BERT-based Financial Sentiment Index and LSTM-based Stock Return Predictability

• DOI: 10.48550/arxiv.1906.09024
• Publication Date: 2019-01-01
• Authors: Joshua Zoen Git Hiew; Xin Huang; Hao Mou; Duan Li; Qi Wu; Yabo Xu
• Keywords: Predictability; Sentiment Analysis; Stock (firearms)

Traditional sentiment construction in finance relies heavily on the dictionary-based approach, with a few exceptions using simple machine learning techniques such as Naive Bayes classifier. While current literature has not yet invoked rapid advancement natural language processing, we construct this research textual-based index well-known pre-trained model BERT developed by Google, especially for three actively trading individual stocks Hong Kong market at same time hot discussion Weibo.com. On one hand, demonstrate significant enhancement of applying financial analysis when compared existing models. other combining two commonly-used methods it comes to building literature, i.e., option-implied and market-implied approaches, propose more general comprehensive framework analysis, further provide convincing outcomes predictability stock return LSTM (with feature nonlinear mapping). It is significantly distinct dominating econometric influence which are all nature linear regression.