FinBERT and LSTM-based novel model for stock price prediction using technical indicators and financial news

Abstract

Stock price movement is highly nonlinear, volatile, and complex. Traditional machine learning techniques are employed by researchers for stock price prediction, but due to shallow architecture, adequate accuracy is not achieved. In this paper, recently introduced bidirectional encoder representations from transformers (BERT) and long short-term memory (LSTM) hybrid model is utilised for stock price prediction. BERT model is used for financial news sentiment analysis. The sentiment score is merged with technical indicators of stock prices. In our approach, FinBERT is used which is specifically trained on financial corpus. Stock market prices were highly fluctuated in March 2020 due to lockdown. Thus, it is essential to predict high variation which existing works have not experienced due to lack of availability of highly fluctuated dataset. In our approach, the effect of financial news on stock indexes is analysed. Experiment analysis validates that our proposed approach outperforms existing approaches significantly.