Title: Confident risk premiums and investments using machine learning uncertainties
Authors: Rohit Allena - University of Houston (United States) [presenting]
Abstract: Ex-ante standard errors of risk premium predictions from neural networks (NNs) are derived. Considering standard errors, we identify precise stock-level and portfolio-level return predictions and provide improved investment strategies. The confident high-low strategies that take long-short positions exclusively on stocks with precise risk premia significantly outperform traditional high-low trading portfolios out-of-sample. Optimal mean-variance portfolios incorporating (co)variances of expected return predictions also outperform existing strategies. Economically, time-varying standard errors reflect market uncertainty and spike after financial shocks. In the cross-section, the level and precision of risk premia are correlated, thus NN-based investments deliver more gains in the long positions.