COMPLETE SUBSET AVERAGING FOR QUANTILE REGRESSIONS

We propose a novel conditional quantile prediction method based on complete subset averaging (CSA) for regressions. All models under consideration are potentially misspecified, and the dimension of regressors goes to infinity as sample size increases. Since we average over subsets, number is much larger than usual model which adopts sophisticated weighting schemes. use an equal weight but select proper leave-one-out cross-validation method. Building upon theory Lu Su (2015, Journal Econometrics 188, 40–58), investigate large properties CSA show asymptotic optimality in sense Li (1987, Annals Statistics 15, 958–975) check finite performance via Monte Carlo simulations empirical applications.