A war‐related factor model derived from textual analysis of media news reports explains the cross section of expected stock returns. Using a semisupervised topic model to extract discourse topics from 7,000,000 New York Times stories spanning 160 years, the war factor predicts the cross section of returns across test assets derived from both traditional and machine learning construction techniques, and spanning 138 anomalies. Our findings are consistent with assets that are good hedges for war risk receiving lower risk premia, or with assets that are more positively sensitive to war prospects being more overvalued. The return premium on the war factor is incremental to standard effects

Persuasive communication functions through not only content but also delivery—facial expression, tone of voice, and diction. This paper examines the persuasiveness of delivery in startup pitches. Using machine learning algorithms to process full pitch videos, we quantify persuasion in visual, vocal, and verbal dimensions. We find that positive (i.e., passionate, warm) pitches increase funding probability. However, conditional on funding, startups with higher levels of pitch positivity underperform. Women are more heavily judged on delivery when evaluated in single‐gender teams, but they are neglected when copitching in mixed‐gender teams. Using an experiment, we show that persuasion delivery works mainly through leading investors to form inaccurate beliefs

We reconsider trend‐based predictability by employing flexible learning methods to identify price patterns that are highly predictive of returns, as opposed to testing predefined patterns like momentum or reversal. Our predictor data are stock‐level price charts, allowing us to extract the most predictive price patterns using machine learning image analysis techniques. These patterns differ significantly from commonly analyzed trend signals, yield more accurate return predictions, enable more profitable investment strategies, and demonstrate robustness across specifications. Remarkably, they exhibit context independence, as short‐term patterns perform well on longer time scales, and patterns learned from U.S. stocks prove effective in international markets

We provide the first systematic evidence on the link between long‐short anomaly portfolio returns—a cornerstone of the cross‐sectional literature—and the time‐series predictability of the aggregate market excess return. Using 100 representative anomalies from the literature, we employ a variety of shrinkage techniques (including machine learning, forecast combination, and dimension reduction) to efficiently extract predictive signals in a high‐dimensional setting. We find that long‐short anomaly portfolio returns evince statistically and economically significant out‐of‐sample predictive ability for the market excess return. The predictive ability of anomaly portfolio returns appears to stem from asymmetric limits of arbitrage and overpricing correction persistence

This paper studies the role of trust in incumbent lenders (banks) as an entry barrier to emerging FinTech lenders in credit markets. The empirical setting exploits the outbreak of the Wells Fargo scandal as a negative shock to borrowers’ trust in banks. Using a difference-in-differences framework, I find that increased exposure to the Wells Fargo scandal leads to an increase in the probability of borrowers using FinTech as mortgage originators. Utilizing political affiliation to proxy for the magnitude of trust erosion in banks in a triple-differences specification, I find that, conditional on the same exposure to the scandal, a county experiencing a greater erosion of trust has a larger increase in FinTech share relative to a county experiencing less of an erosion of trust. Estimating treatment effect heterogeneity using generic machine learning inference suggests that borrowers with the greatest decrease in trust in banks and the greatest increase in FinTech adoption have similar characteristics

Nearest neighbor (NN) matching is widely used in observational studies for causal effects. Abadie and Imbens (2006) provided the first large‐sample analysis of NN matching. Their theory focuses on the case with the number of NNs, M fixed. We reveal something new out of their study and show that once allowing M to diverge with the sample size an intrinsic statistic in their analysis constitutes a consistent estimator of the density ratio with regard to covariates across the treated and control groups. Consequently, with a diverging M , the NN matching with Abadie and Imbens' (2011) bias correction yields a doubly robust estimator of the average treatment effect and is semiparametrically efficient if the density functions are sufficiently smooth and the outcome model is consistently estimated. It can thus be viewed as a precursor of the double machine learning estimators

We develop FinBERT, a state‐of‐the‐art large language model that adapts to the finance domain. We show that FinBERT incorporates finance knowledge and can better summarize contextual information in financial texts. Using a sample of researcher‐labeled sentences from analyst reports, we document that FinBERT substantially outperforms the Loughran and McDonald dictionary and other machine learning algorithms, including naïve Bayes, support vector machine, random forest, convolutional neural network, and long short‐term memory, in sentiment classification. Our results show that FinBERT excels in identifying the positive or negative sentiment of sentences that other algorithms mislabel as neutral, likely because it uses contextual information in financial text. We find that FinBERT's advantage over other algorithms, and Google's original bidirectional encoder representations from transformers model, is especially salient when the training sample size is small and in texts containing financial words not frequently used in general texts. FinBERT also outperforms other models in identifying discussions related to environment, social, and governance issues. Last, we show that other approaches underestimate the textual informativeness of earnings conference calls by at least 18% compared to FinBERT. Our results have implications for academic researchers, investment professionals, and financial market regulators

We study the role of a relatively new type of external firm monitor, an on‐site government‐appointed Corporate Monitor, and assess whether such appointments reduce firms' propensity to violate laws. Using a sample of deferred and nonprosecution agreements, we first document the determinants of Monitor appointment. We find firms that voluntarily disclose wrongdoing and have more independent directors are less likely to have Corporate Monitors, whereas those with more severe infractions, mandated board changes, and increased cooperation requirements are more likely to have Monitors. We find such appointments are associated with an 18%–25% reduction in violations while the Monitor is on site, however, the effect does not persist after the Monitorship ends. Using a semisupervised machine learning method to measure changes in firms' ethics and compliance norms, we find that the reduction in violations is associated with changes in ethics and compliance that also do not persist. Finally, we document that firms under Monitorship experience a persistent reduction in innovation, highlighting a previously unexplored cost of these interventions. Overall, our results suggest that, although Corporate Monitors on site are associated with fewer violations, firms revert to previous levels of violations following Monitors' departure

This paper combines a data rich environment with a machine learning algorithm to provide new estimates of time-varying systematic expectational errors ("belief distortions") embedded in survey responses. We find that distortions are large even for professional forecasters, with all respondent-types over-weighting their own beliefs relative to publicly available information. Forecasts of inflation and GDP growth oscillate between optimism and pessimism by large margins, with biases in expectations evolving dynamically in response to cyclical shocks. The results suggest that artificial intelligence algorithms can be productively deployed to correct errors in human judgement and improve predictive accuracy.Institutional subscribers to the NBER working paper series, and residents of developing countries may download this paper without additional charge at www.nber.org

A sovereign creditrating is a function of hard and soft information that should reflect the creditworthiness and the probability of default of a country. We propose an alternative characterisation for the subjective component of a sovereign credit rating – the parts related to the ratee’s lobbying effort or its familiarity from a United States point of view – and apply it to S&P, Moody’s and Fitch ratings, using both traditional ordered-logit panel models and machine learning techniques. This subjective component turns out to be large, especially for the low-rated countries. Countries that are rated as investment grade tend to be positively influenced by it, and vice versa. Subjective judgment in credit ratings does have predictive value: it helps in identifying chances of sovereign defaults in the short-term. Still, the impact of subjectivity in sovereign ratings on borrowing costs is very limited on average