Conference Agenda

We study the interactive vocal communication of European Central Bank (ECB) presidents during press conferences and Q\&A sessions, with a particular focus on informal small-talk exchanges, and examine the extent to which vocal cues embedded in these interactions affect European financial markets. To our knowledge, this is the first study to analyze the informational content of small-talk exchanges during ECB press conferences. Leveraging recent advances in deep learning and machine learning, we extract high-dimensional vocal embeddings from ECB presidents’ prepared statements, responses to journalists’ questions, and informal conversational remarks. We show that subtle vocal cues contain predictive information for market prices beyond what is captured by spoken textual content alone. In particular, informal remarks—such as greetings, personal asides, and casual comments—exhibit the strongest predictive power. In general, our findings highlight the importance of how policy messages are delivered vocally, rather than solely what is said, for the transmission of monetary policy in Europe.

We examine how the use of Artificial Intelligence (AI) in writing fundraising campaigns influences donation behavior. Analyzing over 117,000 real GoFundMe campaigns, we find that campaigns more similar to AI-generated text raise more money. We corroborate this result with a pre-registered online experiment, in which participants were shown campaigns written by either humans or AI. Again, we find that AI-written campaigns attract more donations than human-written ones. However, AI usage disclosure leads to opposite results: When donors knew that AI was used, donations dropped even below the level of human-written campaigns. Disclosing the use of AI in campaign creation may induce perceptions of inauthenticity or manipulation, thereby reducing donor trust and willingness to contribute.

We develop a dynamic model of payment fraud detection in which fraudsters strategically adapt their methods in response to detection technologies, causing model performance to decay over time. We characterize the socially optimal detection level and show that in competitive markets full liability induces overinvestment: each PSP ignores how its detection effort accelerates fraudster adaptation market-wide, leading to excessive investment relative to the social optimum. When PSPs retrain models to counter decay, competition creates a ratchet effect—detection accelerates decay, increasing retraining frequency beyond the socially optimal level. Optimal liability allocation sets full reimbursement for both sending and receiving PSPs, complemented by a regulatory fee that internalizes the social cost of fraud and eliminates free-riding in detection efforts. Using Reddit discussion data and bot-detection Twitter data, we document strategic adaptation: fraudsters reallocate effort across scam types following detection, and detection performance deteriorates as fraud becomes more sophisticated, independent of overall fraud rates.

We study pricing efficiency in decentralized prediction markets by comparing market-implied probabilities from Polymarket with benchmarks derived from option-implied risk-neutral distributions extracted from the derivatives market. We study Bitcoin prediction bets and find that, although Polymarket prices broadly track option-implied benchmarks, they show systematic mispricing driven by complexity, behavioral factors, and market frictions. Mispricing is most pronounced in tail events, during periods of high volatility, major macroeconomic shocks, and reflects the influence of sentiment, attention, and blockchain-specific risks. These results reveal both efficiency and behavioral distortions in prediction markets.

We develop CovenantAI, an artificial intelligence-powered covenant monitoring methodology, to examine whether institutional investors strategically trade around covenant violations in leveraged loan markets. Documenting a persistent decline in loan prices during the 100 days preceding violations—with pronounced drops 20 days prior—we find cumulative abnormal returns of -0.84% during the [-20,-1] event window. Price effects are most severe for loans amended post-violation or remaining in technical default. Covenant violations significantly increase downgrade and bankruptcy probabilities, particularly among non-investment-grade loans held by Collateralized Loan Obligations (CLOs). We document substantial cross-sectional heterogeneity in CLO constraints driven by overcollateralization ratios and CCC-rated loan holdings. Loans predominantly owned by constrained CLOs exhibit steeper pre-violation price declines and significantly more negative abnormal returns. Our evidence demonstrates that constrained institutional investors preemptively divest loan positions in anticipation of covenant violations, with trading intensity reflecting both violation severity and investor-specific portfolio constraints.