Research On Stock Index Futures Price Forecasting Method With Fuzzy Time Series

In the real world, many data is incomplete, inaccurate and vague, especially financial data. Using the classic time series prediction will lead to a decline in the fitting effect. Futures market is more complex, with the stock index futures as the product of a modern capital market by various internal and external factors influence.Stock index futures closing price records may lose some useful information, so the introduction of fuzzy time series model, which is more efficient and accurate for forecasting are obtained. Consideration of available and complete data, this paper selects January 2011 to November 12, 4, to 2015, the stock index futures consecutive day closing price as the research object. Use classical time series ARIMA model to predict, according to three kinds of classical fuzzy time series model of the sample data to predict. On the basis of the improvement in the classical fuzzy time series model, calculate the membership degree matrix introduced non parameter Gauss kernel function and the optimal window width adjustment, change weights of each fuzzy set. Change the first order fuzzy time series to the second order, that is,considering T, T-1, T-2 period of membership of the data fuzzy sets. Calculate the relation matrix introduced Markov model. Relationship matrix is transformed to the transition probability matrix. And calculate the predicted values with the weighted method.Firstly, Make deep research in the fuzzy time series model, such as fuzzy set theory, fuzzy time series development present situation, research review, theoretical model, the ARIMA model theory, Markov chain model, non parameter of the Gaussian kernel function. According to the theoretical model, establish R language program, respectively with three kinds of classical fuzzy time series model and the improved fuzzy time series model to predict. And use Eviews of ARIMA model to forecast the final trading day of nearly a month forecast data. The song’s fuzzy time series model prediction error more than 2% are 20 samples, accounted for 83.33%;Lee’s fuzzy time series model prediction error more than 2% are 10 samples,accounted for 41.67%; ARIMA model error above 2% have four samples, accounted for 16.67%; second order Markov fuzzy prediction error more than 2% have two samples, only 8.33%. Proof the improved based on Markov chain of fuzzy time series model to predict more effectively, and better than the other four models. Fuzzy time series of development provides is a new way of thinking, and widens the fuzzy timeseries application direction.