Why prediction markets are rarely useful
Why prediction markets are rarely useful
I used prediction markets for years. They are entertaining but I have to admit nearly none were actually useful to me. I have two ideas why.
I use Manifold for over a year now and Metaculus since 2018. I had a lot of fun. This newsletter is not about having fun though. This is about making better decisions. Looking back, prediction markets were rarely useful to me. Entertaining for sure, but not useful.
It is fun to gamble on Zelda actors, who will be Person of the Year, or Oscar awards. It feels comfortable to have good estimates about upcoming elections and wars. However, none of that was directly useful for my decision making. The fact checking angle might be useful if the right opportunity comes up.
My best personal example was vacation planning last summer. There was a risk that an oil catastrophe might ruin the beach and thanks to this market, I was slightly more confident that it will not happen.
In theory, prediction markets can be very useful. For example, to manage projects or monetary policy. I’m not personally involved in those decisions though, so this is not useful to me.
1. Usefulness would violate privacy
Decisions which matter to me personally nearly all depend on my private data.
For an example, I might have to decide about a surgery. The usual way is to discuss it with a doctor. They might give me information like “in 70% of cases like yours, it helped”. That is only the base rate. A prediction tailored to me with all my medical context would be useful. However, I will not publish all this sensitive information to the internet.
Another example could be picking a school for my kids. We consider a lot of personal aspects in this decision. Again for privacy reasons, this is not for the internet public.
Privacy applies not just to humans but to organizations too. Usually, it is rather called confidentiality there. Hal Varian said:
We had a prediction market [referring to Prophit in 2007]. I’ll tell you the problem with it. The problem is, the things that we really wanted to get a probability assessment on were things that were so sensitive that we thought we would violate the SEC rules on insider knowledge because, if a small group of people knows about some acquisition or something like that, there is a secret among this small group.
The general point is: We value privacy/confidentiality more than forecast accuracy. The price for getting information from a prediction market is too high. Since it isn’t about the cost, it would not help to make it cheaper.
Robin Hanson suggested hiring markets as a promising approach. Maybe I’m biased from living in Germany, where we have strong privacy laws, but this also requires a lot of personal information made available to lots of people. Data protection and compliance officers would be horrified about this. Maybe dating markets, like Manifold.love? It is somewhat similar to hiring and maybe people are fine with publishing very private information in dating profiles.
Anonymizing data might be feasible but comes with additional risk. Once the data is out there, a little bit too much could be enough to identify you. Also, the platform would know you, so they must be very trustworthy.
2. Forecasts are less useful once public
Some information is only useful if it is exclusive. Robin Hanson describes it with examples like this one:
A G7 government official and advisor to their head of state: “The prediction market experiment was a success, but we will not proceed with the programme as it interferes with our ability to shape the narrative around the direction of government policy.”
Apparently, the forecasts were useful but the side effects were not acceptable. They want a good forecast but they want it for themselves first.
A more personal example could be buying a house or a car. Maybe I need to estimate the cost of some repairs before buying it. If these costs were public information, they are not particularly useful to me–they are just factored into the price. I need the information exclusively to identify a bargain.
This problem seems impossible to solve with prediction markets. The probability is determined by the price and the price is essential for market participants to trade. You cannot hide the price. However, a forecasting platform like Metaculus, which does not rely on traders interacting with each other, you could hide the aggregate forecast. It is a common practice to do this initially so participants predict independently.
This could be turned into a monetization method: Sell the option to delay publication of the forecast. For some bucks, you could be the only one to see the current aggregate forecast and the rest of the world only sees the probability from an hour or a day ago. In other words, pay to hide.
There are no problems, only opportunities
I like to finish on a constructive note. Maybe someone wants to tackle those two problems and discovers a novel approach for applying prediction markets:
Offer anonymized prediction markets on sensitive topics.
Sell exclusive access to forecasts.
This could make them actually “useful” for many people. They would use a prediction markets even if it is no fun at all.
Manifold Markets from the point of view of a bayesian investor.
Robin Hanson considers that maybe markets would be even better if not everybody has access.
Probably until next week, confidential readers! 😊