A BERT-Based Vector Autoregressive Network for Sentiment Analysis of Financial News

Stock market forecasting is a difficult problem in the field of financial analysis. The intrinsic information contained in financial news has a great impact on the stock market performance. In this paper, we propose a BERT-based vector autoregressive network (BVANet), which quantifies financial news sentiment by BERT and then combines it with market performance to construct a financial time series vector autoregressive (VAR) model to achieve stock prediction eventually. The results show that BVANet has improved results in extracting news sentiment information and model prediction compared with traditional algorithms, and the sentiment of news has predictive effect on market performance. This study can provide a practical reference for the application of natural language processing in financial prediction.