Which Variables Predict and Forecast Stock Market Returns?

Abstract

Changes in stock returns arise from changes in expected future cash flow growth and expected future discount rates. However, which variables proxy for those changes remains unknown. This paper considers twenty-five variables that are arranged into five groups and examines both in-sample predictability as well as out-of-sample forecasting. Existing research typically considers either one, or a small selection of variables, prominent within this is the dividend price ratio and interest rates. We consider variables that span the categories of financial ratios, macro, labour market and housing variables as well a group referred to as others, which incorporates measures of sentiment and leverage. In-sample results show that significance arises in variables across these five groups. Of note, price ratios, GDP acceleration, inflation, unemployment and consumer sentiment feature prominently, with the purchasing managers index, housing variables and leverage also represented. Thus, predictive variables appear across the different categories. In conducting out-of-sample forecasts, we utilise a range of forecast performance measures and consider single model and combined forecasts. The results show that, with one exception, the combined model forecasts outperform the single model forecasts across all measures. This supports the view that a range of variables from across the economy can help predict future stock returns.