Beyond regression: evaluating different semi-parametric approaches and machine learning tools in the difference-in-difference design.

The contribution of the thesis to the repeated cross-sections Difference-in -Difference (DiD) literature is threefold: first, it shows that the commonly-used DiD regression is severely biased under realistic scenarios and proposes alternative corrections; second, it presents a semi-parametric estimator robust to heterogeneity both in the treatment group and time dimensions; finally, it compares through Monte Carlo simulations the empirical performance of the proposed estimators with those suggested by the literature, in particular with the semi-parametric doubly robust DiD of Sant’Anna and Zhao (2020). The estimators are also modified to allow for machine-learning first-stage estimates, following the literature of Chernozhukov et al. (2018). Results show that different semi-parametric estimators outperform regression, even if corrections provide substantial benefits. Following Sequeira (2016), the thesis estimates the effect of tariff reduction on bribing behavior by analyzing trades between South Africa and Mozambique during the period 2006–2014. Contrarily to the replication in Chang (2020), the thesis provides substantial proof that the effect is close and even lower in magnitude than the one of the original paper. Still, the contribution reinforces the evidence that tariff reductions tend to weaken bribing behavior.