Favourites priced 0.90-0.995: the edge shrinks from +$0.98 to +$0.12
Over 767 outcomes, favourites at 0.90-0.97 won less often than priced. At 0.97-0.995 the edge was +$0.98 per bet at the mid price but only +$0.12 at the executable ask.
What we asked
Shortly before a market resolves, the favourite is often priced at 0.90 to 0.995. We asked whether those prices are too low, meaning the favourite wins more often than the price says. The scenario is a lag at resolution: the result is effectively known, but the price has not yet reached $1. We tested it for a taker, a trader who buys at the price someone else is already offering.
Data
We observed 767 favourite outcomes in three price bands, recording the price and whether the favourite won. Of those, 512 were in the top band, 0.97-0.995. For 351 records we had both the mid price (halfway between best bid and best ask) and the ask (the lowest price a seller will accept), which let us compare the two. We did not record the observation period or which markets were included.
Result
| Band | Win rate against price | Result |
|---|---|---|
| 0.90-0.94 | 89% against 92% | -$368 on $100 bets |
| 0.94-0.97 | 92% against 96% | -$386 on $100 bets |
| 0.97-0.995 | 510 of 512 won (99.6%) against 98% | +$0.98 per bet at the mid; +$0.12 per bet at the ask |
In the top band the spread, the gap between bid and ask, was 0.8 cents. It removed 88% of the edge. One oracle reversal per 500 bets, costing $0.20 per bet on average, would turn the remaining result negative.
Why the edge disappears
In the two lower bands, the crowd was not wrong: the favourites won less often than their prices implied. An early run of 29 wins in 29 turned out to be luck on a small sample, about 1.03 standard deviations above what the prices predicted.
In the top band the win rate did exceed the price, and the lag at resolution looks real. But the profit appears at the mid, which is not a price a taker pays. At the ask, the spread takes most of it. The oracle is the mechanism that reports the final outcome. A single reversal against the favourite wipes out the small gain from many winning bets.
Limits of this measurement
-
767 outcomes is a modest sample. We did not keep a breakdown of how many fell into each of the two lower bands.
-
The ask comparison rests on 351 records, not all 767.
-
We did not record the observation period or confidence intervals. The figure of one oracle reversal per 500 bets is a stress scenario, not an observed rate: in the 512 top-band bets, 2 lost.
-
The per-bet figures are in dollars per $100 bet. We recorded the stake explicitly only for the lower bands, so for the top band this is our reading.
What it means for a trader
For a taker, we found no profitable favourite band. Below 0.97 the favourites were priced slightly high. Above 0.97 the win rate beat the price, but almost all of that gain was the spread. Whoever has a resting bid, a maker, may collect the lag instead, but we did not test that.
Sources
- sourcePolymarket CLOB API
- sourcePolymarket Gamma API
Polytrading is an analytics tool. It does not trade for you, hold your keys, or promise returns.