| 简介： | This paper revisits the models of predictive quantile regressions with persistent predictors. The nonstationary predictors are allowed to be informative with non-zero slope coefficient when estimating the conditional quantiles of the dependent variable. Under this framework, a rather surprising limit theory is developed: (i) the intercept estimator in quantile regression diverges with √n-rate, and (ii) the slope coefficient estimator diverges arbitrarily fast, thereby leading to a spurious quantile prediction. The new limit theory raises a serious empirical concern when predicting conditional quantile of financial returns using persistent predictors. |