Regulatory Disclosure Sequences and Corporate Distress Risk: Evidence from Korean DART Filings
Year: 2026
Type: article
Abstract: This study examines whether sequences of regulatory disclosure events contain incremental information about future corporate distress beyond accounting fundamentals and market-based signals in the Korean capital market. Using major-event disclosures filed with Korea’s Data Analysis, Retrieval and Transfer System (DART), the electronic disclosure system operated by the Financial Supervisory Service, from 2015 to 2024 and distress outcomes observed through 2025, we construct a firm-year panel of 19,924 observations covering 2,798 firms. Out-of-time tests show that disclosure-sequence features outperform financial-only baselines and that combined sequence-financial models substantially improve watchlist ranking. The best non-market 365-day model reaches an area under the receiver operating characteristic curve (ROC AUC) of 0.906 and top-decile lift 6.90, while the non-market 730-day calibrated model reaches ROC AUC 0.900 and top-decile lift 6.50. To address whether equity prices already summarize the relevant information, a market-price extension built on a single, independently collected daily price source covers 96.9% firm-year anchors and adds 50 trailing price, volume, volatility, drawdown, and market-adjusted return controls benchmarked to the Korea Composite Stock Price Index (KOSPI) and the Korea Securities Dealers Automated Quotations (KOSDAQ) index. The calibrated sequence-financial-market random forest reaches 365-day ROC AUC 0.9082, average precision 0.1599, and top-decile lift 7.08, and 730-day ROC AUC 0.9050, average precision 0.1751, and top-decile lift 7.13; the robust 730-day calibrated market model reaches top-decile lift 7.61. Robustness checks excluding weak listing-status proxies and SPAC-like dissolution cases preserve the core non-market result, with 730-day ROC AUC 0.895 and top-decile lift 6.42. A supplementary no-lookahead large language model context ablation on 30 benchmark cases shows that disclosure-only prompts reach 0.867 accuracy, while adding pre-anchor market and graph-retrieval context lowers the false-positive rate to 0.200 and separates post-hoc + 20 trading-session excess returns for cases predicted as distress versus normal restructuring (-12.0% versus + 5.6%). The findings suggest that ordered regulatory disclosures reveal distress-related information not fully captured by financial ratios or trailing market prices, while the LLM and price-reaction screens remain descriptive case-review layers rather than causal event-study evidence. Because that ablation covers only 30 benchmark cases, it is reported as an explainable-review diagnostic for post-screening the model’s top-decile watchlist, not as quantitative validation on a par with the out-of-time panel results.
Source: Asia-Pacific Financial Markets
Author: Yuyong Kim
Institution: University of Wisconsin–Madison
Language: English
Subfield: Accounting
Domain: Social Sciences
Open Access status: hybrid
Estimated APC paid: $3,390
OpenAlex ID: W7225787678