TIM COHEN: The futures market is everywhere

There is a well-known joke about predictions: that economists have predicted nine of the last three recessions.

In fact, the joke is incorrect. A 2018 IMF study across 63 countries between 1992 and 2014 found economists actually did correctly forecast that recession years would be different from other years, but they missed the magnitude of the recession by a wide margin. However, because life is perverse, the most recent example was in 2022 when a huge chunk of economists (and Wall Street) loudly predicted a 2023 recession in the US that never happened.

We have been fascinated by predictions for aeons, and just one example from classical times is the Oracle of Delphi. The high priestess of the Temple of Apollo at Delphi would deliver prophecies while in a trance-like state induced by inhaling fumes from a chasm in the temple floor; dramatic, but not usually totally useful. When Croesus asked if he should attack the rising Persian empire, the Oracle replied: “If you go to war against Persia, you will destroy a great empire.” He did. His own.

As physicist Niels Bohr is reputed to have said: “Prediction is very difficult, especially if it’s about the future.” But if there is one thing that has changed dramatically over the past year, it’s the emergence of prediction markets, such as Polymarket and Kalshi, which have just exploded onto the scene. Polymarket’s weekly trading volume is now just under $2bn a week; it was $54m for January 2024. I am not making that up.

And now we have dual global events of enormous significance that have economic implications for almost every country: the US-Israeli bombing of Iran and the subsequent spike in the oil price. So, the question is, how did the prediction markets do? The short answer is extraordinarily, creepily well. Almost too well. And also, not.

Gemini, Google’s AI, reports that on the timing of the strikes, prediction markets significantly outperformed traditional news cycles and expert pundits in anticipating the “when”. On Polymarket, the contract for “Will the US strike Iran by February 28?” saw a huge spike in volume just 24 hours before the first Tomahawk missiles were launched.

This has led to an investigation into whether individuals with ‘advance knowledge’ of the Pentagon’s plans used the markets to enrich themselves

But here is the problem: reports surfaced on March 2 that six specific accounts made more than $1.2m in profit from bets funded via crypto wallets just hours before the raids. This has led to an investigation into whether individuals with “advance knowledge” of the Pentagon’s plans used the markets to enrich themselves.

On “leadership outcomes” — the polite description for whether Ayatollah Khamenei would end up dead — the prediction markets were notably bold. While Western intelligence was initially tight-lipped about the success of the strike on the supreme leader’s compound, Kalshi and Polymarket users had already driven the “Yes” contract for his death to above 90% by the morning of March 1. Official confirmation of his death didn’t arrive via state media until later that day, meaning the “wisdom of the crowd” (or perhaps just well-informed leakers) beat the official wire services by nearly six hours.

But here is where things went wrong. Most oil price prediction contracts were in the $95-$100 range. They were caught off guard when Brent crude surged to nearly $120 on March 9 following the Iranian retaliatory strikes on Bahraini desalination plants and refineries. But markets correctly anticipated a “de facto closure” of the Strait of Hormuz, with probabilities for a shipping halt rising to 85% as early as February 25, reflecting a far more cynical (and accurate) view of the region’s stability than the G7’s public statements at the time.

So, what do we make of all this? First, in a deeper sense, the utility of prediction markets is coming to the fore. The argument in favour of prediction markets is that when a geopolitical situation flips from “tense” to “kinetic”, the useful information is rarely sitting neatly in one place. Prediction markets, in theory, pull together lots of small, distributed clues (shipping notices, satellite chatter, diplomatic leaks, insurance premiums, troop movements) into one price — often beating conventional “moderately sophisticated” benchmarks in published research.

But the problems are also acute. In a war scare situation, signals can themselves change behaviour: media amplification, corporate panic buying, posturing by political actors and retail herding. That feedback loop can make the market look prescient when it’s partly self-referential, especially for thinly traded “rare event” contracts.

And then there is insider trading and national security leakage. A market on “US strikes within seven days” is basically an invitation to test whether anyone with privileged access is trading. But this is easy to overstate. Rules obviously forbid it, enforcement is nontrivial, and the reputational/ethical cost of “profiting from war probabilities” is real.

There is also something called “fat-tail geopolitics”. This is when the dramatic outcomes of low-probability events get hidden in “the most likely scenario” thinking. The Iran oil episode is a reminder that the world jumps discontinuously.

All in all, prediction markets, it turns out, are not entirely dissimilar from the Oracle of Delphi: they tell you something valuable, no doubt, but the question remains: precisely what?