Fading consistently bad traders
Find wallets that reliably lose, bet against everything they buy. On paper it returned +38%. At real order books, the trade you'd actually place earns almost nothing.
The hypothesisSkill may not persist, but anti-skill might. Wallets with long, consistently losing records are reliable contrary indicators — mirror-image copy trading: whatever they buy, take the other side.
The test
The roster: wallets with at least 20 resolved buys and a clean negative ROI, scored honestly at their own entry prices. An out-of-sample backtest faded every subsequent buy from the cohort, first at midpoint prices (the naive method every backtest uses), then re-priced at what taking the other side actually costs: buying the opposite outcome token at its live executable ask.
The result
The midpoint backtest was beautiful: +38% ROI, robust to spread assumptions of a few cents. The executable version collapsed. Bad traders overwhelmingly buy longshots; fading a longshot means buying the favorite side at 90¢+, where the fade's theoretical edge compresses into +2–3% gross per trade — before the taker fee and before crossing a spread that is proportionally enormous on the pennies of margin left. The cohort's losers really do keep losing (their forward longshots ran about −65% for them); the problem is that the market already charges nearly full price to bet against them.
Conclusion
A textbook specimen of the most common corpse in this lab: an edge that exists at midpoints and dies at asks. "Fade the dumb money" is directionally true and practically unharvestable — the order book stands between you and other people's mistakes, and it charges toll. Never trust a backtest that fills at the midpoint; almost none of the genre's published results survive the repricing.