L_OPS - 23 9/ POLICY RESEARCH WORKING PAPER 2391 Are the Poor Protected Time-series data for Argentina suggest that action to support from Budget Cuts? propoor social spending is warranted at times of fiscal Theory and Evidence for Argentina contraction. Social spending in general-and social spending targeted to the Martin Ravallion poor in particular-took a heavy hit at times of fiscal austerity. The World Bank Development Research Group Poverty and Human Resources July 2000 POLICY RESFARCH WORKING PAPER 2391 Summary findings Adjustment programs often emphasize protecting social So fine targeting may be a mixed blessing for the poor, spending-especially propoor spending-from cuts. Yet bringing greater vulnerability to cuts, possibly when help the incidence of fiscal contraction-and hence the case is most needed. There is a strong case for action to for action to protect public spending on the poor at a protect propoor social spending at such times. time of overall fiscal austerity-is an empirical question, An externally financed workfare scheme in Argentina which Ravallion addresses using data from Argentina. was far better targeted than other social spending but still Aggregate budget cuts in Argentina in the 1980s and had to ensure that a small but relatively well-protected 1990s typically brought proportionately greater cuts in share of the benefits went to the nonpoor. social spending. "Nonsocial" spending was protected. The program was clearly subject to the same political But proportionate cuts for types of social spending that economy constraints that influenced the incidence of past matter more to the poor were about the same as the cuts fiscal contractions in Argentina. The program expanded for those that tend to favor the nonpoor. Absolute cuts into poor areas when the budget increased but retreated were in fact greater for "social insurance" that matters from poor areas when the program was cut. It was the more to the nonpoor. program's disbursements to nonpoor areas that were But spending on targeted social assistance and protected. Still, given the low wage rate offered, the employment programs was more vulnerable to aggregate direct benefits from the program were still likely to have spending cuts than were more universal social services. favored the poor, even after the cuts. Social spending was clearly exposed to fiscal contraction, but this was somewhat less true of propoor spending on things that also benefited the nonpoor. This paper-a product of Poverty and Human Resources, Development Research Group-is part of a larger effort in the group to better understand the incidence of social spending. The study was funded by the Bank's Research Support Budget under the research project "Policies for Poor Areas" (RPO 681-39). Copies of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Patricia Sader, room MC4-773, telephone 202-473- 3902, fax 202-522-1153, email address psader@worldbank.org. Policy Research Working Papers are also posted on the Web at www.worldbank.org/research/workingpapers. The author may be contacted at mravallion@worldbank.org. July 2000. (36 pages) The Policy Research Wlorking Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, evenr if the presentations are less than fully polished. The papers carry the namnes of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center Are the Poor Protected from Budget Cuts? Theory and Evidence for Argentina Martin Ravallion World Bank and Universite des Sciences Sociales, Toulouse' Keywords: Fiscal incidence, poverty, social spending, budget cuts JEL classifications: E62, H22, I38 I Address for correspondence: mnravallion@worldbank.org or (until 7/2000): Martin Ravallion, ARQADE, Universite des Sciences Sociales, Manufacture des Tabacs, 21 Allee de Brienne, 31 000 Toulouse, France. The author is grateful to Martin Rama and Dominique van de Walle for helpful comments. 1. Introduction It is now common for macroeconomic adjustment programs to call for a pro-poor shift in the composition of public spending - in combination with an overall fiscal contraction, as usually called for to assure macro stability.2 Donors have been particularly keen to support new public anti-poverty programs and "social funds" that claim to target extra assistance to the poor at times of crisis and adjustment. The case for such action to protect pro-poor social spending rests on the answers to a number of questions: Do cuts tend to fall more heavily on the social services that matter most to the poor? When cuts are borne by the poor, do they gain similarly from expansions? Do add-on "crisis programs" help the poor? What happens when such programs are also cut? It has been noted that existing theory and evidence is woefully inadequate for addressing such questions (see, for example, Agdnor, 1998). One argument that has been made in the literature is that targeting spending to the poor can undermine political support for the taxation needed to finance that spending; the poor might even end up worse off (Gelbach and Pritchett, 1997; De Donder and Hindriks, 1998).3 Broad political support for greater targeting may however be possible when there is an exogenously imposed spending cut, which brings tax savings to the non-poor (Ravallion, 1999a). Even when the poor have no power over how cuts are implemented, it is theoretically possible that they will be protected from cuts without further intervention. The outcome depends on the preferences of those in power, notably the extent to which they gain directly from public spending on the poor, and (less obviously) how quickly the marginal utility of their spending on the poor declines relative to the marginal utility of spending on themselves. Nor is it clear that the poor will be powerless even when they are a minority. They may be 2 See for example the World Bank's recent adjustment loan to Argentina (World Bank, 1998). 3 Also see the discussions in Besley and Kanbur (1993), Sen (1995) and van de Walle (1998). 2 able to form a small but influential special interest group, represented by Non-Governmental Organizations, or they may be able to form a coalition with non-poor sub-groups who see it as in their interests to not have the burden of cuts fall on the poor. This paper first establishes that the incidence of cuts is unclear on theoretical grounds even when the poor are a powerless minority. So it is an empirical question. Good data are now available for addressing the issue in Argentina, which has undergone a number of sharp fiscal contractions over the last two decades. The paper studies two main sources of data. Firstly, section 3 uses aggregate times series data on public spending allocations to see how the composition of spending changes with aggregate contraction and expansion. Secondly, the paper turns to a new data set (constructed for the purpose of this paper) on one of the programs explicitly introduced by the Government of Argentina (with support from the World Bank) to deal with the effects of a macro crisis on the poor. Section 4 describes the program and how its performance in reaching the poor is to be measured, while section 5 tests how the program's performance in reaching poor areas was influenced by spending cuts. The paper's conclusions are then summarized in section 6. 2. Are budget cuts passed onto the "powerless poor"? In settings in which the majority of voters are not poor, one might well argue that the poor will be obliged to bear a disproportionate share of a budget cut on the grounds that they are the least powerful. However, there are some obvious problems with this reasoning. If the poor have little or no power, and power is all that matters to the allocation of public spending, then the same reasoning would suggest that the poor gained little from public spending before the cuts - in which case they can have little to lose from cuts. It might be argued instead that the non-poor care positively about spending on the poor (either through altruism, the existence of negative externalities of poverty, or other spillover effects, such as arising from public good nature of some types of public spending). But then surely the non-poor will want 3 to protect spending on the poor from cuts, and will do so without further intervention. Nor is it clear that the relative power of different socioeconomic groups can be treated as exogenous to public decision making. Spending on things like basic education and income support is arguably one of the ways that the balance of political power might shift in favor of the poor. To explore these issues more formally, let us suppose that total public spending is to be allocated between equal numbers of poor (who receive GP) and the