Constantin Charles, London School of Economics and Political Science
Pengfei Sui, Chinese University of Hong Kong-Shenzhen
Abstract: We propose a framework in which investors draw on similar past periods when forming their beliefs or deciding on their actions. Motivated by this framework, we construct a measure that assigns weights to past months based on their similarity to today. Using our measure, we construct similarity-based beliefs and show that they can explain return expectations from surveys as well as higher-moment beliefs implied by stock options and the VIX. We also show that similarity can explain the repurchasing decisions of individual investors. Our results show that similarity is a useful principle for understanding beliefs and actions in financial markets.
Abstract: Applying large language models to more than 46 million StockTwits posts, we distinguish messages expressing investors’ own outlook from those describing others’ expected actions and extract sentiment associated with these posts. This yields firm-week measures of own and subjective sentiment. We find that references to others are pervasive and typically more optimistic than investors’ own views. Retail order imbalance increases with own sentiment but decreases with subjective sentiment, indicating contrarian trading against beliefs about others. Subjective sentiment also positively predicts short-horizon returns, especially where contrarian retail trading is strongest, consistent with contrarian liquidity provision slowing price adjustment and generating return continuation.
Discussant: J. Anthony Cookson, Pennsylvania State University
Christian Heyerdahl-Larsen, BI Norwegian Business School
Zeshu Xu, BI Norwegian Business School
Abstract: Using survey data, we show that stock market participants are systematically more optimistic than nonparticipants, and that entry and exit coincide with substantial belief revisions. Motivated by these facts, we develop a model in which experience-based beliefs generate endogenous participation cycles. Inexperienced cohorts enter the market with leverage during periods of elevated valuations and exit following downturns, while experienced cohorts typically remain invested because the market-clearing risk premium gravitates toward their beliefs. These interactions generate procyclical participation, endogenous feedback between participation and learning, and welfare losses. When disciplined by realized shocks, the model reproduces fluctuations in participation, entry, and exit.
