Frequent contributor Bennett Haselton is still thinking about prediction markets, and giving away money. He writes: "In an article last December I described a problem with prediction markets, where even markets with cap on betting limits could be manipulated by a single trader willing to spend a lot of money to distort the marketplace odds. So I offered a $100 cash prize to be split between readers who collectively came up with the best solution to the problem. Here's an idea that I think would work." Read on for the rest.
In November I wrote an article arguing that prediction markets like Intrade -- where users can bet on the odds of, say, Obama or Romney becoming president -- were a useful tool for aggregating the wisdom of crowds, but could be manipulated by someone placing a large bet in order to create the illusion that "the markets" were favoring their candidate. If the fake "market odds" were reported in the news, it could have the effect of causing more supporters to switch to that candidate, thus increasing the true odds of their victory and creating a self-fulfilling prophecy before the markets had the chance to correct themselves. The solution, I thought at first, was to have a cap on the amount that individual users could bet (which is one of the rules at the Iowa Electronic Markets), and make it illegal for a single mastermind to pay large numbers of third parties to make bets in order to circumvent the single-bettor limit.
As I admitted in a follow-up article, it turns out this regulation would not work after all. The problem is that as long as long as overseas betting markets like Intrade have no limit on wagers, a market manipulator could place a huge bet on Intrade to cause the odds to shift on that market -- for example, changing the odds of Obama-to-win from 4:1 to 6:1. Meanwhile, the odds in a domestic prediction market with a betting limit -- call it CappedEx -- would initially stay at their non-manipulated value of 4:1. But then "arbitrage players" could spot the difference in the odds being offered, and make opposing bets in the two markets in a way that would be guaranteed to make a net profit. (The details are spelled out in my last article, but basically, any time two markets are offering different odds of an event happening, you can pick appropriate amounts to bet in the two markets so that you're guaranteed a profit whether the event occurs or not.) These arbitrage players would continue making opposing bets in the two markets until the odds being offered in the two markets converged onto the same value -- at which point, the market manipulator has successfully manipulated the odds in the capped market, even without ever placing a bet there. Essentially, the market manipulator has hired all of those arbitrage players and paid them to make bets on his behalf, but done so indirectly to avoid violating laws against hiring an army of bet-placers.
I should be clear about the two different time frames being discussed here. If a manipulator places a large bet on Intrade, causing the odds on Intrade to diverse significantly from the odds on CappedEx, then the arbitrage players should cause the odds on the two markets to converge to the same value very rapidly -- plausibly in less than one minute. (Whoever spots the difference first, gets guaranteed free money. It would be easy to write a bot that could watch for any divergence in the odds in the two markets, and place guaranteed-profit bets as soon as a gap appeared.) Then, as political observers noticed that the odds have shifted (without any real-world event in the news that could plausibly explain the shift), another wave of bettors would take advantage of the distorted odds, to bet on the side of the event whose odds had been artificially lowered by the market manipulation. (The odds favor making such a bet, although it's not as good as a guaranteed profit.) As enough people made these opportunistic bets, the market odds would correct themselves to their original values. However, this second wave of betting would probably take a few hours, because it requires humans to think critically about the events. (One likely case of manipulation managed to shift the odds for a few minutes for just $20,000, so it's not unreasonable to think that a million dollars or two -- still small change by the standards of presidential candidates, especially when it's not subject to spending limits -- could distort the market for several hours.) The danger is that the market manipulation could cause the odds to shift in the capped market almost instantly, but the market correction would not take place until several hours later, and in that time the damage (in altering people's perceptions, and possibly creating a self-fulfilling prophecy) would already be done.
It didn't seem like there was any obvious solution to this problem. The U.S. government could ban its citizens from betting on foreign uncapped markets, but it would be too easy for a U.S. citizen to coordinate with an overseas partner to place the arbitrage bets together and split the profits. Or the U.S. could try to ban prediction markets entirely (capped or uncapped), but many economists argue that they're a useful tool for assessing the wisdom of crowds to assess the odds of an event. You could ban media reporting on the odds given by prediction markets (to try and avoid the self-fulfilling-prophecy problem), but that would probably be unconstitutional in the U.S., and unenforceable anyway if people could get their news from overseas.
So in my last article I offered up to $100 to be split between readers who came up with the best arguments for how to stop prediction markets, even markets with individual betting limits, from being shifted by manipulators who place large bets on foreign markets and then count on arbitrage players to pass on the effects to the capped markets. (I've offered cash prizes to readers who submit winning ideas before, and it usually doesn't take this long to get to the follow-up and pay out the prizes. Some follow-up articles that I submitted got lost in the editors' spam filters, sigh, and then there were some other articles in the pipeline that had to go out first. If I offer prize money for ideas and you submit a winning idea, normally you'll get your money much faster.)
Before reading any further, you might want to stop and try to think of what you would consider to be the best solution to this problem (even if the prize money has already been allocated), and then compare it to what we came up with.
... And, welcome back. Here's what I think is the best answer so far: For each event that the capped markets allow users to bet on, the capped market should also be required to monitor the odds that any overseas uncapped markets are offering on the same event. Then if there has been any recent time period where the odds on the overseas markets differed significantly from the odds on the domestic market (significantly enough to indicate manipulation -- and, similarly, significantly enough that the difference probably motivated arbitrage players to place bets to close the gap), then the reported odds should appear with a disclaimer saying, "There was a recent divergence in the odds on capped vs. uncapped markets, so the odds displayed here may have been manipulated, and should be regarded skeptically." This would help to avoid the self-fulfilling prophecy problem, if people are less likely to regard the manipulated odds as a reflection of reality.