non-poor ((G3f), who finance public spending out of their own income Y". Both G' and GP give direct utility to the non-poor. One can think of G( as social insurance (formal pensions and unemployment compensation), which benefits the non-poor directly and exclusively, while GP is spending on basic social services (education and health) which benefits the poor directly, but also yields an external gain to the non-poor. Utility of the non-poor is U" (Y' - G, G", GP) where G = G' + GP. To keep the model simple, assume that the function U" is additively separable in its three arguments. Let us also make the standard assumption that U" is strictly increasing and concave in after-tax income, Yn - G . Assume also that (at given Y' - G ) the function U" is strictly increasing and non-convex in both G' and GP and strictly concave in at least one of them, i.e., Up > 0, U, < O and U' > 0, Un < 0 (using subscripts for derivatives), but that Up, and U,, are not both zero. Utility of the poor is UP (YP, GP), which is increasing and strictly concave in both arguments and also additive separable. Concavity in GP assures that the poor do in fact prefer less variability in their allocation from the public budget. The allocation of spending maximizes: U'(Y' - G G', GP) + AUP(YP,GP)(1 where A is an exogeneous non-negative number giving the relative power of the poor in decision making over public spending. We can write the solutions in the generic forms: 4 G' = G'(Y', A) and GP =GP(Y',A) (2) (Notice that separability of utility and exogeneity of A together imply that the solution is independent of the income of the poor.) Aggregate spending is: G = G(Y-, A) (3) Using this equation to eliminate yn from (2), we can write down the following equations for how the spending allocation will vary with total spending in equilibrium: G' = F'(G, A) and GP = FP(G, A) (4) The income of the non-poor is subject to shocks. Spending decisions can respond to the state of nature, so maximizing (1) also maximizes expected utility (over a known distribution of possible incomes for the non-poor). However, government spending cannot be saved and nor is (further) borrowing possible in the low income states. So an income decline for the non-poor requires fiscal adjustment to maintain macroeconomic stability. How will the change in spending be allocated between the poor and non-poor? Let us start with the simplest case in which the non-poor have full power over the budget (A = 0). In a neighborhood of the optimal G" and GP, it is readily verified that the derivatives of (4) w.r.t. G are given by: nn rG > 0P 20 (5.1) unn +upIp +U'n Up "'- 0 (5.2) pp with at least one of these inequalities being strict (further details are contained in an Addendum available from the author). These equations tell us the incidence of the spending cuts. It is clear that spending on the poor need not bear the burden of cuts, even though the non-poor have full power over the spending allocation. Indeed, it is evident from (5.2) that if U' is linear in G" (i.e., Unn, = 0 ) then the poor will bear none of the spending cut required to 5 restore equilibrium after an income decline for the non-poor. Nor will they gain anything from "trickle down", here interpretable as higher public spending stemming from higher incomes of the non-poor. Even a very small degree of interdependence could be sufficient for this outcome. The above model is sufficiently general to allow an arbitrarily small positive share of spending going to the poor. By contrast, if U" is linear in GPthen the poor will bear the full amount of the retrenchment, and all of an increment to spending will go to the poor. More generally, the poor will bear absolutely more of the cut in total spending (GP - G' increasing in G) if and only if U' is greater than U,7,.. Whether the poor tend to gain more from expansions than they lose from cuts depends on the curvature of the relationship between spending incidence and total spending. Since the slope of that relationship depends on second derivatives of utility, the curvature will depend on third derivatives. The possibility of an asymmetry in the effects of cuts versus expansions is then even more difficult to predict in this model. These uncertainties about the incidence of spending changes are found in more general models. Suppose now that A is a positive constant. Equation (5.2) is now: TP p = U"n (6) U' +U' +AUP nn pp pp Assuming that Un, Upp and UP are all independent of 2, it is evident from (6) that the right hand side is strictly decreasing in A . The lower the power of the poor, the greater the extent to which the changes in total spending affect the amount going to the poor. However, a more general result (in which U', Un and Up'p are free to vary) is illusive, since it will depend on third derivatives of the utility function. Relaxing the assumption that A is constant, let us assume instead that A = -(GP), which is strictly increasing and concave in GP. Greater spending on the poor raises their 6 power over how public spending is allocated, such as by providing resources needed to lobby govemments for more pro-poor spending. Equation (5.2) becomes: FGP = Mn (7) Un +Up +AUp +lpyUP+22pU( Even assuming again that U' , Un, and UP are unaffected (as would be the case with quadratic utility functions), the sign of (7) is ambiguous.4 Assuming that A(GP) UP (YP, GP) is concave in GPand that Unn, Up and UP are unaffected by the change in the distribution of power, the right hand side of (7) is lower than (5.2); the poor will be more protected than if the non-poor have all the power (2 =0). Intuitively, what the "power effect" does in this case is make the marginal (weighed aggregate) utility of spending on the poor increase more sharply as spending on the poor falls, attenuating the impact of spending changes on the poor. However, this need not hold if ApUp is sufficiently large. When spending on the poor has a large impact on their political power (a high P ), a budget cut will undermine their relative power, further magnifying the impact of the cut. A low marginal utility of spending for the poor will help counteract this effect; and a high marginal utility will magnify it further. One can extend this model in a number of ways. One might expect A to vary with incomes of both the poor and non-poor. The initial income loss of income for the non-poor (that led to retrenchment) will presumably raise the relative power of the poor, helping to mitigate the effect on their power. Of course, if the income loss is shared by the poor then this will weaken their power; the final outcome will depend on how the distribution of income changes, and how this affects the relative power of the poor in decision making. 4 This arises from the well known fact that the product of two positive, increasing, and concave functions need not be concave; indeed, A(Gp) UP (YP, GP) could be convex in GP even though A(GP) and UP(YP,GP) are both concave in GU'. 7 These arguments have clearly not exhausted the theoretical possibilities. For example, I have said nothing about the possibility of the poor forming a coalition with a subset of the non-poor so as to protect the poor from cuts. However, the above discussion provides adequate warning against presuming that the poor will necessarily bear the brunt of cuts in situations in which the non-poor hold the balance of power over the budget. The need for specific actions to protect the poor must then be deemed an empirical question. The rest of this paper addresses that question using data for Argentina. Evidence will be drawn from very different sources. In the next section, time series evidence on aggregate budget allocations across types of spending will be studied. After that, the paper turns to evidence for just one anti-poverty program. Although the data will be very different, in both the aim will be to examine how the composition of spending varies with total spending i.e., to estimate empirical models motivated by equation (4). 