Artificial intelligence is reshaping financial markets, yet the limits to its rationality remain underexplored. This paper documents information overload in Large Language Models applied to financial analysis. Using earnings forecasts from corporate calls and market reaction predictions from news, we show that predictive accuracy follows an inverted U-shaped pattern, where excessive context degrades performance. Larger LLMs mitigate this effect, increasing the optimal context length. Our findings underscore a fundamental limitation of AI-driven finance: more data is not always better, necessitating empirical tuning to determine the right amount of context for each task.
Decision-Making Under ComplexityDisclosure and Transparency
Using a large sample of U.S.-based public corporations, we evaluate the business case for board racial diversity. Our analysis shows that board racial diversity is positively associated with firm performance and value, and negatively associated with realized risk. However, causal tests that use mandated board racial diversity in California and the acceleration of the BLM movement as shocks to the demand for racial minority directors indicate that forced increases in board racial diversity have no effect on firm performance, value, risk, or other traditional corporate finance policies. Firms compelled to select racial minority directors find candidates with qualifications similar to those of existing directors, suggesting that in these settings board racial diversity itself does not impact performance and value outcomes. Overall the evidence suggests that the business case is insufficient on its own to either justify or oppose mandated racial diversity on boards.
with Sarah Bekessy, Mark Burgman, William Geary, Oğuzhan Karakaş, Hannah Layman, Annalisa Tonetto, and Brendan Wintle.
Biodiversity metrics built for corporate finance aggregate the harms of a supply chain into a single score, and a single score cannot tell an investor which species, in which places, its capital puts at risk. We treat that as a governance problem rather than a measurement problem, and we separate its two halves: visibility, what a firm knows about where its impacts fall, and transparency, what it discloses, to whom, and to what end. We do not claim that biodiversity is uniquely hard to measure; we claim that it is distinctively damaged by aggregation, because species and places are not fungible in the way a tonne of CO2e is. Answering the call by Hockerts and Searcy (2023) to sharpen the discourse on corporate sustainability, we ask under what organizational and political conditions biodiversity measurement becomes ethically meaningful. Aggregation widens moral distance; we appraise nSTAR, a species-level extinction-risk metric that currently resolves supply chains only to country and sector, weighing its advantages against its assumptions, its biases, and the organizational constraints that limit it. Spatial resolution, we conclude, is necessary but not sufficient: no metric collapses the distance between the boardroom and the biosphere, and a record that makes that distance transparent still needs a standard-setter that makes not-knowing costly, an owner with standing to instruct, and disclosure that reaches the places where the harm falls.
Corporate GovernanceDisclosure and TransparencyEnvironmental and Social Risk
I provide evidence that activist investors create value through the efficient reallocation of human capital by nominating directors at the firms they target. Stock returns are significantly higher at targeted firms when a new director is appointed and takes on a monitoring role to improve governance. Examining director biographies extracted from proxy filings, I show that effective activists match directors to firms that bring new skills and additional prior experience to the board. Consistent with directors' monitoring role and the reduction of agency problems, firms with new directors are more likely to replace the CEO and curb investment levels. Whereas prior research claims that value creation in activism is limited to takeovers, this paper shows that the human capital of directors also predicts returns for firms that are not acquired. Instrumental variables tests that exploit exogenous variation in director appointment rates indicate that the observed relationship is causal.
Corporate GovernanceDecision-Making Under ComplexityStakeholder and Institutional Pressure
The evolution of firms is not necessarily uniform. Exploring how this affects credit risk models, we find that firm life cycle provides additional explanatory power not captured by age. Firm age has an ambiguous effect on default risk and its impact during periods of high volatility is insignificant. Unobserved firm heterogeneity is an important determinant of credit default swap spreads and, when accounted for, riskier growth firms command a lower spread compared to mature firms that commonly benefit from the lowest spreads. Firms that age well by maintaining a growth profile are rewarded with lower cost of capital.
Decision-Making Under ComplexityDisclosure and TransparencyPolicy and Market Design
Matching observations across multiple datasets for corporate governance research carries with it some well-known and other less known challenges. This paper highlights best practices for carrying out merging the widely used BoardEx, CRSP and Compustat databases accessed via Wharton Research Data Services (WRDS), utilizing their respective company and security identifiers. A companion Github repository of the associated SAS code is made available to assist researchers in consistently applying the discussed matching techniques.
This paper provides guidance on the appropriate calculation of financial ratios for capital markets-based empirical research in finance and accounting. Practical examples are provided employing the widely-used Compustat database, and all calculated ratios are made available for download in a dataset. A companion Github repository of the associated SAS code is offered to assist researchers in consistently calculating key financial ratios.