How Well Do Financial and Macroeconomic Variables Predict Stock Returns: Time-Series and Cross-Sectional Evidence

Abstract

Recent evidence of mean reversion in stock returns has led to an explosion in the development of forecasting variables. This paper evaluates the relative performance of these many variables in both time-series and cross-sectional setups. We collect the different measures and compare their forecasting ability for stock returns, and we examine the forecasting variables' ability to reduce pricing errors in the conditional C-CAPM. A key result of the analysis is that the traditional pricedividend ratio performs surprisingly well compared to the many new forecasting variables. We also find that at short and mid-range horizons Lettau and Ludvigson's (2001a) consumption-aggregate wealth variable offers the strongest forecasting ability, although this variable's predictive ability is sensitive to the sample period chosen. At longer horizons, price-normalized variables such as the traditional price-dividend ratio, the price-consumption ratio of Menzly et al. (2004), and the price-output variable of Rangvid (2006) outperform the other variables. These variables also turn out to be superior in reducing pricing errors in the conditional C-CAPM. Thus, the same set of variables dominate in both time-series and cross-sectional settings.

Keywords: Return predictability, C-CAPM, conditional asset pricing

JEL Classification: C52, G12