A0907
Title: Unified inference for predictive quantile threshold regressions
Authors: Xinling Xie - Southwestern University of Finance of Economics (China) [presenting]
Abstract: The purpose is to test the presence of the episodic predictability of a persistent predictor in a predictive quantile regression model. The null limiting distribution of the conventional Wald test for the predictability is shown to depend on the unknown nuisance parameters (i.e., the persistence degree of the predictor and the quantile endogeneity). Thus, the inference for the regime-specific return predictability becomes infeasible. The robust inference is achieved with the aid of the instrumental variable filtering technique, and the induced test statistic is asymptotically Chi-squared under the null hypothesis, regardless of the persistence of the predictor and possible intercept shifts. In addition, it is shown to enjoy non-trivial asymptotic power against a class of local alternatives. Monte Carlo simulations demonstrate the effectiveness and robustness of the newly proposed test. The empirical application to the U.S. stock returns reveals the existence of episodic and heterogeneous predictability across various quantile levels.
