charlieyan.com — working papers with PDFs, and the libraries below, in one place.
Quantitative research on options — surface dynamics, execution cost, and the accounting that decides whether an edge survives contact with a spread. Most of what I publish is the negative result and the machinery that produced it, because in this work a wrong number that reconciles is more expensive than one that crashes.
| exitkit | Twenty-seven position-exit policies behind one interface, with a drop-in adapter for backtesting.py. Exit logic is thin in open source; this makes it something you compare rather than hardcode. |
| runseal | Content-addressed provenance for research runs — which exact data, code, config and environment produced a number, and whether any of it changed since. |
| chainmill | Parallel ingest of zipped option-chain archives into a queryable store, with resumable builds and a coverage line on every run. |
| opentape | Twelve backtest bug classes, twelve runtime invariants, thirteen synthetic reproductions. |
Each repository pairs the paper with the code behind its results and a data-availability statement that says plainly whether you can run it.
- Dollar-Correct, Time-Wrong — a backtest can reconcile to the cent and still be wrong by 2–12× in risk-adjusted terms. Twelve defect classes with runtime-checkable invariants.
- Persistence, Premium, and the Illusion of Residual Reversion — SVI smile residuals converge less than a scale-matched null. The tape's P&L was variance risk premium, not signal.
- Microprice Drift and the Theta Wall — posted option depth carries real short-horizon information and cannot be monetised, because the toll is flat in horizon while the move grows with √t.
- A Formal Theory of the Delta-Matched Risk Reversal — the structure is near-pure differential vega; vanna is the tax on the harvest, not its source. Attribution closes to 100%.
- Margin Is Not a Footnote — collateral is a separate stochastic process from P&L and can end a strategy independently of its returns.
- How Much Tail Prediction Could We Have Detected? — a power accounting for a 1,450-test search, and a pre-registration gate that removes premium-confound false positives.
- The Expiry Corner as a Logarithmic Clock — a boundary-level error law for Carr randomization of the American put.
Pre-register the bar before running. Report three accounting lines — mid, friction-aware, and full cross — and decide on the last one. Print the coverage line with every result. Treat a Sharpe above 3 as an audit trigger rather than a discovery.