Using Markov-switching models with Markov chain Monte Carlo inference methods in agricultural commodities trading

Abstract

In this work, the use of Markov-switching GARCH (MS-GARCH) models is tested in an active trading algorithm for corn and soybean future markets. By assuming that a given investor lives in a two-regime world (with low- and high-volatility time periods), a trading algorithm was simulated (from January 2000 to March 2019), which helped the investor to forecast the probability of being in the high-volatility regime at t ? 1. Once this probability was known, the investor could decide to invest either in commodities, during low-volatility periods or in the 3-month US Treasury bills, during high-volatility periods. Our results suggest that the Gaussian MS-GARCH model is the most appropriate to generate alpha or extra returns (from a passive investment strategy) in the corn market and the t-Student MS-GARCH is the best one for soybean trading.