Optimising Anti-Poverty Transfers With Quantile Regressions

Under perfect observation of incomes, designing such scheme boils down to solving an optimisation program under constraints, which can be achieved with well-defined methods. In contrast, when incomes cannot be perfectly observed, the schemes are usually based on predictions of living standards using ancillary regressions and household survey data to predict the unobserved living standards of households. In this paper, we study the poverty minimisation program under imperfect information. We show why using predictions of living standards helps to deal approximately with an otherwise intractable problem. Then, we propose a new approach to the practical optimisation procedure based on improved predictions of living standards in terms of the targeting problem to be solved. Our new empirical methodology to target direct transfers against poverty is based on observable correlates and on estimation methods that can focus on the poor: the quantile regressions. We illustrate our results using data from Tunisia.