Prediction markets for internet points?

by paulfchristiano Sideways View4 min read27th Oct 20198 comments

40


Using real money in prediction markets is all-but-illegal, and dealing with payments is a pain. But using fake money in prediction markets seems tricky, because by default players have no skin in the game.

Here’s a simple proposal that I think might work reasonably well without being too hard to try:

  • Create a service that tracks Internet Points. Anyone can quickly sign up with a Facebook account and is initially given 1000 points. Points are non-transferrable and are permanently associated with your Facebook account.
  • Run prediction markets in which people wager points rather than dollars. You can initially copy questions and verdicts from existing markets and bookies. (Ideally you can just fork an existing market implementation.
  • Keep the point tracking simple and transparent. Allow anyone to observe anyone’s point total, and the history of bets that led to those totals. Avoid issuing new points. Try to create a sense of stability. No leaderboards, just observing your friends’ (or strangers’) point totals. (Maybe allow people to deactivate their account to hide their point total page, but if they reactivate they still have their old point total.)
  • Limit maximum exposure to 50% of net worth. (But people can “exit” a market by taking a bet on the other side—betting on a 5 year question doesn’t necessarily tie up your points indefinitely.) Encourage people to optimize their log returns rather than expected returns.
  • Make it easy to link to pages for particular markets so people can see the current spread on a given question; make it very easy to quickly create a new account and bet on that market (2 clicks—login with facebook, and allow service to access your profile info).

It may be that a close-enough service already exists, though I’m a bit skeptical.

If you are at all interested in doing this, you have my blessing, this is a pretty generic idea (similar to sites that have existed in the past, e.g. Foresight Exchange). I’d also be happy to pay a bounty for an implementation I think is reasonable, I’d pay at least $1k for something that seems OK and up to $10k for something I think is actually really good.

Why I think this would be good

The barriers to entry in these markets would be tiny compared to existing prediction markets—within 30 seconds of seeing someone mention a market on Facebook, I could have created an account and placed a bet. I would likely bet in this kind of market if a few of my friends did.

Some (though far from all) people would take these markets seriously. For events in the next year or so, I think it’s likely that these markets would quickly converge to better predictions than existing political prediction markets—if prices stayed crazy I think there are plenty of people who would be excited to step in to fix them even if they didn’t take Internet Points very seriously (after all, they might be worth some reputation in the future, the cost is super low, and betting on things is fine). Note that people who step in early to correct mispricings would be able to exit the market after sanity was restored, so wouldn’t even need to tie up their points very long, and that they would have an incentivize to advertise the “easy win” to others after they’d bet in order to free up their capital.

If Internet Points are simple enough for people to really understand, and people expect them to stick around, I think they could potentially capture a big chunk of the altruistic upside from medium-stakes prediction markets.

If the overall workflow is very smooth, I think there is a reasonable chance that the site could spread quite rapidly (spreading just as well to new users as to people who have already made an account).

Note that playing this game can still be a fine experience even if you’ve lost most of your points—going from 250 to 1000 points is no harder than going from 1000 to 4000, it’s just that the numbers are smaller.

Fake accounts and private bets

Ideally you’d have exactly 1 account per person. But creating Facebook accounts is quite cheap. Since every account is granted 1000 points, that’s a problem. Some bad things that could happen, and what you might do about it:

  • Someone could lose some bets, and then make a new account. But then their points won’t be associated with their identity (and most people wouldn’t start a new FB account to get more internet points), so as long as you mostly care about “your” points then this isn’ ta big problem.
  • Someone could make fake accounts and try to take their points. To avoid this, you probably want to make it impossible to transfer points except in public markets—the worst you can do is have some sock puppet accounts make bad bets and then take the other side, but if we structure markets appropriately this might be hard to do discretely. We can also make it against the terms of service and have simple mechanisms to try to detect it (e.g. by disallowing new FB accounts until we have a better system in place).
  • Someone could make fake accounts and use them to influence markets, if they care about the market odds (e.g. if they want their preferred political candidate to look like they have a better chance in the general election). This can be partly addressed by simple measures above (like prohibiting new FB accounts).

It would be nice to fix these problems in a more robust way, so that market odds are more trustworthy, we can allow private bets, etc. That seems like a big project, over the long run I think the principled solution is basically a credit network: I indicate that I’m willing to trust my friends for IOUs, and then all bets need to be implemented as a string of IOU exchanges. This means that the social network can eventually fragment into different pieces (with the fake accounts basically living in their own part of the social network, which has very few trust connections with real accounts), and the market odds will potentially be different for different people. That makes everything way more subtle to think about, and the implementation seems way more complex, but it seems to me like it should work over the long run and is where you’d really like to be.

(You could also integrate that system with identity verification services who receive a high degree of trust, and then a service itself could effectively be deactivated—have all of its trust exhausted—if it certified a bunch of users who lost money.)

Subsidies

A basic problem with this mechanism is that there is no subsidy (except for noise traders). It’s not as bad as existing prediction markets, where the house takes a huge cut, but not as good as most “real” applications of prediction markets where someone interested in the info is willing to pay for it.

If I’m betting with someone reasonable, at least one of us is losing expected log-points (since the log bakes in risk aversion). So the market can’t work with participants who trust each other’s rationality, unless people just love gambling.

The most natural way to fix this is to subsidize markets. But doing this makes the governance and interpretation of Internet Points harder—does someone with 4000 points have a great history of losing bets, or did they just collect a subsidy? who are we effectively subsidizing, and how do we make that call?—so I’d prefer avoid it. If necessary, probably the nicest way to do it would be for the house to provide a modest amount of liquidity at the prices implied by existing markets. Existing prediction markets are sufficiently irrational that I think this would provide a significant incentive to participate.

If prices on Internet Point markets are used seriously as forecasts, it also introduces an additional set of incentives—some people want to manipulate forecasts (e.g. to make it look like their preferred candidate would be more likely to win the general election). So the average truth-seeking trader expects to get money by betting against them, and this can make the whole system work.

In some sense this is actually the only case where you really needed prediction markets anyway—if everyone is being reasonable then we can just talk it out and share our predictions, and that naive mechanism only really breaks down when some participants believe other participants are being predictably unreasonable (for whatever reason). For better or worse, this is a pretty typical situation.

40