Commentary: The odds racket in prediction markets

NEW DELHI: A multi-billion-dollar industry is thriving around a contemporary obsession with scientistic soothsaying. Prediction pays.

Want odds on the prospect for Donald Trump to take military action against Cuba? Or the chances for success in the latest round of talks on Iran? No problem.

The idea behind such markets is simple. Electronic platforms open markets on all manner of prospective issues, framing an eventual outcome in a binary yes-no fashion. If it happens, one share in that market pays its holder one dollar; if it does not, then nothing. As traders - basically anybody with a crypto account - join in and buy or sell, the price of a single share varies. Because of the set-up, at a given point of time it is read off as the probability the market assigns to that outcome at that moment.

As an example, the “Khamenei out as Supreme Leader of Iran by March 31?” market assigned a 27 per cent probability to that outcome hours before Khamenei’s death in an Israel-US operation on Feb 28. The peak probability was 44.5 per cent on Jan 14, around the time Donald Trump asked Iranians to “keep protesting - [and] take over your institutions.”

Much of the current criticism of platforms such as Polymarket and Kalshi, two major prediction market exchanges, stem from well-founded concerns that insiders stand to make a quick buck off their knowledge of sensitive secrets.

Such understandable worries obscure a larger set of issues. As politicians openly flout their scepticism of experts, prediction market forecasts are alluring for their apparent transparency and the warranted “wisdom of crowds”. Some evangelists are professionals who see prediction markets adding value to intelligence tradecraft.

At a time when security establishments around the world find themselves struggling to offer their consumers direct, apparently uncluttered forecasts, the pull of market-based mechanisms that dramatically simplify complex contingencies into single numbers is clear.

Whether it is sensible to give in to this siren call is a different question altogether.

MARKET EFFICIENCIES?

Before Polymarket entered the scene in 2020, there were two earlier waves of interest in prediction markets.

The first started in 1988, when University of Iowa researchers created an electronic exchange to test the extent to which American electoral outcomes could be forecasted. Around the same time, a few private companies sought to use prediction markets for in-house use.

The second, between 2000 and 2003, saw the Defense Advanced Research Projects Agency (DARPA) - the US military’s blue-sky research arm - funding prediction markets research, including one on a “Policy Analysis Market” that would focus on Middle East-specific outcomes. Curiously, around the same time, a private Dublin-based website allowed anyone interested to bet on Saddam Hussein’s ouster by a specific date.

At the heart of such efforts were three interrelated claims. First, prediction markets behave exactly like any other financial market. Second, the price of an instrument in a financial market changes only if new information about it comes to light (the “Efficient Market Hypothesis”). And third, the price of an instrument at a given time reflects all the information about it spread across innumerable players and institutions. In other words, markets provide a clean mechanism to aggregate dispersed information (the “Hayek Hypothesis,” so named after the late Austrian free-market champion Friedrich Hayek).

According to the second and third claims, a company’s stock price would change only if new information about, say, the company’s earnings or the myriad conditions that affect it emerges. The price is what is because it reflects nothing less than the totality of information about the company.

The Efficient Market Hypothesis is deeply suspect. Some people buy a stock, for example, not because they possess any extra insights about it, but simply because others around them are doing so. Nevertheless, it inspires prediction market aficionados because of a single, odd episode.

Within hours of the explosion of the US space shuttle Challenger on Jan 28, 1986, the stock price of aerospace contractor Morton Thiokol fell by around 12 per cent. It would be much later that an independent commission would find the cause of the explosion to be a faulty part manufactured by that company.

After performing rigorous statistical tests, two researchers in 2003 found the stock market’s ability to pinpoint Morton Thiokol as illustration of its efficiency, though they were unable to discover how the market came to its conclusion. As a counterpoint, however, in 2003, when another US space shuttle exploded, the markets moved swiftly to punish the then-owner of Thiokol, even though the disaster had nothing to do with faulty parts.

A crucial test for market efficiency is whether the same instrument trading on different exchanges is priced similarly. If different exchanges have different prices of, say, a given stock at a given time, then one can - through a simple combination of buying and selling - make free money. Clever traders make money off such arbitrage all the time. Markets are always efficient in theory, but only sometimes in practice.

Research on prediction markets around the outcome of the 2024 US presidential elections shows how different exchanges priced similar bets differently. More recently, Polymarket and Kalshi offered different odds - and therefore, different prices - for whether Mojtaba Khamenei would become Iran’s new Supreme Leader, allowing some to profit off the mismatch.