3. Social spending in Argentina during fiscal expansions and contractions While methods of measurement differ, it is widely agreed that in the mid-1990s less than 30% of the population is poor by Argentinean standards (World Bank, 1999). The level of "social spending" averaged 56% of total government spending in the period 1980-97 (Government of Argentina, 1999). Less than half of this went to "social services" (education, health, water and sewerage, housing and urban development, social assistance, and labor programs); the remainder can be labeled "social insurance" (pensions, public health insurance, unemployment insurance). Spending on social services is believed to be pro-poor, in that the poorest x% of households receive more than x% of spending, but this is not so for social insurance (Gasparini, 1999; Llach and Montoya, 1999). Table 1 reproduces recent estimates of the incidence of public spending on social insurance as distinct from the social services. The results confirm that social service spending is more pro poor than social 8 insurance. Access to social insurance in Argentina typically requires that one has a job in the formal segment of the labor market, which is less than half of the workforce, and relatively few of the poor (World Bank, 1999). There is evidence that spending on social services has responded more to changes in national income than has social insurance. Wodon and Hicks (1999) study the effects of changes in GDP on targeted spending on social services (about 4% of total government spending). They find that the ratio of targeted public spending to the number of poor had a positive elasticity (of about three) to GDP; in recessions, there were more poor people, and less was spent on them. The political regime is also likely to matter. For a sample of Latin American countries (including Argentina), Brown and Hunter (1999) find that democracies are more likely to protect social spending in a recession, but that authoritarian regimes are more inclined to expand social spending when the crisis is over.5 Argentina has been a democracy since 1982. The data series I will use start in 1980, and so it is reasonable to ignore the change in regime. In this setting I examine how changes in the government's total budget affected the level and composition of social spending. Did budget cuts have similar effects to budget expansions? Were the categories of spending that are known to matter to the poor more protected than other types of spending? Figure 1 offers a direct test for whether social spending has been protected from budget cuts. The figure plots the time series of changes in the log of total public spending ("Gasto Puiblico Consolidado") and the log of social spending (both in 1997 prices) as compiled by Government of Argentina (1999) and covering the period 1980-97.6 Unlike 5 This study does not, however, control for the level of total spending, so it is unclear whether the identified effects operate through the composition of spending or its aggregate level; "non-social spending may well behave the same way as social spending. 6 Both time series are highly serially correlated; indeed, augmented Dickey-Fuller tests do not reject the unit root hypothesis for either variable at even the 10% levels. So the following analysis will focus on changes from year to year rather than levels. 9 most other compilations of public spending data, this one includes all levels of govermnent. In this and other respects, considerable care appears to have gone into constructing the data. There is clearly little sign that social spending was protected from cuts. Indeed, there is sizable co-movement, with indications that (if anything) social spending was more volatile than other types of spending. One can see quite large proportionate declines in social spending in every year in which total spending falls. On the other hand, one often sees smaller (and sometimes negative) changes in social spending when total spending rises. Letting Gs denote social spending at date t, and Gt total spending, one can test whether the elasticity to an increase in total spending differs from that for a decrease using the regression: AlnGs = a + [yl,5, + Y2(1- 3,)]A ln G, + et (8) where a is a time trend, yr is the elasticity when total spending increases, y2 is the elasticity when it falls, and et _ I(AInG G) takes the value unity when AlnG, > 0 and zero otherwise. The data in Figure 1 yield an estimate of 0.14 for y1 . This is not significantly different from zero (the standard error is 0.37). On the other hand, the estimated elasticity to a decrease in total spending (r2) is 2.14, which is significantly greater than one (the standard error is 0.26). Social spending responds elastically to aggregate cuts, but the responsive to fiscal expansions is not statistically significant.7 The constant term of 0.086, which is significant (t-ratio=2.83), indicating a sizable independent trend increase in the share of social spending. By contrast, non-social spending was well protected; the elasticity when total spending fell was 0.09, and not significantly different from zero (a standard error of 0.22); on the other hand, the elasticity to an increase was 1.68 (standard error of 0.41). 7 The average elasticity (constraining y, and Y2 to be equal) is 1.366, with a standard error of 0.21; however, the restriction that ri =Y2 performs poorly (t=3.16). 10 While these elasticities are of descriptive interest, their causal interpretation requires that we believe that changes in total spending are uncorrelated with the error term e* in (8). To test for the causal effect on social spending, I assume that the elasticity to higher spending is in fact zero (on the grounds that the OLS results are so strong that it is difficult to believe they are not robust in this respect). Under this assumption, I use lagged values of both social spending and other spending as instruments for cuts in total spending. The resulting 2SLS estimate for the elasticity of social spending to a cut in the total budget is 2.28 with a standard error of 0.27. Again, not only is social spending not protected, its elasticity to cuts exceeds one, implying a fall in the share of social spending during fiscal contractions. This protection of "non-social" spending does not however mean that the non-poor shift the cuts to the "powerless poor". As was noted in section 3.1, social spending in Argentina includes types of spending that matter more to the non-poor than the poor, such as social insurance. Also, there may be pro-poor changes in the composition of social spending, dampening the marginal impact on the poor.8 To see how the composition of social spending changes with cuts, Table 2 reports estimates of equation (8) for various categories of social spending. The same pattern is evident in almost all spending components; social spending responds elastically to cuts in the total budget, but does not respond significantly to budget increases. The only exception is housing and urban spending, which is not significantly different for an increase in total spending versus a decrease, and is not significantly different from zero. The elasticity to budget cuts is very similar for social services as social insurance.9 The table also gives the 2SLS estimate of the elasticity to budget cuts. The most notable 8 There is evidence (for India) that spending composition is not homogeneous in the level of spending (Lanjouw and Ravallion, 1999). 9 The difference between social insurance and social services in the elasticities is not significant (t-0.3 0); nor can one reject the null that the parameters are jointly the same (F=0.28). 11 difference is that spending on employment programs becomes highly elastic to cuts; these include the Trabajar Programs we will study in depth in the next section. This unifonnity in the elasticities to budget cuts between social insurance and social services is inconsistent with the idea that cuts will simply be passed onto the categories of spending that matter most to the poor. As we have seen from Table 1, middle (and upper) income groups are likely to benefit relatively more from pensions and (formal sector) unemployment compensation than the poorest quintile. The non-poor might also be expected to resist cuts to these categories of spending given that formal social insurance spending is heavily pre-committed, and hence harder to cut. Yet we find that the proportionate cuts are just as great for social insurance as social services; the absolute cuts are in fact higher for social insurance, given that it accounts for a higher share of the budget. A possible explanation for this result is that the benefits to the non-poor from social services are tied to consumption by the poor. It is hard to cut spending on schools without also hurting the non-poor. Inability to finely target many social services thus helps protect the poor from differentially higher cuts, even though the non-poor benefit proportionately less from this type of spending than from social insurance. This begs the question: did more targeted categories of social services receive heavier cuts? Government of Argentina (1999) provides a classification of social service spending according to whether it is "targeted" or "universal". The targeted programs are housing and urban programs, social assistance and employment programs; on average, these account for 17.7% of spending on social services. Table 2 also gives the estimnates of yl and 72 classified this way. The elasticity to total spending cuts is not any higher for the targeted components of social services; indeed, if anything, the elasticity is higher for universal social services. However, this conclusion is sensitive to the classification of "targeted" spending. Table 2 also gives separate estimates for two of the components of targeted social service 12 spending, namely social assistance and employment. Both have high elasticities to a fall in total spending. If one excludes housing and urban from the targeted component we also find a relatively high elasticity (Table 2). So these results do offer some support for the conclusion that targeted social spending is more vulnerable to fiscal contraction. 