Adaptive trading rules for commodity futures

This paper extends and refines an adaptive filtering model utilizing entropy-based decision rules. The initial adaptive model chose filter sizes by considering both current profitability and previous forecasting success, where forecasting success is measured by a statistic derived from information theory. Alternative specifications of the historical information vector are employed to examine the sensitivity of various components of the model. Although the original adaptive model was a substantial improvement over existing models, it lacked stability; that is, there was more stability in the underlying information of the process than there was in the net profits produced by the adaptive model. Hence this further study. The information-based adaptive model we developed was run on a large data base composed of 15 years of soybean and wheat data. Our results allow us to reject the Martingale hypothesis of commodity price movement.