Title: Testing in predictive quantile regressions with time-varying volatility
Authors: Robert Taylor - University of Essex (United Kingdom)
Paulo Rodrigues - Universidade Nova de Lisboa (Portugal)
Matei Demetrescu - CAU Kiel (Germany) [presenting]
Abstract: While stock return predictability has received considerable attention in the literature, predictability tests are geared at detecting whether the conditional mean of the return series of interest depends on putative predictors or not. To help decide on quantile predictability, we discuss tests based on the Langrange Multiplier principle. The LM approach leads to a simple auxiliary regression, for which inference can be conducted using instrumental variable estimation. We therefore obtain simple linear IV-based tests that are robust to conditional and unconditional heteroskedasticity. Moreover, we provide an analysis of the behavior of the proposed tests under a factor structure of the regressors, where the common and idiosyncratic components may be either near-integrated or stable autoregressions or both. We find the proposed tests to perform well in finite samples compared with alternatives based on quantile regressions and resampling.