Research
One question runs under all of these: whether an apparent edge survives honest accounting. Surfaces, execution cost, attribution and the arithmetic that decides what a result was actually measuring.
Most of the answers are no, and those are the ones worth writing down. A signal that converges less than chance, a search whose null was never interpretable, information that is real and cannot be monetised, a foundation model that does not beat the market's own quote.
Each paper pairs with a public repo holding the code and a data-availability statement that says plainly whether you can run it. 3 of 8 reproduce end to end with no market data at all.
Working papers · 8
- The Market Already Knew: A Pre-Registered Falsification of TimesFM-3 on the SPY Implied-Volatility Surface
A time-series foundation model does not improve on the market's own implied-volatility surface. Registered on the model's release day, before any arm ran.
- Microprice Drift and the Theta Wall: Why Order-Book Depth Cannot Time Option Execution
Posted option depth carries real short-horizon information and still cannot be monetised — the toll is flat in horizon while the move grows with √t.
- Margin Is Not a Footnote: Reg-T Feasibility as a First-Class Backtest Output
Collateral is a stochastic process separate from P&L, and can end a strategy independently of its returns.
- 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 — refuting the standard vanna-centric account.
- How Much Tail Prediction Could We Have Detected? A Power Accounting for a 1,450-Test Search
A null from a large signal search is uninterpretable unless the searcher reports what the search could have detected.
- Persistence, Premium, and the Illusion of Residual Reversion in SPY Options
SVI smile residuals converge less than a scale-matched null. The tape's P&L was variance risk premium, not signal.
- Dollar-Correct, Time-Wrong: Twelve Backtest Bugs That Survive Reconciliation
A backtest whose P&L reconciles to the cent can still be wrong by 2–12× in risk-adjusted terms.
- The Expiry Corner as a Logarithmic Clock
A boundary-level error law for Carr randomization of the American put, with every numerical claim re-derivable from the published verifier stack.