Research On Futures Price Forecast And Quantitative Strategy Based On HMM

Security price forecasting is an important research issue in quantitative investment,with the development of computational algorithms and computing power,machine learning has been used in the field of quantitative investment,which provides a new solution to the non-stationary and nonlinear time series prediction problem.Hidden Markov Model(HMM)is an important time series analysis method in machine learning,which can reflects the characteristics of the dynamic change of time series.At the same time,the HMM model requires fewer training parameters,which can effectively reduces the overfitting problem of the model,reduces the amount of calculation when training model,it has many advantages to deal with time series problems.The paper trains the price prediction model of gold futures based on HMM,and constructs a quantitative strategy of gold futures for backtesting.Firstly,the paper uses the forward-backward algorithm in the HMM model to perform pattern recognition on the time series,finds the sequence in the historical sequence most similar to the current sequence,and uses it as an approximate replacement of the current sequence,and then obtains the predicted price of the current sequence.Secondly,aiming at the stable linear relationship between the close price of the gold futures main contract and the secondary main contract,which is the characteristic of the cointegration,the cointegration and the HMM model are combined to construct a cointegrate-HMM model.Gold futures main contract and the secondary main contract can respectively obtain a predicted price for future directly through the HMM model,and the predicted price of one contract is converted to the predicted price of the other contract through the cointegration relationship,each contract can get two predicted prices,the two predicted prices are averaged to obtain a more accurate predicted price,which can increases the accuracy of model prediction,and then the paper constructs a quantitative trading strategy for gold futures based on the cointegrate-HMM model.Lastly,in view of the fact that there is much noise in the price sequence,the paper uses the Butterworth filter to decrease noise,and the noise-reduced sequence is backtested using the cointegrate-HMM model,which can reduces the number of incorrect opening of the strategy,decreases the volatility of the strategy,and increases the capability of the strategy to obtain stable returns.The research result of this paper shows that: firstly,the HMM model has a high accuracy in predicting the future rise and fall of gold futures contracts.After applying the model to the quantitative trading,the strategy backtest results are better than the benchmark.The HMM model has a good effect in quantitative trading.Secondly,combining the cointegration theory with the HMM model can significantly improves the accuracy of model predictions and enhances the ability to obtain profit and avoid risks of the strategy.Thirdly,after the noise reduction of the Butterworth filter,the strategy filters out many invalid transactions,improves the winning rate of the strategy,and enhances the ability of the strategy to obtain long-term stable returns.