Missing Public Funds and Targeting: Evidence from an Anti-Poverty Transfer Program in Indonesia November 24, 2011 Daniel Suryadarma, ANU and Chikako Yamauchi, ANU and GRIPS
Introduction Loss of public resources due to corruption or mismanagement can hinder targeting performance & public spending efficacy Only a fraction of public funds are spent on the intended purposes Direct measure of disbursement and receipt Reinikka and Svensson (2004) [80%] and Olken (2006) [18%]
Issue 1: Local government A local government in poor areas plays an important role in distribution Increasingly popular usage of communitybased distribution scheme Involve beneficiaries in local resource management Shift in the attitude towards collective action (e.g., Ostrom) Local capture, incompetent local government Relatively little is known about local characteristics - particularly those of local government - associated with more successful delivery of public resources
Issue 2: Targeting problem Missing public funds are often not accounted for, when targeting performance is studied Targeting problem = how much public resources are received by intended beneficiaries (the poor)? Measure = [funds received by the poor] / [total funds received by all] Total (denominator) might be already reduced by corruption or mismanagement
This Paper Quantify the amount of missing funds in Indonesia s Inpres Desa Tertinggal (IDT) Examine targeting performance with and without consideration of missing funds Investigates pre-existing local conditions associated with more successful distribution of public funds
Overview of the Findings On average 70 percent of public funds reached intended beneficiaries Without taking this into account, conventional targeting measure suggests pro-poor distribution, while with the loss, it implies slightly pro-rich distribution Districts which initially had many organized village governments exhibited high receipt
Contribution 1 First evidence linking the literature on targeting and the literature on public resource delivery
Contribution 2 First evidence empirically showing association between pre-existing local government s capability and the efficiency of public resource delivery 3 rd evidence using objective measure Previous studies used subjective corruption indicators Perception may underestimate the true level of corruption (Olken, 09) Rich explanatory variables Pre-existing indicators for village governance Panel outcome variable
Confirm the importance of lost public resources to explain that: Effect of public spending on growth and other outcomes is insignificant (Landau (1986), Filmer and Prichett (1999) Countries perceived to be corrupt show worse outcomes (Barrow (1991), Mauro (1995), Azfar and Gurgur (2007)) Effect of public spending depends on perceived level of corruption (Rajkumar and Swaroop (2008) and Suryadarma (2008)) Other explanations Contribution 3 limited redistributive capacity (La Porta et al. (1999) ) differential budget allocation (Mauro (1998) and Gupta, et al. (2001))
Background
Indonesia Indonesia 2008 Corruption Perception Index (Transparency International) 126th out of 180 countries tie with Eritrea, Ethiopia, Honduras, Uganda..
IDT: grants to poor villages Inpres Desa Tertinggal (IDT, 1994-97) Presidential Instruction for Left-Behind Villages Indonesian Government Grant of 20 million rupiah A fund for business loans Treasurers of groups of eligible households in Poor villages (20,000) Non-poor villages (40,000)
Selection of eligible households National guideline Select poor households (eligible for IDT loans) in each selected village A village head and a local government agency (LKMD, Village Community Resilience Board) facilitate the selection We explore characteristics of these local institutions Eligible households are formed into community groups pokmas
Central Government Province Government District Government Flow of money Subdistrict Government [2] approval certificate Flow of documents Flow of money [1] proposal Village head Financial institution [3] [4] receipt Treasurers of pokmas
