An attempt to promote prediction markets

I had a project last month to promote prediction markets in the Middle East. The idea was to publish some op-eds about applications of prediction markets in the US, build some traction. Unfortunately, the 3 newspapers I sent it to declined to run it. It’s possible I need connections or a reputation to publish a piece like this. But it’s also possible the piece is just badly written.

Anyway, I post it here for your critique.

Die US spies bet on Russia’s invasion of the Ukraine? Yes, and that is good.

The Russian invasion of Ukraine has seen major successes by the US and UK intelligence communities, alongside embarrassments for the Russian and French agencies. One factor in American and British success might be their use of prediction markets. Prediction markets are systems for betting on future events (usually in reputation points, not money). Despite their resemblance to the stock market and sports gambling, prediction markets solve critical problems intelligence services face.

Days before the invasion, Biden stated that he “was convinced” by his intelligence services that Putin would invade. It was a bold position: in my university in DC we regarded an invasion as inconceivable: international wars are vanishingly rare, an invasion would unit Russia’s enemies, and it would trap Russia in a bloody and expensive insurgency. Even the French army were surprised by the invasion, prompting the resignation of France’s head of military intelligence last week.

During the invasion Russian intelligence faced its own embarrassment. Russia’s invasion planners anticipated a rapid collapse. Accordingly, they used rapid, unsupported armored thrusts and aerial landings behind Ukrainian lines. When Ukrainian resistance did not crumble, this invasion plan left Russian units isolated in front of long, insecure supply lines. Leaked Russian intelligence shows they believed support for the Ukrainian government was weak and would evaporate quickly. That failure has prompted a purge of the Russian intel agency FSB, whose leaders are now under house arrest.

The Russian story demonstrates some key challenges facing an intelligence service. States pay intelligence officer’s salaries partly to learn what will happen in the future. The inherent uncertainty of political and economic outcomes makes this exchange difficult. You can never be 100% certain about a future event, especially in a foreign country. Suppose an advisor states “the Ukrainian military will collapse with 80% probability”. If the collapse does not occur, how can we know if the analyst was wrong or just unlucky? You need a system for judging predictive ability across many statements. And what if some analysts believe an event is likely and others disagree. Leaked documents reveal some FSB agents dissented from the invasion plan. How can you aggregate the predictions of multiple analysts together?

A prediction market uses betting to solve both these problems. A prediction market is similar to a stock market, in that participants buy and sell assets linked to a future payoff (options and futures are literally prediction markets). Unlike stock markets, the assets in a prediction market are tickets that pay if certain events happen. For example, a ticket might pay 1 dollar if Russia invades Ukraine. If someone buys the ticket for 30%, we know they think an invasion has at least a 30% chance. Prediction markets naturally track performance over time; after a few bets my account balance will tell me just how competent I am. In non-monetary markets, the difference between prediction and accuracy can be easily calculated. Moreover, bets require clearly defined outcomes to attract participants. The bet “will Russia win their invasion” invites acrimonious debates but “will 100 or more Russian troops enter Lviv before January 1” is clear to everyone.

The US intelligence community started its main prediction market in 2010, the creatively named Intelligence Community Prediction Market (ICPM). Only people with top-secret clearances can access it, so we know very little about it We know that war incidence is a common question type, and the ICPM features precisely worded questions. The market is reasonably popular, with 190,000 predictions and 4,300 users as of 2018. The ICPM does not use real money, but a system of play money with no redeemable value. Real money would incentivize the participants to intentionally mislead one another to get a better price, which obviously damages the whole enterprise. High scores are not even directly linked to commendations or promotions.

Did the ICPM give the US an edge in the Ukraine crisis? With no public access to the system, it’s hard to say. We know that the public prediction markets like Metaculus performed quite well. Metaculus’s “will Russia invade?” question increased form 20% in December to 70% by the middle of February. But intelligence community had plenty of other resources. Public statements show the US had an abundance of information Russian plans, troop movements and false flag attacks. Undoubtedly the invasion was discussed at length by high-ranking staff, who likely rely less on the ICPM.