In a clean efficient-market universe, prices should not change when new information is absent. But a February spike in a market on the Second Coming of Christ by the year’s end was not because theologians had discerned anything significant, but because a secondary, derivative, market on the direction of the Second Coming market had opened up. “And now there is almost US$50 million riding on the chances of a Second Coming,” a commentator observes acidly.

In financial markets, derivatives can and often do change the price of the underlying instruments. For platforms like Polymarket, that is a formidable challenge to their claim that prices and probabilities are solely due to news - that they are “News 2.0”.

FROM SECURITIES TO SECURITY

Can prediction market probabilities serve as valuable intelligence indicators, as some claim?

On paper, this is an interesting idea. Say a market sees an increase in the probability of a salient foreign political outcome. Based on this, intelligence services could assign more collectors and analysts to the issue, for example.

But in practice, this is useful only when the market shows clear upward or downward patterns. For example, the infamous US$63 million Khamenei-out bet showed no discernible trends when it was active between mid-December 2025 and the end of February. Lead time is important. To see a sudden spike in probability hours before the market is set to expire is useless.

Consider the claim that in-house prediction markets could help with aggregation of information and expertise spread across different agencies and departments. Again in theory, and according to the wisdom-of-crowds logic, the only way that an in-house market would do its job well is if you allow analysts from different desks to trade on that question. Squaring this permissibility with security requirements around compartmentalisation is very hard.

Then there is the issue of what in-house analysts would be betting with. Real money is out of the question, given obvious problems. But analysis shows that the accuracy of non-sports prediction markets improves when traders trade with cash and not play money.

QUANTIFYING THE UNQUANTIFIABLE 

Children are taught probability in terms of packs of cards and rolls of dice. In such situations, the odds of an outcome are verifiably measurable. For example, assuming that you have an untampered pack of playing cards, the probability of drawing a heart is always a little less than a quarter, 24.1 per cent to be precise. (It is not exactly 25 per cent because of the two jokers in the pack.) We learn to trust the odds based on the frequency with which they occur.

But intelligence analysts cannot rely on such an interpretation. Most such things don’t repeat themselves often or even, ever. So probabilities in terms of frequencies and fixed structures - such as a deck of cards always having 54 cards - make very little sense in the context of world affairs, much as Donald Trump might deploy the “hold the cards” analogy in his commentary.

When talking about probabilities in the context of strategic affairs, we innately leap into the realm of subjective probability, where the odds assigned to an outcome reflect the degree of belief the assignee has in the outcome. It is what bridges uncertainty (where we are unsure of an outcome) and risk (where odds can be objectively assigned to the outcome, independent of who does the assigning).

But here be dragons.

Some uncertainties can never be reduced to consistent numerical subjective odds. Furthermore, subjective probability assignments are often driven by cognitive biases and shortcuts.

A CONCEPTUAL MESS

Then there is the question of the basis of odds assignment. The US intelligence standard is not just the consistent assignment of numerical probabilities, but also specifying one’s confidence in the evidence that led to that assignment.

When it comes to prediction markets, it is impossible to attach such confidence levels to probabilities, for the simple reason that there is no clear way of knowing why a given trader is buying or selling. Is it because of insider information? Social media vibes? We do know, as a matter of mathematical fact, that the overall probability that a prediction market assigns to an outcome is not the average degree of belief of traders in the market. So what is it?

The fine garb of clean-looking streams of probabilities does much to conceal conceptual mess.

The mess is compounded by the fact that in matters of war, the manipulation of shared risk is par for the course. Plainly, countries often do many things that seem to apparently heighten or lower the risk of conflict. If prediction market players factor in such “information”, the ensuing market probability is going to be useless insofar as it faithfully reflects the probability of an outcome.

Some intelligence analysts have maintained the distinction between “forecasting” and “fortune-telling.” The former includes a clear explanation of how an outcome can come to be, and a clearer defence of the basis of that judgement. The latter is an unexplained, unsubstantiated prognosis.

It is in this sense that prediction markets are fortune tellers. They must be treated as such.

Abhijnan Rej is a security analyst, focused on counterterrorism and counterinsurgency, political warfare, forecasting and computational modelling of security risks. This article was first published in The Interpreter.