts BH in a particularly difficult situation with the onset of the global economic and financialcrisis. Giventhat BHhas a high current account deficit and, givenits orthodox fixed-exchange rate regime, there are few macroeconomic policy options at its disposal to deal with economic downturns. Inthe face o f economic contraction, targeted social transfers are needed to help protect the poor from the adverse effects o f the crisis, and public investments are neededto stimulate economic activity andjobs. With the current allocation o f public spending, BHwill find itself handicapped for both o f these crisis responses. Moreover, there i s a risk that the authorities will be burdened with a set o f claims and entitlements for (untargeted) social assistance benefits that would, if unchecked, absorb an even greater proportion o f the (reduced) GDP as the economy contracts. This would leave authorities little room to maneuver to implement investment programs and other measures that would help the economy exit from any recession sooner. Ibid. 15 1.6. HIDDEN COSTS OF UNTARGETED BENEFITS -LABOR MARKET IMPLICATIONS There is some evidence that untargetedbenefit schemes in BH also produce disincentives and distortionsin the labor markets. It is estimated that 57 percent o f people capable of work inBHare economically inactive-some 1.5 million-in addition to those who are registered as unemployed, or, some 500,000 people as o f December 2007.5This state o f affairs cannot be the sole consequence o f untargeted benefits, which are not generous enough to produce unemployment and economic inactivity by themselves. In the "grey" economy, workers are frequentlywillingto work for lowwages if remittances and untargeted social protection benefits supplement their generally unstable or insufficient earnings. In this way, both employers and employees profit jointly thanks to untargeted social benefit transfers-with the real losers being those who need to depend on social assistance payments as their sole source o f income. At present, one-quarter of all registered unemployed in FBH appear to be demobilized soldiers-as shown by the data presentedinTable 1.3. Moreover, the veteran benefits system in BH extends to include beneficiaries of 20 to 40 percent disability range who are capable of working. Indeed, the system i s dominated by this category o f beneficiaries-as illustrated in Figure 1-16.For instance, in August 2007, in FBH, there were 26,155 such beneficiaries on the rolls (one-half o f all disabled veterans' benefit claimants), accounting for KM1.5 million o f the total monthly expenditures on disabled veterans' benefits that month. Table 1.3: Demobilized Soldiers as Percent of Total Number of Unemployed inFBH,2002-08 2002 2003 2004 2005 2006 2007 2008 22.7 20.5 18.9 17.4 17.6 23.8 24.5 Source: FBHEmploymentBureau. Judging from the figures available, in August 2007, total payments to lower-category disabled veterans and the unemployed demobilized soldiers totaled KM14.5 million. It i s worth noting that the two types o f beneficiaries in question were, technically, capable o f productive employment. Figure 1.16 illustrates what this meant interms o fthe size o f the individual benefits paid out and how they compare with the social assistance benefits paid out to civilians who are unable to work. As indicatedfrom the datareleasedby the EmploymentBureaus of FBHandRS. 16 Figure 1.16: Veterans Beneficiaries, by Disability Category Veterans beneficiaries by cisability catepry I II 111 M w VI VI1 WII D[ x Source: WB Benefit Survey. The right-based survivor dependents' benefits are available to spouses, children, parents, and grandparentso f killed soldiers, according to defined conditions, which have nothing to do with employment status. Inboth RS and FBH, more thanhalf o f the beneficiaries under the veterans' protection system are members o f the families o f killed soldiers, and this constitutes a survivor family assistance scheme as much as it is an aid to disabled veterans. Moreover, the recent changes to the FBHlegislation have awarded the survivor benefit rights even to widows who are younger than 45 years o f age, thus providing further disincentives to employment and formal economic activity. 17 SECTION2: IMPROVED TARGETING SOCIAL ASSISTANCEBENEFITS: OF PROXY-MEANS-TESTING MECHANISM ON THE 2007HBSSURVEY BASED This section develops a variety o f technical tools that could further contribute to the debate regarding possible ways to improve the targeting o f non-insurance benefits in Bosnia and Herzegovina (BH). It calibrates a new proxy-means-testing (PMT) model for BHusing the 2007 Household Budget Survey (HBS).The results indicate that the PMT model i s comparable to, and by some measures an improvement over, previous BH PMT models. It represents a substantial improvement over the present means-tested programs operated in BH in 2007. Currently, the targeting efficiency, as measured by the share o f recipients who are inthe bottom quintile o f the population o f the BH means-testing (MT) programs, i s around 30 percent, while the forecasted targeting efficiency o f the P M T model is above 55 percent. It also appears to be comparable to the performance o f PMT models inother countries. This update o f the BH scoring formula includes a larger number o f sensitivity tests. The tests checked the accuracy o f the model across different specifications o f the regression and for each domain (region) where HBS 2007 is representative. The parameters o f any PMT model change over time, and countries revise the PMT weights after some years (as i s done in Armenia and Chile, for example). Above all, the targeting accuracy o f any program implemented in BH will depend not only on the design, but also on the quality o f implementation. 2.1. RATIONALEFOR TARGETING OVERVIEW OFTARGETING AND INSTRUMENTS A. Why Target? Targeting is a means o f increasingprogram efficiency by increasingthe benefit that the poor can receive within a fixed program budget. The motivation for targeting arises from three policy considerations: (a) objectives o f reducing poverty and protecting the poor; (b) limited resources (budget constraints); and (c) opportunity costs, or tradeoffs between the number o f beneficiaries and the level o f transfers (Coady, Grosh, and Hoddinott 2004). Simply put, the rationale for targeting involves concentrating scarce resources on those who need them most. B. Whom to Target? Whom to target is generally determinedby need, that is, economic status (poverty, risks o f poverty), but it can also relate to other aspects associated with vulnerability such as age (elderly, children), ethnicity (historically excluded groups o f the population), or disability. Policy choices policymakers make in determining whom to target based on measures o f need include the following: 18 0 Narrow vs. broader targeting. In many countries, targeting based on "need" focuses social assistance resources on a rather narrow definition o f "the poor" (as in Brazil, Mexico, and the United States), with higher benefits for the extreme poor and a gradual reduction in benefits as incomes rise. There i s some evidence that the political economy o f targeting in those countries favors such narrow targeting. In Brazil, for example, evidence suggests that politicians are penalized for perceived "leakages" o f benefits to the non-poor and have a higher likelihood o f reelection with "stronger" targeting o f the poor (de Janvry and others 2006; Lindert and others 2007; Lindert 2008). In other countries, programs are targeted to a broader definition o f "lower-income groups," possibly in part to bring in a broader political basis for support. Ultimately, the decision o f what "threshold" to set for eligibility depends on a combination o f fiscal and political economy considerations that are specific to each country. Regardless o f the level at which such eligibility thresholds are set, however, the tools used for screening for eligibility (means tests, proxy means tests, hybrid means tests) and for managing benefits (unique registry system) should be standard, common, and transparent for all applicants. 