Text-Based Industry Classification Based on Chinese A Share Annual Reports

Over the past decades, industry classification has been the focus of research due to its potential to be used in industry rotation trading strategies, bringing the return to beat the market. However, the research on Chinese A-shares is fairly scarce, and many of them rely heavily on human interference, leading to difficulties in replicating the experiments to classify firms. This paper proposed a text-based classification method to tackle this issue, which can be used to automatically classify firms. To be specific, we employ hierarchical clustering on firm annual reports combined with financial indicators to classify. The experiment results demonstrated that the proposed method achieved an average R2 of 0.303 by regressing different financial data of the stocks in an industry on the industry mean. By sharp contrast, the commonly used CITIC classification obtained an average of 0.154 only.