Neural Network Quantitative Stock Selection Strategy Based On Investor Sentiment Indicators

According to the noise trading theory in behavioral finance,investors’ emotions have a significant impact on their trading behavior,so traders can make use of emotional information to obtain excess returns.At the same time,with the continuous development of the Internet stock community,many individual investors post comments on the Internet stock review platform.These comments can be regarded as the expectation and evaluation of the future trend of the stock,and can be used as a suitable indicator to measure investor sentiment.This paper mainly studies the comments of investors on Snowball.com.After that,the cross-day trend forecast is constructed based on GRU and residual network based on the stock price data,and the trading strategy is to invest from the stocks predicted to”rise”.The model with investor sentiment indicator is compared with the model without investor sentiment indicator.Finally,according to the backtest results,it shows that:(1)after fine-tuning the Bert model with manually annotated Snowball comments,the model has better performance in terms of accuracy for the sentiment factors constructed;(2)After comparing the strategies with and without sentiment indicators,it is found that the addition of investor sentiment indicators can make the whole strategy perform better.This paper mainly has the following two innovations: First,in terms of Bert model pre-training,the training set used in this paper comes from Snowball news dichotomous sentiment data set and Weibo financial comment dichotomous data,which is closer to China’s language system and financial related corpus.On this basis,this paper also uses manual data labeling,and the accuracy of the improved model has been qualitatively improved.Therefore,this paper uses Bert model to extend the research on sentiment classification of finance-related reviews.Second,in terms of strategy research,this paper builds a model to predict the cross-day trend of stock prices based on the investor sentiment index built by the comments of investors on Snowball and the stock price data,and carries out the strategy of stock selection and investment with the predicted”rising” stock.