An Improved Fuzzy Time Series Model And Its Application In Stock Index

Although there is no precise method for short-term forecasting of the stock market right now,this field always attracts people's attention,and many researchers are still working hard on more accurate forecasting methods.It is a more objective technical analysis method of stock price that using the statistical method to analyze the distribution of historical data and build a model to forecast stock price.At present,the research on the Chinese mainland stock market is mainly based on the classical time series methods.However,the classical time series methods often have strong data requirements which the stock price data are unable to meet.An alternative way is to use the fuzzy time series method.After 20 years of development,this method has been applied in many different fields,including the stock price analysis which is a popular research field of fuzzy time series.This thesis uses an improved fuzzy time series method to analyze Shanghai Composite Index of China.There are four main steps to establish fuzzy time series model:define and divide the universe of discourse,define fuzzy sets and fuzzify the historical data,establish the fuzzy relations,and defuzzification.Based on the previous work,this thesis improves the original methods according to the characteristics of the stock index studied here.First,a new partition method for the universe of discourse is given based on Gaussian mixture clustering model.Second,Gaussian membership function is used to fuzzy the historical data.Then,according to the results of Gaussian membership function,a more flexible defuzzification method with an adjustable parameter is given.This defuzzification method breaks through the limitation of the existing methods,which often only consider the fuzzy intervals' midpoints when carry out the defuzzification procedure.Finally,in the prediction of the testing data,a method for establishing dynamic fuzzy relations is given based on kernel function.Using the given method above,the stock index of Shanghai Composite Index are analyzed.The advantages and disadvantages of the given method are evaluated according to the prediction results.According to the characteristics of technical analysis and the news of the international market,the reasons for the prediction error are also analyzed.From the prediction results,the trend of prediction obtained by the given method is consistent with that of the stock index.Even when the stock index fluctuates violently,the given method still guarantees high prediction accuracy.For the unobserved stock index in the test set,when there is no new fuzzy relation,the given method also has a good prediction ability.Compared with the existing fuzzy time series methods,the improved method presented in this thesis has lower prediction error.