4. Tracking expansion and contraction in an anti-poverty program The above results do not suggest that social spending provided a good safety net for Argentina's poor at times of fiscal adjustment, given how exposed that spending was to aggregate cuts, and how little of it went to the poor at normal times. This provides a strong motivation for looking to alternative programs that might better reach the poor in a crisis. We shall now study one such program in depth. The program was picked because of the unusually rich data available, and the fact that these data cover a period in which the program both expanded and contracted. We will examine how well the program performed in reaching the poor in a crisis, and see how its performance changed with both aggregate expansion and contraction, exploiting the fact that this happened differently in different provinces. 4.1 The Trabajar Programs The Government of Argentina introduced the Trabajar Program in 1996, in the wake of a sharp rise in unemployment, and evidence that this was hurting the poor more than others. In May 1997 the unemployment rate for the poorest decile of households (ranked by household income per capita) in Greater Buenos Aires was 40% versus 17% on average. The Trabajar Programs also followed a period of declining social spending (Figure 1). The program's aim was to reduce poverty by providing relatively low wage work on community projects in poor areas. The central government pays for the wage cost, and local or provincial governments cover the non-wage costs. Within provincial budget allocations, proposals for sub-projects compete for central funding according to a points system. Three 13 versions of the program have been tried since then, Trabajar I, II and III. In terms of design, Trabajar II (TII) and III are more similar to each other than either are to Trabajar I (TI). There were substantial design changes between Trabajar I and II. The inter-provincial allocation of spending was reformed, moving away from a largely political process to an explicit formula based on the estimated number of poor unemployed workers in each province. TII also put greater emphasis on creating assets of value to poor communities. Poverty measures were included in the center's budget allocation rules and in the selection criteria for sub-projects. The poverty focus was also made clearer to provincial administrators. TIII was very similar to its predecessor in design. The main difference was that greater emphasis was placed on the quality of sub-projects, to assure that the assets created were of value to the communities. The World Bank has supported TIl and TIII by loans (disbursed against the wage payments), and through technical support on program design, monitoring and evaluation. All results quoted for Tlll in this paper relate to the first 16 months of its operation, up to November 1999. (TIII is ongoing at the time of writing, and is scheduled to finish in mid 2000.) From the point of view of this paper, an important difference between the three versions of the program is in the level of funding. In Trabajar I, disbursements by the center (covering wages for participating workers) averaged $77 million per annum; for TIl this rose to $160 million per annum, and it then fell to $98 million per annum under TIII. As we will see, there were also differences in levels of funding between sub-periods. Survey-based impact evaluation methods have been used to assess the gains to participating workers and their families from TIl and TIII. Propensity-score matching methods were used to construct a comparison group to surveyed Trabajar participants from an identical national sample survey implemented at the same time. Income gains were then estimated by comparing incomes of the Trabajar participants with the matched comparison group. The results have indicated that Trabajar jobs are well targeted to the poor; Figure 2 14 gives the concentration curve for worker participation in the program. This was estimated by locating the families of a random sample of 3,500 Trabajar workers within the national distribution (based on a sample of 22,000 families).' 0 For example, 76% of people living in the households of participating workers had a household income per capita that placed them amongst the poorest 20% of Argentineans nationally. How does this incidence of income gains compare to other social spending in Argentina? Table 3 gives the concentration curves for the Trabajar program and both aggregate social service spending and social insurance, all on a household basis (to assure that the Trabajar concentration curve is comparable with the numbers in Table 1). Since there is very little variation in the Trabajar wage rate, the concentration curve based on participation is also the benefit incidence curve for gross wage payments." It can be seen from Table 3 that the direct income gains from the program were far better targeted than social insurance and social services as a whole. The program's targeting also appears to be better than any other targeted programs in Argentina. Amongst the programs for which incidence calculations are given in World Bank (1999, Table 3.7), the next best performance was for programs directed at pregnant mothers and children, for which 70% of the benefits went to the poorest quintile of households, which was itself an unusually good performance compared to other programs. Of course, targeting performance is only one factor in assessing the performance of such programs in reducing poverty (Ravallion, 1999b). A program such as Trabajar is designed to help in one dimension of poverty, while programs directed at health and nutrition of the poor help in quite different dimensions; both types of programs can have important 10 Identical surveys were used for the program participants and the national sample, and the surveys were implemented at approximately the same time; for details see Jalan and Ravallion (1999). "1 Again this does not net out foregone income, though nor do the standard benefit incidence calculations in Table I take account of behavioral responses. However, as noted above, factoring in foregone income mainly affects the concentration curve below the 20kh percentile (Jalan and Ravallion, 1999). 15 roles. Nor does targeting performance tell us anything about net income gains. Using propensity score matching methods, Jalan and Ravallion (1999) estimate the net income gains from the Trabajar program, allowing for foregone income. The net income gains to participating workers represented 50% of the gross wage gains on average (Jalan and Ravallion, 1999). (Factoring in foregone income mainly affects the concentration curve below the 20th percentile.) Such calculations relate solely to the benefits from the work provided by the scheme. There are also indirect benefits from the assets created. While non-poor people are unlikely to find the Trabajar wage attractive, they would no doubt like to have the scheme producing things of value in their communities. (There is negligible cost recovery.) How well did the program perform in assuring that the work was provided in poor areas? How did this change when the program expanded and contracted? One can monitor the extent to which the program reached poor areas, by tracking the geographic distribution of disbursements and comparing this to the poverty map of Argentina. By doing so within a period of budget expansion then contraction, and comparing the results across provinces, we will be able to test for budget effects on this aspect of the programs' poor-area targeting performance. The following section outlines the method. 