What ClawdBot REALLY thinks about the polymarket and kalshi data that was recently dropped here:
There it is. The honest truth from 355K real trades.
Here's what the data actually says:
(continued in comments)
The longshot bias is real in the CALIBRATION data (price vs resolution frequency), but that doesn't translate to a tradeable edge because:
1. The 2% fee eats most of the small edges
2. Buying NO at 95Β’+ has terrible risk/reward β you risk 95Β’ to make 3Β’
3. The only viable
[-3.3%, +5.3%]. Not statistically significant. With $100 over 50 trades: median outcome is $102.90 (+$2.90). Only 59% chance of profit.
β Mid-Range Fade / Deep Fade: Catastrophic losses. -90% and -52% per trade.
Bottom line for your $$$:
You need it to resolve NO 97%+ of the time just to break even. The actual NO resolution rate is only 43-48%. Dead strategy.
β οΈ Buy Favorites (YES at 80-95Β’): The ONLY strategy with any pulse. +1.14% per trade, 91.4% win rate. But the 95% confidence interval crosses zero:
β Fade Longshots (Buy NO at 90-99Β’): MASSIVE LOSER. -56% per trade. Win rate only 43% β meaning those "longshots" actually hit way more than the NO price implies. This is the OPPOSITE of what we expected. The reason: when you buy NO at 95Β’, you're paying 95Β’ to win $0.98.