0 Chronic vs. transient poor. Another aspect o f "whom to target" involves whether to targetthe chronic or transient poor. This dependspartly on the objectives o fthe particular safety net program, but is also particularly relevant in times o f crisis. Fiscal constraints mean that not all can be served as much as needed, giving rise to competing pressures. The logic of a crisis response program i s to address the income losses caused by the crisis. However, while the newly poor are often politically vocal, they are not necessarily the poorest (Grosh and others 2009). The chronically poor are likely to become poorer as a result o f the crisis and may be most at risk o f suffering irreversible losses. These choices o f target group also affect the type o f targetingmechanism adopted, with "proxy- means testing" more appropriate for depicting chronic poverty, but less sensitive to changes ineconomic status (for example, crises). C. How to Target? A number o f mechanisms exist for channeling resources to a particular target group. Some require some sort o f assessment o f eligibility for each applicant (individuals or families). Others grant eligibility to broad categories of people based on single characteristics such as geographic location (geographic targeting) or demographic category. Needs-based targeting (where the target group i s "the poor") generally adopts applicant screening methods (for individuals or families), but sometimes also combines these with geographic targeting. This review focuses on needs-based targeting via applicant screening methods (for individuals or families). An important aspect of targeting is the need to design program parameters (benefit levels, entry and exit criteria, and so forth) such that they avoid creating opportunities for "masquerading" or changing behaviors to become eligible for benefits or incentives for reducing adult work effort. There are several methods for screening applicants (individuals or families) for eligibility, including: (a) means-testing (MT), (b) proxy means-testing (PMT), and (c) hybrid means-testing (HMT). The choice among methods generally depends on administrative capacities, degree of formality or "measurability" o f incomes, and variation in other observable characteristics associated with "need." Table 2.1 provides an overview o f these measures, the types o f data that are collected, and their respective advantages and disadvantages, based on international practice. 19 Currently, BH uses income and asset tests (means-testing, MT) to determine eligibility for the child allowances and social assistance program. Usually, countries with a large formal sector use verified income and asset-tested programs (VMT). This targeting method i s found in most Organisation for Economic Co-operation and Development (OECD) countries, with notable examples in Australia, France, the U.K.,and the United States. The success o f the means-tested programs depends on extensive verification o f information, which covers two aspects: (a) the identity o f the applicant and familyhousehold composition, and (b) the income and assets o f the assistance unit. The information submitted by applicants is verified based on documentary evidence (the applicant presents documents and invoices), andvia automated computer matches. At the other extreme, countries with a large informal sector use indirect methods of estimating welfare, especially based on a proxy means test (PMT). PMT-based programs determine eligibility based on a multidimensional index o f observable characteristics highly correlated with the welfare (consumption, income) o f the household. Typically, these include information about location, housing quality, possession o f assets/durables, education, occupation and income o f the adults, and a variety o f others (disability, health, and so forth). The variables are aggregated into a composite score (index) using weights determined using a regression model. Eligibility i s determined by comparing the score o f each household with an eligibility threshold. First developed in Chile, then used extensively in much o f Latin America, PMT programs are now spreading to other parts o f the world, such as Armenia, Georgia, Indonesia, the Philippines, and Turkey. Between these two extremes, there are intermediate solutions that combine the elements o f means-tested and PMT programs. We call this intermediate targeting method a hybridmeans test (HMT). Under the HMT model, programs assess the welfare of the applicant based on a per capita income indicator that i s the sum o f verifiable income (from wages and social protection transfers) and the estimated unverifiable income. This model i s being developed in some transitioneconomies, notable examples o fwhich are Bulgaria, Kyrgyzstan, and Romania. Targeting those "in need" involves not only an assessment o f "means" (incomes, proxies, imputedincomes) but also a "threshold" cutoff to distinguish between those who are eligible and those who are not. Such a threshold can be determined empirically-for example, a poverty line estimated using costs o f basic food and non-food consumption. Or it can be determined more broadly to allow for inclusion o f the near-poor (vulnerable) or lower-middle-income groups, depending on the objectives o f the program and the political calculus for acceptability o f the reforms/program. Regardless o f the level o f the threshold for eligibility, the "tools for targeting" should be standard, common, and transparent for all-namely, a consistent measure for estimating "means" (HMT, PMT) and a single registryo f applicants. E. ImplementationMatters.Finally, indeveloping a targeting system, implementationmatters. Beyond eligibility criteria and estimates o f "need," programs need to be supported by adequate administrative capacity at all levels, registry, and information technology (IT) systems, oversight and controls mechanisms, clear institutional roles, and so forth. These are crucial inputs to strengthening the safety net in Bosnia, though the remainder o f this Note focuses on the design aspects. 20 0 0 0 0 . rl N 0 0 1 0 0 0 2.2. HOUSEHOLD DATA SURVEY USEDTO DEVELOP TARGETINGTOOLS A. The HouseholdBudget Survey (HBS). We use data from the 2007 HBS to develop options for tools that could be used to target social safety net programs in BH. For the foreseeable fwture, only HBSs-not Living Standards Measurement Surveys (LSMSs)- are likely to be conducted. A new round o f an expanded HBS/Survey o f Income and Living Conditions (S1LC)-light is planned to be conducted in 2010. It is therefore important to continue updating the baseline PMT model based on the HBS, subject to further verification and improvement as data from future rounds o f the HBS are made available. Inaddition, inits current form, the HBS is a rich source o f information on both household consumption patterns and the demographic and socioeconomic characteristics o f members o f the household, physical characteristics o f dwellings, and ownership o f durables (see Annex C for more information). The availability o f information on these variables makes the HBS a suitable database for calibrating a PMT model. B. Shortcomingsof the HBS data for PMT/HMT Simulations.The first shortcoming o f the HBS i s that, in its current form, it is unable to provide information on a number o f indicators on "non-monetary" measures of living standards. Unlike the LSMS, the HBS does not have detailed modules on, for example, access to education or health services, agricultural activities, or labor market activities. The current HBS-based model i s therefore unable to capture certain information that was used in previous models on agricultural activities, unlike the Bosnia PMT models of Braithwaite (2003) and CEPOS (2006) or the P M T models in Russia (World Bank 2007), or war-related variables, such as Bisogno and Chong (2001) and Braithwaite (2003) (for a review o f previous models, see Annex C). Second, the 2007 HBS resolves only some o f the 2004 HBS's lack o f disaggregated information on social assistance benefits received. For instance, two growing non- insurance and non-income-tested programs, NWI and CVW, are lumped together under one category-Center for Social Works (CSW) benefits-in the HBS questionnaire (see Annex F for the actual social protectionmodule used inthe HBS questionnaire). In2004, the income module only asked survey respondents whether they receive "other fees and additions," including unemployment benefits, disability benefits, social and humanitarian benefits, and others. We are now able to better assess the targeting performance o f the social assistance system and then compare it with PMT simulations. The 2007 HBS also has an improved capability to monitor living standards, including revisions to the reference periods associated with expenditures on selected goods (including utility expenditures) and an updating o fthe samplingframe. Third, the income data in the HBS is severely underestimated, which prevents us from simulating an HMTmodel using the 2007 HBS data. Inorder to calibrate an HMT model and predict whether income is a good proxy o f consumption, the household survey data should have high-quality income data. The quality o f income data, generally a difficult variable to collect in household surveys, is a function of, first, the level o f informality in the economy, and second, how the income question was asked. In BH, the level o f 22 informality in the economy is high. In addition, the HBS questionnaire is not detailed enough to capture self-employed and agricultural incomes. 