4.2 Assessing poor-area targeting performance Each provincial government's optimal allocation to a household is unobserved, but it depends on the household's level of poverty. That may in turn depend on where the household lives, but I assume that the poverty rate in the area where it lives does not matter to a household's allocation independently of its own level of welfare. In other words, there is no "poor-area bias" in that a poor person living in a poor local-government area expects to get the same amount from the program as an equally poor person living in a rich area of the same province. (The allocations need not be identical, but only equal in expectation; random 16 deviations are allowed.) The same holds for the non-poor. This assumption can be thought of as a form of horizontal equity within provinces (Ravallion, 2000). Let us consider how to measure each province's performance, making this assumption of horizontal equity in expectation within the province. The central government allocates a total budget of G per capita across Mprovinces such that Gj per capita is received by provincej. After that, each province decides how much should go to the poor versus the non- poor. The chosen allocation by provincej is G,' per capita for the non-poor and for the poor. Provincej comprises Mj local government areas, which I call "departments". The per capita allocations to department i (=1,.., Mj) within provincej can be written as: Gn = G; + e,, and GP = G, + (9) where the £s are departmental deviations from province means. Total disbursements to the poor and non-poor must exhaust the budget. This creates an accounting identity linking total program expenditure per capita to the poverty rate in a department. Let GY denote program spending in the i'th department of thej'th province, and let the corresponding poverty rate be H, - the "headcount index", given by the proportion of the population that is poor (for which the overall poverty rate in the province is Hi). Then: GUJ = HUG#P + (1-HHj)G, (10) Using equation (9) we can re-write (10) in the form of a simple linear regression across all departments in provincej: Gjj - Gj = Tj(Hjj - Hj) + v,j. (11) where v. = g6,' + (e,'P -g,l)H,j (12) 17 and Tj = GJP - GJ. is the absolute difference between the average allocation to the poor and that to the non-poor in provincej. If Tj is negative then the program favors the non-poor in absolute terms; if Tj is positive, then the program favors the poor, and the higher the targeting differential, the more provincial spending favors the poor. How can the targeting differential be estimated? Under the horizontal equity assumption, the error tern in (12) has zero mean for any given province and is uncorrelated with H, since the es are zero-mean errors within any given province and are uncorrelated with Hy (and its squared value). Thus H, is exogenous in (11) and so one can estimate TJ from an OLS regression of G, on Hi across all departments within a given province.12 Provincial performance in reaching poor areas can thus be measured by the regression coefficient of spending per capita on the poverty rate, estimated across all departments in each province. Call this the "targeting differential" (TD) for provincej. This is given by: Mj. . (Gei - G,)(Hi - Hj) i=1 (13) Y. (Hu - Hj)2 One can similarly define a national inter-departmental targeting differential, by calculating (13) over all departments nationally (ignoring province boundaries). The targeting differential can be interpreted as a measure of absolute progressivity, namely the difference between per capita spending on the poor and that on the non-poor. A TD of zero indicates that there is no difference in Trabajar spending on the poor versus non- poor. A positive TD means that the program favors poor areas; a negative coefficient means it favors non-poor areas. Poverty is measured by the proportion of the population deemed to have unmet basic needs (IJBN), based on the 1991 census. 1Z Equation (12) indicates that the error term will not be homoskedastic. Standard errors of the targeting differential were corrected for heteroscedasticity. 18 The overall targeting differentials across all 510 departments were $41, $110 and $76 per capita for TI, TII and Tll respectively; all three are significant at the 1% level. To help interpret these numbers, compare the poorest department, namely Figueroa (in Santiago Del Estero province) where the incidence of unmet basic needs is 75.5%, with the least poor department, namely Chacabuco (in Chaco province) where the poverty measure is 3.3%. The expected difference in spending was $30 under TI, $79 under TII, and $55 under TIII. So the expansion to the programn between TI and TII was associated with a more pro- poor allocation of funds geographically, while the contraction between TII and TIII came with a less pro-poor allocation. Next we will see if this aggregate correlation is borne out when we compare provinces over times. 5. Program spending and poor-area targeting across provinces With the extra degrees of freedom made possible by exploiting the changes in the inter-provincial allocation of spending, it is possible to test for statistically significant effects of fiscal expansion and contraction on the program's targeting perfonnance. The better information system for TII and Tlll allows a breakdown of the aggregates into sub-periods by province. Intervals of five months were chosen. A data addendum is available from the author giving the detailed breakdown of the aggregate targeting differentials by these intervals, as well as program spending per capita for each five-month period.13 To assess the effect of the cuts on targeting performance, one can regress the province and period-specific targeting differentials on program spending per capita across provinces, pooling all five-month periods and all provinces. The targeting differential will, however, vary across provinces according to other factors, such as the strength of provincial concern for the poor, how poor the province is as a whole, the history of the provincial efforts at 13 This is an extended version of the data used in Ravallion (1999). The latter paper only used data for TIH. Adding TII more than doubles the number of degrees of freedom in the data. 19 targeting the poor, and the capabilities of local managers. It is not implausible that some or all of these variables will also be correlated with program spending. So their omission will yield a biased estimate of the effect of cuts on targeting performance. However, this problem can be dealt with by treating these differences in provincial targeting performance as provincial fixed effects when estimating the impact of changes in program spending. Given these considerations, the test for the effect of changes in program disbursements on targeting performance takes the form of a regression of the province and date-specific targeting differential on aggregate spending per capita in the province and a set of province-specific effects. The regression is thus: Tj, = a + 3Gj, + NJ + p1j, (=1,..,22; t1,2,3,..) (14) where T is the targeting differential for provincej at date t, Gjt is spehding by provincej at date t, qj is the province-specific effect and p1jt is an innovation error, representing random, idiosyncratic, differences in targeting performance uncorrelated with spending. As discussed above, the aggregate spending allocation Gj, is, however, allowed to be endogenous in that it is correlated with the province effect ij. This regression can be used to estimate a counter-factual targeting differential, which controls for differences over time in program spending. In particular, define the budget- neutral targeting differential (TD ) as the value of TD if program spending did not vary over time within provinces, and was given by the mean spending of TII and TIll. This is identified by simply re-writing equation (10) as: Tjt = (Gjt-Gj)+T* +±1jt (15) Thus Tj is the expected value of the targeting differential for provincej when program spending does not vary over time in that province. By regressing Tjt on spending expressed as a deviation from the overall (five month) mean spending per capita for TII and TIII, and a 20 complete set of province dummy variables, one can then estimate Tj* by the regression coefficient on thej'th dummy variable. For example, the coefficient on the province dummy variable for Cordoba in TII can be interpreted as the estimated targeting differential for that province under Trabajar II if it had its mean budget allocation across TII and TIll. Table 4 gives the results, both for the combined sample and split between TII and TIII. When the regressions for TII and Tlll are combined, allowing all coefficients to differ between TII and TIII, a joint test convincingly rejects the null hypothesis that the budget- neutral TDs are the same for the two programs. I also tested whether the estimated value of j was different when spending increased