Prediction Markets Explained

Why prices aren’t probabilities, and what traders need to know.

​Prediction markets​ had a breakout moment in 2024, with bets based on the US election attracting multi-billion-dollar volume (and huge media attention).

While things have calmed down since then, volumes are still far higher than pre-election averages. Prediction markets are here to stay.

Prediction markets open up a whole new asset class, allowing investors to translate superior forecasts of real-world events directly into profits.

But there’s still a tremendous amount of misinformation about these markets, with potentially serious implications for anyone trading event contracts.

Today, I’ll cut through the noise and explain how prediction markets really work.

In the first half of this issue (free), I’ll explain event contract mechanics and describe why prediction markets can be so valuable.

In the second half (paywalled), I dive into the finer details, exploring the two most important misconceptions of prediction markets:

1. Why event contract prices aren’t probabilities
2. Why prediction markets may not be an accurate forecasting tool.

The second half gets technical — but don’t be scared. I’ll break things down simply, and if you’re serious about trading in prediction markets, these are details you absolutely want to understand.

Let’s go 👇

Since you have the All-Access Pass, you get the full issue. ✅

Let’s go 👇

​​Brian Flaherty​ purchased his first mutual fund at 15. After graduating from UVA with a degree in Economics, he began advising institutions and high-net-worth investors as a strategist at a wealth management firm. Today, Brian helps investors uncover the best opportunities and make intelligent use of their capital. You can follow him on ​LinkedIn​.

How do prediction markets work?

Prediction markets allow investors to trade a unique type of asset: event contracts.

Event contracts can be created on almost any topic. The most popular are politics, economics, sports, weather, and technology. (Some markets are off-limits, including death. Kalshi recently shut down a market ​betting on the fate of Luigi Mangione​).

Event contracts have a binary yes/no payoff based on a specific future event. This is what separates prediction markets from traditional betting platforms.

See, in sportsbooks and casinos, you’re betting directly against the platform. Event contracts, however, are a form of peer-to-peer betting in which you trade directly against other bettors.

(This is similar to ​Kutt​, the ​P2P betting startup I recently wrote about​).

Event contracts are traded on a market, and need standardized terms.

Here’s how it works:

1. Each contract has two sides, Yes and No. You open a position by buying either Yes or No. There is only one winner.
2. The price of Yes and No fluctuates as new bettors place bets.
3. All contracts have a $1 payout, which gets distributed entirely to the winning side.
4. Some events have multiple contracts, such as a presidential election with several candidates. But there is still only one winner.
5. You can sell your position before or after the contract ends.
6. You can only sell a position you own — there is no shorting.

(Technical note: Prediction markets don’t actually have separate Yes or No contracts. “Under the hood,” buying Yes is the same exact thing as selling No.)

For example: Suppose I wanted to bet that RFK Jr will be confirmed as Trump’s Health Secretary.

• Right now, I could buy one ​Yes​ contract on Kalshi for about 74¢. If I’m right, I’ll get paid $1 when the contract settles, netting me a profit of 26¢ for a 35% return.
• I could also bet ​against​ RFK’s confirmation by buying a No contract for 27¢. This is riskier, but if I’m right, I could earn a 170% return (73¢ profit on 27¢ outlay).

Now, I must address a near-universal misconception that many people hold about prediction markets:

Prediction market prices do not precisely reflect the collective beliefs of an event occurring.

Yes, prediction market prices approximately reflect collective beliefs, and those approximations can be very close to realized outcomes.

But there are some technical reasons why prices stray from true probabilities, especially at the tails. You can already see evidence in the fact that the two RFK contract quotes add up to $1.01 — which shouldn’t be possible if prices reflect exact probabilities!

I’ll explain this in more depth later on.

A short history of prediction market regulation

Before we go any further, we need to talk about the importance of CFTC, or the ​Commodity Futures Trading Commission​.

CFTC regulation isn’t just what sets apart legitimate derivatives exchanges from gambling platforms. It’s what lets prediction markets operate in the first place.

The past decade has not gone smoothly for several prediction market companies:

• In 2012, Irish prediction market InTrade was taken to court by the CFTC for violating various rules. Six months later, the platform was gone for good.
• In 2022, ​Polymarket settled a $1.4m suit​ with the CFTC for offering noncompliant event contracts.
• Seven months later, prediction market pioneers PredictIt also found themselves in hot water. The New Zealand-based platform was banned from servicing US Investors by the CFTC.