Take-up rate was high Province 1994 1995 Aceh 100.00 92.40 Sumatera Utara 100.00 100.00 Sumatera Barat 100.00 100.00 Riau 100.00 100.00 Jambi 100.00 100.00 Sumatera Selatan 100.00 99.10 Bengkulu 100.00 100.00 Lampung 100.00 100.00 DKI Jakarta 100.00 100.00 Jawa Barat 100.00 100.00 Jawa Tengah 100.00 100.00 Yogyakarta 100.00 100.00 Jawa Timur 100.00 100.00 Kalimantan Selatan 100.00 100.00 Kalimantan Barat 100.00 100.00 Kalimantan Tengah 100.00 100.00 Kalimantan Timur 100.00 100.00 Sulawesi Selatan 100.00 92.42 Sulawesi Utara 100.00 92.47 Sulawesi Tengah 100.00 100.00 Sulawesi Tenggara 100.00 100.00 Bali 100.00 100.00 NTB 100.00 100.00 NTT 100.00 100.00 Timor Timur 100.00 99.73 Maluku 100.00 100.00 Irian Jaya 99.71 71.38 Average 99.99 98.06
Scope Scope 41% of communities were funded at least once Total grant value per village = Rp.45 million 90 percentile annual PCE = Rp. 999 thousand 28% of households received a loan Average yearly loan size (Rp.124 thousand) = 27% of recipients annual PCE (Rp.462 thousand)
Village-level corruption Major allocation decisions were made at the village level The share of receipt measures a loss which occurred at the village level between financial institutions and households in designated communities treasurers and village officials can be involved compulsory saving or fees, collective purchase
Data & Amount of Missing Funds
SUSENAS 1993 & 1997 Household socio economic survey Repeated cross-section representative at the district level IDT data indicate which villages received IDT between 1994 and 1996 PODES 1993 Census of villages Data Sources
% entitled funds actually received % entitled funds actually received = Value of funds received/ Value of funds disbursed Value of funds disbursed to each district = 20 million rupiah * number of villages funded Value of funds received in each district = Total value of loans reported to have been received by households in IDT villages Measure of resource delivery Any households in IDT villages are regarded as beneficiaries Targeting within IDT villages is evaluated in the next section Households creditworthiness is not taken into account
More about value of funds received Has anyone in the household been a member of pokmas? Has anyone received a loan from IDT? How much? What year? E.g.1: a rich household without a pokmas member receives funds -> unlikely reported in SUSENAS E.g.2: a rich household with a pokmas member receives funds -> likely reported
Estimation of value of receipt Sampling weight for households in IDT villages = reciprocal of the probability for a household of being selected for interview given being in IDT village = Total number of households in the 1993 PODES / Number of sampled households in the 1997 SUSENAS Poorer districts contain a larger number of sampled households Weighted regressions are used with the number of sampled households as a regression weight
0.05 Fraction.1.15 Overall share of receipt Mean = 0.71 (N = 263) 0 1 2 3 4 5 6 7 8 9 10 Overall share of funds received as IDT benefits (1994-96)
0 Fraction.02.04.06 Log of overall share of receipt * 100 0 2 4 6 8 Log of overall share of funds received as IDT benefits * 100 (1994-96)
Targeting Performance
Targeting performance measure Predict per capita household expenditure PCE hat 97 = β hat 9394*X 97 Rank households in a district Define the poor as the bottom 20% Measure the share of funds received by the bottom quintile If 20% -> universal allocation If > 20% -> better targeting than universal allocation
Targeting w. & w/out missing funds Average share of overall IDT funds received by each district s poorest 20% households (N=251) (1) Baseline (2) Scenario 1 (3) Scenario 2 Σ i=poor r i / Σ i=all r i [Σ i=poor r i + 0.2*(D-Σ i=all r i )] / D Σ i=poor r i / D 27.4% 24.9% 20.1% p-value=.107 p-value=.000 r i = value received by household i D = value of disbursement Significantly worse than (1), the mean indicates worse than universal distribution
Correlates of Receipt: Conceptual Framework
