更强的针对性与减少开支之间是否一致？

Is More Targeting Consistent with Less Spending? WPS-2079 Martin Ravallioni March 1999 ko1r/z?dAawk, lc9cYHS&ee/1 W/J ff-ashi,zg.ofSC, 2&1133, US,4 23 Deceln/her 19-98& revsed 16-Feh1ra1y 1999 Abstract Economists often advise governments to target their spending better when cuts are called for. This paper asks whether that advice is consistent with a political-economy constraint that limits the welfare losses to the non-poor from spending cuts. A simple theoretical model shows that the answer is unclear on a priori grounds, and so will depend on the specifics of program design and financing. A case study for a World Bank-supported social program in Argentina illustrates how cuts can come with worse targeting performance; the allocation to the poor falls faster than that to the non-poor. Some lessons are drawn for how the poor might be better protected from cuts. Keywords: Targeting, poverty, political-economy, Argentina JEL classifications: H53, 138 I This paper draws on the excellent information system of the Trabajar II Project Office in the Ministry of Labor of the Government of Argentina. The financial support of the World Bank's Research Committee (under RPO 681-39) is gratefully acknowledged. Some preliminary results from this work were presented at the 1998 Congress of the International Institute of Public Finance held in Cordoba, Argentina, and the author is grateful to participants for their comments. Bill Easterly, Polly Jones, Martin Rama, Dominique van de Walle and the journal's referee provided useful comments. These are the views of the authors, and should not be attributed to the World Bank. FILE COPY  1. Introduction The impact on the poor of cuts in public spending has been of widespread concern. It is often recommended that cuts should be combined with better targeting, so the poor do not suffer. If there is a broad political constituency in support of protecting the poor from cuts then the task should not be too difficult. But is it possible to target more, while spending less, when the political support of the non-poor is crucial, and cannot be counted on? It is common to find benefit "leakage" to the non-poor from programs ostensibly targeted to the poor. This suggests scope for finer targeting. However, a degree of leakage may well be crucial to the political sustainability of social programs (Gelbach and Pritchett, 1997; de Donder and Hindriks, 1998). By the same token, the feasibility of combining cuts with better targeting is unclear on a priori grounds given that the non-poor may also have the power to protect themselves from the welfare burden of spending cuts. While there have certainly been cases in which cuts were combined with better targeting, possibly these were all situations in which the political economy constraint was not binding. Nor does the (limited) evidence available support a presumption that targeting tends to worsen with program expansion, or improve with cuts. For example, comparisons of the regional differences in participation rates in various social programs in India suggest that it is the non-poor who capture the early benefits, with larger marginal gains to the poor only emerging later on (Lanjouw and Ravallion, 1999). The benefit-incidence studies I know of that have tracked changes in incidence over time have found improvements in targeting with program expansion (Hammer et al., 1995; van de Walle, 1995). To investigate this issue, the paper begins with a simple model of the political-economy of spending cuts (section 2). The model identifies conditions under which the poor will lose from 2  cuts that are constrained to have limited welfare impacts on the non-poor. After describing how targeting performance is to be measured (section 3), the paper presents evidence on how cuts to a social program in Argentina affected its performance in targeting the poor (section 4). The concluding section summarizes the results, and discusses possible implications for efforts to protect the poor from public spending retrenchment. 2. Targeting and Spending Cuts The model is as follows. There is initially some allocation of program spending between given numbers of "poor" and "non-poor" people. Spending on each poor person is O 7 and it is Ga for the non-poor. Aggregate program spending per capita is: 6= GVH+ GZ'(I - IV) (1) where His the proportion of the population that is poor. The allocation a", Gryields a benefit to each non-poor household of 3(6, 6G). The Bfunction is smoothly increasing in both arguments with first derivatives 8,,, which I call the "direct marginal benefit", and ,, the "indirect marginal benefit"; the latter allows for interdependence, in that the non-poor can gain (indirectly) from spending on the poor. The cost of the program to each non-poor person is C(6). (This includes the taxes or fees for financing the program plus efficiency losses from raising the revenue.) The function Cis assumed to be smoothly increasing. So the net gain to the non-poor from the program is 3(6', Gl) - C((6). However, C7is assumed to be an adequate measure of the gain to the poor. The allocation is assumed to be efficient in that any re-allocation of the budget would hurt one of the two groups. For this to be the case, a reallocation from each poor person to each non-poor person - holding total spending constant - must benefit the non- poor on balance. This implies that g A> 8(1- h'. 3  Total spending is now to be cut, and the non-poor have the power to block any proposal that makes them worse off. So B(G, GC) - C(6) is to be held constant. This constraint, and equation (1), then determine how the aggregate budget cut is distributed between the poor and non-poor. Let GI(6Y) and G(G) be the allocations from the total outlay, 6, consistent with the political-economy constraint. The targeting performance of the program is measured by the absolute difference in the benefits going to the poor versus the non-poor. Let this difference be _7 6) ... 6`0 ( 6) - G" ( 6) (2) which I will call the "targeting differential"; the benchmark of a uniform allocation (with equal outlays per capita to the two groups) is deemed to be "untargeted". Under the political economy constraint, the effect of a change in total spending on the targeting differential is given by: 7C " +1 , Co (3) This has the sign of . + B - C., given that the allocation is efficient. If the marginal cost to the non-poor is zero then targeting will deteriorate with cuts, and improve with higher outlays. This is intuitive; a low marginal cost will mean that the cuts do not entail much saving to the non- poor, who will then require a larger slice of the program benefits to compensate. The above analysis relies heavily on a binding political-economy constraint preventing the cuts from reducing the welfare of the non-poor. Alternatively, one can consider the allocation that maximizes A(G, C) + e0" (for some e? O)at given 6, and ask how this allocation varies with 6C Assuming that A(G6, G') is strictly quasi-concave, there will be interior solutions for Cl and G6 as functions of a However, the effect on the targeting 4  differential is also ambiguous in this case. For example, if B9= G7 + /(6G) (for some strictly increasing concave function h) then the poor are fully protected from cuts ( 60a = 0) and l6. = -1 /(I - H) < 0; by contrast, if B= Z(C) + Cr? then the allocation to the non-poor is fully protected ( 66 = 0 ) and so F. = 1/ H > 0 . The main task of the empirical work will be to estimate the effect of changes in total spending on a measure of targeting performance motivated by the above discussion. 3. Measuring Targeting Performance To identify the effects of program cuts on targeting performance we need a consistent measure of performance observed over various levels of program spending, with suitable controls for other incidental factors influencing performance. I will focus on a central government anti-poverty program for which the job of making spending allocations is decentralized to provincial governments. Each provincial government allocates its budget across the local government areas ("departments") within its boundaries. The measure of targeting performance I will use is the province-specific regression coefficient of the inter-departmental budget allocation on a (pre-determined) measure of poverty by department.2 This regression coefficient can be interpreted as an estimate of the targeting differential defined in the last section. That interpretation rests on an assumption about the behavior of provincial governments. It is assumed that a province's preferred allocation to a household depends on that household's level of poverty, but that it does not depend on the level of poverty in the household's department of residence independently of that. (The household's own poverty may nonetheless depend on where it lives.) The actual amount received by a 5  household can deviate from the province's optimal allocation; but the deviation is independent of the level of poverty in the area of residence. This assumption assures that there is horizontal equity in expectation within a given province, in that a person living in a poor department expects to get the same amount from the program as an equally poor person living in a rich area within the same province.3 To see the implications of this assumption for measuring targeting performance, consider again the model of section 2, which is now interpreted as applying to each of the provincial governments. The central government allocates a total budget of Gper head of across the ff provinces such that