Research

Abstract: This paper examines the predictive power of corporate disclosures in macroeconomic forecasting. Leveraging advanced language models, I analyze the prevalence and tonal polarities of 31 accounting topics across 890,277 10-K and 10-Q filings. Firm disclosures exhibit superior performance in long-horizon forecasts, with their predictive strength primarily driven by discussions on corporate income taxes, revenue performance, mergers and acquisitions, and repurchase agreement activities. These topics transmit information to macroeconomic outcomes through firms' capital expenditures and dividend policies. Moreover, firm disclosures provide particularly strong forecasts of GDP and Investment during periods of economic downturn, outperforming the FRED-MD dataset.

Presentation: Econometric Society European Summer Meeting (EEA-ESEM) 2026; International Association of Applied Econometrics Annual Conference 2026; 13th Nordic Econometric Meeting; 7th ECONDAT Fall Meeting 2025.

Abstract: This paper examines relationships between sections of the same 10-K filing as a source of information for detecting financial misreporting. We find that semantic consistency between the Risk Factors and Management's Discussion and Analysis sections is positively associated with the likelihood of misreporting. Motivated by this finding, we develop LongFinBERT, a financial-domain language model that processes an entire 10-K in a single pass to capture cross-section relationships for misreporting detection. In out-of-sample tests, LongFinBERT outperforms established financial predictors and alternative language models, with the largest gains among firm-years with the highest predicted misreporting risk. The improved statistical prediction performance translates into meaningful additional misreporting detections and resource savings for the regulator. We also find that misreporting-related information is distributed across 10-K sections, and capturing textual cross-section relationships provides additional predictive power. In a second application, we use LongFinBERT to revisit investor reactions to year-over-year disclosure changes. In this context, we find that the previously documented decline in market reactions derived from word-count models was temporary and that reactions are strongest during recessions.

Presentation: 47th Annual Congress of the European Accounting Association 2025; Financial Econometrics and Machine Learning Conference 2024; Financial Fraud, Misconduct and Market Manipulation Conference 2024; 17th International Conference on Computational and Financial Econometrics 2023.

Abstract: Cyber risk depends on both the likelihood that a breach enters the public record and the severity of its financial consequences. Yet firms retain discretion over whether those consequences are quantified. Using U.S.-listed firms from 2005 to 2025, we separate breach observability from loss quantification. We find that larger latent losses are less likely to be quantified, so visible losses understate conditional log-dollar severity. Quantification is negatively associated with CEO unearned equity incentives and salary share, consistent with strategic disclosure. We construct a firm-level disclosure-risk exposure from the reporting-margin selection correction and show that it is priced conditionally on public-record breach likelihood: within the highest breach-probability bin, portfolios sorted on reporting-margin exposure earn a value-weighted Fama-French six-factor alpha of up to 0.78% per month. The loss-quantification process therefore shapes measured cyber exposure and contains return information beyond breach likelihood.

Presentation: 39th Australasian Finance and Banking Conference (scheduled); 2026 New Zealand Finance Meeting (scheduled); Financial Econometrics and Machine Learning Conference 2026 (scheduled); Financial Fraud, Misconduct and Market Manipulation Conference 2026.