How narrowly should anti-poverty programs be targeted? Simulation evidence from Bolivia and Indonesia

A key question in the design of anti-poverty programs is to what extent they should be targeted. Empirical evaluations of targeted transfer schemes and simulation exercises often point to further gains that can be had from targeted transfers vis-à-vis universal transfers or from more narrow targeting. Theoretical work, on the other hand, has identified hidden costs associated with targeting - including politico-economic constraints on budgets - but these are frequently ignored in empirical work. In this paper we first argue that common targeting measures can be interpreted as preferences that attach specific weights to true and false positive rates. Based on data from Bolivia and Indonesia, we show that targeting based on an imperfect poverty classifier based on proxy means tests results in very distinct 'optimal' beneficiary shares when these measures are used as a decision criterion. Implications from poverty simulations are sensitive to assumptions about the political economy relationship between the beneficiary share and the available budget. In fact, in many situations, optimizing targeting measures will be misleading when the actual goal is to maximize the effect on poverty.