Regardless, this is why CFTC regulation is so important. Without it, the agency has a clear path to probe and potentially penalize prediction markets. The big winner from all of this is ​Kalshi​.

This is huge. Kalshi investors don’t have to worry about the big bad agencies coming to shut them down, take away their investments, or smash their liquidity.

What’s the purpose of prediction markets?

A more direct way to hedge

Prediction markets clearly offer value to people who just want to gamble on a wide range of events. But they can also provide value to savvy investors as a risk management tool.

Let’s say you believe China will invade Taiwan in the coming year. You could express that view by going long American defense companies, or short ​TSMC​.

But both of these positions are imprecisely related to the underlying event — they’ll also be influenced by dozens of other factors.

In contrast, event contracts can offer a financial instrument precisely tied to the event you want to trade. This is a much cleaner way to speculate or hedge a specific outcome.

Back in 2022, we recorded a ​podcast​ with ​Tarek Mansour​ and ​Luana Lopes Lara​, co-founders of Kalshi. As Tarek put it, this risk management idea was one of the key reasons behind Kalshi’s creation:

When I was at Goldman in 2016, I was working at this exotics desk. And one of the main things we’d done that summer was ​structured products​ for institutions that wanted to get exposure to or hedge themselves against Brexit… It was a lot of different financial instruments. And the question was like, why is there no direct way of doing it?

But prediction markets aren’t just valuable to individuals — they can benefit society.

Prediction markets can be more accurate than polling

Aggregating the insights of many people tends to be more accurate than relying on the prediction of just one individual, an idea known as the ​wisdom of the crowd​. (This is the basic logic behind polls and surveys.)

But prediction markets take this a step further by forcing forecasters to back up their beliefs with actual cash.

Prediction markets aren’t perfect, but ​academic research indicates they can be more accurate​ than traditional polling in forecasting political events.

Calibration tests from Kalshi show that when aggregating over many events, prices mostly align with the proportion of realized event outcomes.

(NOTE: Forecasting individual events is much different — in the paid section I’ll explain why).

Companies like Google and Ford have even deployed ​internal prediction markets​ to better forecast things like project timelines and expected sales.

That doesn’t mean that prediction markets are without flaws, however.

Now let me show you why these markets can be controversial and misinterpreted — starting with the three reasons why prices aren’t probabilities.

Why prices are not probabilities

It’s tempting to treat prediction market prices as the “collective belief of the probability of the outcome.”

This idea is further entrenched because some markets even quote probabilities as if they were equivalent to prices.

But this isn’t quite right. There are at least three valid, rational reasons that prediction market prices should systematically deviate from true underlying probabilities.

Reason 1: The risk-free rate

Prediction markets seem to exhibit a persistent longshot bias, with things that very clearly aren’t going to happen nonetheless trading at positive prices.

For example, let’s look at the Kalshi market for the ​number of Fed rate cuts in 2025​. Specifically, observe that markets seem to expect a 2% chance of seven Fed rate cuts this year (?!)

Now, I’m a smart guy, and I feel very confident that the Fed won’t cut rates 7 times this year. Heck, there are only 7 more scheduled Fed meetings for the entire year, and we’d have to see truly rapid economic deterioration for cuts to occur at every single meeting!

However, that doesn’t necessarily mean I should buy the No contract! Here’s why:

• For every contract I buy at 97.9¢, I could earn 2.1¢ in profit for a return of 2.15%.
• This contract doesn’t settle until the end of the year. If I bought those contracts on February 1st, my annualized return would be roughly 2.34%.
• Why would I ever accept a risky return of 2.34% when I can currently earn 4.17% risk-free on a 1-Year Treasury?

Basically, because of the risk-free rate, it’s not always worth it for investors to take the other side of longshot trades. That can result in persistent price distortions that make improbable outcomes appear more likely than they truly are.

This also helps explain why long-shot contracts tend to fall to 0 as the end of the contract approaches: arbitraging small price-probability discrepancies becomes more attractive on a time-value basis.

The risk-free rate has a fairly straightforward impact on contract prices, but it’s actually a subset of a more complex issue: the existence of alternative investment opportunities.

Reason 2: Opportunity cost

Let’s continue with our Fed rate cut example.

Suppose that you are absolutely certain that the Fed will cut 7 times this year — there’s just no doubt in your mind.

Does that mean you should buy the Yes contracts? Possibly, but it’s not a foregone conclusion.

