Which Prediction Market Moves First? Price Discovery Between Polymarket and Kalshi

Published: November 2026 | OddsReference Research

When two venues list the same contract and its price jumps, one of them usually gets there first. Our price discovery page measures which one, how long the other takes to follow, and how often they move together. This article explains the method; the live discovery page carries the current numbers.

Why Does Price Discovery Matter?

Price discovery is the process by which new information gets into a price. When the same event trades on two venues, the one that moves first is where the information entered, and the other is following. Knowing which is which tells you where to look for a fresher reference price.

Prediction markets are fragmented. The same election, rate decision, or game can trade on Polymarket, Kalshi, and several smaller venues at once, each with its own users, fees, and liquidity. Those prices are usually close, but they are not set in one place. Someone reacts first, and the rest adjust.

That ordering is a measurable property of the data, and it is separate from any claim about which platform is better. We report it neutrally: the page shows the split between venues for each pair, and the split changes as the underlying markets change.

How Do We Find Markets Listed on Both Venues?

We match contracts that appear on at least two venues. A shared outcome is one with price history from two or more sources in our database, aligned on the same hourly timestamps over the last 14 days. Only hours where both venues have a price are compared.

Matching happens at the outcome level, not the headline level. A multi-candidate market such as a nomination field is split into one binary contract per candidate, and each candidate is paired across venues separately. The page reports how many shared outcomes fed the latest run, and any pair of venues with enough overlap is included, not only Polymarket and Kalshi.

Input	Setting
Price series	Hourly closing prices per venue
Look-back window	14 days
Unit of comparison	One outcome priced on two or more venues
Minimum for a pair to appear	At least 20 move events
Refresh	Every six hours

What Counts as a Move?

A move is a net change of at least 8 points in the consensus price within 24 hours. The consensus price is the average of the venues’ hourly closes. Windows do not overlap, and we take the earliest qualifying window first, so one long repricing is counted once instead of many times.

We measure the move on the consensus, not on either venue alone, so neither venue defines the event it is then judged on. An 8-point threshold on a 0 to 100 scale is large enough to filter out ordinary drift and small enough to capture real news: a poll release, an injury report, a data print.

Each move event is stored with its start and end, the signed size, the category, and both venues’ prices at the beginning and end. The recent-moves list on the live page shows the latest ones, so you can check the method against specific cases.

How Is the First Mover Decided?

Inside each move window, a venue “arrives” at the first hour its own price has covered at least half of the consensus net change, in the direction of the move. The venue that arrives earliest is the first mover. If both arrive in the same hour, the event is simultaneous.

The gap between the two arrival hours is the lead time. A venue that never covers half the move inside the window counts as arriving last. We use the halfway point because it is robust: the very first tick can be noise, and full convergence can take days, but half of a large move is a clear signal that the venue has reacted.

The page summarizes events per pair as three counts: venue A first, venue B first, and simultaneous. The first-mover share excludes simultaneous events, so it answers a narrow question: when one venue was strictly earlier, how often was it this one? The median lead is the typical size of that gap in hours.

How Do You Read the Lag-Correlation Chart?

The lag-correlation chart shows how closely hourly price changes on one venue line up with changes on the other, shifted forward or back by up to six hours. A peak at zero means the venues move together. Bars heavier on one side mean one venue’s changes tend to show up on the other later.

Concretely, each bar is the correlation between venue A’s change in hour t and venue B’s change in hour t plus k, pooled across every shared outcome for the pair. Positive k asks whether A’s move predicts B’s move k hours later. If the bars on the right of zero are taller than the mirror bars on the left, A leads.

We summarize the shape as an asymmetry score: the sum of the positive-lag correlations minus the sum of the negative-lag ones. Positive favors the first venue in the pair, negative the second, and a value near zero means no consistent direction.

Read the chart in three steps:

Find the tallest bar. At zero, most of the co-movement is within the same hour.
Compare the two sides. Imbalance is the lead-lag signal; symmetry is not.
Check the sample. A pair with few shared outcomes produces noisy bars, and the page shows the counts.

