A1497
Title: How to bet on winners (and losers)
Authors: Andre B M Souza - ESADE Business School (Spain) [presenting]
Christian Brownlees - UPF (Spain)
Abstract: The construction of long-short portfolios is cast as a statistical decision problem in which the investor seeks to buy the top-performing stocks (the ``winners'') and sell the worst-performing ones (the ``losers'') on the basis of stock characteristics. We derive the optimal portfolio selection rule implied by a loss function that accounts for different types of misclassification errors in portfolio construction. This approach leads to a return classification problem and the optimal rule buys or sells stocks based on their probabilities of being winners or losers, conditional on the stock characteristics. When returns are generated by an additive regression model and misclassification costs satisfy a symmetry condition, the optimal rule simplifies to the conventional sorting procedure based on expected returns. An empirical application using U.S. stock data shows that portfolios constructed using the optimal rule achieve higher Sharpe ratios compared to those built using conventional methods. Our results demonstrate that predictive signals in the cross-section of stock returns go beyond expected returns, and that properly optimized portfolio selection rules based on these signals can generate substantial economic value for investors.
