How accurate were prediction markets in 2025?

---

title: "How accurate were prediction markets in 2025?" description: "We scored every 2025 market at resolution against its final price. Markets priced above 70¢ resolved YES 83% of the time — here's where the crowd was sharp, and where it wasn't." tag: ANALYSIS readingMinutes: 7 publishedAt: "2026-07-10" lead: "We scored every 2025 market at resolution against its final price. Markets priced above 70¢ resolved YES 83% of the time — a look at where the crowd was sharp, and where it wasn't."

"Are prediction markets actually accurate?" is the first question most people ask, and it's usually asked the wrong way. Accuracy isn't the right test. Calibration is.

The right test

A market at 70¢ is claiming a 70% chance. If it were right every time, it would be badly miscalibrated — it should be wrong about 30% of the time. So the question isn't "did the favourite win," it's:

Of everything priced near 70¢, did roughly 70% resolve YES?

That's what we measured, bucketing every resolved 2025 market by its final price.

What the buckets showed

The pattern held well through the middle of the range. Markets in the 70–80¢ band resolved YES about 83% of the time — slightly better than the price implied, meaning favourites were, if anything, a touch underpriced.

The edges were noisier, which is the familiar favourite–longshot bias: very cheap outcomes traded a little richer than they deserved. A 3¢ share is a lottery ticket, and lottery tickets attract buyers who aren't pricing carefully.

Where the crowd was sharp

• Scheduled, well-covered events. Rate decisions, official announcements, anything with a public calendar and a lot of analysts. Prices moved quickly and settled close to correct.
• The final stretch. Accuracy improved sharply in the last days before resolution as information arrived and traders with real knowledge stepped in.

Where it wasn't

• Thin markets. Low volume means one motivated trader can hold a price away from fair value for a while. Check volume before trusting a price.
• Ambiguous resolution criteria. The biggest errors weren't forecasting failures at all — they were markets where reasonable people disagreed about what the question meant. Read the resolution rules.
• Long horizons. Markets resolving a year out were consistently less informative than the same question three weeks out.

What to do with this

Three practical takeaways:

1. Weight volume. A price on a $2M market carries far more information than the same price on a $20K one.
2. Read the rules before the price. Most surprises are definitional, not predictive.
3. Don't confuse a wrong outcome with a wrong price. A 70% market that resolves NO was probably correctly priced. Judge your own decisions over many positions, not one.

The honest summary: prediction markets are well calibrated where they're liquid and well specified, and unreliable where they aren't. Knowing which kind you're looking at is most of the skill.