Forecasting the Equity Premium in a Bayesian Setting
The subject of this Master thesis is the predictability of the equity risk premium. It aims to increase the predictability by summarizing information from a large set of variables into a small number of factors. The variables include macroeconomic variables, technical indicators, and measures of investor sentiment. Technical indicators provide most predictive power at short forecast horizons, whereas sentiment measures provide complementary value at longer forecast horizons. A market timing study confirms that this predictability also translates into economic value when implemented as a trading strategy. Next, time-varying volatility and regime-dependent specifications of the equity premium's volatility process increase the predictability further, especially during the post- 2008 period. Time-varying volatility performs especially well at short forecast horizons, whereas regime-dependent models yield additional improvements at longer forecast horizons. The thesis also tests the value of assuming regime-dependent relations between the predictors and the equity premium. This does not generate additional improvements. The results support the hypothesis that the countercyclical pattern of equity premium predictability is at least partially driven by the time-varying nature of the volatility process. It does not support the hypothesis that it is driven by a regime-dependent relation between the predictors and the risk premium. The research takes estimation uncertainty of all models into account by making use of Bayesian MCMC methods.