2.3. EMPIRICAL FRAMEWORK THE PMTAND HMTESTIMATION FOR This section presents the results o f simulating a PMT model, which predicts the consumption o f each householdbased on a limited number o f variables, and explains why anHMTmodel cannot be calibrated withthe HBS due to the underreportingof income in the data. These tools build on four previous documented efforts to calibrate proxy-means- testing (PMT) models for BH(see Annex C for more information). A. New PMT Model UsingtheHBS 2007 This section calibrates a new PMT model or scoring formula for Bosnia and Herzegovina (HB). It draws on the 2007 Household Budget Survey (HBS) and builds on previous efforts to design a PMT model, including the World Bank 2004 PMT estimation. Means- tested and hybrid-means-tested models are not calibrated based on the HBS data set because o fthe weak income data. Following the literature, the choice o f explanatory variables for PMTs (and the imputed proxy aspects of HMTs) i s guided by its statistical association with per capita consumption and its verifiability (that it, that it can be potentially cross-checked against other sources o f information, or may be physically inspected or verified by a social worker, or that households are arguably less able or less likely to provide misleading or false information). The exercise starts from a large set of variables, which are then reduced to a much smaller subset using stepwise estimation techniques, that is, a subset o f variables selected on the strength o f their statistical association with per capita consumption andthat together these variables maximize the fitness o f the PMT model. Our variables can be broadly classified under one o f the following categories: 0 Household demographic and socioeconomic characteristics, such as the number o f members, their ages, the number o f dependents, gender of the head o f household, and the educational attainment o f household members. It also includes labor market activities, such as the employment status, occupation, and sector o f employment o f the head o f household, the number o f employed members o f the household, and the occupational status o f the spouse. O f these characteristics, labor market activities may be the most difficult to verify, given the existence o f a large informal sector. 0 Housing characteristics, such as the availability o f certain facilities (water, sanitation system, phones, and so forth), the types o f appliances used, the manner by which heat is supplied, the year the dwelling was constructed, the number o f rooms, construction type (multifamily, individual, other), whether owned or rented, and so forth. 23 Ownership of selected durable goods, such as ownership o f vehicles, telecommunications equipment, or selected appliances. This can be potentially assessed against administrative data, such as data on vehicle registration, or by visual inspection. 0 Location, such as a household's entity o f residence or whether the household lives ina rural or urbanarea. 0 The affordability of selected expenditures, such as utility (water, heat, electricity, gas) expenditures, which inprinciple can be verified by the respective utility company. Selected income sources, such as whether the household receives pension income. Pension receipt and/or the level o f pension income should be easily verifiable with the administrative records o f the PensionFund. B.A Note on HMT andIncome Data Quality To calibrate an HMT (a mix o f proxy- and income-tested eligibility) model, we need two conditions for the survey data: (a) income and consumption are highly correlated; and (b) income i s not severely underestimated. These conditions are necessary to make the argument that income is a good approximation o f the true welfare o f the households (which is the theoretical concept o f "permanent income," best approximated by consumption). In Ukraine, for instance, it was found that these conditions hold (Tesliuc and others 2009). InBHandRussia, on the other hand, this is not the case. InHBS 2007, income levels are one-third o f the corresponding consumption levels (see Annex D). First, we have done a number of tests to assess whether the quality of the income and consumption data collected inHBS is adequate. Income data, unfortunately, seem heavily underreported (Figure 2.1 and Annex D). Income is generally a difficult variable to collect in household surveys, but in BH there are several reasons why income might be underreported: (a) the income question on informal and self-employed income i s not detailed enough, and experimental evidence from other countries has shown that less- detailed questions lead to underreporting compared to more-detailed questions. The question posed to survey respondents i s only, "Do you have income from own company, craft, agricultural holdings, or freelance (employers and self-employees)"; (b) an agricultural income module is lacking; and (c) the recall period for income and consumption are different. Income uses annual recall while consumption uses biweekly recall; thus, when comparing the two, income will tend to have more o f a downward bias thanconsumption. 24 Figure 2.1: Household Monthly Per Capita Income and Consumptionover Quantilesof Income Distribution A. Entire Population B. Householdswith Hard-to-verify Income ess Percent Source: Authors' calculations using the 2007 HBSdata. Second, we use the HBS data to estimate the share o f verifiable and unverifiable income (Figure 2.2). Verzj?abZe income considers (a) salaries and benefits o f public sector employees or employees on a permanent contract, (b) all public transfers except pensions from abroad, and (c) savings and rent o f business premises. Hard-to-verzfj income includes salaries and benefits of all those not on a permanent contract or in the public sector, salaries paid by foreign employers (even though the sector is highly formalized, access to information is not readily available), and all rents and interests except savings and business premises. Data from the HBS 2007 indicate that a slightly larger share o f the incomes earned in urbanareas can be verified, such as wages and social protection transfers (Figure 2.2). In rural areas, however, unverifiable income dominates. The share o f unverifiable income does not decrease with higher welfare status, suggesting that income-based programs will be subject to errors. Hybridmeans-testing models combine information on (a) verifiable income, and (b) proxy indicators that are incorporated into a prediction model to estimate the share o f incomes that are hard to measure. Thus, because of the heavily underreported income data, we cannot use the HBS to simulate an HMT program for BH. In addition, the high level o f hard-to-verify income suggests that income-based programs might run into difficulties, as i s the case for the current FBHprograms. 25 Figure 2.2: Composition of Household Income Over the Distribution of Welfare A. Entire PoDulation B. Urban Areas C.RuralAreas Source: Authors' calculations using2007 HBS data. 2.4. RESULTS: PROXY-MEANS-TEST MODEL USING THE 2007 HBS A. Baseline PMT Model Annex F Table AF.1 presents the results o f the stepwise regression analysis. The final model consists o f 25 variables, derived from an initial set o f about 50 variables. For the dummy variables representingthe entity o fresidence, the omitted category is Brcko.6The indicators o f heating source are in relation to "other" sources o f heat. Every variable i s significant at the 1 percent level. A positive coefficient indicates that a household or dwelling characteristic is associated with higher per capita consumption; a negative coefficient, conversely, indicates that a characteristic is associated with lower per capita consumption. The signs o f the coefficients make intuitive sense or are consistent with existing analyses o f poverty in BH, though the PMT model should not be interpreted in any causal senseq7 For example, the 2003 Poverty Assessment and the 2005 Poverty Update suggested that poverty i s lower (and thus per capita consumption is higher) among female-headed households' and among those with relatively more educated heads o f households and that Brcko District is an autonomous region, which, though part o f the country, i s separate from the two Entities that comprise BH. 'Thecoefficient estimates o f the PMT model should not be interpreted in any causal sense, that is, that possessing a certain characteristic leads to higher poverty. Nor should we expect that the coefficients estimates and their signs would be necessarily consistent with our prior expectations, given the likelihood o f co-linearity or the strong statistical association between independent variables. A coefficient estimate with an unexpected sign (for example, car ownership associated with lower predicted per capita consumption) may, infact, serve a useful practical purpose. That is, it can be an important deterrent against any attempt to provide false information to the scoring formula or to the system. 