Trading and filtering futures spread portfolios: Further applications of threshold and correlation filters

This study researches the topic of trading futures spreads, that is, trading the pricing differential between two futures contracts. We trade an equally weighted portfolio of three oil spreads using four trading models: the fair value cointegration, Generalised AutoRegressive Conditional Heteroskedastic, Moving Average Convergence Divergence and Neural Network Regression (NNR) models. The motivation of this research is twofold. First, the profitability of spread markets has been tested further than the traditional fair value model. Second, the correlation filter has been extended by investigating the effect of combining inputs from both a threshold filter and a correlation filter. The results indicate that the best model type for trading spreads is the NNR, with an out-of-sample annualised percentage return of 10.76 per cent and drawdown of –1.52 per cent, resulting in a leverage factor of 7.0964 in the case of the hybrid filter. Further, the results show the hybrid filter to be a sound development, proving to be the best out-of-sample filter on the occasion it is selected.