March Madness – Were the Markets Right?

March Insanity has all the time been a check not simply of faculty basketball groups, however of the world’s prime forecasters. The event turns thousands and thousands of followers into informal statisticians.

Each spherical turns into a working argument not nearly what’s going to occur, however how seemingly it’s to occur. That distinction issues. Once we look again at predictions in politics or financial information that folks publicly received mistaken, we regularly concentrate on what they missed, what data they didn’t account for. However that framing misses one thing essential. Even when one thing solely occurs a small share of the time, in a single trial that final result can nonetheless happen. Low likelihood doesn’t imply unimaginable, it simply means unlikely.

So after we ask whether or not prediction markets “received March Insanity proper,” we’re actually asking one thing deeper: had been the possibilities themselves correct? Not whether or not each favourite received, however whether or not the chances mirrored actuality.

Did the Prediction Markets get March Insanity Proper?

If each recreation had been priced at 90% and each favourite received, that wouldn’t imply the market was good, it will imply it was mistaken. A 90% likelihood implies that upsets ought to occur about 10% of the time. In the event that they by no means do, the mannequin is underconfident.

So to reply whether or not or not the markets received March Insanity proper, we are able to’t depend on uncooked wins and losses. We’d like a solution to consider whether or not the possibilities themselves made sense.

Enter the Brier Rating…

What’s the Brier Rating?

The Brier rating is a manner of evaluating how good probabilistic predictions really are. As an alternative of asking “did you get it proper,” it asks “had been your possibilities correct over time?”

Give it some thought like predicting the climate. Think about on daily basis you give a likelihood that it’s going to rain, 20%, 70%, 90%. Over time, these possibilities ought to match actuality. Whenever you say 70%, it ought to rain on roughly 70% of these days. Whenever you say 20%, it ought to solely rain 20% of these days.

The Brier rating measures how nicely that alignment holds. For every prediction, you evaluate the likelihood to what really occurred, sq. the distinction, and common it throughout many observations. Decrease scores are higher, like in golf.

What makes this convenient is that it punishes you for being confidently mistaken. Saying “90% probability of rain” on a sunny day hurts rather more than saying “55%” and being barely off. Over time, somebody whose possibilities match actuality, somebody nicely calibrated, will outperform somebody simply making daring guesses.

That’s the important thing thought for March Insanity evaluation. The query isn’t whether or not the market picked each winner. It’s whether or not, throughout dozens of video games, higher-probability groups really received extra typically than lower-probability ones.

So had been the markets proper?

To reply this, I put collectively a dataset of all 63 video games within the 2026 event utilizing a mixture of handbook assortment and automatic evaluation. Meaning there may be definitely room for human error within the inputs, and this must be considered much less as a definitive measurement and extra as a directional illustration of how nicely prediction markets carry out.

For every recreation, I recorded the pre-game likelihood implied by market pricing and in contrast it to the precise final result. From there, I calculated the Brier rating throughout the complete event.

The end result: An general Brier rating of 0.1536.

To place that into context, a totally uninformative mannequin that assigns 50/50 odds to each recreation would rating 0.25. An ideal forecast would rating 0. So a end result round 0.15 means that, on common, the possibilities had been meaningfully higher than random and broadly aligned with how typically these outcomes occurred.

Breaking it down by spherical provides some texture:

This sample makes intuitive sense. Early rounds, the place there are a lot of heavy favorites, are simpler to cost and lead to decrease (higher) Brier scores. Later rounds are nearer matchups, with extra uncertainty, and the errors enhance.

The weakest stretch got here within the Last 4, pushed partially by outcomes like UConn beating Duke at roughly 32% implied likelihood after which beating Illinois at round 45%. These aren’t unimaginable outcomes, however they’re outcomes that ding your rating.

March Insanity Odds Variations between Sportsbooks

Whereas I didn’t evaluate the above prediction market Brier scores to sportsbooks, there’s one other layer right here that helps clarify comparatively how prediction markets fared.

Based on evaluation from Residents Capital Markets, Kalshi really provided barely higher pricing than main sportsbooks throughout the event, with a mean “vig” of 4.13% in comparison with roughly 4.3–4.5% for rivals.

Which may sound trivial, however in a thin-margin setting like sports activities betting, even a fraction of a % issues. In relative phrases, that suggests Kalshi’s take was about 7% decrease than sportsbooks, that means costs had been barely extra environment friendly.

Last Verdict – Had been the Markets Proper?

Prediction markets are nonetheless comparatively new to sports activities at scale, and this type of evaluation remains to be early. The dataset right here is imperfect, the methodology is easy, and higher variations of it will come.

However even with these limitations, the takeaway is pretty clear: the markets weren’t simply guessing. They had been, in a measurable sense, getting the possibilities proper.