Seleção de Carteiras: Escolha entre modelos baseada em persistência de performance

This essay presents an alternative to the problem of choosing between strategies for building investment portfolios. We propose a new portfolio selection procedure, dividing the sample into three equal parts (for estimations initiations, training, and evaluation outside the sample) in which, at each point of time, the strategy with the best performance is chosen in a window of p recent observations for a given criterion. We considered as criteria the mean, variance, and Sharpe ratio, aiming to construct sequences of allocation choices that best adapted to the different contexts and databases analyzed. Results indicate that the suggested approach was capable of generating allocation sequences with good performance in terms of average return and Sharpe ratio.