Risk Aversion and the Forecasting Performance of Implied Expected Returns in Bull and Bear Markets

In order to obtain more accurate return forecasts, this paper proposes several methods to improve the almost always positive implied return forecasts. Firstly, we estimate a constant and time-varying volatility risk premium and link these to the risk aversion, which is an input for implied return forecasts. Secondly, because implied return forecasts are almost always positive, forecasting accuracy is low during bear markets. To remedy this problem, we implement a dynamic forecast combination approach based on the output of Markov switching models, or on the level of volatility. We nd that a constant volatility risk premium only marginally aects forecasting performance. On the other hand, a time-varying volatility risk premium reduces the root-mean-squared forecast error (RMSFE) from 53.66h to 53.08h. The forecast combination approach based on a Markov switching model leads to further considerable improvement and achieves the RMSFE of 51.21h. This improvement mainly comes from more accurate forecasts in bear markets where the RMSFE decreases from 72.07h to 62.30h.