8This phenomenon runs counter to the experience o f other countries and is not well understood, even in BH, based on our consultations with our counterparts. Nonetheless, this statistical pattern holds up across various BH household surveys: the 2001 LSMS, the 2004 LSMS, and the 2004 HBS. They are also consistent with some recently published analysis o f gender and poverty in BH (Smajic and Ermacora 2007). 26 poverty rises with the number of household members. These patterns are confirmed by the regression results in Annex F Table AF.1. In addition, the ownership of selected durable goods (cars, appliances, and so forth) is, as expected, positively associated with per capita consumption. Housing characteristics also have the expected signs: the use of firewood and coal stoves, typically associated with poorer families in remote areas, is negatively associatedwith per capita consumption. The R2of the baseline model is equal to [0.488], compared with the [0.496] of the 2004 model.' This measure of the model's goodness-of-fit is an improvement over revious BH PMT models. The model in Braithwaite (2003), for example, yielded an R'= 0.35, while Bisogno and Chong (2001) obtained an R2= 0.32 intheir best model." This is also broadly comparable to or higher than the R2in the older PMT literature covering other countries. For example, the models for Latin American countries in Grosh and Baker (1995) yielded R2values up to, at best, 0.41;Grosh and Glinskaya (1997) obtained R2= 0.2 in Armenia; and the final model for Egypt in Ahmed and Bouis (2002) yielded R2= 0.43. However, the baselinePMT model for BHdoes not perform as well as a few recent PMT models calibrated in some countries in the ECA and in other regions. For example, the PMT model for the Targeted Social Assistance program inGeorgia had an R2= 0.62. The one calibrated for the Republic of Kalmykia inthe RussianFederation has an R2= 0.59. Similarly, the PMT models considered in Sri Lanka obtained R2 values up to 0.59 (Narayan and Yoshida 2005). Figure 2.3 plots the actual per capita consumption values against those predictedby the baseline PMT model for 2004 and 2007 models. 9 In statistics, the coefficient of determination, R2,is used inthe context of statistical models whose main purpose is the prediction o f hture outcomes on the basis o f other related information. It is the proportiono f variability in a data set that is accounted for by the statistical model. It provides a measure o f how well hture outcomes are likely to be predicted by the model. Inregression, the R2coefficient o f determination i s a statistical measure o f how well the regression line approximates the real data points. An R2 o f 1.0 indicates that the regression line perfectly fits the data. lo The R2inCEPOS (2006) is not reported. 27 Figure2.3: BaselinePMT Model-Actual and Predicted Values A. 2004 Estimates B. 2007 Estimates N Source: Authors' calculations usingthe 2004 and 2007 HBSdata. B. SomeSensitivity Testsfor PMT Using the poorest half of the population. Following the PMT literature such as Braithwaite, Grootaert, and Milanovic (1999) and Grosh and Baker (1995), one could argue that using only the poorest half o f the population may be a better basis for calibrating a targeting model. The literature argues that this is far more realistic than making use o f the full population, since members of the upper class and upper-middle class are not likely to bother to participate insocial assistance programs. Inpractice, then, the proxy means test will likely apply only to those near the lower end o f the welfare distribution. Following this line o f argument in this section, we calibrate the baseline PMT model using only the poorest 50 percent and the poorest 40 percent o f the population, respectively. The new regression results are in the fourth and fifth columns of Annex F Table AF.1. The associated errors o f inclusion and exclusion are also the second and third columns o f Table 2.4. Calibrating a more parsimonious PMT model. The baseline PMT model follows the standard structure o f existing PMT models, that is, it uses information on household and housing characteristics. However, as previously suggested, some characteristics lend themselves more conveniently to inspection or verification than others. Employment status is, for example, difficult to verify in an environment where a large informal sector exists. What happens if we exclude this variable from the baseline model? The results suggestthat the predictive power diminishes only slightly; the R2drops from 0.49 to 0.46. The errors o f exclusion and inclusion are broadly similar to the baseline model. Urbadrural variations. One could also assess how well the baseline PMT model performs across geographic boundaries. The urban area P M T model yields an R2= 0.458 while a rural area PMT model produces R2 = 0.46 (Annex Table AF.3). These are associated with errors of exclusion o f 79 percent and 57 percent, respectively, and errors 28 o f inclusion o f 2 percent and 6 percent, respectively. Thus, the urban model is relatively better able to predict the non-poor, but the rural model i s relatively better at identifying the poor, similarly to the 2004 estimates. Entity-level models. Considering the vastly differing results in program performance between the entities, we calibrate a PMT model for FBH and RS separately as a sensitivity check. The estimated regression for each entity remains basically unchanged from the national model without gaining any further precision (Annex F Table AF.2). The R2stays at .5 for each entity. The current programs' distribution o f beneficiaries (percent o f beneficiaries who constitute the poorest 20 percent o f the population) in the RS is 39.4 percent for child protection and 46 percent for all other social assistance, and inthe FBHit is 20.8 percent and 31 percent, respectively. Underthe simulated PMT for each entity, the predicted distribution o f beneficiaries i s 57 percent and 54.31 percent" in FBH and RS, respectively, which is the same level as the predicted performance of the national model o f 55.4 percent. C. How well does the simulated PMT identijy thepoor? A further test o f our baseline PMT model is to see how well itpredicts the poor and non- poor using the variables listed in Annex F, Table AF.1. For the purposes o f this simulation, we identify the poor as those with monthly consumption per capita less than KM385.69 (the relative poverty line used by the Bosnia and Herzegovina Agency for Statistics [BHAS] in their 2007 poverty profile). Using actual monthly per capita consumption, data from the 2007 HBS indicate that the poverty headcount i s about 18 percent and that there i s essentially no extreme poverty (0.6 percent o f the population). Inthis section, we compare actual consumption per capita-and actual poverty status- with the predicted consumption per capita-and predicted poverty status-to see how well the baseline P M T model performs in identifying the poor and non-poor. The results are inTable 2.2, under the first column. The results suggest that the baseline model correctly predicts 39.1 percent o f the poor and 95.3 percent o f the non-poor. This implies that the (a) error o f exclusion-the percent o f poor predicted as non-poor-and (b) the error of inclusion-the percent o f non-poor predicted as poor-are 61 percent and 4.7 percent, respectively. In other words, the model performs very well in identifying the non-poor. It does not perform as well in identifying the poor, but i s comparable inthis respect to previous PMT models for Bosnia (Braithwaite 2003) or models calibrated for other countries (see, for example, Braithwaite, Grootaert, and Milanovic (1999) on baseline PMT models for Bulgaria, Hungary, and Poland, which correctly identified only 47, 13, and 17 percent o f the poor, respectively). It is also comparable to the 2004 HBS estimates. ~ l1Percent ofpredictedrecipients that are inthe bottom20" percentile of the Entities' distributions. 