Research Based On Time Series And Neural Network Prediction Of Stock Index

Since China's Shanghai Stock Exchange and Shenzhen Stock Exchange been set at the end of the last century 80's early 90's,China's securities market has been development rapidly after many years of development.With economic development as well as the market matures,people pay more attention to the stock market and the stock has become an important means of investment.However,there too many types of stocks,and stock indices appear more accurate for people to grasp the overall development trend of the stock market.With stock index,investors can observe and analyze the development of the entire stock market dynamics,getting the formation of a comprehensive on the market to determine.With the purpose of grasping the situation in economic development,as well as reducing investment risk,the forecasting of stock market has always been a hot academic research topics-scholars at home and abroad have been making efforts to explore the inherent law of the stock market,and explore a variety of forecasting methods for the stock market.In this paper,drawing on domestic and foreign stock price index forecast,we select the corresponding.impact factor based on the characteristics of stock price index,combined with statistical forecasting methods and artificial intelligence forecasting methods,establishing a GARCH model and a BP neural network model.And make an empirical analysis of Shanghai Composite index based on the models we have already established.Firstly,we make ARCH effect analysis of the Shanghai Composite Index,and establish the GARCH(1,1) and the GARCH-M(1,1) model to analyze the volatility of its rate of return.The results show that the Peak thick tail and volatility clustering characteristics of Shanghai Composite Index, and its conditional variance sequence is "long memory" type with obvious characteristics of sustain,yield the conditional variance series is smooth,thus the sequence is predictable. Then we establish the BP neural network model which based on principal component analysis,through studying,training and forecast testing of large sample,finding the effect is good,and the results also prove the feasibility of the established forecasting model. The innovation of this paper is mainly reflected in the use of BP neural network model which based on principal component analysis.Artificial neural networks do not need accurate mathematical models and assumptions to forecast and it has an important role in many areas.However,too much variables may affect the accuracy of models, so we use principal component analysis to reduction the dimensions of factors we have chose,then put the extracted principal component factor into network model in order to get more accurate forecast results.Finally,by observing the line graph and calculate the mean square error,to further test the applicability and accuracy of this model.