When Do Polymarket Markets Decide? Half Wait Until the Last 15 Minutes

TL;DR: 50% of Polymarket markets lock in within 15 minutes of resolution. The number of different liquidity providers explains ~1% of when markets decide. The type of market (sports, crypto, politics) explains 10x more.

Half of all Polymarket binary markets reach certainty within 15 minutes of resolution. Out of 397,000 resolved markets since January 2025, the median time between the last uncertain trade and the final result is 14.6 minutes.

We set out to test whether having more different liquidity providers — the people placing resting orders on each side — affects how quickly a market reaches its verdict.

The hypothesis: too few providers means thin price discovery, too many means sustained disagreement, and somewhere in the middle should be optimal.

The data has a different story. The number of liquidity providers barely matters. What does matter — 10x more — is what the market is about.

The Data

397,498 resolved Polymarket binary markets, each with at least 10 trades and at least one trade on the winning outcome below 90 cents. For each market, we measure lock-in timing: how long before resolution did the winning outcome's price last dip below 90 cents? A market that locked in 5 days before resolution "decided" early. One that locked in 2 minutes before resolution was uncertain until the end.

Half of all markets lock in within the final 15 minutes. Another 20% take 1–6 hours — a second cluster of markets that "know" a few hours early. The remaining 30% spread across days to weeks, with 10% locking in more than 5 days before resolution.

More Providers, Slightly Later Lock-in

At first glance, markets with more liquidity providers lock in faster. But that's misleading — those markets also have more trades overall (correlation: 0.96), and more trades is the real driver.

To separate the two, I held trading volume constant and asked: among markets with the same amount of trading, does having more different providers change anything?

Yes, but barely. Markets with 8 unique providers lock in about 4 minutes before resolution. Markets with 500 unique providers lock in about 1.7 hours before resolution. The curve rises, but across the full range of provider diversity, the difference is small in absolute terms — and it explains only about 1% of the total variation.

What Actually Matters: The Type of Market

The type of market explains 10x more than the number of liquidity providers. Sports markets lock in at the final whistle. Crypto price targets lock in when the price hits. Political markets wait for official results.

Adding market-type categories to the model explains 6.8% of additional variation beyond trade count alone. Provider diversity adds just 0.6%.

The chart shows signed ΔR² — the additional variance explained by provider diversity, with the sign indicating direction. Positive means more providers predicts later lock-in; negative means faster. (Raw ΔR² is always positive; we sign it to show whether diversity delays or accelerates convergence.)

In sports and crypto (87% of markets), more providers weakly predicts later lock-in — consistent with more disagreement. In politics, weather, tech, and entertainment, more providers weakly predicts faster lock-in. The pooled result is positive only because sports and crypto dominate the sample.

A Single-Character Bug That Changed the Finding

The most useful discovery was not about liquidity at all.

To classify markets by topic, I built a keyword matcher over market titles. The first version used '% vs %' (space after "vs") to catch sports markets. Polymarket titles mostly use "vs." (period). One character. This silently misclassified ~100,000 sports markets into the "other" bucket, inflating it from 5% to 35% of the sample and creating a fake signal — the misclassified sports markets mixed with genuinely unclassifiable markets produced an apparent provider-diversity effect of 2.7% within "other" that did not exist.

If you do prediction market research that uses title-based topic classification — and most empirical PM work does — this is a reminder to check your classifier against sample titles. A one-character difference in a LIKE pattern changed the headline finding of a 400K-market analysis.

What This Means

For traders: counting liquidity providers won't help you predict when a market converges. Watch the market type and the event schedule instead.

For researchers: when a market decides is a different question from whether it's right — and the answer depends almost entirely on the domain, not the market structure.

For market designers: a sports market and a political market have fundamentally different convergence dynamics. One-size-fits-all parameters may be leaving efficiency on the table.