Korean Forensic Finance
Detecting financial-statement manipulation in Korean markets with public DART data — scoring models, enforcement datasets, and the platform behind them.
8 articles
Four Signals, One Score: How kr-anomaly-scoring Flags KOSDAQ CB/BW Manipulation The kr-anomaly-scoring library turns four CB/BW manipulation signals into a single 0–4 integer score per issuance. This post explains each signal's threshold, the rationale behind each parameter choice, and why additive composition produces a better priority queue than a probability estimate. Read article The Beneish M-Score, Reimplemented for Korean IFRS The Beneish M-Score is a 27-year-old fraud-screening formula calibrated on 1990s US GAAP filers. Apply it to Korean IFRS data and ~19% of KOSDAQ companies break the math. This library handles the structural differences. Read article Pricing Convertible-Bond Dilution Without SEIBRO: Black-Scholes on the DART Filing Korean convertible bonds and bonds with warrants are a known dilution vector on KOSDAQ. The natural data source for analyzing them — SEIBRO — is gated behind a separate API key that has been broken since early 2026. This library reframes the question as an option-pricing problem and answers it from DART data alone. With caveats. Read article Splitting a Forensic-Finance Monolith into Four Repos: Why and How What started as one repo grew into a coupled mess of ETL, scoring, and statistical validation. The split into four targeted libraries — kr-forensic-core, kr-dart-pipeline, kr-anomaly-scoring, kr-stat-tests — was about being honest with the dependency graph. Read article 240 Korean Accounting Violations, Coded Once So Researchers Don't Have to Code Them Again The Financial Supervisory Service and Securities & Futures Commission publish enforcement decisions in PDFs and HWP files. The text is in Korean. The structure varies by source. This dataset codes them — 240 violations across two regulators, with a bias-validated taxonomy and DART-linked Beneish ratios for the named cases. Read article 60 Calendar Days Is 38 Trading Days. The 37% Error Hiding in Korean Market Code. Korean exchanges observe roughly fifteen non-weekend holidays a year. Naively shifting a date by N calendar days to build a price window silently drops a third of the sessions you thought you had. This library is the three functions that fix it. Read article An MCP Server for Korean Forensic Finance: Eleven Tools, One Conversation krff-shell is the delivery layer of the forensic-accounting toolkit — a CLI, a DuckDB query layer, per-company HTML reports, and an MCP server that exposes eleven forensic-finance tools to Claude. The point is to make the infrastructure usable by people who do not write SQL. Read article 3,949 Companies. Four Numbering Systems. One Table. Korean government agencies assign different IDs to the same company. DART, the stock exchange, the tax authority, and the corporate registry each use their own code — and no official table links them. This one does. Read article