Writing
Technical write-ups covering each project — how the analysis works, what data it uses, and what decisions shaped its design.
Showing 1–5 of 11 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 17-Month Cascade: A Repeating Pattern in Korean District Housing Distress A statistically consistent 17-month lag between lease-lien spikes and forced auction spikes, documented across 228 Korean districts using 16 years of public court registry data — and what it means for lenders expanding into 비수도권. 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