Skip to content
TwapTrade

Prediction markets vs polls

Updated

A poll measures what a sample of people say they think or plan to do at one moment; a prediction market price reflects what traders, with money at stake, expect to happen. Markets update within minutes and give a single probability, while polls explain who thinks what and why. They work best together, and neither is right every time.

Prediction marketOpinion poll
What it measuresTraders' expectation of an outcomeWhat a sample says it thinks or plans
OutputA price read as a chance (63¢ ≈ 63%)Shares of opinion, with a margin of error
Money at stakeYes, traders gain or loseNo
How often it updatesContinuously while the market is openWhen a new poll is published
Explains whyNoYes, through its questions and breakdowns
Main weak spotsThin trading, shared blind spotsSampling error, non-response, turnout guesses

Polls and prediction markets are often quoted side by side, but they answer different questions. A poll tells you what a sample of people say right now. A prediction market tells you what traders, with money on the line, expect to happen. Knowing the difference helps you read both.

What each one measures

A poll asks a sample of people a question and reports their answers, with a margin of error. The American Association for Public Opinion Research (AAPOR) describes election polls as a snapshot of opinion at one point in time, not a forecast of the result.

A prediction market asks what will happen. Traders buy Yes or No shares priced between 1¢ and 99¢, and a winning share pays $1. The price works as the crowd's estimate of the chance: a Yes share at 63¢ means traders see about a 63% chance. See how to read prediction market odds.

So a poll showing a candidate ahead 48% to 45% and a market giving that candidate a 62% chance aren't in conflict. One is a share of the vote; the other is a probability of winning. A small, steady lead can still mean a good chance of winning.

Money at stake

Poll respondents don't lose anything for being wrong. Traders do, which gives them a reason to weigh all the information they can find, polls included, and to trade against prices that look off. If enough people think a price is too low, their buying pushes it up.

The flip side is that a market reflects the people who trade in it, who aren't a cross-section of the public. When few people are trading a question, one large trader can move the price for a while, and a crowd can share the same blind spot.

Speed of updating

A market price can move within minutes of a debate, a court ruling or a resignation, and it keeps moving until the market settles. A poll takes days to run and publish, and it describes the days when the interviews happened.

Speed cuts both ways. Markets can overreact to noise, and a sharp price swing isn't proof that anything important has changed. Polls are slower, but they're designed to measure a defined group of people with known methods.

What the research says

A widely cited study of the Iowa Electronic Markets compared its prices with 964 national polls across the US presidential elections from 1988 to 2004. The market was closer to the final result 74% of the time, and its advantage was largest more than 100 days before election day.

Other work is more mixed. A 2009 study of the 2008 US elections found raw market prices were, on average, slightly less accurate than poll-based forecasts, though markets came out ahead once both were adjusted for known biases. AAPOR also points out that a poll's reported margin of error covers only sampling error, not problems such as non-response or wrong turnout guesses, so the real uncertainty is bigger than the headline margin. For more, read are prediction markets accurate?

When each is more useful

  • Polls are better for understanding what people think about issues, how groups differ, and why opinion is shifting. A market price can't tell you any of that.
  • Markets are better for a single, up-to-date probability of an outcome, including questions polls don't cover, like an interest-rate decision, a match result or the weather.
  • Both together are most useful when they disagree. A gap between a polling average and a market price is a prompt to look for the reason: a new poll, a change in the rules, or traders reading the same numbers differently.

Reading them side by side

  • Check a poll's dates and sample, and whether you're looking at one poll or an average of many.
  • Check how actively a market is traded. A price that rarely changes hands is a weaker signal.
  • Treat a 70% price as 70%, not as a sure thing. If a market is well calibrated, outcomes priced at 70% still fail about 3 times in 10.
  • Follow election markets on our US elections odds page and compare them with the latest polls.

Questions

Are prediction markets more accurate than polls?

Sometimes. A long-running study of the Iowa Electronic Markets found market prices closer to the result than polls 74% of the time, but research on the 2008 US elections found raw market prices slightly less accurate than poll-based forecasts. Neither is right every time.

Why can a candidate lead in the polls but have a lower chance in the market?

Because they measure different things. A poll measures vote share today; a market prices the chance of winning on the day, taking in things like the electoral map, turnout and how far polls have missed before.

Do prediction market traders use polls?

Many do. New polls are one of the things that move election markets, alongside news, results and other data.

If a 70% favorite loses, was the market wrong?

Not necessarily. A 70% chance means the other side should win about 3 times in 10. You judge a forecast by how it does across many questions, not one.

Can one big trader move a prediction market?

In a thinly traded market, yes, for a while. Other traders can trade against a price they think is wrong, which tends to pull it back, but it's worth checking how actively a market trades before reading much into its price.

More comparisons