I've lately been thinking about how a world with widely-used prediction markets might look like. In this post, I expand on some ideas I stole from Robin Hanson.
Aside from legal, financial and user-acquisition concerns, prediction market providers face two major problems:
Both of these are really hard problems when you want to scale for a large amount of markets. In a proper system, anyone should be able to create new markets. Having paid personnel to go through all of them is not feasible. Even worse, for many topics specialist expertise is required to judge the answers.
Manifold, for instance, does manual review of all real-money (sweepstakes) markets. There are about 40 of them at the moment, and almost all are about US elections. This makes sense, as such questions are easy to resolve based on government-provided data.
Ideally a prediction market system would be mostly automatic, and require minimal administrative oversight. Let's start the examination from the second point:
Somewhere in Robin Hanson's writings [1] there's an idea of filtering court cases by doing a prediction markets about the results, and probabilistically processing only some cases in the actual court, likely by random sampling weighted with uncertanty of the case. Rest of the cases can be resolved to the prediction. The most important feature of this system is that it's quite scalable and can have predicatable timelines.
The same idea could be used for prediction market resolutions. The market platform employs a small number of professional resolvers, or arbitrators. Complex cases can be referred to neutral external arbitrators as well. The for each market we can have a metamarket "Would an arbitrator rule that this question was resolved correctly?". In the vast majority of the cases the market should be around 99% yes. Again arbitrators should sample these cases and issue corrective resolutions.
To have enough participation, some incentives must be set up. Maybe some of the profits from the primary market can be redirected here? Or maybe the resolver is required to bet for "yes" as a part of the resolution?
One question that springs to mind is whether losers of the original market would try to overturn the result of the metamarket. However, markets with more liquidity attract more traders. As Hanson puts it in Shall We Vote on Values, But Bet on Beliefs?
However, market speculators would have already corrected for an average expected desire to bias the price, and have increased their willingness to collect info and trade to match expected variation in such hidden desires. So every case where a manipulator raises the price too high must be matched by an equally likely case where the price is made too low, even if manipulators wanted on average to raise it. As discussed above, the more manipulation is expected, the more accurate are prices. This is also why one needn’t worry about a nation’s enemies hurting it via adding misleading trades. On average, adding noisy trades increases price accuracy.
Metamarkets could also be useful in ensuring that submitted markets are of good quality, i..e content moderation. A market platform could, for instance, run meta-markets about the was-resolved-correctly market resolving above certain value. Or maybe suggesting edge cases by opening secondary markets about them?
Other moderation like "is this market legal/sensible/interesting?" could also be done with a market asking if site curators would approve it.
My worry here is that the metamarkets won't get enough participation. In the current world, even most primary markets do not get enough attention to be too accurate. I expect this problem to mostly disappear if serious-real-money markets become widely available. Fortunately these tools are mostly optional, and can be used to alleviate problems that only appear with scale. For smaller marketplaces manually reviewing everything, or using informal feeback or report button for correctness should be enough.
I'd be grateful for a link, in case anyone can find this ↩︎