Predicting stock returns: A risk measurement perspective

This paper proposes a new and efficient model selection strategy to obtain significant stock returns predictability from a risk measurement perspective. The risk interval is defined as the distance between the current actual return and the returns' historical average. The model selection strategy involves switching stock return forecasting models according to different risk intervals from the mean reversion and extreme value theory. This new strategy generates encouraging results in the empirical analysis. A mean-variance investor can realize sizeable economic gains by allocating assets through this new approach relative to competing forecasting models. Furthermore, the strategy performs robustly under alternative settings from both statistical and economic perspectives.

Automated summary

This paper introduces a new model selection strategy for predicting stock returns from a risk measurement perspective. Empirical analysis shows that this strategy is effective and can lead to substantial economic gains for investors. Additionally, the strategy performs well under alternative settings both statistically and economically.