Evaluating quantile treatment effects with Machine learning. An application to the informal sector wage gap

We explore how using the Gradient Boosting Machine (GBM) method to compute propensity scores may improve the Quantile Treatment Effect (QTE) estimation in the case of a binary treatment and a high dimensional dataset. We use the proposed method to estimate the wage gap between workers in the informal and formal sectors in South Africa at different quantiles of the wage distribution. Interesting results emerge. The earnings differential between informal and formal workers is negative. However, it significantly decreases when the propensity scores are estimated with GBM, compared to alternative methods, showing that our strategy provides more accurate estimates of the QTEs when using mixed-type data.