A hybrid model of dynamic time wrapping and hidden Markov model for forecasting and trading in crude oil market

Abstract

In this study, a hybrid model of hidden Markov model (HMM) and dynamic time wrapping (DTW) is proposed to predict the return of crude oil price movements and trading. First, three indicators are used as inputs of HMM to determine the market state for each month; next, DTW algorithm is applied to match similar price sequences which have the same market state in historical time series, and then to calculate expected returns; Finally, it forecasts the crude oil spot price direction and executes related simulation trading. For design of the trading strategy, we adopt different parameters such as trading thresholds and position-closing thresholds for each market state, and the particle swarm optimization algorithm is applied for parameter optimization of our trading strategy. In experiments, the proposed method is applied for direction forecasting and simulation trading of WTI and Brent crude oil market. Experimental results show that the proposed method yielded the best forecasting and trading performances in average. For instance, in the WTI market, the proposed method produced a hit ratio of about 62.74% and a yield of 34.3% profit per year, and a Sharpe ratio value of 2.274. Furthermore, experimental results of the proposed method were significantly superior to other benchmark methods, demonstrating that the proposed method is not only good at direction prediction and profit making, but also return/risk ratio.