A table view carries the same values for anyone who wants the numbers.

Why Do Fast Sports Moves Show Up as Simultaneous?

Because our series are hourly, any two price changes inside the same hour look identical. In-game sports prices can swing several points in minutes after a score or an injury, so both venues often cross the halfway mark within one hourly bucket, and the event is recorded as simultaneous.

That is a limit of the instrument, not a finding that the venues are equally fast. We count those events as simultaneous rather than guess an order, and the page splits results by category once a category has enough events. Slower-moving categories such as politics, economics, and long-dated questions give hourly data much more room to resolve an order.

If you trade live games, treat the hourly view as a floor on speed differences, not a measurement of them.

Does Moving First Mean Being Right?

No. Price discovery measures who was early; it does not measure who was correct. A venue can react first to a rumor that fades, or to a single large order, and the other venue’s slower reaction can turn out to be the better one. Accuracy needs resolved outcomes.

That is the job of our prediction market scorecard, which scores each venue’s prices against what actually happened using Brier scores and reliability curves. The two pages answer different questions and are best read together. A venue that is early and well calibrated is a strong reference. A venue that is early and poorly calibrated is fast but noisy.

Our guide to calibration explains why calibration is the right test of whether a price means what it says. Timing and calibration are independent properties, and neither implies the other.

How Can a Trader Use This?

Use the page as a map of where fresh information shows up. If the following venue lags in a category, the leading venue’s price is the better live reference for that category. During the lag window the two prices can also sit apart, and that gap is where cross-venue spreads tend to close.

A few practical readings:

What the page shows	A reasonable reading
One venue is first in most non-simultaneous moves in a category	Use it as the reference price for that category
Lag-correlation bars are symmetric	No consistent leader; check both venues
Most events are simultaneous	Hourly data cannot separate the venues; do not infer a leader
Wide gap between venues right after a move	Possible spread that is closing, subject to fees and depth

Two cautions apply. A gap that exists during a lag window may be too small to survive fees, and it may only be executable at sizes the thinner venue’s order book cannot absorb, so check liquidity before treating a spread as an opportunity. And fee structures, access rules, and contract terms differ by venue, which our platform comparison lays out side by side.

To see current prices for the same events, the Odds Reference dashboard shows every market in one view. The methodology page states the rules above in full.

Frequently Asked Questions

Does the faster venue always set the price?

No. The first mover is the venue that reached half of a large move earlier, which says nothing about permanent leadership. Which venue leads depends on the pair, the category, and the window, and it can change between snapshots. The live page recomputes every six hours so the current split is always visible.

Why can one hour hide the order of two price changes?

Our price series are hourly closes. If both venues cross the halfway point of a move inside the same hour, there is no way to say which went first, so the event is recorded as simultaneous instead of guessed. Fast in-game sports moves hit this limit most often.

What happens if one venue never reaches the halfway point?

A venue that never covers half of the consensus move inside the 24-hour window is treated as arriving last. The other venue is then the first mover, provided it got there. This keeps stalled or thinly traded listings from being skipped, since a failure to follow is itself a lag.

Is an early price move more accurate than a late one?

Not necessarily. Moving first measures timing, not correctness. A venue can react early to news that proves misleading, or to a single large order. Accuracy is measured separately, against resolved outcomes, in our scorecard of Brier scores and calibration curves.

Key Takeaways

Price discovery asks which venue moves first when a shared market reprices; the venue that moves first is where the information entered.
A move is an 8-point consensus change within 24 hours, and the first mover is the first venue to cover half of it. Same-hour arrivals count as simultaneous.
The lag-correlation chart shows direction through asymmetry: heavier bars on one side of zero mean that venue’s changes show up on the other later.
Hourly data cannot order very fast moves, so in-game sports events mostly land as simultaneous.
Moving first is not the same as being right; pair the discovery page with the scorecard before choosing a reference price.