A Study On The Predictability Of Securities Price Based On Investor Sentiment And NARX Dynamic Neural Network

The financial securities market is full of uncertainties,and modern financial theory is based on the assumption of this stochastic uncertainty.The securities market is a dynamic,non-linear and ambiguous complex system with rapidly changing.But the traditional linear time series forecasting model can not adapt to the fuzzy uncertainty of financial market.In recent years,with the development of computer technology and behavioral finance,investor behavior research base on psychology and artificial neural network research based on artificial intelligence technology has been developed rapidly in theory and practice.In this study,firstly,we introduce the investor sentiment of behavioral finance,which reflect the irrational factors of investors in the securities market.We construct the emotional index which can measure the investor sentiment to a certain extent.We select five variables as proxy indicator of investor sentiment,that are A-share market turnover(TURN),A-share volume(VOL),consumer confidence index(CCI),A-share market active account number(ASA)and Baidu Index(BDI).And we construct the investor sentiment index(SENT)by referring to Baker and Wurgle's method of constructing BW sentiment index.Secondly,this paper introduces the NARX dynamic neural network forecasting model,and creatively adds the investor sentiment to the neural network model as an input to the neural network.Finally,this paper constructs the NARX dynamic neural network prediction model based on investor sentiment.Dynamic neural network based on the investor sentiment is the combination of the irrational factors of the investor sentiment and the neural network,which has strong adaptive learning ability.This model is applied to the forecast of the securities market price.The empirical results show that the accuracy of price prediction is superior to the traditional linear ARIMA model.