ECONOMETRICS: Unified Inference for Long-Horizon Predictive Regressions; Dr Seok Young Hong (Nanyang Technological University)

Abstract

We propose a unified procedure for testing the predictability of asset returns based on the empirical likelihood (EL) method via both sample-splitting and a two-stage approach. We make several novel methodological contributions. First, we allow the predictor variable in our unified test to be mildly integrated or mildly explosive in addition to the usual persistence classes permitted in the empirical likelihood literature: stationary, locally integrated, and unit root cases, thereby covering virtually most possible scenarios that the econometrician may face. Second, we allow for heteroscedasticity in the error term, and relax the usual regularity conditions imposed previously. Lastly, we propose to utilize a two-stage approach in employing the EL method, addressing the efficiency issue of the sample splitting approach. Our robust procedure is applied to investigate the long-run predictive regression model. We conduct an empirical application on the US stock market, reporting evidence of predictability with the T-bills and the inflation rate. A simulation study confirms our proposed test performs very well in finite samples, exhibiting robust size and power properties.