We develop machine learning methods to predict individual stock crashes by directly estimating the probability of extreme negative returns, offering a more robust alternative to existing moment-based crash proxies. We evaluate a wide range of machine learning models and introduce Bayesian optimization with Gaussian processes for efficient hyperparameter tuning. Simulations and empirical analysis using U.S. data demonstrate the high accuracy of our approach. Feature importance analysis provides economic rationales for key crash predictors. Stocks with higher estimated crash probabilities yield significantly lower excess returns, even after controlling for commonly used asset pricing factors.
We examine intraday return predictability in commodity futures and its relation to information diffusion and trading frictions in a retail-dominated setting. Using one-minute data on 36 Chinese commodity futures contracts from 2018 to 2024, we apply machine-learning models to lagged cross-sectional returns and document robust and economically significant predictability. Predictive power is strongest at the one-minute horizon and decays rapidly thereafter, consistent with information frictions theories. Sparse linear methods such as Lasso and Elastic Net consistently outperform more complex nonlinear algorithms, suggesting that predictive relationships in intraday returns can be represented effectively by sparse linear specifications. We find that return predictability is stronger for less volatile, less liquid contracts with higher open interest, greater price discreteness, and more intensive speculative retail trading, yet weaker for nearby contracts under tighter constraints and near price-limit events. Our findings extend high-frequency asset pricing research, highlight the advantage of simple models in extracting intraday signals, and carry implications for trading strategies and market regulation.
Climate-related news has been shown to be financially relevant, and theory predicts that unexpected increases in climate concerns widen the green–brown return spread. Yet entered on its own in a standard factor model, climate attention does not significantly explain weekly green-minus-brown returns. We show that this null is partly an artifact of averaging across ambiguity states. Using weekly S&P 500 returns over 2011–2021 sorted by Refinitiv ESG scores, the marginal pricing effect of a climate-attention shock varies significantly with the level of clean-energy transition ambiguity. We proxy for this ambiguity with ClimateVoV, a novel option-implied vol-of-vol of clean-energy options that captures second-order uncertainty specific to the transition, distinct by construction from general market-wide uncertainty (VIX, VVIX). The marginal effect grows from near-zero at low ambiguity to approximately 8 basis points per week at the 75th percentile. The statistically significant relation is carried by unexpected decreases in climate concerns, not increases. This pattern is consistent with the partial release of worst-case transition discounts on brown firms when concerns ease under elevated ambiguity, compressing the green-minus-brown spread. Restricting the analysis to environmental-impact news yields the strongest interaction. The effect appears in disclosure-based (Refinitiv) but not risk-based (Sustainalytics) ESG sorts. The result survives rate, oil, and policy-uncertainty controls, while a placebo constructed from growth-tech options produces no significant interaction.
This paper examines the environmental consequences of short-sale constraints by investigating their impact on plant-level toxic chemical releases. Leveraging the implementation of Regulation SHO as a natural experiment, we find that removing short-sale constraints leads to a significant increase in toxic chemical releases. This effect remains robust across various robustness and endogeneity checks. Our analysis suggests that heightened short-selling pressure alters managerial incentives, prompting a shift toward short-term stock performance at the expense of long-term environmental performance. These findings reveal an important but previously overlooked environmental benefit of short-sale constraints, highlighting their potential role in promoting corporate environmental responsibility. The study offers new insights into how financial market regulation can influence corporate behavior, with implications extending beyond shareholder welfare to broader societal well-being.
Firms that maintain total payouts, including dividends and share repurchases, despite reporting losses face elevated subsequent financial distress risk. Using 59,936 firm-year observations from 1991 to 2023, we show that the association between payout activity and subsequent distress changes significantly during loss years, with the associated odds ratio approximately 18% higher than during non-loss years. Among loss firms, those that maintain or increase payouts have a higher likelihood of subsequent distress relative to firms that reduce payouts, with increases associated with a 5.9 percentage point higher distress probability. A battery of identification tests, including an instrumental-variable strategy, supports a causal interpretation. The adverse consequences are concentrated among financially constrained firms, and partial payout reductions do not eliminate the elevated distress risk. Mediation analysis indicates that approximately 80% of the effect on firm failure operates through financial distress. Market reaction and investor confidence tests provide little support for signaling, while cross-sectional evidence points to governance frictions and managerial discretion. Additional tests provide little support for strategic leakage, and we find modest evidence of a 57% higher hazard of forced CEO turnover.
