Research On High Frequency Stock Forecasting Based On Investor Sentiment

Stock market forecasting has been considered a very important practical issue in the economic field,and many experts and scholars have introduced investor sentiment into stock market forecasting.The research on the high-frequency stock market can avoid risks more accurately and obtain returns.The investor sentiment index constructed from data such as financial news and related economic variables are all low-frequency data and cannot be applied to the prediction of high-frequency stock data.And due to the non-linear impact,lag and longterm dependence of high-frequency investor sentiment on stocks,it is difficult for traditional economic models to use it to predict.Therefore,this paper uses stockholders’ stock reviews to construct a high-frequency investor sentiment index,and proposes a high-frequency stock market sequence prediction model AEformer based on investor sentiment.This article selected the BERT model to classify the stock review data,constructed a minute-level high-frequency investor sentiment index,and used Granger causality test to verify it.Combining the BERT model with Autoformer through Asymmetric Embedding,a multimodal stock forecasting model AEformer based on investor sentiment and Asymmetric Embedding is proposed.Through the ablation experiment,the effect of each part of the Asymmetric Embedding layer on stock price prediction using investor sentiment is analyzed.Finally,the role of other Transformer class models of Asymmetric Embedding is analyzed.The experimental results show that the classification effect of the BERT model in this text multi-classification task is better than other models,and its accuracy,precision and other indicators have increased by more than 4%.In the high-frequency stock forecasting,adding investor sentiment can improve the prediction accuracy by an average of 10% compared with not adding investor sentiment forecasting.In all forecasting tasks using investor sentiment,the AEformer model has no obvious advantages over other models in the short term,but it is better than other models in medium and long-term forecasting.Asymmetric Embedding not only improves the AEformer model,but also plays an active role in other Transformer models.