Evaluating the impact of public policies : a general framework and applications to agricultural and forest policies ; L'évaluation de l'impact des politiques publiques : caractérisation des enjeux et exemples de politiques agricoles et forestières

Abstract

In this PhD thesis, we describe the evaluation problemas a decision problem: a policy is evaluated by comparing the realized outcomes to the counterfactual ones that could have happened had the policy been different. We study how well the statistical potential outcome framework and the econometric simultaneous equationsmodel performat describing and encoding public policies. The econometric approach to causality is preferred because it allows the modelling of interventions as modifications. The usual treatment effect parameters are linked to the decision problems they allow to solve. New parameters, solving problems in more general models, are defined. We then take an example of the practical problem of identifying the causal effect of public policies from observed data.We study the impact of investment subsidies on the demand for labor and materials. We derive the behavior of the within estimator under different identifying assumptions. Tests are developed using a modified version of this estimator to determine whether there is an impact of the subsidies or if the observed data exhibit only selection effects. The sensitivity of evaluation results to the welfare functions used is finally investigated. We show that getting rid of developed countries agricultural policies increases aggregate welfare in Brazil whereas at the same time it decreases the welfare of the poor. ; Cette thèse aborde le problème de l'évaluation des politiques publiques comme un problème de décision. Lesmérites respectifs des approches statistique (utilisant lesmodèles à résultats potentiels) et économétrique (utilisant les systèmes d'équations simultanées) sont analysés au regard de leur capacité à envisager cette décision. L'approche économétrique est retenue car elle permet notamment de modéliser simplement les interventions publiques. Les paramètres d'effets causals moyens habituellement estimés dans la littérature sont reliés aux problèmes de décision qu'ils permettent de résoudre. De nouveaux paramètres plus largement pertinents sont ...