Working Papers

Portfolio Choice with Non-Fungible Brokerage Cash  (SSRN) 

with Ning Zhu

Revise and Resubmit, Journal of Financial and Quantitative Analysis

Investors perceive cash in savings accounts (“cold cash”) differently from cash recycled in brokerage accounts (“hot cash”). We propose a novel “temperature” framework for dynamic and repeated mental account refreshing process. We find individual investors buy stocks more cautiously with colder cash. Exploiting a quasi-natural experiment of the Chinese IPO-lottery reform, we show the effect is causal. An online experiment indicates the differential loss aversion is a potential channel. Building on observational and experimental findings, we propose a model featuring preferences with temperature-dependent sensitivity to future gains-and-losses, which generates our empirical patterns and provides a cash-temperature perspective for other puzzles.

Risk Revisited (SSRN) 

with Yucheng Liang

This study explores how stock price path shapes investors’ risk perceptions. In four experiments, we present participants with real and fabricated price charts and elicit risk perceptions. We document that a parsimonious set of three features – recency, clustering, and sign – can explain large proportions of risk perception variations. We find that the effects are partially mediated by perceived volatility, which is not fully explained by Bayesian inference. Additionally, a significant portion of the effects occurs independently of perceived volatility. Using observational data, we show that these three features explain cross-sectional variations in stock returns and predict mutual fund negative flows.

Dual Beliefs: Subjective Learning of Trading Talent (SSRN)

This paper studies how investors evaluate themselves and its implications. In my model, investors form subjective beliefs about both the stock currently held in portfolio and their trading talent and update their beliefs through learning with fading memory. I calibrate the memory decay parameters to account-level trading records and show that beliefs updating for talent is about 7 times more sensitive to return signals than that for the stock in portfolio. Consequently, the model predicts that stock replacing typically happens after a good performance of the existing stock, providing a dual learning perspective on the disposition effect. This framework also accounts for the widely documented performance-contingent trading intensity and investor attrition, which cannot be reconciled with the decreasing-gain property of the standard Bayesian learning.