Stock price movement prediction based on Stocktwits investor sentiment using FinBERT and ensemble SVM

• DOI: 10.7717/peerj-cs.1403
• Publication Date: 2023-06-07T08:20:37Z
• Authors: Jin-Xian Liu; Jenq-Shiou Leu; Stefan Holst
• Keywords: Sentiment Analysis; Stock (firearms); Ensemble forecasting

Investor sentiment plays a crucial role in the stock market, and recent years, numerous studies have aimed to predict future prices by analyzing market obtained from social media or news. This study investigates use of investor media, with focus on Stocktwits, platform for investors. However, using Stocktwits price movements may be challenging due lack user-initiated data limitations existing analyzers, which inaccurately classify neutral comments. To overcome these challenges, this proposes an alternative approach FinBERT, pre-trained language model specifically designed analyze financial text. ensemble support vector machine improving accuracy movement predictions. Then, it predicts SPDR S&P 500 Index Exchange Traded Funds rolling window prevent look-ahead bias. Through comparing various techniques generating sentiment, our results show that FinBERT analysis yields best results, F1-score is 4-5% higher than other techniques. Additionally, proposed improves predictions when compared original series experiments.