Portfolio of global futures algorithmic trading strategies for best out-of-sample performance

Abstract

We investigate two different portfolio construction methods for two different sets of algorithmic trading strategies that trade global futures. The problem becomes complex if we consider the out-of-sample performance. The Comgen method blindly optimizes the Sharpe ratio, and Comsha does the same but gives priority to strategies that individually have the better Sharpe ratio. It has been shown in the past that high Sharpe ratio strategies tend to perform better in out-of-sample periods. As the benchmark method, we use an equally weighted (1/N, naïve) portfolio. The analysis is performed on two years of out-of-sample data using a walk forward approach in 24 independent periods. We use the mean reversion and trend following datasets consisting of 22,702 and 36,466 trading models (time series), respectively. We conclude that Comsha produces better results with trend-following methods, and Comsha performs the same as Comgen with other type of strategies.