versus decreased; there was no significant difference (the coefficient on the interaction effect between Gj, - Gj and I(Gjt - Gj), where Iis the indicator function, had a t-ratio of-0.38). There is no difference in the absolute value of the effects of spending cuts versus increases. The regression coefficient of the targeting differential on program spending is 3.13 for the combined samples. So a $10 cut in spending reduced the targeting differential by $3.13 on average. For TII, the regression coefficient of the targeting differential on program spending is 3.55. For TIll, the estimated regression coefficient rises to 10.22. So not only has targeting performance deteriorated in the change from TII to TIII, but the effect of changes in program spending on targeting performance has increased under TIll. The budget neutral TD for TII is positive in 18 of the 22 provinces, and significantly so (at the 5% level or better) in 14 of those; there is one province (Buenos Aires) in which the budget neutral TD is significantly negative. Under TIll, the province effects are now positive in all except one province, and are statistically significant in 18 provinces. There is a high correlation between the budget-neutral TD's (r=0.88). However, it is notable that the budget-neutral TD's are generally higher for TIII. The weaker targeting performance of TIII largely vanishes if one controls for the difference in budget allocation. 21 Indeed, the targeting performance of TIII would generally be better than that of TII if both had the same disbursement rate over time for each province. This comparison of the budget-neutral TDs suggests that other factors were operating to mitigate the effects of lower spending on poor-area targeting performance. The main design change that might have helped is that project size was restricted under TIII; TII allowed projects with up to 100 workers; this was cut to 40 under TIII, and this constraint appears to have been binding. The larger projects are thought to have been more common in less poor cities and towns. Another factor identified in discussions with the central project management is that greater efforts were made (starting during TII, but probably with lagged effects) to train municipalities and NGOs in how to prepare viable project proposals. Discussions with the central program administrators also indicated that targeting performance might have improved with tighter administration and supervision of the program under TIII, and a clearer understanding by all parties involved of the programs' aims. The design changes clearly helped mitigate the effect of higher participation of Buenos Aires under TIll. On balance, Table 4 suggests that overall poor-area targeting would have improved slightly between TII and TIII if not for the cut in spending. The model in section 2 offers some clues as to why we observe a deterioration in targeting performance with cuts, and an improvement with program expansion. A long- standing concern about any program such as Trabajar is that poor municipalities have a harder time raising the cofinancing required for the sub-projects. A provincial government that wants to influence which municipalities participate can readily do so through its ability to propose and confinance projects. In some provinces, it is clear that the provincial government is active in proposing projects in the capital city so as to placate vocal well- organized groups. The workers involved may well be just as poor as those in poorer 22 municipalities outside the capital city. However, to assure maximum impact on poverty it is still preferable for the assets created by the program to be in poor areas. The political economy of the program's operation in most provinces entailed that the cuts were borne heavily by poor areas. The cofmancing requirements allow considerable provincial discretion in the geographic allocation of program spending. Discussions with a number of the provincial project managers and staff have suggested that it was politically difficult in some provinces to assure that the cuts came only from non-poor areas. This reflected (in part) the fact that the program was already favoring poor areas, and so there was little slack for cutting heavily elsewhere while still leaving sufficiently broad participation. Given these pervasive local political-economy constraints, we can begin to understand why lower disbursements resulted in worse performance in reaching poor areas. When a program such as this is cut, there is little obvious saving, via project financing, to non-poor areas. The program has negligible cost-recovery from non-poor areas, even for sub-projects in those areas. Low cost-recovery (at the margin) of program benefits in non-poor areas leaves the poor more exposed to cuts. Also it is not implausible that marginal benefits to the non-poor were quite high; the initially high degree of targeting implied low allocations to non-poor areas and so probably high marginal benefits. The fact that the program provided work to poor neighbors in non-poor areas presumably also entailed indirect benefits to the non-poor. Under these conditions, sub-projects in non-poor areas would have to be protected from cuts to avoid a welfare loss to the non-poor. One can argue that all this helped assure this program's success in helping the poor in the crisis. While the program was clearly well targeted (to both poor workers and poor communities), it was almost certainly not a political equilibrium to assure that only poor areas participated. The other side of the coin to good targeting, was that the (relatively modest) spending on the non-poor had to be protected from cuts. 23 6. Conclusions Even when they have the power to do so, it is not obvious that it will be in the interests of the non-poor to shift the burden of fiscal adjustment to the poor. And even a minority poor can have political influence. As this paper has demonstrated, the incidence of fiscal contraction, and hence the case for action to protect public spending on the poor at a time of overall fiscal austerity, is an empirical question. This paper has tried to address that question using various data sets for Argentina. Aggregate budget cuts in the 1980s and '90s typically resulted in proportionately greater cuts in social spending; it was "non-social" spending that was protected. However, the proportionate cuts were about the same for types of social spending that matter more to the poor as for those that tend to favor the non-poor. The absolute cuts were in fact greater for "social insurance" that matters more to the non-poor. However, spending on targeted social assistance and employment programs was more vulnerable to aggregate spending cuts than more universal social services. While social spending as a whole was clearly exposed to fiscal contraction, this was somewhat less true of pro-poor spending on things that benefited the non-poor too. Fine targeting may thus be a mixed blessing for the poor; a higher mean may come with greater vulnerability to cuts - and quite possibly the cuts will come at times when help is most needed. There is a strong case for action to protect pro-poor social spending at such times. The paper studied one program that attempted to do so. The program was introduced (with support from the World Bank) to help compensate poor unemployed workers and their families for the effects of a macroeconomic shock. The design features of the program - providing low wage work targeted to poor areas - helped assure that the program was far better at reaching the poor than the pre-existing components of social spending in Argentina. 24 However, the program was clearly subject to the same constraints in the political economy that influenced the incidence of past fiscal contractions in Argentina. The program expanded into poor areas when the budget increased, but it retreated from poor areas when the program was cut. It was the program's disbursements to non-poor areas that were protected. Given the low wage rate offered, the direct benefits from the work are still very likely to have favored the poor, even after the cuts. So the design features of the program undoubtedly helped protect the poor from cuts. In conclusion, the time series data for Argentina suggest that action to support pro- poor sending at times of aggregate fiscal contraction is warranted. Social spending in general, and targeted social spending in particular, took a heavy hit at times of fiscal austerity. The add-on program studied here was able to achieve far more pro-poor targeting than pre-existing social spending. The new program was clearly not immune to the same underlying forces in the political economy that help protect spending on the non-poor from aggregate fiscal contractions. But the program helped the families of poor unemployed workers at a time of need; given the pattern of past public spending, it appears unlikely they would have received such help otherwise. 