What Moves Prediction-Market Prices

A prediction-market contract is a bet on a defined outcome that pays $1 if it resolves yes and $0 if it resolves no. Before resolution, it trades somewhere in between, and that price is a running estimate of probability. The interesting question for anyone building strategies is not what the price means, it is what makes it move. Five forces do most of the work: news, polls, large orders, liquidity, and the clock. This post walks through each and how to turn it into a signal you can trade systematically.

The stakes have gone up. Prediction markets are no longer a niche. According to a Pew Research Center analysis of data from The Block, combined monthly global volume on Kalshi and Polymarket rose from less than $5 billion in September 2025 to about $24 billion in April 2026.

News: the fastest input

News is the primary driver. A contract on a Fed decision, an inflation release, or an election result reprices the moment fresh information lands. On a Polymarket election page, prices can move fast when major news breaks, such as a candidate withdrawing or a no-confidence vote toppling a sitting prime minister.

What makes prediction markets distinct is speed. Traders increasingly watch them alongside Treasury yields and the dollar because platforms like Polymarket and Kalshi have become real-time indicators of how participants interpret a surprise inflation report or a Fed speech, often reacting within seconds, before slower financial indicators fully adjust.

For a systematic trader, the signal is the gap between an event and the market's reaction to it. If your feed knows a number before the order book does, you have a window. That window is small and getting smaller, so the engineering problem is latency: a strategy that ingests a data release, maps it to the contract's exact resolution rule, and acts before the price finishes moving.

Polls: signal, but not the way people assume

Polls feed prices, but the translation is not one to one. A new poll might barely move a national polling average yet trigger a sharper response on a prediction market, because traders read it as evidence about turnout in a decisive state rather than a nationwide shift.

The two measure different things. Pollsters ask a sample how they intend to vote and estimate wider opinion, while a prediction market records the prices its traders create, and the group placing those trades bears little resemblance to a representative sample. Some participants live outside the country, follow politics unusually closely, or focus on details they think the market has missed.

The systematic edge here is a poll-to-price model: quantify how much a given pollster, sample size, or state release historically moved a contract, then trade the residual when the actual reaction over- or under-shoots your estimate.

Large orders and liquidity: where prices lie to you

This is where careful traders make or lose money. In many prediction contracts, liquidity is still thin, and a relatively modest order can move prices significantly, creating the impression that probabilities have changed dramatically when only a handful of participants have traded.

Research on prediction market microstructure confirms this: in liquid markets, large trades have minimal impact on price and the accuracy of price discovery improves; in thin markets, a single large trader can move the price sharply and create the appearance of a broad shift that is not there. A market with limited liquidity produces a noisy price that overstates the information content of any single order.

There is a subtlety worth knowing. Standard microstructure theory says large trades carry private information and should sharpen prices. But a 2026 study of Kalshi and Polymarket (arXiv:2602.19520) found that in political markets on Kalshi, large trades are associated with amplified underconfidence: prices are pushed further toward 50%, not toward truth. The effect is statistically significant on Kalshi (Δ=0.53, 95% CI [0.29, 0.75]) but does not replicate on Polymarket, pointing to platform-specific microstructure rather than a universal pattern.

So a large print is ambiguous: it can be informed flow to follow or a whale distorting a shallow book to fade. A systematic approach separates the two by watching depth, spread, and whether the move holds after the order clears. That means reading the full order book, not just last price.

Resolution proximity: the clock is a signal

Time to resolution changes both accuracy and pricing behavior. As an event approaches, more information arrives and prices converge toward 0 or 1, so the final price before resolution is almost always the most accurate. Research on Brier scores confirms it: predictions made closer to resolution tend to score better as the market rapidly incorporates breaking information.

Far from resolution, markets behave differently. A 2026 analysis of Kalshi and Polymarket (arXiv:2602.19520) found a universal horizon effect: at long horizons prices are compressed toward 50%. The gap is large enough to matter in practice: a 70-cent political contract one week before resolution corresponds to a true probability of approximately 83%, not 70%, with politics the most underconfident category across both exchanges. That systematic understatement of the favored outcome is a tradable, testable bias, not folklore.

Sentiment: real but slippery

Prices reflect atmosphere as well as fact. Even in busy markets, participants may share information sources or assumptions, and a prevailing mood can hold a price for a time before reality corrects it. Sentiment is the hardest driver to model because it looks like signal until it snaps back. Treat it as a reason for wider risk bands, not a standalone entry.

Turning drivers into signals

A systematic trader does not trade a hunch about news. They encode each driver as an input, combine them, and size the position against a fixed risk budget. Roughly:

• News: latency-sensitive event handlers that map a release to a contract's exact resolution rule.
• Polls: a model of expected price impact per poll, trading the residual when reaction diverges.
• Order flow: order-book depth and spread filters to tell informed flow from thin-book distortion.
• Resolution clock: horizon-aware recalibration to exploit long-dated price compression.
• Sentiment: a dampener that widens risk bands rather than a primary entry.

The unglamorous part is testing whether any of it survives contact with a live book. This is where Banger fits. You write the logic as a Python banger.Strategy, paper-trade it against the live order book to see how it behaves under real spreads and depth, then run it with a declarative risk envelope: per-trade cap, daily loss stop, max open positions, and a kill switch. Banger never custodies funds. You bring your own venue keys.

pip install bangertrades
banger run strategy.py --paper

Start in paper mode. The order-book distortions described above are exactly the failures that look fine in a backtest and cost money live, so the paper stage is where you find out whether your news handler is really front-running information or just chasing a whale in a thin market.

One caution on calibration

Prediction markets are strong probability estimators in aggregate, often matching or beating experts, but they are not oracles. They aggregate well when liquidity is deep, resolution criteria are clear, and there is historical precedent, and worse when those conditions are absent. Build your signals for the conditions you are actually trading in, not the ones the marketing implies.