Futures Price Forecast Based On Corrected HMM

As a significant component of financial system,futures market,together with stock market and foreign exchange market,constitutes a relatively complete financial market.In recent years,China’s financial market has experienced ups and downs,and volatility is fierce.As an important financial tool for price discovery,hedging and risk aversion,stock index futures play an increasingly prominent role in the financial market.Therefore,reasonable prediction of stock index futures prices can better help investors avoid risks and adjust investment strategies.With the continuous expansion of the scale of China’s futures market and its coverage of more and more fields,the futures market has become volatile and constantly presents many complex phenomena that cannot be explained by classical financial theories.It is difficult to capture the price change characteristics of stock index futures using traditional classical time series models.Although traditional machine learning methods can effectively process nonlinear data,they are prone to over fitting when the data volume is small,and the models built are generally complex,and they also face the problem of optimizing a large number of parameters.In this context,this paper takes the IF300 stock index futures price data as the research object,appropriately modifies the traditional Hidden Markov Model,proposes an improved Hidden Markov Model,which provides investors and regulators with reference for investment and supervision.The paper’s research content principally includes the processing of raw data,the determination of the number of hidden states of the model,the estimation of model parameters,empirical analysis and the comparative analysis of the accuracy of other model prediction results.① In the aspect of original data processing,we preprocess the four-dimensional data of the opening price,closing price,high price and lowest price of the selected samples,construct three-dimensional relative number series such as price volatility,and use the volatility as the input variable of the model after noise removal;② In order to explore the impact of sample size on the prediction accuracy of the model,this paper divides the sample data into training set and test set according to three proportions of 7:3,8:2 and 9:1;③ Determine the number of application hidden states that best fit the AIC and BIC criteria;④ Through Baum Welch algorithm,the test set is trained to obtain the parameters of the model λ=(A,μ,σ,π);⑤In terms of futures price prediction,the prediction method of futures price has been improved.In the prediction method,the state transition probability is used as the weight to calculate the probability distribution of the observed value,and the volatility interval is divided into equal widths.The probability of the observed value in each volatility interval is calculated.The median of the maximum probability volatility interval is taken as the prediction value of the volatility at the next time,The predicted value of volatility is converted into corresponding price data through the price conversion formula;⑥ In terms of comparison of prediction results of other models,traditional HMM,ARIMA and LSTM are used to predict futures prices,and MAPE is used as an evaluation index to compare and analyze the prediction accuracy of each model to verify the accuracy of the improved HMM model in stock index futures price prediction.The empirical results show that the improved HMM is superior to the traditional HMM in forecasting futures prices.Specifically,for the 1-day weighting,5-day weighting,20-day weighting and 50 day weighting of traditional HMM,the prediction accuracy of improved HMM has increased by 73.71%,64.42%,47.44%and 59.99%respectively.At the same time,compared with the prediction accuracy of ARIMA model and LSTM model,the improved HMM accuracy has increased by 91.96%and 40.65%,indicating that the improved HMM model proposed in this paper can improve the prediction accuracy to a certain extent,which has certain reference significance for investment decision-making.The possible innovations of this study mainly include:① The original data of futures prices are used to construct corresponding volatility indicators.When there is no information loss,the data dimension is reduced from four dimensions to three dimensions,which reduces the model’s complexity,and the volatility series has a smaller and more stable range of change compared with the original price series.The volatility is used for model parameter estimation after eliminating the influence of accidental factors through noise removal;② Instead of predicting the future by finding similar patterns of likelihood values in the historical data,the prediction method Instead,the probability distribution of the observations is recalculated by using the state shift probability values as weights,which avoids the situation that the prediction effect of historical matching similar patterns is greatly reduced when the predicted data is beyond the range of historical data;③ The robustness of the median is better than the average.When forecasting the volatility,the median of the maximum probability volatility interval is selected as the volatility value at the next time.