The key assumption here is that if a real-world event happens that changes the probability of, say, an Obama victory, then the market odds in both the capped and uncapped markets should shift at about the same time to reflect that new probability. On the other hand, if the odds have shifted significantly in only one of those markets, that could be taken as a sign that that market was being manipulated. Arbitrage players would still be free to make opposing bets in the two markets to narrow the gap, so the odds in both the capped and uncapped marketplaces would still change in the short term, but in the regulated capped market, the odds would be reported with a disclaimer that they're not reliable. After a few more hours, opportunistic bettors would make bets taking advantage of the distorted odds, and the market would correct itself.
This idea did not come from any particular reader but came up as the result of the back-and-forth I had with several people.
A few readers also had interesting ideas for regulations that could fix the problem if they could be applied to all markets. For example, Nathan Dykman suggested that in order to wager larger amounts, you would have to wager that your candidate would win by a larger margin (e.g. if you can bet $1,000 that Romney would win by 1 million votes or more, or you could bet $10,000 that Romney would win by 10 million votes or more -- so that large "manipulative" bets would stand out more obviously). Andy Jobe suggested "staggering" bets so that high rollers could only bet large amounts by placing lots of small bets in sequence, paced slowly enough that the market would probably detect the manipulation attempt and start correcting for it, before all of your bets went through. Jonathan Pearson suggested mandating that markets report the number of people making particular bets as well as the market odds, so that single large manipulative bets could be easily spotted. Ben Griffin suggested simply requiring disclosure of large bets by certain people (as he put it, the headline "Saudi Prince believes that Romney will win the election. What does he know that we don't?!" contains more useful information than "Romney's odds of victory looking better at Intrade").
I think these points are all correct, but the problem with all of these ideas is that they only work if all of the relevant markets are regulated. And if you allow that assumption, then the problem becomes trivial -- because you can just require an individual betting cap in all of the markets. On the other hand, if there's at least one market anywhere in the world that is beyond the reach of your regulations, then they don't have to disclose any statistics about their bettors or follow any other rules that you make. Then when a manipulator places a large bet in that unregulated market, when the arbitrage players place their many small corresponding bets in your domestic regulated market, the detection mechanisms described above, won't do anything to stop that -- those bets in your regulated market look like real bets because they are real bets.
By contrast, if you require domestic capped markets to monitor the overseas uncapped markets, and disclose if the uncapped odds have diverged recently from the capped odds, this still works even if the regulations only apply to your domestic capped market. People can still place manipulative bets on foreign markets, but if the media reports the current "market odds" by looking at the capped market, those odds will be harder to manipulate without getting caught, because they'll run a disclaimer if manipulation has been detected recently. (Of course if the media gets their "odds" from the overseas uncapped market, and reports those odds as literal truth even when the domestic capped markets are running a disclaimer saying that those same odds have recently been manipulated, we can't do anything about that. The hope is that news agencies, no matter how lazy they may be, will at least choose to report accurate information if it takes the same effort as reporting inaccurate information, and thus would prefer getting their information from the domestic capped market, where they can easily check if there's a disclaimer saying the odds have been manipulated recently.)
Some interesting points made by other readers:
Carl Pearson mentioned that if campaigns had to start diverting attention to prediction market manipulation in addition to all of their other business, this might hurt small third-party candidates more than big campaigns -- because smaller campaigns have fewer available resources to put towards handling new kinds of problems. (True, I think, but only if the markets can be manipulated. If they can't be manipulated, and they're just a barometer of what people are thinking will happen, then you don't need to waste campaign time fighting on that front.)
Michael Mendenhall pointed out that even in a capped market, the cap should be high enough to create a high "signal-to-noise" ratio. If the cap is too low, the market odds will reflect the betting of more uninformed people who use the betting as a low-cost opportunity to cheer for what they think should happen instead of what they think will happen. (On the other hand, if the cap is too high, then the market is too easily to manipulate.)
Marc Beaupré argued that prediction markets can probably never be stamped out anyway, because anonymous payment protocols like Bitcoin make it possible for crypto-anarchists to place best on unregulated darknets where they can ignore caps and disclosure requirements all they want. I'm not sure that's true (how do you place a bet in an anonymous peer-to-peer market -- who enforces the payment from the loser to the winner, depending on the outcome?) but it actually doesn't change the main thrust of my argument -- you can still have a regulated, capped domestic market, which is where the media could go for accurate information about the current market odds. So a manipulator could throw their Bitcoin money away on an unregulated peer-to-peer betting network, but it wouldn't do them any good.
Splitting the $100 in prize money, all 7 of the readers credited here get $15. There may be a simpler idea that we missed, or a different reason why this proposed idea would not work. Either way, I'm always grateful for the high quality of the comments that get emailed to me as part of these contests. Eventually I'd like to run some article contests for people to email ideas for a follow-up article, but without offering prize money, to see if that affects the quality of the submissions. It would be impossible to run a precisely controlled experiment (because you can't write a single article where half of your readers are eligible to submit ideas for prize money, and the other half are expected to submit ideas for free), but if we run contests for a large number of articles, and about half of those contests involve cash prizes while the other half offer only acknowledgement, it should eventually become clear if there's a difference in the quality of submissions. It may be that, unlike prediction markets, idea-improvement contests work just as well when there's no money involved.