Joint work with Karen Lai (SAFTI) and Diego Ronchetti (Audencia Business School)

Bio: Halis Sak is an Assistant Professor at Shenzhen Audencia Financial Technology Institute (SAFTI), WeBank Institute of Fintech, Shenzhen University. Before joining the institute, he was a Visiting Assistant Professor at the Department of Finance of HKUST. His research and teaching address machine learning, empirical asset pricing, risk management, and Monte Carlo simulation. His academic works appeared in journals such as Review of Finance, Quantitative Finance, and European Journal of Operational Research.

Abstract: We extend the shock-based textual analysis of Didisheim et al. (2026) to the Chinese A-share stock market, where retail investors dominate, short-sale constraints are binding, andlimited attention is pervasive. Using BGE-M3 embeddings for firm-month text over the2006–2023 sample, we compare the return predictability of news shocks, analyst-report shocks,and a combined signal that integrates both. Applying their news-shock methodology toCSMAR news data yields no robust return predictability, with only one of four specificationsmarginally significant at the 10% level. This finding is consistent with the possibility thatinvestors in the retail-dominated Chinese market face difficulties processing news text inreal time. However, when we apply the same shock framework to sell-side analyst reportsproduced by information intermediaries that digest and interpret news for less sophisticatedinvestors, we find substantial return predictability. Purifying analyst shocks by controllingfor firm characteristics and contemporaneous CSMAR news yields evidence consistent withanalysts adding interpretive value beyond merely repackaging public information, thoughthe design does not rule out residual private information or other unobserved sources ofanalyst-report novelty. Our findings are consistent with an important intermediary role foranalysts in translating public information into price-relevant signals, although the underlyinginterpretation mechanism is not uniquely identified.

Keywords: Analyst reports, news text, LLM, Chinese A-share market