25 References Ag6nor, Pierre-Richard. 1998. "Stabilization Policies, Poverty and the Labor Market: Analytical Issues and Empirical Evidence," mimeo, Research Department, International Monetary Fund. Besley, Timothy and Ravi Kanbur. 1993. "Principles of Targeting", in Lipton, Michael and Jacques van der Gaag (eds.) Including the Poor. Washington D.C.: The World Bank. Brown, David S., and Wendy Hunter. 1999. "Democracy and Social Spending in Latin America", American Political Science Review 93(4): 779-90. De Donder, Philippe, and Jean Hindriks. 1998. "The Political Economy of Targeting", Public Choice 95: 177-200. Gelbach, Jonah and Lant Pritchett. 1997. "Redistribution in a Political Economy: Leakier Can be Better", mimeo, World Bank, Washington DC. Gasparini, Leonardo. 1999. "Incidencia Distributiva del Gasto Publico", Fundacion de Investigaciones Economicas Latinoamericanas (FIEL), Buenos Aires, Argentina. Government of Argentina. 1999. "Caracterizaci6n y Evoluci6n del Gasto P(iblico Social, Secretaria de Programacion Economia y Regional," Buenos Aires. Jalan, Jyotsna and Martin Ravallion. 1999. "Income Gains to the Poor from Workfare. Estimates for Argentina's Trabajar Program," Policy Research Working Paper 2149, World Bank, Washington DC. Lanjouw, Peter and Martin Ravallion. 1999. "Benefit Incidence and the Timing of Program Capture," World Bank Economic Review, 13(2): 257-274. Llach, Juan Jose and Silvia Montoya. 1999. "En Pos de la Equidad. La Probeza y la Distribuci6n del Ingreso en el Area Metropolitana del Gran Buenos Aires: Diagn6stico y Alternativas de Politicas," mimeo. 26 Ravallion, Martin. 1999a. "Is More Targeting Consistent with Less Spending?," International Tax and Public Finance, 6: 411-419. Ravallion, Martin. 1999b, "Appraising Workfare", World Bank Research Observer, 14: 31-48. Ravallion, Martin. 2000. "Monitoring Targeting Performance when Decentralized Allocations to the Poor are Unobserved," World Bank Economic Review (in press). Sen, Amartya. 1995. "The Political Economy of Targeting" in D. van de Walle and K. Nead (eds) Public Spending and the Poor: Theory and Evidence, Baltimore: Johns Hopkins University Press. van de Walle, Dominique, 1998. "Targeting Revisited", World Bank Research Observer Development, 13(2): 231-48. Wodon, Quentin and Norman Hicks. 1999. "Protecting the Poor During Crises Through Public Spending? Framework and Application to Argentina and Mexico", mimeo, Poverty Group, Latin America and the Caribbean Region, World Bank. World Bank, 1998. Argentina: Special Structural Adjustment Loan (SSAL), Washington DC: World Bank. World Bank. 1999. Poor People in a Rich Country: A Poverty Report for Argentina, Washington DC: World Bank. 27 Figure 1: Total public spending and social spending in Argentina 1980-97 (changes in logs) 0.2 - 0.1 0.0 - 100 - -0.3- -0.4 ____ ~Total spending ---Social spending -0.5. I I I I I I I I I 80 82 84 86 88 90 92 94 96 Figure 2: Concentration curve of participation in the Trabajar I program 80 0 0 0 20 40 60 80 100 Proportion of people in Argentina (ranked by household income per person) 15 Table 1: Incidence of social spending and taxes in Argentina 1996 Shares of spending attributed to quintiles of households ranked by income per person: 1 2 3 4 5 Total (poorest) Social Services 29.8 18.8 21.7 16.8 13.0 100 Social Insurance 9.9 20.6 19.5 23.6 26.5 100 Total Social Spending 21.8 19.5 20.8 19.5 18.4 100 Taxes 7.1 10.7 14.9 20.1 47.2 100 Income shares 4.0 8.4 13.2 21.2 53.2 100 Source: World Bank (1999), quoting Gasparini (1999); estimates for urban Argentina in 1996. 29 Table 2: Elasticities of social spending to total public spending in Argentina Spending Sub-categories Share of Elasticity to a change in total 2SLS category total public spending estimate of spending in the elasticity (%/0) totIncrease n Decrease m for a decrease total spending total spending in spending Social insurance 32.38 0.070 2.050* 2.240* (0.396) (0.368) (0.340) Social care 21.24 -0.129 2.243* 2.449* (incl. pensions) (0.533) (0.569) (0.509) Health 8.36 0.321 1.698 1.773 (0.561) (0.376) (0.438) Work (incl. unemploy. 1.88 0.704 2.904* 3.637 comp.) (2.068) (0.937) (1.691) Social services 23.30 0.246 2.255* 2.343* (0.543) (0.332) (0.389) Social services (excl. housing and 21.43 0.386 2.327* 2.589* urban) (0.481) (0.320) (0.367) Sector classification of social services Education 11.47 0.270 2.283* 2.328* (0.591) (0.398) (0.435) Health 4.83 0.740 2.098* 2.678* (0.578) (0.333) (0.443) Housing and urban 1.87 0.444 0.551 (0.524) (1.158) Social assistance (incl. 1.76 0.377 2.992* 3.650* family allowances) (0.912) (0.611) (0.797) Employment prograns 0.15 1.698 2.740* 4.515* (1.239) (0.591) (1.355) Targeted/universal classification of social services Targeted 4.09 -0.488 2.200* 1.587 (0.876) (0.554) (0.726) Targeted (excl. 2.22 0.220 3.009* 3.422* housing and urban) (0.834) (0.568) (0.695) Universal 19.21 0.398 2.267* 2.505 (0.542) (0.343) (0.402) Total 55.68 0.138 2.140* 2.277* (0.368) (0.260) (0.272) Note: Regressions of the change in the log of each spending category on the change in the log of total public spending, with intercepts, estimated on annual data for 1980-97. 2SLS estimator uses lagged total spending and lagged social spending as the instruments; the dummy variable for whether total spending has decreased is used as its own instrument. (The F-test for the first stage regression was 4.43, significant at the 3% level.) White standard errors in parentheses; * indicates significantly different from one at the 5% level. 30 Table 3: Selected points on the concentration curves Poorest x% of Proportion of households Cumrulative share of Cumulative share of households ranked with Trabajar II participants benefits from social benefits from social by income per with an income per person services insurance person; that places them amongst x= the poorest x% nationally 20 76 30 10 40 92 49 31 60 97 70 50 80 99 87 74 Note: For comparability with Table I the figures for Trabajar participants are households not people (Figure 2 is people not households). Sources: As for Table 1, except for Trabajar participation which is from Jalan and Ravallion (1999). 31 Table 4: Budget effects on poor-area targeting of Argentina's Trabajar Programs Full sample Trabajar II Trabajar III Variable coefficient t-ratio Coefficient t-ratio coefficient t-ratio Program spending 3.13 4.81 3.55 5.32 10.39 4.44 (deviation from time mean TII+TIII) Budget-neutral Targeting Differentials Buenos Aires -5.62 -2.50 -8.35 -2.38 -3.78 -0.43 Catamarca 49.48 3.34 20.38 2.12 93.31 9.54 Chaco 10.07 2.13 6.73 0.60 31.11 3.02 Chubut 31.53 3.92 29.89 2.89 39.99 4.46 Cordoba 144.60 10.25 131.35 6.94 161.51 18.35 Corrientes 24.68 4.64 19.16 2.51 41.25 4.38 Entre Rios 15.27 3.12 16.29 1.96 22.68 2.50 Formosa 10.38 1.82 6.54 0.51 26.74 2.81 Jujuy 61.23 4.59 46.80 8.58 92.46 9.28 La Pampa 6.37 1.36 11.15 1.16 8.01 0.89 La Rioja 3.97 0.43 -1.82 -0.09 26.64 2.62 Mendoza 29.98 4.17 34.67 2.50 31.64 3.54 Misiones -2.10 -0.29 -15.69 -1.68 23.62 2.47 Neuquen -8.07 -1.55 -6.32 -0.66 6.79 0.68 Rio Negro 52.33 4.28 59.11 2.60 54.82 5.97 Salta 67.30 10.81 64.22 6.20 86.70 8.63 San Juan 50.50 6.73 63.15 8.69 48.72 5.23 San Luis 37.08 6.11 30.34 3.55 61.68 5.94 Santa Cruz 9.33 1.21 4.62 0.30 26.50 2.81 Santa Fe 18.52 2.95 30.05 4.23 16.54 1.79 Santiago Del Estero 22.53 3.97 20.09 2.06 43.67 4.12 Tucuman 46.22 5.23 60.32 4.63 46.90 4.76 no. observations 132 66 66 R-squared 0.778 0.813 0.903 S.E. of regression 0.265 0.209 0.176 Mean dep. variable 0.307 0.328 0.276 F-statistic 17.38 8.493 8.568 Note: The dependent variable is the targeting differential given by the regression coefficient of Trabajar spending per capita at department level for each province and time period on the incidence of unmet basic needs per capita. The observation period for each of lTI and TIII was divided into three five month-intervals (one six month interval for TIll, converted into a five month equivalent); a statistical addendum with details is available from the author. The t-ratios are based on White standard errors. 32 Addendum Derivation of equations 5.1 and 5.2 The necessary conditions for a maximum of U'[Y' - (G" + GP), G',GP] with respect to Gn and GPare that U1y[Yn- (Gn + GP)] = Un(Gn) = Up (GP) (recalling that utility is assumed to be separable). The implicit solutions give Gn and GPas functions of Yn. On differentiating w.r.t. Y' we have: U;Y ayn + Unn ay =Un U~(Al) UnaG +Upn aGP= Un (A2) where G = + GP. Equations (Al) and (A2) imply that: Un (aGn /aG )Un (aGP /aG)=O (A3) n dy,, a7yn pp ayn ayn Also: lgGn wG aGP aG (-I"/@ )+(G I-G)=1 (A4) ayn ayn ayn ayn) (since G G" + GP). Solving (A3) and (A4) and noting that - =G,/IYn (A5.1) aGa / yn rP aGpl/ ayn (A5.2) we get equations (5.1) and (5.2). 