B0582
Title: Minimum distance estimation of quantile panel data models
Authors: Blaise Melly - University of Bern (Switzerland) [presenting]
Martina Pons - University of Bern (Switzerland)
Abstract: A minimum distance estimation approach is proposed for quantile panel data models where the unit effects may be correlated with the covariates. The estimation method is computationally straightforward to implement and fast. A quantile regression is first computed within each unit and then applied GMM to the fitted values from the first stage. The suggested estimators apply (i) to group data, where data is observed at the individual level, but the treatment varies at the group level, and (ii) to classical panel data, where we follow the same units over time. Depending on the variables assumed to be exogenous, this approach provides quantile analogues of the classical least squares panel data estimators such as the fixed effects, random effects, between, and Hausman-Taylor estimators. A more precise estimator is provided for grouped instrumental quantile regression than the existing ones. The asymptotic properties of the estimator are established when the number of units and observations per unit jointly diverge to infinity. An inference procedure that automatically adapts to the potentially unknown rate of convergence of the estimators is suggested. Monte Carlo simulations show that the estimator and inference procedure also perform well in finite samples when the number of observations per unit is small. In an empirical application, it is found that the introduction of the food stamp program increased the birth weights only at the bottom of the distribution.
