A0298
Title: Flexible extreme marginal quantile treatment effect in high dimensions
Authors: Jing Zhou - University of East Anglia (United Kingdom) [presenting]
Abstract: The marginal quantile treatment effects are investigated when the quantile level approaches 0 or 1. When the quantile level approaches the ends, quantile regression cannot accurately model the tail distributions. To overcome this limitation, we propose an alternative approach that uses extreme quantile models to estimate the marginal effect in the presence of a continuous covariate shift. Such models use an extreme value index to model the tail of the distribution function. This method estimates an extreme value index at intermediate quantile levels and extrapolates to the tails where the quantile level is close to zero. By extrapolating, we aim to estimate the extreme treatment effects consistently and obtain the corresponding asymptotic distribution. Further, when the number of parameters is nonnegligible, we consider regularization to identify the relevant covariates among hundreds of variables for the extreme quantile treatment effects.
