Stock Price Index Prediction Based On Sequence Approximate Matching In Fuzzy Time Series

With the continuous development and perfection of China's stock market system,the proposal of stock price index satisfies the objective requirements of reformation and opening of China's capital market,and also forms a "barometer" reflecting the development of real economy and finance.Through the study of stock price index,the understanding of China's overall economic development is deepened from the capital market aspect.This thesis mainly uses fuzzy time series model to forecast stock price index.Ever since being proposed in 1993,this model has drawn considerable attention for its capacity in dealing with the fuzzy part inside time series which is hard to observe or quantify,and now is widely applied in the aspects of prediction of enrollments,temperature and stock price index.After developing for over twenty years,under the initial analytical framework,many researchers have improved the origin model in data pre-processing,interval split-up,fuzzy relationship formation,and so on.These improvements target on improving prediction accuracy,or lowering computing complexity.Based on the basic idea of using training set to form fuzzy logic and test set to forecast,this thesis has used three methods to spilt up intervals:automatic clustering algorithm,k-means algorithm,FCM algorithm,and compared their pros and cons through prediction accuracy in interval split-up part.In fuzzy logic formation part,this thesis creatively proposes a new sequence approximate matching algorithm based on longest common sub-sequence algorithm.By searching for the best approximate matching sequence of test set in training set,a better prediction of test set data is attained,and the superiority in fuzzy time series model prediction of the new algorithm is also proven.