A hypothetical strategy of backing the favorite on prediction market Polymarket in every scored 2026 US primary would have lost 4%, even though those candidates won 87% of the time, according to blockchain data provider Bitquery’s scorecard dated Oct. 9.
The disconnect between forecasting accuracy and profitability highlights a pricing inefficiency on the platform. Bitquery analyzed 273 primary election contracts across all 50 states using Polygon trade data, excluding nine markets due to early settlement, missing prices, or missing settlement records.
The study measured trades during the 24 hours before each voting day, using the last trade within 30 days if no trade occurred in that window. For runoff races, Bitquery used the runoff date. Outcome shares on Polymarket pay $1 when a contract wins and $0 when it loses.
Price Tiers and Win Rates
Candidates priced at 90 cents or higher won 177 of 182 races, a near-perfect track record. Yet those high-confidence favorites, by design of the payout structure, generate smaller returns per dollar staked than cheaper winners. Candidates priced between 50 and 90 cents won 71% of the time, with an average price of 77 cents.
The 238 of 273 total races won by the highest-priced contract in each market underscores broad forecasting accuracy across the dataset. Bitquery noted that “general elections attract more money and polling than small, local contests,” cautioning that the methodology may not apply equally to November races.
Methodology and Limitations
The study excluded Polymarket’s US app and other venues such as Kalshi. Polymarket charges a 0% to 1% fee on politics markets. Bitquery’s reference prices represent averages of trades during the specified window, not guaranteed buy quotes. The methodology does not specify full adjustment for spreads, slippage, or fees, meaning the reported 4% loss may not account for the full cost of executing such a strategy in real time.
Bitquery did not name individual races or candidates analyzed, nor did it disclose which nine markets were excluded or provide sample size and confidence intervals for the 71% win rate among mid-tier favorites.