Factors Related to Public Resource Delivery General level of political awareness Living standard Median PCE Village residents education & information exposure Average number of years of education among adults aged 20-60 Share of adults aged 20-60 who read newspaper last week Share of adults aged 20-60 who listened a radio last week Village government s capability LKMD s self-reported organization skills Share of villages with female heads / educated heads Average age / tenure of village heads
Factors Related to Public Resource Delivery Inequality in political awareness / bargaining power Within village coefficient of variation in PCE Homogeneity in citizenship (Hafindahl index) Demand for credit Average village population (number of households) Shares of villages which had a bank, other public credit programs, good inter-village road Share of villages which were funded for infrastructure Geographical characteristics Average distance from district capital Density (population per hectare) Share of urban communities
What s organized LKMD? Village Community Resilience Board Lembaga Ketahanan Masyarakat Desa (LKMD) National institution in charge of the implementation of national programs at the village level members = local residents, appointed by the village head Less organized (1) does not exist (2) only exists in very basic form Organized (3) exists and is able to develop and conduct work projects utilizing grants from the national government matched with contributions of community members (4) exists and forms village development plans, keeps reports in order, and has well-functioning sections
Correlates of Receipt: Empirical Strategy
Correlates of overall share of received funds: island-level fixed effects model Y ij = α 1 + β 1 X ij + γ 1 D j + u ij i: district j: island X ij : Pre-determined covariates D j : island fixed effects u ij : error term, assumed to be independent across districts Weighted by the original sample size Tobit and linear models are used
Islands? I Sumatera, Java, Sulawesi, Kalimantan, the group of Bali and Nusa Tenggara islands, and the group of Eastern islands
Changes in correlates of recept: island fixed effects model Y ijt = α 2 + α 96 2 D 96 t + α 97 2 D 97 t + β 2 X ij + β 96 2 [X ij *D 96 t ] + β 97 2 [X ij *D 97 t ] + D j + γ 96 2 [D j *D 96 t ] + γ 97 2 [D j *D 97 t ] + u ijt t: year ( t = 1995-1997) D 96 t = 1 if year = 1996 D 97 t = 1 if year = 1997 Weighted by the original sample size Tobit and linear models are used
Changes in correlates of receipt: district-level fixed effects model Y ijt = α 2 + α 2 96 D t 96 + α 2 97 D t 97 + β 2 96 [X ij *D t 96 ] + β 2 97 [X ij *D t 97 ] + γ 2 96 [D j *D t 96 ] + γ 2 97 [D j *D t 97 ] + µ i + u ijt t: year ( t = 1995-1997) D t 96 = 1 if year = 1996 D t 97 = 1 if year = 1997 µ i : district-level fixed effects Weighted by the original sample size Linear model only
Correlates of Receipt: Results
Correlates of overall receipt Outcome = Overall share of funds received by households (1) (2) in designated communities Tobit Island FE Log of median PCE (in thousand Rp) -0.378-0.360 Average coef of var within village in household PCE [standardized] 0.056 0.050 Average year of education among adults aged 20-60 0.081* 0.078 Log of average number of households -0.066-0.041 Log of average density (population per hectare) -0.049-0.061 Share of urban communities -0.960** -0.695* Log of average distance from the district capital -0.220*** -0.208*** Share of villages with organized government 0.549** 0.505** Average age of village heads -0.015-0.012 Average tenure of village heads 0.039 0.029 Share of villages with female heads 1.186 0.408 Share of villages with heads who attained high school or above -0.630* -0.540 Share of villages funded under infrastructure programs -1.555** -1.527** Number of districts 263 263 Number of districts censored at zero 15 15 Log likelihood -289.41-252.62 F-stat 2.25 1SD increase (0.3) in the share of organized villages: 15 ppt increase (21 % of the mean) Significant level: + = 10%, * = 5%, ** = 1%.
