Publications

Identifying market states is technically classifying the market into different groups. Accurate market states facilitate understanding of market dynamics. This study uses public information texts and employs a Markov Switching Vector Autoregression (MS-VAR) model to identify market states and analyze their impact on market volatility and returns. Market states capture the time-series characteristics of policy text semantics. During market state transitions, volatility decreases significantly, aligning with existing research and validating the method. Key findings include: 1. At the overall market, market states influence return volatility by moderating economic policy uncertainty, and return level through cash flow channel instead of discount rate channel. 2. At the industry level, market states affect volatility across sectors, with stronger effects in mining, hospitality, personal services, education, and healthcare, indicating higher policy sensitivity in these sectors. 3. This effect is more pronounced when policy texts contain more emotional language.

We propose a novel methodology for deriving investor sentiment from market transaction data, through an approach that models the actual decision-making process of investors. Our sentiment index outperforms the traditional Baker-Wurgler index in predicting returns, with less susceptibility to macroeconomic shifts. It enhances predictive power by forecasting cash flows, discount rates, and market volatility. The sentiment indices retain the advantages of pre-synthesis variables, resulting in superior forecasting accuracy. Notably, our indices contain information from Baker-Wurgler’s indices, indicating that micro-structural market data encompasses macroeconomic signals.