The Googler presented a puzzle today: his prediction markets give accurate internal forecasts and yet no one values them. When an executive considered a market on whether Stadia will be active in one year, the exec doubted the usefulness, remarking that he could decide which way it goes.

Austin presented a compelling diagnosis: that prediction markets only reveal a tiny bit of information, the answer to a very precise question, which is rarely useful. Most of the work, according to Austin, is in asking the right question.

Perhaps what prediction markets need is a way to come back with more than a point estimate. Could you ask a question with a multi-dimensional answer? Can bettors somehow answer a complex question like: what is the reason to keep Stadia alive in one year?

Enter a new kind of prediction market: where the possible answers are submitted not by the asker, but by the traders themselves. The asker and creator of the market can choose which submitted answer wins, and traders who own shares in it split the pool for the question.

It’s tricky to bet in such a market when the eventual answer may not exist yet. We’ll need ways to bet “none of the above”, and pay out correctly depending on the options available when you purchased it.

If we want greater expression on the part of the market creator, we could allow multiple correct answers, or even a weighted selection of answers that sums to 100%, and pay out accordingly.

Imagine how a creator can learn in detail the answer to their question: the strategy to save Stadia that they would have told themselves to pursue last year. The market can solve for the information that you really wanted to know, and tell you now.

Free answer implementation