Leveraging Natural Language Processing on Chinese Broker Research Reports for Stock Selection

With the explosive growth of text documents and the huge interest in text-mining techniques, investors are increasingly looking for natural language processing (NLP)-based features to provide an edge in stock selection. Although there has been an abundance of research on English documents, other languages have not seen the same attention. The authors apply these techniques to Chinese documents to generate features for stock selection strategies in the China A-shares market. They first introduce a novel dataset of Chinese broker research reports. The authors then provide an overview of the challenges in processing Chinese text data and detail the text-mining techniques they use to extract meaningful insights to separate outperformers from underperformers. They find that NLP models show significant promise in the Chinese market, with various types of features carrying impactful information. Indeed, this article suggests that sentiment and complexity features are predictive of future stock performance. The authors also find that context-aware approaches using BERT-based models are superior to bag-of-words approaches. Finally, they show that combinations of complexity and sentiment features can yield even stronger results.