My work.
Major projects and smaller experiments, each one pressure-tested through walk-forward validation with realistic cost and slippage before any capital is committed.
Recent projects.
June 2026LLM news-sentiment sector grader
A daily alternative-data engine that grades sectors from the news and regulatory flow, Polygon news plus SEC 10-K and Form 4 filings, scored by a custom LLM pipeline and mapped through a 2,700+ ticker lookup. Backtesting 231 sector-days showed the signal is concentrated: strongly predictive for cyclicals like Consumer Discretionary, inverted for defensives, so the grader weights news by sector instead of uniformly.
Strait of Hormuz oil event study
A live 47-day event study tracking how the February 2026 closure of the Strait of Hormuz rippled through US oil producers that do not operate in the Middle East. Daily Polygon closes for baskets of domestic E&P and LNG names are charted against the S&P 500 with event markers, showing how foreign supply-chain risk can lift competing domestic companies even when the broad market is down.
TSMC earnings-call risk language vs. return
An NLP study of nine TSMC earnings calls (2024–2026) scoring how heavily management leaned on US–China risk language, H20 export bans, export controls, tariffs, then testing whether that tone predicts the stock’s drift. Cautious language lines up with weaker 5-day returns (r ≈ −0.52), with the Q1 2025 tariff-pause rally as the telling exception.