29 Table 2.2: PMT Errors of Inclusion and Exclusion 2004 2007 Baseline Using Using Baseline Using Using Using Poorest Poorest Using Poorest Poorest PMT Model allHHS 50%a 40%" allHHS 50%a 40%a Percent correctly identified Poor identified as poor 41.0 56.4 71.6 39.1 59.9 73.6 Non-poor indentified as non-poor 94.5 90.0 82.4 95.3 88.2 79.5 Percent incorrectly indentzjled Poor identifiedas non-poor (error of exclusion) 59.0 43.6 28.5 60.9 40.1 26.4 Non-poor identifiedas poor (error of inclusion) 5.5 10.0 17.6 4.7 11.8 20.5 When the model is calibrated only on the poorest 50 percent or 40 percent o f the population, its ability to predict the poor correctly i s now substantially higher-the percent o f poor correctly predicted as such rises to 59.9 (better performance than for 2004) when usingthe poorest half o f the population and to over 70 percent when only the poorest 40 percent i s considered. Because policymakers are likely more interested in reducing errors o f exclusion, targeting only those at the bottom o f the welfare distribution represents an improvement in our PMT model's performance. However, the errors o f inclusion also rise as a result. When only the bottom 40 percent i s considered, the percent o f the non-poor correctly predicted falls from 95.3 percent to a little over 88.2 percent. One drawback of the comparisons based on inclusion and exclusion errors i s that they are sensitive to the coverage o f the program. In general, the smaller the program, the higher the errors o f exclusion and the lower the errors o f inclusion, all else being held equal. One way to avoid such fallacies o f composition is to present information about the coverage and the incidence o f beneficiaries for a program o f a given size by decile, as in Table 2.3. Given a total poverty rate o f 18 percent, we present the simulated distribution o f beneficiaries and the coverage for a program that targets the poorest 20 percent (quintile) of the population, using the three PMT and HMT models calibrated earlier. This is done by considering only the lowest 20 percent o f the predicted values for 30 Figure 2.4: Incidenceand Distribution of InclusionErrorsof EachModel A. Incidenceof InclusionError B. Distribution of InclusionError Source: Authors' calculations using2007 HBS. The distribution o f simulated beneficiaries is progressive, or strongly pro-poor, for all the PMT model variations (Figure 2.4). For the baseline PMT model, 33 percent in 2007 (compared to 36 percent in 2004) o f the projected recipients belong to the poorest decile of the population, with another 21.1 percent from the second decile; overall, 55.4 percent o f the beneficiaries o f the simulated programs belong to the poorest quintile. The PMT model predicted distribution remains as strong as was estimated with the 2004 data. For 2004, using prediction based on the subsample o f the 40 percent or 50 percent poorest will improve this indicator, but only marginally (to 56.1 percent and 55.9 percent, respectively) and this remains the case for 2007, as well. Any of the PMT model variations is a substantial improvement over the current distribution of benefits as found with the 2007 HBS data. F o r the overall social safety net: Overall, the distribution o f social protection benefits is regressive inBH.Those inthe poorest quintile (representing 20 percent of the population) receive less than 17 percent o f total social protection benefits (similar for social insurance and total social assistance benefits-Figure 1.1); F o r existing means-tested benefits: In the RS, those in the poorest quintile receive 48 percent o f social assistance benefits and 35 percent o f child protection allowances (Figure 1.13). The FBHtargeting accuracy i s much lower-1 7 percent for child protection and 25 percent for other social assistance. The simulated PMT model compares favorably in terms of targeting accuracy o f the performance of other countries operating means- and proxy-means-tested programs (Figure 2.5). However, this comparison should be qualified: the targeting accuracy of any programs implemented in BH will depend not only on its design, but also on the quality o f implementation(Castaneda and Lindert 2003). 31 Figure2.5: Share of Beneficiariesinthe Bottom Quintile -InternationalComparisons 80 60 s 40 20 0 ECA Source: HBS 2007 actual and simulated results, and Tesliuc and others 2009. D.Limitations of thePMT model The PMT model has several limitations stemming from the imperfect HBS data set on which predictions are made. Due to limitations o f the HBS dataset, the baseline PMT model seems not to perform as well as the models calibrated by CEPOS (2006) in correctly identifying the poor, with errors o f exclusion rangingfrom 33 percent to 44 percent and errors o f inclusion ranging from 4 percent to 7 percent. This i s because of the difference in data sets that were used for calibrating the models. CEPOS used the LSMS 2004 data set while this PMT model calculated by the World Bank used the HBS 2007 data set. This mainly reflects the usefulness o f LSMS-type data over HBS-type data for any welfare analysis. In addition, the CEPOS sample size is about halfthe size ofthe World Bank sample. The HBS data set has further peculiarities that might make the PMT simulations in BH less precise. First, the HBS 2007 sample is scattered across the whole year; hence, consumption i s measured in different months over the year. Consumption i s highly seasonal. This seasonal effect reduces the fit o f the PMT regression. Second, the variance o f the per-adult-equivalent (PAE)12 consumption i s typically less than that o f the per- capita (PC) consumption. It means that PAE consumption discriminates less than PC consumption. PC consumption picks up better correlates like household size and l2 For BH, the Bosnia and Herzegovina Agency for Statistics (BHAS) uses the OECD scale for adjusting the composition o f households to account for economies o f scale and better measurement o f consumption o f the individual within the household. 32 composition while PAE already adjusts for household composition. As discussed in section 2.2, some important predictors have not been collected in HBS 2007, such as agricultural income and assets and employment characteristics. Hence, the PMT model has less explanatory power on the right-hand side of the regressionand thus less power to approximate the real datapoints. 33 m d SECTION3: CONCLUSIONS AND RECOMMENDATIONS This section presents conclusions and recommendations regarding non-contributory social assistance transfers inBosnia and Herzegovina: 0 Public spending on such transfers is extremely high (4 percent o f GDP) and unsustainable, particularly inthe face o f the impending economic crisis. 0 Transfers are biased toward rights-based benefits for veterans/survivors and non-war invalids. Although these rights-based transfers reflect the post-conflict situation, and likely serve important political and social stability functions, they are regressive, transferring a higher share o f benefits to those in the middle and upper quintiles than those inthe poorest quintiles o fthe population. 0 Coverage o f the poor by non-contributory transfers i s quite low, meaning that the poor will receive limited protection with the onslaught o f the looming economic crisis. As a result, high spending on social transfers also buys negligible poverty impacts. 0 High spending on social transfers also likely crowds out resources for public investment-which will further cripple the governments' abilities to respond to the economic crisis or stimulate economic activity. 0 There i s also evidence to suggestthat transfers may dampen adult work effort. This situation is unsustainable, inefficient, and inequitable. An overhaul o f non-contributory social assistance transfers i s needed. Political and social constraints do need to be taken into account, and radical measures to terminate (regressive) veteran-related benefits are unlikely to have much political support. Nonetheless, there are many ways in which BH could reform its programs and systems to strengthen and develop a true social safety net that i s (a) less o f a burden on public resources, (b) more efficient, and (c) better targeted to the poor. Specifically, it i s recommended that the government consider a three-pronged approach with measures to: 0 Improve and introduce targeting mechanisms to better channel resources to the poor.The rationale for targeting benefits to the poor is to concentrate scarce resources on those who need them most. This concept i s particularly important in times o f crisis, where the poor are even more vulnerable to adverse effects o f the economic downturn. From a technical perspective, developing and introducing improved targeting mechanisms could be done fairly quickly, with a rollout o f revised eligibility mechanisms possible over a period o f 6 to 12 months. This would involve an assessment o f 35 institutional and implementation aspects o f existing enrollment criteria and processes in each Entity (Republika Srpska [RS] and the Federation o f Bosnia and Herzegovina [FBH]) and further diagnostics on proposed mechanisms to reform such criteria and processes. Such tools could be applied on a pilot basis for certain civilian and possibly war veterans' benefits inan initial phase. The tools developed inthis paper could provide important inputs to these diagnostics. The results o f the proxy-means-testing (PMT) modeling exercise indicate that the targeting accuracy would be substantially improved following reforms that would lead to an introduction o f PMT formulas as a means o f assessing the applications for some non-insurance benefits. From apolitical perspective, policymakers would need to determine the pace at which such reforms could be rolled out (rapid reforms versus a more gradual approach), the thresholds for eligibility to be established (more narrow focus on the poor versus a broader definition o f low-income groups), and which programs would be selected for targeting based on need. Such political decisions would need to strike a careful balance between fiscal pressures and political support for such reforms, and should be accompanied by a strong consultative process and communications strategy to improve awareness in BH of the need for such reforms. 