The polls are broken

Front of mind: prediction markets can’t fix opinion polls if they simply copy them

Author

Robert Gilhooly

Senior Emerging Markets Economist

Duration: 2 Mins

Date: Oct 15, 2026

November’s US midterm elections will shape the balance of power in Washington and determine how much of the president’s agenda can pass Congress. Polls point to a Democratic wave, and prediction markets agree.

But recent elections have humbled pollsters and forecasters like me, because the polls can get it wrong.

I see prediction markets as an independent check only when traders use other information. If ‘thin’ markets simply follow the polls, I think they may repeat the same error.

The 2024 US presidential election is a case in point.

A result that defied the polls

Commentators later explained Donald Trump’s decisive win through Joe Biden’s late withdrawal, Kamala Harris’s short campaign and the assassination attempt. Yet before election day, polls still suggested a close race.

The Selzer poll drew attention before election day, suggesting that Harris was leading. In the end, Trump carried Iowa by 13 points – a big miss.

Aggregated state polls were often within their margins of error. The problem was that errors moved in the same direction. Aggregation reduces random noise, not shared bias. Forecasts built on the polls were blinded to that risk.

Aggregation reduces random noise, not shared bias

Recent primary polls offer little reassurance: results in Minnesota, Wisconsin and Michigan missed by 10 to 25 percentage points. That matters because the same polling evidence is shaping expectations for November’s midterms.

Another mirage?

Polls now suggest a Democratic wave, even in Republican strongholds. Prediction markets increasingly back the Democrats too.

High petrol prices, falling presidential approval and the usual backlash against incumbents support that view. Even so, I wonder whether polls and prediction markets have become too confident about Democratic gains.

The economist in me is still wary about two things.

First, the Senate map favours the Republicans. Democrats need to gain four seats to take control, so only a few ‘toss-up’ states need to break to the Republicans to stop the Democrats taking the Senate.

Second, polling is harder when fewer people answer unknown calls or disclose their choice. It may be that people don’t want to say who they are voting for in an increasingly polarised world. If ‘shy’ voters are more prevalent in one party, polls can be systematically skewed.

There’s a prediction market for that

Prediction markets are everywhere now, bringing the ‘wisdom of crowds’ into an array of economic, political and cultural events.

Prediction markets correctly favoured Trump in 2024. And I think they can provide a useful check on the polls when traders draw on evidence beyond polling data.

But I’m cautious this time round because individual Senate races are often ‘thin markets’, with fewer traders and less independent information. Published polls can therefore become the main input into prices: polls shape traders’ expectations, and market odds then repeat the same signal. I think prediction markets can challenge the polls only when they contain genuinely independent information and assessment.

Otherwise, they risk giving uncertain evidence false confidence, and I suspect the race may be closer than the polls and prediction markets suggest.