In my first article about funding public goods, I mentioned in passing the Wall Street performer protocol, which involves bonds that pay out when a certain public good is provided. In this article, instead of talking about them as bonds, I’m going to think of them as bets — bets on whether the public good will be provided.
But the curious thing is this: people who want to help fund the public good do so by betting that the good won’t be provided. How does that work?
Well, it shouldn’t be too surprising, when you compare it to how insurance works. You don’t want your car to be damaged or stolen. But if you buy insurance for it, you’re effectively betting that it will be damaged or stolen; if it is, you win the bet, and the insurance company pays you; if it isn’t, you lose the bet, and the insurance company keeps the premiums you’ve paid.
One thing to notice about this is that it shifts incentives. If the financial costs of damaging your car (or having it stolen) are diminished by this bet, you might be less careful to avoid damage and theft. But the insurance company is suddenly interested in your welfare; if they could, they’d be willing to spend money to reduce the chances that your car is damaged or stolen, as long as the reduced chances outweigh their cost. And they might even offer lower premiums if you agree to take certain precautions, like using a steering-wheel lock.
This shift of incentives is the idea behind the Wall Street performer protocol. If lots of people want a public good to be provided, and bet that it won’t be provided, then someone must be betting that it will be provided. And whoever’s betting that it will be provided has an incentive to ensure that it is provided.
There are already markets for bets on future events. Bookmakers, for example, often take bets on the outcomes of sporting events. But there are more general prediction markets, as they sometimes call themselves, like iPredict.
As a somewhat arbitrary example of how iPredict works, they have a contract on whether Edward Snowden will leave Russia this year. It’s currently trading at around $0.07, which means the market estimates that there’s roughly a 7% probability that he will leave Russia this year.
If you think the event is more likely than that (say, 10%), then you can buy a contract for around $0.07, and if he does leave Russia this year, then the contract pays out at $1; you paid $0.07 for something that you estimate (if you’re risk-neutral) is worth $0.10.
But if you think the event is less likely than that (say, 5%), how do you bet against it? You can think of it like this: You turn up to iPredict with $1 and create another copy of the Snowden contract, which you can try to sell for around $0.07. If the year ends, and Snowden is still in Russia, the contract closes at $0, meaning that you can buy back for $0 the $1 contract you created.
In practice, you don’t need to turn up with the whole dollar — only the net $0.93 you will have paid once you’ve created the contract and sold it for $0.07. And the bit about buying back your dollar for $0 happens automatically if the contract closes at $0. Either way you look at it, you paid a net $0.93 for something that you estimated was worth $0.95 (since you think there’s a 95% chance you get the dollar in the end).
In this way, the prices of such binary contracts at iPredict reflect the probabilities that the stated events will happen.
With this understanding of prediction markets, we can see how they might be used to fund public goods. Suppose there’s a contract for a particular bug to be fixed in Firefox. The order book currently shows someone willing to buy 20 of the contract for $0.70, and various people willing to sell a total of 400 of the contract for $0.75.
If you’re a programmer, and you’re willing to fix the bug for $100, then you can spend $300 now to buy the 400 contracts, fix the bug, release the fix to the world, wait for the contract judges to confirm that you really have fixed it, and collect the $400 that the contracts are now worth, for a net income of $100.
If you’re not a programmer, and you’re willing to pay $10 for the bug to be fixed, you can put in an order to sell 40 more of the contract for $0.75, or even 50 more of them for $0.80, or whatever. (You might even try to do choose your sale price in a way that has no expected cost to you, assuming you’re risk-neutral. But be careful: your extra offer on the order book might make it 80% likely that the bug will be fixed, but the event that the bug is fixed isn’t probabilistically independent from the event that your order is accepted.)
This scheme is quite flexible. For example, if the contract is for the writing of a complex piece of software, rather than the fixing of a simple bug, you might have skills that pertain to some of the task, but you won’t be able to do it on your own. What do you do?
Well, you could buy some of the contract at, say, $0.20 each, if the market estimates it 20% likely that the project will be finished. Then you do your bit towards completing the project, and release it to the world. Seeing that, everyone increases their estimates of the probability that the project will be completed, to, say, 40%. Then, you can sell your contracts at around $0.40 each, and profit from your work.
Furthermore, this scheme discourages the duplication of effort that can be encouraged by bounty schemes, as I mentioned last time. If you buy some contracts for the provision of a public good, intending to provide that good yourself, you get paid when the good is provided, regardless of whether you were the one to provide it. So it’s in your interests to find out if other people are already trying to provide the good, and to help them do so, even if they get some of the credit; and if they own some of the contracts, too, then it’s in their interests to accept your help, even if you get some of the credit.
But one thing that interested me about the original description of the Wall Street performer protocol was the comment that the bonds would be “interest bearing”. It didn’t explain who should be paid the interest, but I’ve come to the conclusion that it should be the short-sellers of the contract, or, if you look at it that way, the people who showed up with their dollars to create the contracts.
If they earn interest (on the full $1 per contract) at the prevailing market discount rate, then there’s no need to have an arbitrary deadline for the provision of the public good, like the “in 2013” part of the Snowden contract; if the public good is never going to be provided, then that stream of interest should be worth exactly the dollar-per-contract they would get if the contract closed at $0 immediately.
In fact, this idea could be applied to prediction markets in general, and it might help solve the problem of poor liquidity when naked short-selling isn’t an option. For example, it may be the case that Snowden really is only 5% likely to leave Russia this year, but no-one is willing to bid the price down to $0.05, because selling each contract at the current price requires putting aside a little more than $0.93 now in order to get a 95% chance of $1 next year; there may be a better way of earning a small interest between now and the end of the year.
This problem is particularly bad when it comes to really long-term contracts, like whether Graham Henry will be New Zealand’s head of state by 2020. I’m not a fan of rugby, so I don’t know much about him, but it seems to me this event is less likely than 5.7%.