0 Strengthen benefits administration and beneficiary registry svstems. In many countries, modernizing and strengthening benefits administration i s an important "hook" for progress on the safety nets agenda, with many benefits in terms o f improved efficiency; reduced duplications, fraud, and errors; and improved transparency. The importance o f such improvements i s even more apparent in BH, given the high level o f spendingon such benefits (4 percent o f GDP) and the plethora o f non-contributory social transfer benefits in both Entities. In many cases, definitions o f key parameters for these programs are inconsistent (for example, definition o f the assistance unit, benefits calculations, eligibility criteria, and so forth). Registry systems are also weak, with little integration across cantons in FBH or across programs. This creates ample scope for errors, fraud, and duplicate benefits, both within and across programs. The introduction o f automated, unified registries o f social benefits programs is often a crucial first step toward improved efficiency, reduced duplications, and consolidation o f safety net programs. Developing harmonized and consistent parameters i s an important input for unified registries (and usually requires legislative reforms). Improved institutional capacity and information technology i s also usually needed, both at central agency (ministry) levels and at local levels (centers for social work). Oversight and controls mechanisms (such as random-sample spot-checks, internal cross-checks, and so forth) are also usually important for reducing fraud and errors. Such improvements in benefits administration and registries are generally medium-term measures requiring significant technical assistance and institutional%lTinvestments, though some initial actions could be taken in the short run for example, mapping out and assessing informationjlows, institutional responsibilities,and soforth). 0 Focus on rationalizing disabilitv-related benefit schemes. The apparent inability to impose restraint on the rapidly increasing expenditures on disability-related benefits i s a key challenge facing authorities, especially in FBH. Notwithstanding the recent attempts to reform the area o f non-war disability benefit programs whose effects will be limited at best, given the present sociopolitical conditions, it i s difficult to expect that this whole 36 segment can be subjected to radical reforms inthe near future. The proposed measures, if taken in a measured and sequenced fashion, would eventually help the governments tackle the challenges in this segment by virtue o f keeping the focus o f disability benefit programs on the most vulnerable beneficiaries, while achieving savings to bringdown the overall high non-insurance social transfer expenditures. Specific recommendations include: o Clarifying and strengthening systems for certifying and recertifying different levels o f disability (such measures could be introduced relatively quickly, in the short-run, though they would require continuous oversight). o Introducing targeting-related eligibility criteria to better channel increasingly scarce resources to the poor disabled (those most in need). Politically, it would likely be more feasible to start by introducing targeting criteria (for example, some form o f means-testing) to civilian disability benefits (Non-War Invalids benefits) rather than war veterans' benefits. This is also a most pressing concern from the point o f view o f fiscal sustainability, given that these expenditures have ballooned inrecent years inthe FBH.Such measures could be introduced in the short to medium term, once targeting tools are developed. o Improving benefits administration, registry management, and oversight and controls mechanisms for disability benefits (medium-term measures). Reforming safety nets i s usually an iterative and ongoing process that takes place over a significant period o f time. Initial measures could involve the development o f technical tools (for example, targeting mechanisms) and legislative reforms to pave the way for implementing improved benefit targeting, benefit administration, and management. While there are ample technical opportunities for strengthening and reforming the safety net in BH (many of which are discussed in this Note), the Entity Governments will need to strike a careful balance between fiscal pressures for reform (which are increasing under the global crisis) and political support for such measures. Such a balance will need to come into play indecisions about (a) which programs to target, (b) how narrow to target them (setting levels o f thresholds that focus narrowly on the poor or more broadly on "lower-income groups," for example), and (c) how fast to proceed (bold, sweeping, and fast reforms versus a more gradual approach). A strategy for (continued) consultations and clear communication o f the rationale and need for reforms will also needto accompany any technical strategy for improving the system to balance political support for reforms with the fiscal, efficiency, and equity objectives for overhauling the system. 37 m 03 Q, m r-: 3 0 W O r - O O ? . c ? b 2 3 m '? 0 m m 3 3 3 3 c \ l 3 - W w - m - W 2 W m 3 3 3 - 3 3 2 0 d m r-: 3 m 3 vi m ln 3 3 3 - 3 - 2r- d 0 d z 0 -3 - c? 00 m 2m . . . . . . . . . . . . rcl 3 - 3 - 3 3 z m d b W d m ? . W d z . . . . . . . . . . 00 m 3 3 3 3 3 3 .-0 0 Y 3 0 a e N d m d ei .e N m 9 E: 0 2 4x .-zP .-a A z 3 Y 6 Y &0 c iig. .e 0 0 9 c"4 E B w s , Q 0 ._c0 iia3 5w v) 2 m 8 i * 0 3 a eC 2e u 4C e 6 9 9 9 9 9 9 9 9 9 9 9 9 9 ~ O O O O O O O O 0 0 0 4 1 - 3 - 3 3 3 - ~ 0 0 0 0 0 0 0 0 0 0 0 0 3 1 1 - 7 9 9 9 9 9 9 9 9 9 9 9 9 s z z z z z z z s s s s z ~ 0 0 0 0 0 0 0 0 0 0 0 0 c 0 8 .3 Y 0 m O m ~ b m m ~ w m O m m 9 9 9 9 9 9 9 9 9 9 9 9 m b m m m - m m - - 0 0 0 0 0 0 0 0 0 0 0 0 Q w P Q m m b P m o P P m w b m w m b b m m b b 9 9 9 9 9 9 9 9 9 9 9 9 rl 0 0 0 0 0 0 0 0 0 0 0 0 m w .C v) 9 e 0 5 c 0 ex a, P N u, Y m .d -.- d Y E g .I + cd E .I E 0 .I Y cd ru a 0 .I 58a B W P x % z> -a 8 .I Y E & .I - .I Y Ce a 3 8 cd P x t 5C 1 C !eC . C 1C N 9 iC ccc i C L LC i f L F e E 0 5 1a P (1 ta z 2 2 2 z z z z 0 0 0 0 0 0 0 0 p" 4 0 W n e cE YW C .I Y 8 W gPE v1 C i% Icl 0 a 0 .I Y 9 a v a.. Y L v1 cl =P 4 4 0 ANNEXB: SPENDINGONNON-INSURANCE PROTECTIONCASH TRANSFERS SOCIAL Inmillions ofKM 2005 2006 2007 2008" Ministryb TypeC Federation of B H Veteran-relatedbenefits Mil.invalids' allowance 151 151 172.3 113.9 Veterans R Survivor dependents' benefit 150 150 152.8 221.2 Veterans R Demob. soldiers' allowance 0 0 60.7 Od Labor/Cantonse R Medal holders' allowance 0 17 17 19 Veterans R Miscellaneous 0 4.5 0 2.5 Veterans R Total 301 322.5 402.8 356.6 Civilian benefits Social assistance 33.8 31.9 24.7 39.3 Cantons M Child benefits 30.8 33.1 35.8 41.9 Cantons M Non-war invalids 0 30 124.8 157.7 Labor R Civilian victims o f war 0 9 42.3 40 Labor R Miscellaneous 0 25 0 0 Labor Total 64.6 129 227.6 278.9 Grand Total for FBH 365.6 451.5 630.4 635.5 2005 2006 2007 2008 Ministry Republika Srpska Veteran-relatedbenefits Mil.invalids' and survivor benefit 106.6 112.5 130 127 Veterans R Special supplement 0 0 10 12 Veterans R Medal holders' allowance 0 0 0.4 1.9 Veterans R Miscellaneous 5.5 5.3 4.8 3.3 Veterans Total 112.1 117.8 145.2 144.2 Civilian benefits Social assistance 10.4 10 7.5 8 Welfare/Municipality M Child benefits 25.1 37.5 38.4 39.5 Welfare/Fund M Miscellaneous 0 1.5 0 0 Welfare Total 35.5 49 45.9 47.5 Grand Total for RS 147.6 166.8 191.1 191.7 a. 2008 data refer to the plannedibudgeted amounts. Data for all other years refer to the amounts executed. b. Relevant Institutions: Ministry of Labor and Social Policy (FBH), Ministry of Labor and Veterans' Affairs (RS), Ministry o f Veterans' Affairs (FBH), and Ministry o f Health and Social Welfare (RS). c. Type: Right-based benefit (R), Means-tested benefit (M). d. While FBH Budget makes no reference to this line item, it has been estimated that KM170 million would be neededto finance obligations created by the relevant law that introduced this benefit in2006. e. This itemhas beenjointly financed by FBHGovernment and FBHand Cantonal extra-budgetary Employment Funds. f. This item has been jointly financed by the extra-budgetary Child Protection Fund with supports from RS Government. 