These are great suggestions and I would love to see them implemented! I'm curious though how you think prediction markets compare to reputational platforms (e.g. Metaculus) for information processing. It would be really interesting to hear your ideas on UI updates for reputational platforms.
Thanks for the question! To be honest I have been much more interested in play-money prediction markets than reputational platforms, due to the following thought experiment: In principle, we could reengineer the backend of Metaculus to use prediction markets under the hood, while presenting the exact same user experience on the frontend. Users making predictions would be offered access to portions of liquidity pools for each question such that their change in reputation from making a prediction equates to their currency balance change from making a trade according to that prediction. We can imagine that the platform operators write a bot that has near-infinite liquidity, has access to all the user's trades, and uses the same aggregation mechanism that the platform would otherwise use to set the probability as its own prediction. If this bot has ~100% all of the liquidity, then the price shown in the platform will be the bot's probability, so the price that appears in the UI is the same as it would be on vanilla Metaculus.
Given this thought experiment presents the same UI, we can start to ask questions like:
* Why not, instead of giving users access to just the portion of the liquidity pool that reflects how much reputation they could gain, let them access the whole liquidity pool?
* Why give almost all of the starting liquidity to the house bot, rather than, say, 90%? If the bot is really better at aggregating predictions than the best alternative a bot-writing user will come up with, it will maintain its lead. But if someone writes a better aggregation mechanism, then it can outcompete the house and provide better predictions across the board. In the worst case scenario, the total increase in log loss across Metaculus questions would be log(0.9), per the theorem in the Beygelzimer et al paper that I wrote about a few months ago.
I guess to make some concrete suggestions along these lines, I would promote:
* Making it more clear how much reputation stands to be gained or lost from a particular prediction.
* Making it more transparent what the aggregation mechanism is, and allowing users to suggest their own mechanisms (though I think Metaculus has already recently made some movements in this direction, so kudos on that front)
* Incorporating or making it easier to incorporate bots or other automatic prediction mechanisms into the data feed.
These are great suggestions and I would love to see them implemented! I'm curious though how you think prediction markets compare to reputational platforms (e.g. Metaculus) for information processing. It would be really interesting to hear your ideas on UI updates for reputational platforms.
Thanks for the question! To be honest I have been much more interested in play-money prediction markets than reputational platforms, due to the following thought experiment: In principle, we could reengineer the backend of Metaculus to use prediction markets under the hood, while presenting the exact same user experience on the frontend. Users making predictions would be offered access to portions of liquidity pools for each question such that their change in reputation from making a prediction equates to their currency balance change from making a trade according to that prediction. We can imagine that the platform operators write a bot that has near-infinite liquidity, has access to all the user's trades, and uses the same aggregation mechanism that the platform would otherwise use to set the probability as its own prediction. If this bot has ~100% all of the liquidity, then the price shown in the platform will be the bot's probability, so the price that appears in the UI is the same as it would be on vanilla Metaculus.
Given this thought experiment presents the same UI, we can start to ask questions like:
* Why not, instead of giving users access to just the portion of the liquidity pool that reflects how much reputation they could gain, let them access the whole liquidity pool?
* Why give almost all of the starting liquidity to the house bot, rather than, say, 90%? If the bot is really better at aggregating predictions than the best alternative a bot-writing user will come up with, it will maintain its lead. But if someone writes a better aggregation mechanism, then it can outcompete the house and provide better predictions across the board. In the worst case scenario, the total increase in log loss across Metaculus questions would be log(0.9), per the theorem in the Beygelzimer et al paper that I wrote about a few months ago.
I guess to make some concrete suggestions along these lines, I would promote:
* Making it more clear how much reputation stands to be gained or lost from a particular prediction.
* Making it more transparent what the aggregation mechanism is, and allowing users to suggest their own mechanisms (though I think Metaculus has already recently made some movements in this direction, so kudos on that front)
* Incorporating or making it easier to incorporate bots or other automatic prediction mechanisms into the data feed.