Abstract

We take up the challenge of forecasting out-of-sample monthly returns on stock market indices. Recent contributions show that a wide range of popular predictors poorly forecast the US equity premium when the performance is compared to the average predictor. We revisit these findings focusing on three aspects. First, we report results for four major stock markets: US, UK, Germany and Japan. We find similar patterns of forecastability across countries. Second, this paper focuses on the choice of models and, to a lesser extent, on the choice of predictors. This reflects the changing focus that can be found in recent years in the literature. We explore two structural aspects that have attracted attention: structural changes and theory-induced restrictions. In our samples, models incorporating structural breaks fare better than standard regressions and outperform the average the average predictor at longer horizons. Theory-induced restrictions that were proposed in the literature are shown to be rarely binding and hence the performance is similar to standard regressions. One key finding is that the restrictions produce meaningful results when they are combined with models of structural breaks. These joint models actually do add forecasting power, especially at longer horizons. On the contrary, popular standard models underperform in almost all respects. Finally, we evaluate the forecasting performance in terms of both the size of the forecasting error and the quality of sign predictions. One of the key findings of the paper is that the results look more favorable for forecasting models when it comes to forecasting the direction of market movements.

Citations