62 THE FEDERATIONBOSNIAAND HERZEGOVINA ANNEXc:DETAILSOFTHE SOCIAL PACT IN(FBH) OF The FBH Government has recently taken several important steps toward rationalizing its safety net, with the announcement o f the "Social Pact," which essentially sets out a commitment by key stakeholders to push for reforms to rationalize the safety net on a number o f different fronts, including improved targeting, fiscal restraints, improved registries, oversight and controls, improvements to the Medical Examinations Institute (responsible for overseeing certification for disability benefits), and so forth. Specifically, the Social Pact makes a political commitment on a number o f areas including: 0 Section IV: Social partners (institutions listed above), are committed to adjusting the budget spendingfor social protection o f invalids and civil victims of war, with realistic financial possibilities (meaning available resources). In this regard, social partners support reforms inthis sector that will ensure that the protection is primarily targeted to lower-income beneficiaries. Misuse o f these benefits should be prevented by strengthening supervision. Section V: Regarding veterans and war invalids, social partners confirm their commitment that this protection be organized in accordance with a unified set of regulation (war invalids, demobilized soldiers, medal holders) and improved targeting toward beneficiaries in the need. Budget spending for these purposes should be adjusted to financial possibilities. 0 Section VI: Social partners support ongoing reforms of the pension and invalid insurancethat will ensure compliance with EuropeanUnion standards. It is necessary to widen the scope o f the insurance coverage to include categories that were left out heretofore (farmers, and so forth). Favorable conditions for obtaining pensions should be used very carefully and inaccordance with available resources. 0 Section VII: Employment-social partners agree to support employment in the production sector, o f invalids, older individuals, and other hard-to-employ groups. Job brokerage services should include private service providers. While the social pact has no legal standing, it i s an important political commitment-n the part o f a range o f stakeholders-to a much-needed course o f reforms. Another measure, the draft Social Protection Law in the Federation, which is in the final stages o f approval, also formalizes the introduction o f targeting for disabilities and other measures. 63 ANNEXD:DATA DESCRIPTION PASTPROXY-MEANS-TESTING(PMT)MODELS AND Data Description The Household Budget Survey (HBS) was designedto be representativeo f boththe national and the entity levelandthe urbdrural dimensions. For the 2007 HBS, data collection lastedall year, divided into 12 monthly samples. A total of 9,019 households were interviewedof which 7,468 completed interviews were retained. The survey instruments included: a diary of purchases (with a 14-day recording period); a self-consumption booklet (products produced and consumedby the household); and a final interview, covering topics such as non-food expenditure, socioeconomic and demographic characteristics of household members, ownership of durable goods, and housing. Data collection consisted of three visits by enumerators to each household: the first, at the beginning of the reference period, to deliver the diary; the second, at some point in the middle of the referenceperiod, to ensure that the procedure was understoodand carried out; and the third, at the end of the reference period, to conduct the final interview. A monthly list of householdswas provided to each enumerator. The 2007 HBS was the second HBS conducted in Bosnia and Herzegovina (BH) following the 2004 HBS. Prior to 2004, the main source of information on household living standards was a national survey conducted in 1997 (Bisogno and Chong 2001) and the 2001 Living Standards Measurement Survey (LSMS). Three more waves of the LSMS were conducted each year after 2001, though only the 2004 wave included a consumption module. Thus, the first HBS was conductedjust as the panel LSMS was winding down. A new round of the HBS was conducted in2007. Past PMT Models The four previous documented efforts to calibrate proxy means-testing included Bisogno and Chong (2001), Braithwaite (2003), Schreiner and others (2004), and CEPOS (2006). Though they each had specific objectives-for example, Bisogno and Chong (2001) were interested in designing a more efficient use of foreign-aid programs flowing into BH in the late 1990s following the 1995 Dayton Peace Agreement, while CEPOS (2006) was exploring alternative means of alleviating the poverty and social consequences of the introduction of the value-added tax (VAT) in 2005-these calibration exercises were all an effort to identify a limited set of variables that: (a) can be easily measured and verified, and (b) can be usedto predict per capita consumption with a reasonable degree of accuracy. These variables, in turn, could, in principle, be usedto improve the targeting of social assistance. These BH calibration exercises used various household survey data available at the time the exercise was conducted: Bisogno and Chong (2001) used a national "vulnerability ~urvey"'~of about 7,000 households conducted in 1997, Braithwaite (2003) used the 2001 LSMS covering some 5,400 households; and CEPOS (2006) used data from the 2004 LSMS on some 3,000 households. Following the literature on proxy means-testing (PMT), exercises invariably identified a set of variables related to the following: demographic characteristics, educational ~~ l3 The survey is otherwise known as the Food Security Assessment Vulnerability Survey. 64 attainment, and labor market activities of membersof the household; housing characteristics; and ownership of selected durable goods (for example, cars). The choice of variables was also determinedbythe availability ofrelevant informationina given household survey. The agriculture module in the LSMS, for example, allowed the use of variables related to livestock ownership in Braithwaite (2003) and CEPOS (2006). Bisogno and Chong (2001), on the other hand, made substantial use of war-related variables (for example, whether the household lost any of its possessions during the war), while Braithwaite (2003) had one war-related variable. Braithwaite (2003) also used information on receipt of disability, survivors, and war veterans pension, and on the geographic location of the respondent's birthplace. Though none of these PMT models were subsequentlymade operational, though they were based on household surveys that are different from the survey used in this note, their key results nonetheless provide a useful benchmark. 65 ANNEXE:ANALYSISOFTHE UNDERESTIMATIONOFINCOME Income reported in the Household Budget Survey (HBS) data is greatly underestimated compared to reported consumption, which i s the best approximation o f the theoretical concept o f "permanent income." We thoroughly looked at the following aspects o f income to determine ... whether the income data were underreported: Non-response by quintile and income source Medianand means by quintile and by income source Income source by type o f profession, sector o f employment, and type o f contract. The most telling indicator seems to be the ratio o f income t.0 consumption-if we think of consumption as a reflection o f the true welfare o f the household and want to estimate whether income is underreported. We found that public sector employees and permanent contract employees tend to have a higher total income-to-consumption ratio (Table AE.1). Public sector incomes are more regular and more easily verifiable and thus more easily recalled and reported to the HBS enumerator. Some other results remain unclear. For instance, the ratio o f salary income for private sector employees i s slightly higher than for public sector employees, which could reflect better wages, not necessarily better income reporting to the HBS enumerators. Overall, the quality of the income data i s poor. 66 Table AE. 1: Income/ConsumptionRatio in 2007 HBS Data Quintile 1 0.43 2 0.36 3 0.33 4 0.3 1 5 0.27 Professional Status Employer 0.35 Self-employed 0.27 Employee 0.34 Unpaidworker 0.27 Apprentice 0.19 Other 0.28 Typeof Work Contract or Activity Permanent 0.35 Temporary with contract 0.30 Temporary no contract 0.30 Payment 0.27 Seasonal 0.26 n.a.(Not available) 0.27 Sector Public sector 0.34 Privatesector 0.32 Mixedownership 0.34 NGO 0.35 Total 0.33 Looking at the household response rates to a yes/no question of "does your household receive income/pension/benefit" also points out the poor quality of income data. Over 90 percent of households reported not receiving any income during the last 12 months from the following sources: income from own company, property income, and remittances (Table AE.2). Only 56 percent of households reported receiving salaries at local employers in the last 12 months. On average, less than 10 percent of all households reported receiving social insurance or social protection transfers such as war veteran pensions, war disability pensions, work-related disability pensions, pensions from abroad, child benefits, benefits from the Center for Social Work, allowances, and unemployment benefits. 