Prediction markets aren't wrong just because an underdog wins

Prediction markets gave Luiz Inácio Lula da Silva roughly a 20% chance of winning Brazil's presidential election. Then he won.

Cue headlines declaring that prediction markets had failed, missed the result, or got the election spectacularly wrong.

But there's a fundamental misunderstanding at play.

A 20% chance of winning isn't the same as no chance of winning.

If an outcome has a genuine 20% probability, you'd expect it to happen approximately once in five across comparable events.

An underdog winning doesn't automatically mean the original forecast was wrong. It means the less likely outcome occurred.

And that's a distinction that matters whenever people use prediction markets to understand uncertain events.

Why prediction markets are partly to blame

The confusion isn't entirely the public's doing.

Following Donald Trump's victory in the 2024 US presidential election, Kalshi and Polymarket celebrated their markets having favored Trump over Kamala Harris.

Kalshi CEO Tarek Mansour even posted:

Polls 0 Prediction Markets 1

— Tarek Mansour (@mansourtarek_) November 6, 2024

But Harris still had a substantial chance of winning according to those same markets.

Trump's victory didn't prove that prediction markets were inherently more accurate than polls, just as Lula's victory doesn't prove they were inaccurate.

There's a contradiction in celebrating when a favorite wins, then criticizing people for misunderstanding probabilities when an underdog wins.

If prediction market platforms want their forecasts judged as probabilities rather than definitive predictions, their own messaging needs to reflect that.

What prediction market probabilities actually mean

On a prediction market, a contract trading at 80¢ generally implies an approximately 80% chance that the event will occur.

If the event happens, a winning contract settles at $1. If it doesn't, the contract settles at zero.

But that 80¢ price isn't a guarantee. It reflects what market participants are collectively willing to pay, influenced by available information, trading activity, and market conditions.

And even if the 80% estimate is accurate, the outcome won't happen every time.

To assess whether prediction markets are genuinely accurate, you need to look beyond individual results.

Across a sufficiently large sample, events assigned a 20% probability should occur roughly 20% of the time. Those assigned an 80% probability should occur roughly 80% of the time.

That's a more meaningful measure of forecasting quality than simply counting how often the favorite wins.

Why this matters for bettors

The same principle applies to sports betting.

At Dimers, we distinguish between identifying the most likely outcome and identifying a bet that offers value at the available odds.

A basketball team might have a 70% chance of winning, but backing it isn't necessarily a good bet if the odds imply an 85% probability.

Conversely, a team with only a 30% chance could represent value if the available odds imply just 20%.

Just because a team is more likely to win doesn't mean it's a good bet. And losing doesn't automatically make the original probability wrong.

Understanding that difference is essential when real money is involved.

Probabilities aren't promises

With the US midterm elections approaching, prediction markets will again attract attention as traders price the chances of different political outcomes.

Some favorites will win. Others won't.

Neither result, in isolation, proves that prediction markets succeeded or failed.

The challenge for Kalshi, Polymarket, and other operators is communicating that uncertainty clearly, rather than promoting their ability to pick winners.

And for anyone using these platforms, the lesson is straightforward.

Prediction markets don't tell us what will happen. They tell us what the market currently thinks is likely to happen.

Those are two very different things.