Abstract: Based on the background of a typical stock market crash for Chinese stock market in the second half of 2015, this study selects 60 stocks in the Shanghai and Shenzhen 300 Index stocks as a sample through stratified sampling, and crawls the Sina Weibo texts related to the company for calculating the multi-category emotion in social media (or investor sentiment). This study evaluates the impact of investor sentiment on stock market during crash period. Based on the “Emotion-Cognition-Behavior” framework, the study mines the hidden state of “market cognition” for the stock market by using Hidden Markov Model (HMM), and then establishes an indirect relationship between investor sentiment and stock return. The study discovers that investor sentiment indirectly affects the stock market by the market cognition. To be more specific, positive sentiment plays a positive role in adjusting market cognition, while negative sentiment with high arousal (e.g. “Anger”) will worsen the market cognition toward the crash state. The results guide business operators to make scientific emotional counseling strategies to reduce the impact of investor sentiment in social media on the market during the crash period. In addition, this study also provides a new approach for business operators to explore the relationship between sentiment and stock market indicators by hidden states mining.

Key words: stock market crash, Hidden Markov Model, investor sentiment, ordered-logistic regression, social media