resolved Apr 24

Resolved

N/AAs a follow-up to this market, I'm interested in projecting out prediction market usage in the short to medium term. Since Manifold helpfully exposes its analytics (https://manifold.markets/analytics#), I thought we could start by projecting Manifold usage out to the end of the year. In the past week, Manifold roughly averaged 70 daily active users (DAUs). This market will resolve in favor of the numerical answer closest to the average number of DAUs during the aforementioned week on January 2nd, 2023. If for some reason, Manifold no longer exposes its DAUs as of 01/01/2023, I'll do my best to get them some other way but resolve the market N/A if I'm unable.
Apr 24, 9:44am: On reflection, the resolution strategy I described may be a little too winner-take-all to provide good incentives. I'd be open to a strategy that involves choosing multiple, weighted by their distance from the true value if a principled one existed. If I had asked for means and standard deviations, I can imagine giving weight proportional to the density each answer assigned to the true value but with only point estimates, I'm not sure how to do something sensible. If anyone has good ideas, please reply in the comments! I'll informally declare the first week of the market a commenting period during which I reserve the right to update the resolution strategy. After that, starting on May 1st, the resolution criteria will be locked in.

7000

33%

Subsidy/discussion thread

16%

700

10%

70

1.6%

Sort by:

And the easiest way to enact this is to resolve this N/A and recreate, right?

Yup, that's the case at the moment :P
For these kind of exclusively bucketed scalars (and maybe in general), it'd be nice to have an option to additional responses from anyone who is not the market creator.

@austin to be clear, I couldn't actually prevent someone from adding another answer, right? The pre-specified buckets is just enforced by me saying I'll only pick one of the pre-specified answers?

Oh right - one other option is to have pre-specified buckets. @Tetraspace is doing so eg here: https://manifold.markets/Tetraspace/if-kamala-harris-is-elected-preside
I think buckets might be the best approach for soliciting scalar values, for now. You could switch between linear or log scales or some hybrid, to capture outcomes you care about (e.g. 0-1k DAU means Manifold has failed; 1k-5k means underperforming; 5k-10k means on track, 10k-50k means breakout success etc)
Setting up the buckets to be meaningful does impose extra up-front work on the market creator, but that may be a reasonable tradeoff.
Also I just realized, in the limit (with arbitrarily small buckets), you get to input a full probability distribution haha!

Yeah I'm a fan of the Metaculus full distributions thing although I sometimes with it could be constrained to a single distribution to prevent people from overthinking the shape of the distribution. I also considered doing a similarly naive solution but having distance be in log-10 space to deal with the intuition that 141k seems like an equally if not better guess for 70k than 1 user. But I feel like there are some counterexamples where log-10 space also feels unintuitive. E.g., the case of 1M vs. 10K being equally good for a true value of 100K. I guess this is why ultimately a distributional solution is way more principled.

@Austin knowing / being reminded you're targeting this makes me think it's more likely, although I guess it's not a total surprise that DAUs is a key metric.

Yeah, the question of how to frame a good scalar market is one that we haven't fully resolved. If people have a good solution, please let us know!
One approach is to use a YES/NO market where you map 0-100% to different probabilities. E.g. 100% = 10K DAU, and then you'd resolve this market to PROB. This is a fairly straightforward, but I'm not fully convinced it's incentive aligned, yet.
Metaculus does the cool thing where you can input a full probability distribution (eg a normal-ish distribution where you set the mean, variance, and left/right skew) allowing you to place true estimates on scalar values. Maybe there's a zero-sum way we can do this to make it fit in the framework of a prediction market (Metaculus is positive-sum). The probability distribution is tricky for new users to grok, though...

In our seed round http://bit.ly/manifold-seed , we were targeting 10k DAUs in about a year; if we're on track then 7k DAUs should be the right number. (Of course, it's an open question whether we're be on track; at the time of the seed round we had about 100 DAU so it's dropped a bit since. But Jan 2023 is really far away -- farther in the future than Manifold has been around!)

A trader bought Ṁ1 Subsidy/di...ion thread from 14% to 16%

A trader bought Ṁ1 Subsidy/di...ion thread from 12% to 14%

A trader bought Ṁ1 Subsidy/di...ion thread from 9% to 12%

A trader bought Ṁ1 Subsidy/di...ion thread from 7% to 9%

A trader bought Ṁ1 Subsidy/di...ion thread from 5% to 7%

A trader bought Ṁ10 7000 from 17% to 38%

A trader bought Ṁ1 Subsidy/di...ion thread from 3% to 6%

A trader bought Ṁ1 Subsidy/di...ion thread from 0.0% to 3%

A trader bought Ṁ5 7000 from 3% to 18%

A trader bought Ṁ1 7000 from 0.0% to 3%

A trader bought Ṁ5 700 from 4% to 20%

A trader bought Ṁ1 700 from 0.0% to 4%

A trader bought Ṁ1 70 from 0.0% to 4%