33 Supplementary data tables Table Al: Targeting Performance by Five Month Intervals, Trabajar II and.Ill Targeting differential ($ per person per five months) Province Trabajar II Trabajar IH 11.1 11.2 11.3 111.1 111.2 11I.3 5/1997- 10/1997- 3/1998- 8/1998- 1/1999- 7/1999- 9/1997 2/1998 7/1998 12/1998 6/1999 11/1999 (5 month equivalent) Buenos Aires -2.8 -15.4* -7.8 1.6 -3.8 -5.5 Catamarca 33.9 11.6 30.7 51.4* 85.1* 84.2* Chaco 19.3 2.7 18.2 11.3 0.3 8.6 Chubut 33.1 45.4 18.2 55.8* 34.2* 2.5 Cordoba 143.5* 96.9* 159.3* 187.2* 113.3* 167.4* Corrientes 67.6* -0.3 4.8 20.4* 21.2* 34.4* Entre Rios 31.2 -0.6 26.0 22.8 -4.3 16.5 Formosa 7.9 18.5* 6.7 6.1 9.4 13.7 Jujuy 93.2* 27.1 34.2 128.6* 46.9* 37.4 LaPampa 27.8* 17.5* -6.0 3.3 -4.5 0.1 La Rioja -7.5 -1.3 22.4 -1.1 11.2 0.1 Mendoza 64.0* 4.4 43.5* 35.3* 13.5 19.2* Misiones -14.2 -8.8 -8.5 0.6 8.8 9.5 Neuquen -4.3 -11.4 12.6 -23.8 -13.6 -7.9 RioNegro 60.1* 25.1 103.2* 28.1* 69.9* 27.6* Salta 94.4* 39.5* 76.6* 49.6* 81.4* 62.3* San Juan 92.0* 63.1* 45.7* 35.5* 49.7* 17.0 San Luis 78.5 3.1 31.2* 29.6* 29.3* 50.8* Santa Cruz 39.3 -23.3 8.3 3.6 24.0 4.1 Santa Fe 56.7* 25.7* 19.9* 8.1 -3.1 3.8 Santiago Del Estero 88.3* -6.6 3.8 9.2 18.4* 22.1 Tucuman 112.2* 38.6* 47.1* 27.8* 26.3* 25.3 All departnents 71.6* 15.3* 22.9* 24.3* 23.4* 23.9* Note: * indicates significantly different from zero at the 5% level. 34 Table A2: Program Spending by Five Month Intervals, Trabajar II and III Program disbursements ($ per person per five months) Province Trabajar HI Trabajar III 11.1 11.2 11.3 111.1 III.2 I11.3 5/1997- 10/1997- 3/1998- 8/1998- 1/1999- 7/1999- 9/1997 2/1998 7/1998 12/1998 6/1999 11/1999 (5 month equivalent) Buenos Aires 1.09 0.92 1.51 1.23 1.57 1.42 Catamarca 14.24 3.76 4.06 3.42 3.96 3.28 Chaco 17.79 5.16 7.09 6.88 4.02 5.06 Chubut 5.67 2.69 4.83 3.32 2.18 2.40 Cordoba 4.44 2.82 2.38 2.12 1.77 2.55 Corrientes 14.81 2.30 3.38 4.74 2.73 3.83 Entre Rios 7.43 4.13 3.92 4.38 2.46 2.28 Fortnosa 13.36 4.19 5.22 4.86 4.23 3.86 Jujuy 18.96 5.20 6.58 9.41 4.46 4.46 La Pampa 8.54 4.20 6.20 5.60 4.30 4.20 La Rioja 15.13 2.55 4.00 3.06 2.61 2.59 Mendoza 5.89 1.56 2.23 1.72 1.10 1.68 Misiones 12.05 6.07 5.56 4.93 3.83 4.92 Neuquen 12.11 5.83 7.68 6.07 3.54 3.37 Rio Negro 8.90 4.58 5.51 4.13 3.85 3.53 Salta 13.07 3.78 4.55 2.52 3.51 2.51 San Juan 8.92 3.35 2.79 1.86 2.91 1.83 San Luis 16.66 2.47 3.53 2.78 2.30 3.08 Santa Cruz 5.69 2.87 4.05 0.42 2.62 0.37 Santa Fe 7.95 4.46 4.01 3.06 2.36 3.14 Santiago Del Estero 21.38 4.07 3.96 3.87 4.18 5.71 Tucuman 13.65 4.58 5.46 4.12 4.00 3.77 All departments 6.37 2.58 3.05 2.62 2.29 2.44 35 Table A3: Data used for Table 2 Total Social Social insurance Social services public spending Total Pensions Health Unemp- Total Educa- Health Housing Social Employ- Targeted Universal spending etc loyment tion etc assist. ment 1980 77.792 38.861 22.055 13.873 5.4040 2.065 16.807 8.004 3.11 2.095 1.258 0.069 3.422 13.385 1981 82.007 38.959 23.263 13.894 7.075 1.327 15.696 7.644 3.314 1.179 1.342 0.055 2.575 13.121 1982 66.459 25.796 15.243 8.641 5.105 0.800 10.553 5.154 2.207 1.085 0.749 0.036 1.87 8.683 1983 66.148 29.143 16.133 8.823 5.447 1.141 13.009 6.299 2.527 1.712 0.908 0.037 2.656 10.353 1984 68.337 33.334 17.793 10.399 5.686 0.991 15.541 7.909 3.099 1.524 1.289 0.047 2.86 12.681 1985 69.567 34.709 19.711 12.805 5.220 1.021 14.998 7.262 2.890 1.365 1.595 0.052 3.012 11.986 1986 74.143 39.755 22.126 14.003 5.798 1.590 17.628 8.577 3.509 1.715 1.662 0.047 3.424 14.204 1987 82.044 44.236 24.665 16.233 6.267 1.404 19.571 9.641 3.692 1.972 1.885 0.052 3.909 15.662 1988 72.657 37.663 20.913 13.459 5.977 0.778 16.750 8.483 3.202 1.442 1.570 0.045 3.058 13.692 1989 65.252 34.174 20.486 14.574 4.658 0.739 13.689 6.557 2.730 1.045 1.725 0.030 2.800 10.889 1990 61.948 37.878 23.231 16.179 5.687 0.765 14.647 7.171 2.896 1.444 1.197 0.039 2.679 11.968 1991 68.807 43.471 26.776 18.588 6.140 1.466 16.695 8.009 3.497 1.367 1.416 0.058 2.841 13.854 1992 77.301 48.613 29.415 20.725 6.652 1.455 19.198 9.347 4.297 1.120 1.696 0.090 2.905 16.293 1993 82.670 52.511 30.567 20.584 7.267 2.009 21.945 10.792 4.850 1.126 2.026 0.194 3.260 18.685 1994 88.203 57.765 34.047 22.816 8.225 2.226 23.717 11.605 5.380 1.270 2.225 0.279 3.657 20.06 1995 85.393 55.530 33.213 22.083 8.109 2.269 22.318 11.325 5.009 1.121 1.819 0.266 3.216 19.102 1996 83.246 54.532 32.190 21.854 7.705 2.100 22.342 11.260 5.037 1.139 1.926 0.359 3.480 18.862 1997 87.393 56.542 32.116 21.972 7.662 2.005 24.426 12.435 5.263 1.319 2.371 0.471 4.010 20.416 Note: Billion pesos in 1997 prices. Source: Government of Argentina (1999). Policy Research Working Paper Series Contact Title Author Date for paper WPS2365 Leading Indicator Project: Lithuania Stephen S. Everhart June 2000 M. Geller Robert Duval-Hernandez 85155 WPS2366 Fiscal Constraints, Collection Costs, Keiko Kubota June 2000 L. Tabada and Trade Policies 36896 WPS2367 Gender, Poverty, and Nonfarm Constance Newman June 2000 M. Clarke Employment in Ghana and Uganda Sudharshan Canagarajan 31752 WPS2368 Seeds of Corruption: Do Market Harry G. Broadman June 2000 S. Craig Institutions Matter? Francesca Racanatini 33160 WPS2369 How the Proposed Basel Guidelines Giovanni Ferri June 2000 E. Mekhova On Rating-Agency Assessments Li-Gang Liu 85984 Would Affect Developing Countries Giovanni Majnoni WPS2370 A New Model for Market-Based Marcelo Giugale June 2000 M. Geller Regulation of Subnational Borrowing: Adam Korobow 85155 The Mexican Approach Steven Webb WPS2371 Shock Persistence and the Choice Marcelo Giugale June 2000 M. Geller of Foreign Exchange Regime: An Adam Korobow 85155 Empirical Note from Mexico WPS2372 Financial Openness, Democracy, Mansoor Dailami June 2000 W. Nedrow and Redistributive Policy 31585 WPS2373 Reciprocity across Modes of Supply Aaditya Mattoo June 2000 L. Tabada in the World Trade Organization: Marcelo Olarreaga 36896 A Negotiating Formula WPS2374 Should Credit Be Given for Aaditya Mattoo June 2000 L. Tabada Autonornous Liberalization in Marcelo Olarreaga 36896 Multilateral Trade Negotiations? WPS2375 Asset Distribution, Inequality, Klaus Deininger June 2000 M. Fernandez and Growth Pedro Olinto 33766 WPS2376 The Effect of Early Childhood Michael M. Lokshin June 2000 P. Sader Development Programs on Women's Elena Glinskaya 33902 Labor Force Participation and Marito Garcia Older Children's Schooling in Kenya WPS2377 Reforming the Water Supply in Claude Menard June 2000 H. Sladovich Abidjan, C6te d'lvoire: Mild Reform George Clarke 37698 in a Turbulent Environment Policy Research Working Paper Series Contact Title Author Date for paper WPS2378 Disintegration and Trade Flows: Simeon Djankov June 2000 R. Vo Evidence from the Former Soviet Caroline Freund 33722 Union WPS2379 India and the Multilateral Trading Aaditya Mattoo June 2000 L. Tabada System after Seattle: Toward a Arvind Subramanian 36896 Proactive Role WPS2380 Trade Policies for Electronic Aaditya Mattoo June 2000 L. Tabada Commerce Ludger Schuknecht 36896 WPS2381 Savings and the Terms of Trade Pierre-Richard Ag6nor June 2000 T. Loftus under Borrowing Constraints Joshua Aizenman 36317 WPS2382 Impediments to the Development and Thorsten Beck June 2000 E. Mekhova Efficiency of Financial Intermediation 85984 in Brazil WPS2383 New Firm Formation and Industry Thorsten Beck June 2000 E. Mekhova Growth: Does Having a Market- or Ross Levine 85984 Bank-Based System Matter? WPS2384 Are Cost Models Useful for Telecoms Daniel A. Benitez July 2000 G. Chenet-Smith Regulators in Developing Countries? Antonio Estache 36370 D. Mark Kennet Christian A. Ruzzier WPS2385 The Rise, the Fall, and ... the Antonio Estache July 2000 G. Chenet-Smith Emerging Recovery of Project John Strong 36370 Finance in Transport WPS2386 Regulators and the Poor: Lessons Richard Green July 2000 G. Chenet-Smith from the United Kingdom 36370 WPS2387 The Long and Winding Path to Private Antonio Estache July 2000 G. Chenet-Smith Financing and Regulation of Toll Manuel Romero 36370 Roads John Strong WPS2388 The Role of Special and Differential Constantine Michalopoulos July 2000 L. Tabada Treatment for Developing Countries in 36896 GATT and the World Trade Organization WPS2389 Vietnam: On the Road to Labor- Patrick Belser July 2000 H. Sutrisna Intensive Growth? 88032 WPS 2390 The Social Rate of Return on David Canning July 2000 H. Sladovich Infrastructure Investments Esra Bennathan 37698