Gerardo Ferrara

“Portfolio Optimization under Model Uncertainty”

abstract

This study proposes a novel methodology to deal with model uncertainty in forecasting stock returns. My main interest here is to overcome the
tendency of Bayesian Model Averaging to give all of the weight to a single model. A potential solution of this problem is to capture the
nature of the underlying data generating process by sampling from the space of possible combinations. After presenting the methodology in
detail, I show that my approach increases the accuracy of out-of-sample forecasting. Moreover, I investigate the impact of improved forecasting on portfolio performances.