Research On Commodity Futures Based On Hidden Markov Model

After more than two decades of development and exploration,from chaos to maturity,China's commodity futures market has stepped into a benign track which features sound development and highlights economic functions increasingly.Whether in academia or industry,people are sinking a quantitative research method with a high degree of accuracy to forecast the movements of commodity futures,and conduct transactions in proper timings to make more profits.From a financial engineering perspective,this paper applies the Hidden Markov Model(HMM)to China's commodity furthers market to predict prices.HHM is used to describe a Markov process which contains hidden parameters.The state of it couldn't be observed directly,but it could be observed by vector sequences.The method-introducing part of this pater mainly introduces the basic principle of HHM,three typical problems and relative algorithm.In the empirical study part,this paper introduces some common quantitative trading-strategies in commodity futures market and their application situations.Through the analyses,this paper first selects several well-performed investments from the whole commodity futures market,and set several technical-analysis indices into the input observation vector groups of HMM.Then,the rising,falling and volatile situations are respectively analyzed,and the data,in-sample and out-of-sample,are also respectively analyzed with different hidden states numbers.Finally,this paper presents a trading-strategy of commodity futures main contract based on single-day predicting signals given by a HMM with given hidden states numbers,and calculates the cumulative earnings of model's simulative trading,therefore the practicability of HMM for the quantitative trading in China's commodity futures are discussed.This paper finds out that the hidden state of rising trend can be better identified when five technical analysis indices selected from the soybean meal futures market are set into the observation vectors in HMM,while the ability of identifying hidden state of falling trend is slightly weaker.Meanwhile,when five hidden states are chosen,the predicting result of out-of-sample for this model is better.The rolling predicting trading strategy based on HMM with five hidden states can help to gain better profits.