 |
Social programs for poverty alleviation involve eligibility rules and transfer
rules. Both kinds of rules depend on the characteristics of applicants. Often,
statistical scores, such as for proxy-means tests, are used that summarize the in-
formation about these characteristics. However,these methods are often attacked
for their mediocre targeting performances. In this paper, we explore the esti-
mation of conditional and unconditional focused quantile regressions to generate
tted-values of living standards that are plugged into the poverty minimization
program to obtain optimal transfer amounts. Incidentally, we provide a precise
mathematical translation of the intuition of Bourguignon and Fields (1997) to de-
ne these optimal amounts, in terms of the calculus of variation. The use of these
regression methods are suggested by a theoretical analysis of the poverty mini-
mization problem. We illustrate these methods with simulations based on data
from Egypt in 2013. In these simulation results, the Recentered Inuence Function
(RIF) regressions focusing on the poor correspond to the most e¢ cient transfer
scheme. However, most of the gain in poverty reduction is obtained by making
transfer amounts varying across bene
ciaries rather than by varying tted-value
estimation methods. In particular, using focussed RIF regressions instead of fo-
cussed quantile regressions delivers only marginal additional poverty alleviation.
Nonetheless, using focussed RIF regressions centered on the proportion of poor
households generates substantial reduction of the exclusion of the poor from the
program, as compared to using other regression methods.
