Publications

We show that stock returns exhibit predictable patterns before the publication of anomaly trading signals. Moreover, anomaly trading signals derived from financial data are themselves predictable, making it possible to trade before financial statements are released. A trading strategy based on predicted anomaly signals earns an annualized return of 2.80% in the quarter before the signal is released. In recent periods, this return predictability is concentrated in signals that are harder to forecast, and returns are increasingly earned several quarters before signals are released. Our findings suggest anomalies are more anomalous than previously recognized.

We examine the timing of returns around the publication of anomaly trading signals. Using a database that measures when information is first publicly released, we show that anomaly returns are concentrated in the first month after information release dates, and these returns decay soon thereafter. We also show that the academic convention of forming portfolios in June underestimates predictability because it uses stale information, which makes some anomalies appear insignificant. In contrast, we show many anomalies do predict returns if portfolios are formed immediately after information releases. Finally, we develop guidance on forming portfolios without using stale information.

Despite assumptions of mean-variance efficiency that underlie most asset pricing models, investors have shown a penchant for positive skewness. This study documents that the ratio of call option volume relative to total option volume is greatest for stocks with return distributions that resemble lotteries. These results suggest that investors’ preferences for lottery-type stocks might be reflected in the level of call option volume. Perhaps, more importantly, we test whether these preferences affect future spot price volatility. Consistent with our expectation, we find that preferences for lotteries by call option traders directly affect future volatility in the underlying asset.

We infer firm connections from hedge funds' co-search behavior in EDGAR and find that connected firms exhibit strong return predictability: annual alpha is 8.16%. Moreover, this effect is strongest for firms with complex disclosures and sparse analyst coverage, consistent with models of limited attention. We also show that hedge funds frequently churn their network of firm connections---responding to news and capturing fleeting ties among firms---and actively trade based on these connections. These findings highlight a distinct view of firm connectedness and uncover a novel channel of information flow that hedge funds translate into consistent alpha.

Using mutual fund holdings linked to EDGAR requests, we study how fund families research long and short positions. Among families with observable research activity, short positions outperform comparable longs by roughly 100 basis points per quarter and receive about 10% more EDGAR requests. We develop a model of information acquisition with short-side frictions that rationalizes greater research intensity and higher returns, and a sequential-sampling extension predicts, paradoxically, that more-researched shorts can earn smaller absolute returns. The evidence is consistent with this inverse relation: attention concentrates on borderline shorts, while clear winners require less research.

Factor models conventionally form portfolios annually using stale financial statements released months earlier. We construct fresh factors each month using the most recent annual financial statements. Fresh factors modestly improve pricing, but augmenting the stale model with the change in returns from updating stale portfolio assignments significantly improves pricing because the two components carry different pricing content. This reallocates abnormal returns, reverses the interpretation of thousands of event-study CARs, and reclassifies 14.7% of top-decile mutual funds. These findings reveal a fundamental structure in returns: a slow-moving component reflecting persistent fundamental risk and a fast-moving component capturing the arrival of information.

Identity-linked portfolio tilts in delegated management can reflect taste-based affinity or a comparative advantage in monitoring. Using gender alignment between U.S. mutual fund managers and CEOs, we show female-managed funds overweight female-CEO firms by 0.13 percentage points of TNA (about 5% of mean exposure), a tilt that rises after female-manager appointments. The taste-based channel predicts ESG co-tilts and return sacrifice; we find neither. Instead, the tilt is strongest for geographically proximate firms and is accompanied by fewer SEC filing downloads despite larger positions, consistent with substitution away from public-information acquisition toward a monitoring-based comparative advantage.

We develop a novel measure of effort based on unique observations of weekend work activity and revisit fundamental questions of asset management: how does effort relate to incentives, and how does effort affect performance? Mutual fund managers facing competitive incentives work more weekends, particularly following outflows and increased tracking error. Also, heightened effort predicts improved future returns -- especially among funds with competitive incentives, high active share, and low turnover -- and using exogenous variation in local weather we demonstrate a causal link between effort and returns. Finally, we show that the outperformance arises from a mixture of effort, targeted firm-research, and selectively trading the right stocks.

Social media affect financial markets, but public posts by financial media personas are voluntary disclosures. What is not disclosed is therefore usually unobserved. We address this measurement problem by conducting repeated, real-time interviews of "digital twins" built from monitored finfluencers' X accounts under a fixed protocol. The interviews recover stocklevel public-persona belief proxies even when no public recommendation is made. Because the interviews are generated and archived before the relevant return windows, the design avoids the look-ahead bias that arises when LLMs are queried ex post. The evidence shows that information obtained from these digital-twin interviews predicts the cross section of large-cap stock returns in the expected direction. Repeated real-time interviews therefore show how selective disclosure can be turned into measurable panels of market views.