Incorporating textual network improves Chinese stock market analysis

This study adopts the textual network to describe the coordination among the interplay of words, where nodes represent words and nodes are connected if the corresponding words have co-occurrence pattern across documents. To study stock movements, we further proposed the sparse laplacian shrinkage logistic model (SLS_L) which can properly take into account the network connectivity structure. By using this approach, we investigated the relationship between Shenwan index and analysts' research reports. The securities analysts’ research reports are crawled by a famous financial website in China: EastMoney, and are then parsed into time-series textual data. The empirical results show that the proposed SLS_L model outperforms alternatives including Lasso-Logistics (L_L) and MCP-Logistic (MCP_L) models by having better prediction performance. Besides, we search published literature and find the identified keywords with more lucid interpretations. Our study unveils some interesting findings that the efficient use of textual network is important to improve the predictive power as well as the semantic interpretability in stock market analysis.