Assuming that you have a limited amount of capital, you need to find the most capital-efficient way to express your view. And there are plenty of other asymmetric ways to bet on interest rates falling, including:

• Going long high-duration ​Treasury bonds on margin​
• Buying deep, out-of-the-money ​Federal Funds options​

This problem is even more severe if you consider liquidity constraints in prediction markets.

The Kalshi order book (​sign up to view it​) shows me that there are only about 4,000 contracts available for sale at less than 4¢. If I want to do this trade in serious size while preserving my asymmetry, I’ll probably need to look to larger markets.

The bottom line is that even a bettor with perfect information isn’t always incentivized to arbitrage price-probability differentials.

Traders need to find the most capital-efficient way to express their views, which may live on prediction markets, but they may live elsewhere.

Reason 3: Bid-ask spread

Earlier, I mentioned that the RFK contracts didn’t sum to $1. The reason is due to the bid-ask spread.

All financial exchanges have a bid-ask spread, and prediction markets are no different. The bid-ask spread represents the cost of immediacy (aka the fee that liquidity providers earn).

If I wanted to take a round-trip in the RFK market at quoted prices, I would need to buy the Yes at a 74¢ ask and then immediately sell it at a 73¢ bid.

Remember that selling the Yes is equivalent to buying the No. And indeed, selling Yes at 73¢ implies buying No at 27¢.

In other words, the price-implied probabilities at the same instant are 74% and 27%, summing to 101%. This extra 1% reflects the distorting impact of the bid-ask spread.

This effect can be more noticeable in markets with low liquidity. The current market on Kalshi for whether the ​stranded astronauts​ will be home by July 1st somehow implies both a 98% chance that they will return by that date, and also a 7% chance that they won’t — clearly violating basic probability.

Are prediction market prices actually informative?

In the previous section, I argued that prediction market prices can only approximate underlying probabilities.

Now, I’m going to explore a more controversial idea…

Contrary to popular opinion, careful data analysis shows that prediction market prices may not actually offer much useful information for forecasting individual events.

To understand why, we’ll need a brief statistics refresher.

(This next part will get a bit dense, but if you’re truly interested in these markets, it’s absolutely worth your time.)

Calibration vs accuracy

Earlier, I showed a chart showing that Kalshi prices appear well-calibrated to actual event outcomes.

However, calibration is not the same thing as accuracy.

To understand the difference, imagine if I were trying to build a predictive model to forecast the chance of rain in a wet region over the course of a year.

Based on historical data, I see that it tends to rain about half of all days. Therefore, I develop a simple model — every day, I predict a 50% chance of rain.

Over the year, it ends up raining exactly 183 days while being dry the other 182. In the aggregate, my model is extraordinarily well-calibrated — events that I predicted would occur 50% of the time ended up happening about 50% of the time.

But on a day-to-day basis, my model is absolutely useless!

Flipping a coin to determine whether it will rain each day offers no predictive value. In statistical terms, our model has very poor accuracy.

So, how do you measure the accuracy of a predictive model?

One of the best ways is with a Brier Score.

What is the Brier Score?

The Brier Score can be used to determine whether prediction market prices offer useful forecasting information.

You don’t need to understand all the ​intricacies​. I’ll give you the main idea.

A Brier score measures the average error a model has in forecasting individual events (technically, the ​average squared error​).

Here’s how it works:

1. Events that occur are assigned a value of 1, and events that don’t get a value of 0.
2. Suppose our model predicts the likelihood of an event at .90 and the event occurs. This error is small (.10)
3. On the other hand, if the event does not occur, the error will be large (.90).
4. This logic also applies to low predicted probabilities for events that don’t occur (small error) and events that do occur (large error).

Thus, the lower the Brier score, the more accurate our model is at predicting individual events.

Now, let’s return to our 50% rain forecast model — what will our Brier score be?

1. Each day, our error will always be either -.50 (if it rains) or .50 (if it doesn’t).
2. Squaring these errors gives us .25 in either case.
3. Thus, our average squared error (and resulting Brier score) = .25

Okay, great! So this .25 figure is a special score known as our Reference Score.

The Reference Score is your baseline

When trying to predict an event, the simplest model we can use is guessing the ​base rate​.

Suppose that I meet someone new, and I’m trying to guess whether or not they’re an ​accredited investor​.

I know that about 1 in 5 people in the US are accredited. Thus, without incorporating any information about the person, my base rate guess for their accreditation status would be 20% (or .20)

This is exactly the way I came up with my original rain forecasting model. I looked at the historical data, discovered a 50% base rate, and then used that as my forecast.

