There have been many attempts at self-resolving markets, such as resolving to the market price at close, or only doing so with a certain probability, resolving to the quiescent market price with a time limit, or with a different result at the time limit, or with no time limit, resolving to a poll, or a different type of poll, or a third type of poll, or even resolving to a future superintelligence. None of these have been entirely satisfactory.
Will this be a solved problem by the end of 2023? Will Manifold eventually settle on a single type of market that self-resolves robustly and to everyone's satisfaction? (Or a few different types for different situations.)
At the end of 2023, I will pick the self-resolving mechanism that seems the best to me and apply it to determine the resolution of this market. (If it's a method that requires the market stay open for a certain length of time, such as involving a quiescent market price, I'll extend the close date as necessary.)
1,000
3.00🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ244 | |
| 2 | Ṁ203 | |
| 3 | Ṁ193 | |
| 4 | Ṁ125 | |
| 5 | Ṁ71 |
@IsaacKing I don't even think that should count as a self-resolving mechanism at all, but regardless I think the long delay is indeed sufficient to make it non-ideal. It's also ambiguous if there are multiple superintelligences that disagree, and based on the ambiguity of what counts as a superintelligence.
@IsaacKing I'd say public voting, but doing something like
1. using mana (100 mana = one vote, no fractional voting),
2. with a set limit for number of votes (maybe 20?)
3. with a rule that each vote delays the resolution by 24 hours
a. but only people who have voted prior to the delay can vote,
b. and additionally you get disqualified from voting additional times if you have not voted in more than the 4 latest delay periods
That should let people vote, signal confidence, harness the power of accumulated good predictions (given people are using mana/money), makes sure it isn't whalebait, and still ensures the market gets resolved in a somewhat timely fashion.
@Joshua people have run a ton of markets that way as pure gambling (whalebait/altbait) certainly, but I don't remember one that was framed as self resolving a real question. But I think it would be a bad mechanism, certainly worse than a poll. Note in particular that in a market buying no means you think it is lower than the market probability. Consider that the market might be trading at 80%, and equal numbers of people may think it is actually more vs less likely than 80%, and then there's no reason for more people to buy yes than to buy no. So it totally fails unless people buy yes or no as votes, which is dumb, them you just want a poll instead.
@IsaacKing I agree that resolve to poll is the best currently practical method.
And I asked whether it counted as self resolving in one of the first comments on the market, with Isaac responding yes, going a definition that the decision is distributed rather than centralized.
I created this market - https://manifold.markets/qrdlkaggle/would-self-resolving-markets-with-d?r=cXJkbGthZ2dsZQ
I think the way to do it is to use a complex function which can best estimate how likely someone is able to make an accurate ground truth resolution for a market.
This estimate will cap how much they are allowed to directionally bet in a particular market.
The reason we haven't seen this in manifold before is there doesn't seem to be a way to enforce this.
@qrdlkaggle For the markets I created I've added this appendum. Let me know what you think!
I will resolve based on the current price at a random time in a 24h period before the end date to reduce manipulation. The random stopping idea is from this paper - https://arxiv.org/abs/2306.04305
Probably, but the results of a poll can still be manipulated, unless the sample size is very large (and taking a poll with a large sample size is too difficult for many users, and thus not satisfactory).
Also, I think it's debatable whether resolving based on a poll really counts as self-resolving or not. The poll is a factor external to the market, even if its purpose is to determine the resolution.
@JosephNoonan While I completely agree those are issues, in practice I have run many polls where I ask people to vote honestly by honor system, and they pretty much do.
Also agree it may or may not count as self-resolving, but for the purpose of this question it should, because Isaac mentioned it in the description.
@IsaacKing I think more discussion needs to be done around this. Self resolving markets could be huge.
@jfjurchen "A self-resolving market is any where most of the decision-making power is distributed among many individual users rather than centralized in a small number."
@Yev And even if @jfjurchen's method technically counted as a self-resolving method, it certainly wouldn't be my pick for the best one.
What is your definition of self-resolving? Depending on what goals you are most looking at, you might ask for different criteria. For example, the goal might be resolving based on market participants only without any other external data (or perhaps, just requiring that with high probability).
I think resolving to a poll works well enough for most markets today, although they definitely aren't perfect and there are many ways they can be improved.
@jack A self-resolving market is any where most of the decision-making power is distributed among many individual users rather than centralized in a small number.