Testing for Multiple-Horizon Predictability: Direct Regression Based versus Implication Based

Abstract

Research in finance and macroeconomics has routinely employed multiple horizons to test asset return predictability. In a simple predictive regression model, we find the popular scaled test can have zero power when the predictor is not sufficiently persistent. A new test based on implication of the short-run model is suggested and is shown to be uniformly more powerful than the scaled test. The newtest can accommodate multiple predictors. Compared with various other widely used tests, simulation experiments demonstrate remarkable finitesample performance. We reexamine the predictive ability of various popular predictors for aggregate equity premium.