67 Table AE. 2: Percentof HouseholdsReportingNOto ReceivingIncome/Pension/Beneft I Income from (full- and part-time) employment: Salaries of employeesat local employers 44% Meal allowanceandtransportto andfrom work at localemployers 84% Salaries ofthe employeesat foreign employers (internationalemployers) 98% Allowance for living inother town andfees for managementboardmembers 100% Other income from employment(leave pay, bonuses, severance) 94% Incomefrom owncompany, craft, agriculturalholdings 67% Interestfrom savingsanddividends 100% Y Rentsfrom rentingland 100% Rents from rentingresidentialpremises 99% Rents fromrentingbusinesspremises,garages, etc. 100% Rents from rentingeauiument, cattle, etc. 100% Remittancesandreceiptsfrom abroad(exceptpensions) 94% Receiptsincash fromrelatives, friends, etc., in-country 95% Pensionand social transfers: War veterans pensions 99% War disabilitypensions 92% Survivoruensions 87% Old-agepensions 77% Work-relateddisabilitypensions 90% Pensions from abroad 97% Childbenefits 94% Benefitsreceivedfromthe Center for Social Work 98% Allowances (temporary andpermanent) 100% Unemploymentbenefits 100% 68 ANNEXF:STATISTICAL TABLES Table AF. 1: Baseline Model StepwiseregressionResults (Dependent variable: natuml log of per capita consumption) 2004 Baseline 2004Pwrest 2004Poorest 2007 Baseline 2007 Doorest 2007 poorest Ail HH 50% 40% All HH 50% 4'0% Household Charactedstics #of household members - 0 . 2 3 * * * - 0 . 1 1 * * + -0.10"' -0.17"' -0.10"' - 0 . 0 8 * * + - 3 9 . 9 9 - 19.75 -17.74 - 2 6 . 5 1 -14.96 - 1 4 . 2 1 #of children under 14 0 . 1 0 * * * 0 . 0 6 * * * 0.05"' - 1 0 . 0 6 - 6 . 9 1 -6.73 #of children, 14-24 0 . 0 3 * * * 0.02"' -2.94 - 2 . 6 1 # of elderly 65+ 0.03*** - 2 . 6 6 household head wilh postgradeducaSon 0.14*** 0 . 1 3 * * * 0 . 0 9 * * * 0.08"' - 5 . 4 3 - 5 . 8 -3.06 - 2 . 6 householdwith female head 0.09"' - 5 . 8 4 # of employed members 0.08"' 0 . 0 6 * * * 0 . 0 6 * * * 0 . 0 9 * * * 0 . 0 6 * * * 0 . 0 5 * * * - 8 . 4 9 - 7 . 5 1 - 6 . 4 3 - 1 1 . 6 - 7 . 6 6 - 6 . 3 1 Housing Chamctwistics (dumrnyvadables) hot water 0.10'** -4.92 central heating - 0.10'** - 0 . 2 4 * * * -0.26*'* - 2 . 9 2 - 4 . 9 1 -6.42 self-provided heating -0.18"' -0.20"' -3.73 -4.93 single equipmentheating -0.21"' -0.23"' - 0 . 0 7 * * * - 4 . 9 9 - 8 . 7 1 - 4 . 6 7 garage 0.06"' -3.98 balcony 0.05"' 0 . 0 4 * * * - 3 . 6 5 -2.98 garden 0.07"' 0 . 0 5 * * * 0 . 0 6 * * * 0 . 0 5 * * * - 4 . 2 - 3 . 8 1 - 5 . 1 7 - 4 . 1 6 kitchen 0 . 0 4 * * * 0 . 0 5 * * * -2.82 - 3 . 0 1 attic - 0 . 0 4 * * * - 2 . 9 6 69 (Table AF.l continued) Ownenhip of Dutabiw (dummyvallables) video recorder 0.07*+* - 4 . 5 8 car 0.23*** 0.16"' 0 . 1 4 * * * 0 . 2 6 * * * 0.12'** 0.10'*' -15.12 -11.8 -10.46 - 2 0 . 4 2 -10.45 - 7 . 8 4 retigerator 0.10"' 0.10"* 0.09**' 0 . 0 9 * * * 0.09"' -2.92 -3.6 -2.98 -2.64 - 2 . 7 7 computei 0.13*** 0.08**' -6.53 - 5 . 1 7 phone 0 . 0 5 * * * 0 . 0 5 * * * 0.12"' 0.11"' 0.11'** - 3 . 0 4 -3.22 - 8 . 3 - 8 . 1 9 -8.39 dish washer 0 . 2 1 * * * 0 . 1 5 * * * - 7 . 3 9 - 8 . 1 5 vacuum cleaner 0.07"' 0.07"' 0 . 0 7 * * * -3.88 - 3 . 1 8 - 3 . 9 firewood & maistom - 0 . 1 1 * * * - 4 . 1 7 sewing machine 0.10"' 0 . 0 7 * * * 0.05"' 0.07'** 0.07"' 0.05*" - 6 . 1 7 - 4 . 4 - 2 . 7 9 - 4 . 8 6 -5.24 - 3 . 6 8 mobile Dhone 0.15"' 0.08"' 0.07"' 0.19"' 0.11"' 0 . 1 1 * * * -10.46 - 6 . 0 1 - 4 . 7 5 - 1 3 . 9 9 -8.19 - 8 . 0 9 washer 0.06"' 0 . 0 7 * * * 0.12"' 0.08"' -3.76 - 4 . 3 4 - 6 . 4 5 -5.39 HI-FI systems 0.19"' - 5 . 0 5 satellite dish 0.06" * - 3 . 9 2 electric & gas cookers 0.10"' 0.08"' 0.08**' - 5 . 0 9 - 4 . 0 1 - 3 . 9 secondary home 0.21"' 0 . 1 4 * + * - 6 . 8 7 -3.09 Location (dummy varlable) Republika Srpska - 0 . 1 1 * * * -0.08*'* - 0 . 1 1 * * * 0 . 0 4 * * * -4.56 - 2 . 9 6 - 4 . 0 6 -2.94 FBiH - 0 . 1 3 * * * - 0 . 1 1 * * * -0.14"' - 6 . 3 8 -4.35 -5.34 Affordabiiity of selected expenditures logof udlityexpendihres 0.20"' 0.11**' 0.09"' 0.31"' 0.21"' 0 . 2 1 * * * -20.16 -11.4 - 9 . 4 6 -29.12 -20.37 -18.67 Income Source(dummyvadaMe) remims Densioninmme 0.04"' 0.07"' 0.06"' 0 . 0 5 * * * 0.05"' 0 . 0 4 * * * - 3 . 0 6 -5.16 -4.54 - 4 . 6 9 - 4 . 8 1 -4.14 constant 5.22"' 5 . 2 5 * * * 5 . 2 7 * * * 4.63"' 4.71"' 4.62"' -85.72 - 8 6 . 5 -108.92 - 7 8 . 5 6 -77.16 -71.64 Observations 722 0 3173 2486 7440 3686 2 950 R-squared 0 . 5 0 . 3 0 . 2 9 0.5 0.34 0.33 Robust t s t a t i s t i c s i n parentheses s i g n i f i c a n t a t 10%; **s i g n i f i c a n t a t 5%; * * * s i g n i f i c a n t a t 1% 70 Table AF. 2: Entity-level Models Stepwise Regression Results by Entity (Dependent variable: Natural log of per-adult-equivalent consumption) FBH RS Total B I H Household Characteristics Number ofhouseholdmembers -0.27*** -0.26*** -0.27*** [0.01] [0.01] [0.01] Number of children under 14 0.05*** 0.05*** 0.05*** [0.01] [0.02] [0.01] Number ofchildren, 14-24 0.06*** 0.05*** 0.05*** [O.Ol] [0.02] [O.Ol] Headofhousehold:female 0.10*** 0.08*** c0.021 [0.02] Headofhousehold:haspostgradeducation 0.17*** 0.12*** 0.14*** [0.03] [0.04] [0.02] Headofhousehold:employed -0.07*** -0.04* ** c0.021 [0.02] Number of employedmembers 0.09*** 0.07*** 0.08*** [0.01] [0.01] [0.01] Housing Characteristics (dummy variables) Dwellinghas sanitary connection 0.22*** 0.11*** [0.05] [0.03] Dwellinghas central heating -0.14*** -0.08* ** [0.03] [0.03] Dwellinguses single equipmentheating -0.21*** -0.08*** [0.04] [0.02] Dwellinghas a heatingsource 0.10*** -0.09*** [0.02] [0.03] Indoor toilet andbathroom O s l o * * * 0.09*** [0.02] [0.02] Dwellinghas a garage 0.03*** [0.01] Dwellinghas a balcony 0.09*** 0.06*** 0.06*** [0.02] [0.02] [0.01] Multifamily residentialbuilding -0.08* * * [0.03] Age ofthe dwelling o.oo*** o.oo*** [O.OO] [O.OO] Secondaryhome 0.20*** 0.28*** 0.22* ** [0.03] [0.07] [0.03] Ownership of Durables (dummy variables) Phone 0.09*** 0.13*** 0.10*** [0.02] [0.02] [0.01] Washer 0.15*** 0.17*** 0.12*** 71 [0.03] [0.03] [0.02] Vacuum cleaner 0.07*** [0.02] Satellite dish 0.05*** [0.02] Sewing machine 0.09*** 0.07*** [0.02] [0.01] Computer 0.10*** 0.06*** [0.03] [0.02] Car 0.26*** 0.28*** 0.25*** [0.02] [0.02] [0.01] Electric and gas cookers 0.15*** 0.09*** [0.03] [0.02] Firewood and coal stove -0.08*** -0.07*** [0.03] [0.02] Dishwasher 0.20*** 0.17*** 0.18*** [0.02] [0.04] [0.02] Video recorder 0.09*** 0.07*** [0.02] [0.01] HI-FIsystems 0.07*** [0.02] Mobile phone 0.09*** 0.13*** 0.11*** [0.02] [0.02] [0.02] Affordability of Selected Expenditures Log of utility expenses 0.20*** 0.16*** 0.18*** [0.01] [0.01] [0.01] Income Source (dummy variable) Receives pension income 0.06*** 0.04*** [0.02] [O.Ol] Constant 5.09*** 5.55*** 5.28*** [0.07] [0.08] [0.05] Observations 4491 2602 7435 R-squared 0.52 0.5 0.51 Robust standard errors inbrackets *Significant at 10%; *** ** Significant at 5%; Significant at 1% 72 Table AF. 3: PMT Regressionby Urban and Rural Areas PMT PMT DependentVariable Urban Rural Number of householdmembers -0.20*** -0.15*** [0.01] [0.01] Numberof children under 14 0.11*** 0.06*** [0.01] [0.01] Numberof children, 14-24 Numberof elderly 65+ Head of Household:female Head of Household:has postgrad education 0.15*** 0.14*** [0.03] [0.05] Head of Household:employed Head of Household:unemployed Numberof employed members 0.09*** 0.06*** [0.01] [0.01] Employedspouse Dwelling has sanitary connection 0.1o*** [0.03] Dwellinghas central heating -0.09*** [0.03] Dwellinguses self-provided heating Dwellinguses single equipment heating -0.1o*** [0.03] Dwellinghas hot water Dwellinghas a heating source 0.1o*** [0.02] Dwellinghas a kitchen Dwelling has a boiler Indoortoilet and bathroom 0.09*** [0.02] Dwellinghas a garage Dwelling has an attic Dwellinghas a balcony 0.08*** 0.06*** [0.02] [0.02] Dwelling has a garden Multifamilyresidential building Age of the dwelling Primary home Secondary home 0.18*** 0.32*** [0.03] [0.05] Dwelling is rented Refrigerator Phone 0.11*** 0.12*** [0.02] [0.02] Washer 0.17*** 0.14*** [0.04] [0.02] Vacuum cleaner 0.09*** [0.02] 73 Satelite dish Sewing machine 0.09*** 0.05*** [0.02] [0.02] Computer 0.11*** 0.07*** [0.02] [0.02] Car 0.25*** 0.29*** [0.02] [0.02] Electric & gas cookers 0.09*** [0.02] Firewood 8, coal stove Dish washer 0.21*** 0.16*** [0.03] [0.03] Video recorder HI-FI systems 0.23*** 0.11*** [0.06] [0.04] Mobile phone 0.17*** 0.17*** [0.02] [0.02] FBiH 0.06*** [0.02] Republika Srpska Rural area (dummy) Log of utility expenses 0.21*** 0.19*** [0.01] [0.01] Receives pension income 0.05*** [0.01] Constant 5.39*** 5.25*** [0.08] [0.05] Observations 3096 4339 R-squared 0.458 0.46 Robust standard errors in brackets *** p