Market Insights
Thoughts on screening, technical analysis, and the mechanics behind finding good setups.
A Correction I Had to Publish: When the Sample Is an Era That Ended
I recommended a strategy on the strength of the literature. Then I split the data by era and found the effect had decayed by 75%. The literature wasn't wrong — it was describing a period that no longer exists.
A Market With a Known Birthday
Most questions about how quickly a market becomes efficient are unanswerable because you can't see the starting gun. Alberta's betting market opened on 13 July 2026. That's a natural experiment with a timestamp.
An Inefficient Market With a 4.5% Toll and a Bouncer
Sportsbooks post prices instead of running an auction, which really does make them less informationally efficient than equities. It also makes them far more expensive to trade, and they can throw you out for winning.
My Crypto Momentum Backtest Passed Every Test But One
Net Sharpe 1.19. Beat Bitcoin on return and drawdown. Survived Bonferroni, Hansen SPA and a deflated Sharpe ratio. Then it landed at the 52.8th percentile of a portfolio that entered at random.
Proving You Have an Edge Takes Longer Than Having One
At a realistic 2% edge you need about 6,000 settled outcomes to distinguish it from zero. At 100 a week that's over a year. The fix isn't more patience — it's measuring something with less variance.
Stripping the Margin Out of a Quoted Price
Every quoted two-sided price contains a fair value plus a margin you can't observe. How you separate them is a modelling choice — and on a longshot, that choice is an order of magnitude larger than any edge you'd find.
Five Hypotheses, Five Rejections, One Law
I spent a month testing strategies outside equities — crypto momentum, funding carry, sports betting models. Almost everything got rejected, and every rejection failed the same way: the edge was real and smaller than the friction.
The Mean Splits Evenly. The Variance Does Not.
I went looking for a stale pricing formula and found the means were right. The variance was the thing that was wrong — and it was wrong in opposite directions for two quantities derived from the same assumption.
A Real Signal That Loses Money the Harder You Trade It
Cross-venue funding carry is the most-marketed arbitrage in crypto. The bias is genuinely there — 8.5 to 11.4% annualised. The rule that captures the most of it finishes at −15% a year.
Your Backtest Universe Is Lying to You
Of the two million tokens launched in 2024, the ones with an average lifespan of six days don't show up in your data. A universe built from what's listed today measures the returns of the things that didn't die.
Why RSI Actually Matters for Stock Screening
Most traders use RSI wrong. Here's how I use it as a filter to cut through noise and find stocks with real momentum — not just ones that look like they're moving.