A Performance Comparison of a Technical Trading System with ARIMA and VAR Models for Soybean Complex Prices

Both technical trading systems and standard economic time series models are based upon the assumption that current market prices are not independent of past market behavior. This study examines the relative performance of a Channel (CHL) technical trading system with an Autoregressive Integrated Moving Average (ARIMA) model and a Vector Autoregressive (VAR) model in forecasting soybean, soybean meal, and soybean oil prices over the period January 1984-June 1988. ARIMA and VAR models are developed over the time period January 1974-December 1983 and then are used to forecast out-of-sample from January 1984 through June 1988. The CHL trading signals and out-of-sample two month ahead forecasts from the ARIMA and VAR models are used to take positions in the futures markets. The resulting trading returns are evaluated to determine the relative economic performance of the models within the soybean complex. Of these models, the CHL technical trading system exhibits consistent trading returns across the soybean complex. Furthermore, the CHL technical trading system is robust across the two subperiods of the out-of-sample period, one of which is characterized by rising commodity prices and the other by declining commodity prices. These results suggest that in the short run, regularities within a single price series can be used to forecast prices within the soybean complex. Further, technical trading system prove more useful in utilizing such regularities for forecasting than the autoregressive or moving average processes found in either ARIMA or VAR modeling techniques