When will Manifold implement question format for predicting dates?
💎
Premium
19
Ṁ999
Apr 1
66%
Q4 2024 or earlier
85%
Q1 2025 or earlier
Resolved
NO
Q1 2024
Resolved
NO
Q2 2024 or earlier
Resolved
NO
Q3 2024 or earlier

I'm fond of creating markets for predicting dates. Currently the best available mechanism for doing this is to create a multiple choice question with cumulative intervals, like you see in this very question.

This is workable, but has its downsides:

  1. The order between the probabilities is not enforced. It means that when there's not enough bets, "Before 2026" could have higher probability than "Before 2027".

  2. It's easy enough to add new options for future dates, but adding new more fine-grained options for the dates between existing dates is messed up because they will not be sorted properly in the UI. (Edit: I'm wrong here. If you sort alphabetically and write all dates as YYYY-MM-DD.)

Metaculus has its own format of questions for date predictions, but it's a) doesn't translate well to prediction markets, b) (subjectively) is a bit hard to use.

The question will resolve positively as soon as I am able to create a publicly visible date-prediction question with the following requirements:

  1. It predicts the whole probability distribution over dates, not just the median prediction.

  2. It is possible to increase prediction resolution when the event draws near. Either anyone can make predictions for any dates right from the start, or if this is implemented as a type of multiple-choice question, then it's possible to add answers between existing options without messing up their order.

  3. The market should enforce the probability being monotone over time. If X < Y then P(event happens before X) should be ≤ P(event happens before Y).

  4. The options should be dates, not real numbers.

Currently multiple choice questions fulfill requirements 1 and 4, and numeric markets fulfill 2 and 3 (kinda).

One possible solution is as follows:

  1. Date market is a type of a multiple choice question, in which all answers are "Before YYYY-MM-DD".

  2. Answers are automatically ordered by date.

  3. It is possible to add new answers. Their initial estimation is not 50%, but is interpolated from existing neighboring options.

  4. When a bet is made that would push the probability of some date past its neighbor, part of the bet is automatically transferred to the neighbor to push it's probability in the same direction and preserve the ordering or probabilities.

To illustrate 4: suppose there's a market with predictions

Before 2026: 20%

Before 2027: 25%

Before 2028: 30%

If a bet is made on "Before 2027" that would push its probability to 35%, then part of it is transferred to 2028, resulting in probabilities like:

Before 2026: 20%

Before 2027: 32%

Before 2028: 32%

Get
Ṁ1,000
and
S1.00
Sort by:

Great idea! Would be good for any market in which some choices are a subset of others

I like this proposal. I too like this "or before" style of question, and participate in them a lot.

Step four of the proposed solution is a neat feature, but I think it may be a little hard to explain to the user exactly what will happen once they click buy. Manifold would ideally show a little animated graph or table that demonstrates what will happen as you increase or decrease your buy quantity, taking into account any limit orders that may exist. A bit of work to implement, but it would be very useful!

It should, of course, work in either direction. Buying a YES should propagate to later answers, and NO should propagate to earlier ones.

bought Ṁ10 Q4 2024 or earlier YES

Good idea, I do have one nitpick though

Adding new more fine-grained options for the dates between existing dates is messed up because they will not be sorted properly in the UI

Isn’t this solved by Sort: A-Z and then Make Default?

@ahalekelly I don't know how I missed that. :) I'm always doing "Sort: Old" and make sure to add options in order.