The base rate strategy Brier Score is the baseline by which all other models are measured.

Now, if I develop a model that incorporates new information, that information only helps provide more accurate forecasts if the model has a lower Brier Score than the Reference Score.

How to calculate Brier Scores for prediction markets

Note: I would have preferred to use Kalshi for this analysis, since I have some concerns about the ​quality of Polymarket’s trading data​. But Polymarket data is more easily accessible.

Based on this understanding of Brier scores, we can come up with a coherent strategy for determining whether prediction market prices offer useful information in forecasting the underlying events.

There are 4 steps:

1. Find the base rate proportion for all contracts on the market that settle to ‘Yes.’
2. Calculate the Brier score using this base rate as a universal forecast for the likelihood of every event (our reference score).
3. Calculate the Brier score using contract prices to forecast the individual underlying events.
4. Compare the two Brier scores to see if prices offer new predictive value.

The latest data shows that ​26.4% of events​ on Polymarket settle to Yes.

The Brier score for a constant 26.4% prediction with a 26.4% base rate is 0.194. This is our Reference Score.

So here’s the million-dollar question: what is the Brier score of using prices as forecasts?

Throughout a contract’s life on Polymarket, midpoint prices (which split the bid-ask) consistently have a Brier score above or around 0.20.

This is such a monumentally weird finding!

The data implies that, when it comes to forecasting the likelihood of individual events, using prices as probabilities is often less accurate than just guessing the base rate.

We’ve already seen the financial reasons why prices will deviate from underlying probabilities. But the price-probability relationship may be far weaker than many realize…

I must admit, I was really surprised by this finding, but I don’t know of any other way to interpret the data. I also ran this conclusion by individuals far more knowledgeable about statistics than myself, who agreed.

Here is one final, intriguing point that might offer some supporting evidence for this idea:

• If prices contain useful predictive information, then Brier scores should fall over the course of a contract as prices incorporate more and more information.
• The fact that Brier scores stay roughly stable over a contract’s life on Polymarket supports the idea that prices don’t have much useful forecasting value.

It’s important to note that this data is just on Polymarket — and ​some analysis from 2022​ on Kalshi came to the conclusion that prices did contain useful predictive info.

If you’re a statistics expert, please reach out if you have any thoughts on this data!

Join the discussion →

My conclusion & interpretations

One way to interpret the previous section is to view prediction market prices as the sum of information and noise.

The information component contains useful forecasting value. But there’s a noisy error term that’s so large it completely swamps any valuable information…

In the aggregate, noise tends to cancel each other out (the errors have a mean of zero). That allows the information content to dominate over many contracts, revealing well-calibrated price-probability plots.

But for individual contracts, noise is still the deciding factor.

I’m tempted to draw a fascinating conclusion: Investors who are even slightly better informed than the market should have the potential to realize outsized profits.

Is it really that hard to build a predictive model that beats the reference Brier score? I’m not sure! But if you want to go experiment and find out, here are a few prediction market resources worth exploring:

• ​PredictionMarketMap.com​ – a useful roundup of the prediction market universe.
• ​Calibration.city​ – the source of a lot of the data used in this article.
• ​Market Prices Are Not Probabilities​ – goes into a bit more depth on the argument from a trader’s perspective.
• ​A Brief on Brier Scores​ – helpful for understanding Brier scores more formally.
• ​Kalshi​ is a CFTC-regulated market open to US investors, though liquidity is lower than Polymarket.
• ​Polymarket​ is the world’s largest prediction market, although it’s unregulated, technically not open to US investors, and you need crypto to use it.
• ​PredictIt​ is a prediction market run for academic purposes. As a result, traders are limited to $850 positions.
• ​Interactive Brokers​ offers event contracts through ForecastEx and CME.

—

That’s it for today!

Come ​chat with me about this issue​ in the Alts community!

See you there,
Brian

Disclosures

• This issue was written and researched by ​Brian Flaherty​ and edited by Stefan von Imhof.
• This issue was sponsored by ​GrowthSchool​ and ​Kalshi​
• Kalshi had zero editorial input or oversight in this issue.
• After researching this issue, Brian may place some trades in prediction markets. He held no positions in prediction markets at the time of writing or publication.
• This issue contains no affiliate links. It does contain non-affiliate links to Seeking Alpha, an Alts partner.
• Neither Alts nor Altea has any current holdings in any companies mentioned in this issue
• This is a paid issue. To read the full thing you need the ​​​​All-Access Pass​.