SMALL SAMPLE EVIDENCE OF REGRESSION QUANTILE ESTIMATES FOR STRUCTURAL MODELS: ESTIMATION AND TESTING

Abstract

A small Monte Carlo study is conducted to investigate the small sample properties of regression quantiles estimates in structural econometric models with iid errors. Two versions of median regression (the 50th quantile) estimates using instrumented regressors are used, with the first stage estimated by least squares and by median regression. These estimators are compared to two stage least squares, ordinary least squares and simple median regression. It is found that the performance of the two stage regression quantile estimators is equal, if not better, than two stage least squares in terms of bias and that t tests have good size and power properties, under a variety of error distributions.