Don't just put the chart on the screen

What's missing when journalists cite prediction markets

Jan 16, 2026

If we’re going to start using prediction market prices as a complement to news, we should consider how new and still evolving these markets are. Just because a market price is available doesn’t automatically make that market efficient. Some of these markets are more like penny stocks than market benchmarks. When prediction market probabilities are projected as true odds without caveats, that’s not enough context for the viewer. Up front here’s my checklist:

Report volume (ideally cleaned of inferred wash-trading).
Avoid reporting contracts with insufficient volumes - this may vary depending on how important the market is.
Report a measure of liquidity - for example average total liquidity available within 5 cents in either direction of the midpoint, sampled hourly over 24 hours. If there’s no liquidity within this depth, this is a very thin market and should be treated as such.
If there are multiple contracts within the same event, focus on the most liquid contract
Report unusual contract terms. Note past resolutions that were contested
Compare contracts across multiple exchanges.
Provide a link to the specific contract referred to
For on-chain markets, investigate large, impactful trades from one account (or accounts that may be linked). The “French Whale” from last election is a good example of how learning more about the identity or number of traders affects how a reader would parse the market price.

Here is an example from the WSJ following their recently announced partnership with Polymarket:

Traders Bet on the U.S.’s Next Airstrike Target

In this article, several Polymarket price charts are presented, but without volume, open interest, or liquidity data. To their credit, the authors do note that contested definitions matter to traders, and they frame these prices as traders’ implied probabilities rather than as statements of truth.

But let’s dig deeper on the volume/liquidity data that’s missing. As of the writing of this piece (1/16/26) the total volume of the US strike on Colombia by Dec 31 2026 contract is: $135,696. To pick one example from the stock market, yesterday I found struggling mall chocolate retailer, the “Rocky Mountain Chocolate Factory” (RMCF, no position), with an 18mm market cap, that has averaged trading about $80k volume each day over the last 30 trading days, vs the all time total volume of ~$135k since listing on 1/4/26 for the Colombia contract. I’m not saying that volume in a nanocap stock is exactly analogous to a binary option contract’s volume (and note there are a few other dated contracts available on this Polymarket event), but it shows the scale we’re talking about.

The price of a contract in a prediction market like Polymarket or Kalshi is based on the current clearing price of a limit order book. A limit order book is a collection of buy and sell limit orders - each order contains a direction, a price, and a size. When two orders with different directions (buy vs sell) cross prices, a trade occurs. The inside market of the limit order book is the current best bids/offers, it is not the truth value of the world, it is just the current price based on this system.

This is what the order book looks like today for this contract:

This isn’t the kind of liquidity you’d want to make major conclusions upon. If someone wanted to manipulate the price of this contract temporarily, it would not be difficult with a few thousand dollars. In the longer term - much of the real liquidity is not necessarily posted in the order book, which is why displaying total volume traded and open interest is also helpful.

It turns out that sometimes the prices delivered by limit order books end up as good predictions of the true fair probability of some event! But in reality the participants aren’t “searching for truth,” they are trying to make money by trading the thing vs its future price. Noise traders, poorly written contracts, intentional manipulation, and inherent uncertainty can all have effects on market prices.

What claim is being made by showing a stock’s move?

News organizations constantly provide charts or daily price moves of equities. However equity prices hardly claim to be “the truth” of anything other than the current market cap of the company. Sometimes it’s helpful to pair a news event with a market move - like “AAPL shares rose 5% during the event where they launched the new phone.” And similarly we, as a society, have chosen to display the percentage move of the stock at the instant we are reading the article vs the closing auction price of the previous trading day as the default data point whenever the name is written, like: “GS (+4.5%)”. Most high cap stocks trade hundreds of millions or billions of volume per day, and they are claiming much less than what a prediction market contract claims. We should require far more evidence of the truth value of an important geopolitical data point (will we or won’t we invade?) vs the new price of Microsoft.

In general - a contract with more volume traded is likely to be a better reflection of the true fair than not much trading. Papers like this have estimated ~25% of Polymarket volume are likely to be wash-trading and that at least some of this is easy to detect. And importantly also that the percentage of wash trading varies over time. Given that real volume is likely to be correlated with efficient prices, reporters should attempt to clean volume data for wash trades. Kalshi does have similar challenges with liquidity and contract design in some cases, but their fee structure disincentivizes wash trading.

All else equal a market with tighter spreads and higher depth is more likely to be efficient. (I’ve pointed out corner cases where incentives can throw this out of whack). There are many ways to do this, each newsroom should come up with a consistent metric.

When we ask “how good is this prediction market vs other data sources like polls?” Keep in mind the inherent uncertainty of what each contract is trying to model. If we all bet on a fair coin flip, no amount of research can move us from 50-50 as the best possible forecast. Similarly with earthquakes, tornadoes, hurricanes, if looking more than a few days out, there is likely to be an uncertainty floor that cannot be improved upon. That’s fine! How good the market is depends on the difference between the market probability and some uncertainty floor, not vs a omniscient predictor. I would bet that hurricane market probabilities could be quite good a few days out, but earthquakes might rely more on base rates. Academics have studied these for awhile and know this, but Trump being priced at 60% before the election and then winning does not mean that “Prediction Markets called the election,” just like Trump losing would not mean the markets failed.

This is why comparing markets with different types of contracts is not apples to apples. As soon as you start to project real world, multiplayer interactions it gets even more complex: is the true probability that the US invades Greenland this year deterministic if you could map every synapse in Trump’s brain? How do you even model the inherent uncertainty in events that require coordination? We’re going to need more reps for different classes of prediction market contract, and see where they’re more or less reliable.

It is great that we’re using market probabilities to provide context to world events. But let’s not overdo it and overstate the claim. These markets are growing fast, they’re noisy, potentially inefficient, but still useful. Betting on sports and music is fun and all that, but if we start reporting on the probability of WWIII, let’s provide real context.

Two projects I’d love to see someone tackle:

Generate a “wash-trade” cleaned Polymarket volume dataset.

Develop custom liquidity measures for news reporting - could be top of book, depth up to Nth price point etc across various time frames. Make it easy to digest.

Whirligig Bear

Discussion about this post

Ready for more?