The Queue Was the Arbitrage: What 147 Days of Tick-Size Asymmetry Were Worth
For 147 days two of the largest USDC/USDT venues quoted the same asset on price grids differing tenfold, with zero fees on both. The cross-venue gap was there 41% of the time — and was worth almost nothing. A taker bound that collapses with simultaneity, a 91-day order-book replay, and a spread decomposition that shows why: a coarse tick is a queue rent, not a mispricing. Now posted as a working paper on SSRN.
Read the research → Model Validation · Original ResearchTwelve Entry Filters, Zero Survivors
Two live strategies, two bad days, and a 48-hour campaign to teach them to skip their losing trades. Every filter cost money — and the effect-size table explains why: at entry, future winners and losers are statistically identical. A Bayesian reading, and a Deflated Sharpe verdict on the strategy itself.
Read the research → Market Microstructure · Original ResearchHow Much Money Fits Inside a Depeg? A Census of 1,136 Stablecoin Dislocations
Nobody had published a systematic measurement of capacity during stablecoin dislocations. Two years of tick data, a reproducible event definition, a per-event capacity distribution spanning four orders of magnitude — and the finding that 97% of the money flows through the slow events, not the millisecond wicks. Now posted as a working paper on SSRN.
Read the research → Statistics · Live Track RecordThe Track Record That Couldn’t Lose
Bootstrap resampling produced ten thousand simulated years without a single loss — and that certainty is the finding. Sharpe t-tests, the Deflated Sharpe Ratio, and the date on which the record actually graduates.
Read the analysis → Position Sizing · MethodologyThe Kelly Criterion Said Bet 3,000x. The Interesting Part Is Why It's Wrong.
Log-growth sizing applied to a live trading book: why the naive answer is a data-completeness alarm, why the formula's decisive input is unobservable, and the three situations where Kelly earns its seat back.
Read the analysis → Model Validation · Case StudyWhy "Backtest Says It Works" Isn't Enough — An Independent Model Review Case Study
How a plausible-looking trading rule was pressure-tested: selection bias, regime decomposition, out-of-sample splits, PBO overfitting checks, and measuring real fill slippage instead of assuming it.
Read the case study →