Forecasting the equity premium using weighted regressions: Does the jump variation help?

Growing literature documents that jump variations are important for comprehending the evolution of asset prices. In this paper, we provide a novel insight on the jump components. Specifically, we forecast the equity premium using the weighted least squares (WLS) approach that assigns the inverse of variance weight to observations, and detect the role of jump contributions in it. The results indicate that the WLS models with jump-robust variance weights generate superior out-of-sample performance both statistically and economically relative to that with the jump-involved weights, suggesting that eliminating the jump variation in the variance weight helps to predict the stock returns. The predictive source of the jump-robust variance stems from its efficient measure of the continuous price process and forecast error variance reduced. Furthermore, we demonstrate that the jump component in the variance weight should rather be dumped than collected in terms of minimizing the forecast losses.