Working Papers
(with Costas Arkolakis, Dominic Russel, Claire Shi, and Hyunjoo Yang)
Submitted to the QJE, November 2026
As young adults in advanced economies have become increasingly pessimistic about their economic prospects in recent years, gambling-like financial activities have proliferated. We link these trends through a Friedman-Savage (1948) motive: when life goals such as homeownership or marriage seem out of reach, individuals turn to financial gambles. We test the theory using an original survey combined with millions of linked records from Korea’s leading credit bureau, telecom provider, and credit card issuer. Consistent with the theory, individuals who place greater importance on hard-to-reach financial goals are more likely to exhibit risk-seeking preferences and to engage in high-risk investment. Conversely, when goals become attainable, risk-taking declines: exploiting Korea's housing lottery, which quasi-randomly allocates subsidies for home purchase, we show that attaining homeownership reduces interest in cryptocurrency. We conclude with a stylized calibration that shows how goal-driven risk preferences can amplify the welfare costs of overoptimism.
(with John Conlon)
Reject and Resubmit, QJE
Beliefs often depend on contextual cues that influence which states come to mind. Models of associative sampling capturing this idea give a central role to similarity, which is typically unobserved and in principle very high-dimensional. We show theoretically that, despite this potential complexity, the effects of cues depend only on two sufficient statistics—representativeness and resonance—that summarize how similarity distorts beliefs. Further, these statistics are measurable with simple, pairwise similarity judgments, offering a straightforward method of generating and testing quantitative predictions. In a controlled experiment, we deploy this method and show that belief responses to normatively irrelevant cues closely track model predictions. We also show that large language models can approximate human similarity judgments and extend the approach to field settings. Applying the framework to survey data on personal experiences and macroeconomic expectations, we show that similarity-based predictions quantitatively explain much of the heterogeneity in experience effects across a wide range of experiences and expectation domains.
Investor Composition and Overreaction
(with Michael Blank and Johnny Tang)
Do stock price run-ups predictably revert? We develop a model of financial markets with two types of investors: rational investors and "oversensitive" investors who react excessively to salient public news. The model yields a summary statistic for the degree to which a stock price has overreacted to news: the gap in holdings between oversensitive and rational investors. We compute this measure empirically using quarterly institutional holdings data. We first measure each investor's news sensitivity using their tendency to purchase stocks that have experienced positive earnings announcements. Consistent with our model's premise, we find that news sensitivity is a persistent investor characteristic. We next aggregate our investor-level measure to the stock level to compute the asset-level holdings gap between oversensitive and rational investors. A larger holdings gap forecasts less continuation in stock prices and greater reversals in the long-run, especially for extreme price run-ups. Furthermore, our holdings gap aggregates several distinct channels of overreaction, including both price extrapolation and overreaction to non-price information.
Communication under Disagreement
(with Adi Sunderam)
New draft, November 7, 2026
Why do communicators often choose to be vague rather than clear? Why do central banks, firms, and governments form committees with known internal disagreement, and at times issue deliberately ambiguous statements? This paper develops a theory of communication under disagreement, in which a benevolent sender and a receiver hold different beliefs about the state of the world. Unlike standard models, where communication fully reveals information, we show that disagreement allows for sustained ambiguity. The key mechanism is strategic pooling: senders obscure their true views either by mixing messages across different states of the world or by speaking through committees composed of diverse members. Ambiguity arises for two distinct reasons: caution — when the sender agrees with the receiver but is less certain and seeks to temper overconfidence — and contrarianism — when the sender disagrees with the receiver and wants to soften the reaction to an unpopular action. In both cases, ambiguity helps coordinate more moderate responses from the receiver. The framework sheds light on real-world communication practices, including central bank rhetoric, corporate messaging in speculative environments, and political strategies around unpopular reforms.
Models, Not Forecasts: Mental Models and the Belief–Action Gap
(with Hongqi Liu, Cameron Peng, and Tianyu Zhang)
Expected returns are central to portfolio choice, yet subjective expected returns weakly predict when investors trade. We argue that the missing link is investors’ mental models of what drives returns: people act on qualitative internal models of the market that numerical expected returns capture only noisily. We compare two measures of investor beliefs: directly elicited numerical expected returns and a mental-model measure of beliefs that summarizes investors’ views on what they believe are the most important drivers of stock market returns. In our sample, a one-standard-deviation increase in mental-model beliefs predicts a 32.6 percent higher probability of being a net buyer in the next trading session. Subjective expected returns, directional forecasts, and even intended actions are much less predictive. The elicited mental models also predict which stocks investors trade. Controlling for overall investor activity and stock popularity, investors direct more net buying toward firms whose business exposures align with the drivers they are optimistic about. A simple framework in which numerical expected returns are constructed from more primitive mental models rationalizes these findings. The apparent belief–action gap may therefore be a measurement gap: standard elicitations of subjective expected returns may introduce noise into the measurement of the underlying beliefs and mental models that guide action.
Paying for Convexity: Risk Premia and Retail Option Losses
Authors: Giyoung Kwon (University of Chicago); Spencer Yongwook Kwon (Brown University); Jaesang Lee (University of Chicago); Seung Hyeong Lee (Northwestern University); Haesung Kim (Korea Exchange).