Desempenho de modelos de otimização em diferentes horizontes de investimento no mercado brasileiro

Abstract

This paper proposes a comparative analysis of optimization models through different investment horizons in the Brazilian market. The use of models is based on modifications of the sample covariance matrix (such as shrinkage methods and detection of outliers models) and the replacement as measure of risk of the variance by the Value at Risk (VaR) and Expected Loss (ES). The analysis separates in two different times: in-sample, understood in daily returns between January 2003 and December 2010; and another out-of-sample on a daily basis of returns from January 2010 to December 2013. As results, this paper observe that both, in-sample as the investment horizon out-of-sample long-term, models involved in risk measures presented better performance, especially in case of market crash. In short-term and medium-term investment horizons, models of minimum-variance and mean-variance obtained the best results. Economically, the investor can use the models for increased safety in the allocation assets in turbulent periods. In theoretical implications, there is the importance of not using excessively a single optimization model for the management of portfolios.