Polymarket Favorites Were Right 87% of the Time, but Bettors Still Lost 4%

Polymarket's favorites were right 87% of the time in this year's US primary races, and a flat bet on every one of them still lost money. That's the pair of numbers in a new audit from the on-chain data firm Bitquery, which scored 273 races and took each candidate's price on the morning of the vote.

The loss sounds small. Put a dollar on every voting morning favorite, the report says, and you'd have lost 4 cents on the dollar.

Put $1 on every favorite on voting morning and you'd have lost 4 cents on the dollar.

That figure has travelled further than the 87%, which gets the order backwards. The 4 cents describes one staking plan at one set of prices. The 87% describes who won. Read together, they say something neither says alone: Polymarket's primary markets were good at picking winners and worse at pricing the middle of the field.

The report is credited to Gaurav Agarwal, dated 9 October 2026, with figures measured against Bitquery's Polygon data to 8 October 2026.

How Bitquery Priced a Favorite

A candidate's price here is the average of trades in the 24 hours before 12:00 UTC on voting day, which is early morning in the US. Where a candidate had no trade in that window, the audit used the last trade in the 30 days before. Races that went to a runoff were priced on the runoff date.

Two boundaries shape the result:

• Nine markets were left out, for settling before anyone voted, for unusable prices, or for a missing settlement record. That leaves 273 scored races.
• Only Polygon trades count, so Polymarket's US app and rival venues such as Kalshi, which keep their books off-chain, sit outside the sample.

Bitquery's trade data also starts on 1 September 2025. Money totals, including the $36.3 million on these primary markets, count trading since then, so races that opened earlier, such as the Texas Republican Senate primary, carry only part of their history.

The Losses Sat in the Middle of the Board

Favorites priced between 50 and 90 cents won 71% of their races while costing 77 cents on average. That gap is the one that cost money. The strongest favorites, the ones that felt like sure things, merely broke even.

The value showed up further down the board. A dollar on every candidate priced between 5 and 30 cents paid back at least $1.38, which is the kind of result that makes a blanket favorite strategy look careless.

Why Being Right 87% of the Time Can Cost You

An outcome share pays $1 if it wins and nothing if it loses. Buy one at 79 cents and a win returns a thin slice of the stake, while a single loss erases the cost of several winners.

Counting correct calls treats every race as equal. Returns can't: a cheaper winner produces a larger payout for the same stake than an expensive one. That's why the audit puts an equal-dollar figure beside the accuracy rate instead of stopping at 87%.

What the 4 Cents Doesn't Include

The prices are reference prices, not fills. The published methodology does not specify a full adjustment for spreads, slippage or fees, so the 4 cent result is better read as a signal about pricing than as a verified trading loss.

Fees are the sharper edge of that. Polymarket's trading documentation lists a Politics taker fee rate of 0.04, charged through the formula fee = C x feeRate x p x (1 - p), where C is the number of shares traded and p is the share price. Makers are never charged, only takers pay, and the money funds the Maker Rebates Program, according to Polymarket's help centre.

The formula peaks at 50 cents, inside the price band where the audit found its losses. Whether any fee was charged on these historical primary trades isn't established by the report, and at a 4 cent margin that question carries more weight than it would at a wider one.

The Misses an Average Hides

Wisconsin's Democratic primary for governor stands out. Francesca Hong sat at 96 cents on voting morning, and David Crowley won by less than a point. New Jersey's Republican Senate primary went further: Justin Murphy's shares cost 0.4 cents while favorite Alex Zdan was at 79 cents, and Murphy won by 4 points with Zdan finishing third.

House races were harder than the rest. The favorite won 85% of them, against about nine in ten for Senate and governor contests. Money didn't explain the gap: thinner races tended to have near-lock favorites, and those favorites won.

Late movement cut both ways. In 15 races the favorite changed hands in the final week before the vote. The new favorite won 7 of them, the old one won 5, and someone else won the other 3. In the eight biggest upsets, more than twice as many wallets held the losing favorite into the vote as held the winner.

What the Primary Record Says About November

Not much yet, by the audit's own account. Primaries are small, local contests. General elections draw far more money and polling, so the measured primary record may not carry over. Whether the 4 cent figure survives replication is also open.

The more durable point is structural. When a market is sharp about who wins and loose about what a contract costs, accuracy rates keep looking strong while mid-priced favorites keep leaking. That's the number worth watching in November, not the hit rate.