Stock Price Predictability Based On Noise And Investor Sentiment

Noise trading theory is that noise traders in stock market trade based on noise but not signal, which results in the pricing bias from the situation where only exist rational traders. Researches on investor sentiment, developing from noise trading theory, is that investor sentiment influences the stock price by affecting investors’ trade preferences. According to the theories and research conclusions, this paper argues that noise and investor sentiment may contain additional information that the stock price does not contain. Based on this, this paper researches on whether and to what extent noise and investor sentiment influence the stock price predictability, whether and how noise and investor sentiment provide additional predictability.Firstly, this paper measures noise and investor sentiment in Chinese stock market. In aspects of measuring noise, this paper, referencing the variance ratio test method which tests the market efficiency, uses wild bootstrap variance ratio test and sliding window to build noise series. In aspects of measuring investor sentiment, this paper references the academically widely accepted Baker-Wugler investor sentiment to the build investor sentiment series. By reason of the suspension of the IPO in Chines stock market in the observation interval, this paper eliminate the variables containing IPO information in Baker-Wugler investor sentiment and build investor sentiment series by extracting the first principal component of SSE closed-end fund discount, average market capitalization-weighted turnover rate, average new accounts and average trading volume.Secondly, this paper predicts SSE Composite index and then build the stock price predictability series. The prediction methods used are fuzzy time series, BP neural network and hybrid forecast model of fuzzy time series and BP neural network. Specifically, this paper uses Chen’s first-order fuzzy time series model with one antecedent part to predict; uses BP neural network after choosing the training function, learning function and network structure to predict; uses hybrid forecast model by combining the predicted value via entropy evaluation method. This paper generates PI series that describes the predictive effect to build the stock price predictability.Thirdly, this paper researches on the stock price predictability based on noise and investor sentiment. The first step is to explore whether noise and investor sentiment significantly affect the stock price predictability. The empirical results show that noise and investor sentiment have significant positive linear impact on the stock price predictability under the significant levels0.10, which means a higher noise or investor sentiment can be a potential factor for a higher predictability. Though the conclusions do not come into existence under the significant levels0.05, they still have reference value. Furthermore, noise and investment sentiment have significant non-linear impact on the stock price predictability under the significant levels0.05, which is caused independently but not synergistically and can be seen as the adjustment of the linear relationship. Choosing different PI series will change the empirical result. The second step is to explore whether noise and investor sentiment provide additional predictability. By introducing noise and investor sentiment into forecast model, this paper concludes that noise and investor sentiment can improve the predictive effect, especially the BP neural network and the hybrid forecast model. Considering that BP neural network can describe the non-linear relationship in stock price series best, this paper compares the predictive effect of BP neural network before and after introducing noise and investor sentiment and finds that noise and investor significantly improves the predictability. The reason is that noise and investor sentiment contain additional information.