Changes in correlates of receipt Outcome = Overall share of funds received by households (3) (4) (5) (6) (7) in designated communities Island FE District FE 94 benchmark Change in coeff Change in coeff 94 & 95 94 & 96 94 & 95 94 & 96 Log of median PCE (in thousand Rp) -0.392 0.334 0.137 0.596 0.343 Ave coef of var within village in household PCE [standardized] -0.050 0.050 0.146 0.033 0.092 Average year of education among adults aged 20-60 0.132** -0.119** -0.095-0.025-0.056 Log of average number of households 0.062-0.131-0.382** 0.022-0.189 Log of average density (population per hectare) -0.063 0.019 0.001 0.022 0.033 Share of urban communities -0.547 0.653* -0.178 0.493-0.151 Log of average distance from the district capital -0.041 0.019-0.275 0.046-0.237 Share of villages with organized government 0.160-0.082 0.873** -0.137 0.802** Average age of village heads -0.029 0.006 0.035-0.012 0.028 Averate tenure of village heads 0.077-0.033-0.097-0.038-0.101 Share of villages with female heads 2.030-0.206-4.366-0.310-4.798 Share of villages with heads who attained high school or above -0.264-0.041-0.331-0.149-0.209 Share of villages funded under infrastructure programs -0.967 0.340-1.285 1.375-0.075 Number of districts 771 771 Number of districts censored at zero 85 85 Log likelihood -947.24-590.31 F-stat 1.76 1.04 Significant level: + = 10%, * = 5%, ** = 1%.
What to Take Away?
Summary of Findings Conclusions Only 71 percent of public funds reached intended beneficiaries Once missing funds are taken into account, it is revealed that IDT s targeting performance was poorer Share of receipt was 15 ppt (21%) higher in districts with higher share of villages with organized governments Higher record keeping & organizational capabilities Not systematically correlated with other village government characteristics (heads education, gender, and tenure)
Summary of Findings Implications It is important to take into account leakage in evaluating targeting performance Imply that training / monitoring of local government officers might limit the disappearance of public funds and improve public spending efficacy Cannot distinguish dishonesty and incompetence, but Olken (2007) suggests dishonesty is not the only factor Evaluating impact of different types of training programs would be fruitful future research
Accuracy of responses Responses were not used by the central government for monitoring Only the list of participants Responses were not used by the central government to decide the funding status of the village in coming years A separate village census was conducted for the purpose Repayment obligation was likely to be perceived very weakly Repayment rate = 19 percent
Treatment of loans from different sources Treatment in the upper bound Treatment in the lower bound Data source = 1997 SUSENAS 1994 1995 1996 Direct 85.33 73.88 73.21 As it is As it is Rotated 13.77 23.86 23.64 0 0 DK 0.81 2.12 2.20 As it is 0 Direct & Rotated 0.08 0.13 0.95 Divided by two Divided by three Direct & DK 0.00 0.02 0.00 Divided by two Divided by three Rotate & DK 0.00 0.00 0.00 n.a. n.a.
Aligning reference periods Value of receipt refers to a calendar year Value of disbursement refers to a fiscal year 1994 1995 1996 II III IV I II III IV I II III IV 1994/95 receipt = total 1994 receipt + ¼ * 1995 receipt 1995/96 receipt = ¾ * 1994 receipt + ¼ * 1996 receipt 1996/97 receipt = total 1996 receipt
Adjust for lower take up rates Reduced the value of disbursement for provinces where take-up rates were lower than 100 percent
Data Issue: Excess Receipt Differences in observables between villages with R<=1 and R>1 Variable Mean Std Dev Coef {R>1} Citizenship fragmentation 0.670 0.022 0.004 Religion fragmentation 0.561 0.116-0.001 Share of villages with bank 0.217 0.250 0.011 Share of villages with previous credit program 0.296 0.265-0.002 Share of villages with year-round roads 0.844 0.217 0.003 Average village density (population / ha) 4.876 9.366 0.946 Share of villages with advanced LKMD 0.767 0.266-0.010 Median per capita expenditure (PCE) (Rp.1000) 34.986 12.780 0.782 Inequality: 90/10 PCE Ratio 3.373 0.810-0.084