A large number of stock characteristics have been found to predict the cross-section of returns. Return predictability can be driven by risk or mispricing, and the nature of most return predictors remains an open question. I use analysts’ earnings forecasts to determine if a return predictor is linked to mispricing. I find that at least 40% of return predictors from a dataset of 172 significant predictors are related to mispricing, including the momentum predictor from the Carhart four-factor and the profitability and investment predictors from the Fama–French five-factor model. I further study whether the mispricing predictors’ abnormal returns capture the divergence of prices from the fundamental value (build-up predictors) or their convergence back to the fundamental value (resolution predictors). Build-up predictors are less common than resolution predictors but do exist, implying that trading on certain return predictors can exacerbate rather than eliminate mispricing. Momentum is related both to the build-up and the resolution of mispricing.

We study in a highly abstract laboratory setting whether and how subjects in the role of shareholders use the stock market to monopolize product markets. We find that shareholders holding stakes in product market rivals choose compensation packages for subjects in the role of managers that reward the latter to reduce production. Many managers act in accordance with their incentives, thus raising prices and firm profits. Although the experimental environment features no risk, most shareholders actively choose to diversify their portfolio across competitors, which gives them subsequent incentives to incentivize rival firms’ managers to act as part of a monopoly.

In social dilemmas, cooperation failures often arise due to the absence of mechanisms that prevent free-riding and enhance cooperation. Given the critical role these mechanisms play in sustaining cooperation, why are they so frequently missing? To explore this, we conducted an online experiment testing whether individuals choose to implement such cooperation-inducing mechanisms and why they might refrain from doing so. Participants were introduced to the rules of two public goods games, one of which includes a cooperation-inducing mechanism, while the other does not. Regarding the likelihood of successful cooperation, we found that participants were overly optimistic in the absence of the mechanism and overly pessimistic in its presence. As a result, a majority of subjects preferred the game without the cooperation-inducing mechanism. However, when we corrected participants’ beliefs about the actual payoffs obtained in the two games, a majority shifted their preference toward the game with the cooperation-inducing mechanisms in place.

Motivated by a model linking time-varying expectation bias to anomaly returns, we construct a return predictor from the gap between analysts’ earnings growth forecasts and unbiased machine-learning forecasts. Across 179 anomalies, our predictor outperforms the historical-mean benchmark in out-of-sample tests for up to 30% of cases at investment horizons beyond twelve months. Focusing on anomalies that exceed benchmark performance in a validation sample further improves performance. Consistent with our model, the persistence of portfolio-level bias determines whether an anomaly reflects the build-up or resolution of mispricing. A subset of anomalies shifts between these classifications, challenging static taxonomies.

Benchmarks are commonly included in charts depicting an asset’s past returns, and this is mandatory in certain regulated documents in the US and the EU. However, there is little evidence of the impact of benchmarks on retail investors. We hypothesize that the provision of uninformative benchmarks may bias investors through a contrast effect. This effect predicts that investors will have lower (higher) return expectations and invest less (more) in a fund if its returns are shown together with a benchmark that outperformed (underperformed) the fund relative to a baseline with no benchmark. We test this in an experiment and find that benchmarks affect investors’ return expectations and propensity to invest. The results show that benchmarks that underperform the fund increase return expectations and investment, consistent with a contrast effect. Conversely, benchmarks that outperform the fund increase return expectations even more strongly, contradicting the contrast effect.

Using the text of Form 10-K filings, we fine-tune a large language model to explain contemporaneous announcement-window returns and to predict post-filing returns over the subsequent 20 trading days. Out of sample, our model outperforms a range of text-based benchmarks at both horizons. The predictability of post-filing returns indicates that the market does not immediately incorporate all information contained in the filings. SHAP-based interpretability analysis attributes the model’s predictions mainly to segments that discuss financial reporting and R&D activity, particularly clinical development. The same topics drive predictions at both horizons, suggesting an immediate but incomplete reaction to the full content of the filing, rather than sequential processing of different types of information.

We study dynamic choice under risk through the lens of salience theory. We derive predictions on salient thinkers’ gambling decisions and strategy choices. We test our model experimentally and find support for all of our predictions. We also detect a strong correlation between static and dynamic choices, suggesting that salience theory can coherently explain risky choice in both static and dynamic contexts. Our results help to understand when people sell assets, stop gambling, enter the job market, or retire.