Will Manifold set abandoned questions to resolve to MKT, by default?

Basic

Ṁ2611

resolved Jan 3

Resolved

ALL

@MilliOnaire is working on a way to automatically resolve markets to MKT probability, after a certain period of time. I think this is quite a good thing - it means that traders can free up - For abandoned markets, someone else can come in and provide liquidity, and bet the MKT to what they think it should be - If they're wrong, then other traders will seek to profit out of them - It's decentralizing the resolution criteria by using a keynesian beauty contest - what grounds the resolution is both 1) the community opinion on the resolution and 2) the possibility of the creator coming back My current proposal: The auto-MKT-resolve date defaults to 7 days after "criteria date". This is changeable by the creator (but not set in the creation screen, similar to the tagging interface) on "criteria date", creator auto-withdrawals all liquidity. But holders of YES/NO shares can continue to trade on the same market if someone else provides liquidity. One major alternative: Markets could resolve to N/A automatically instead. @JamesGrugett likes this because it prevents the possibility of "gaming" the system - if you think a market will be abandoned, you can manipulate the price right before the auto-resolve. I think gaming is possible, but the lost profits from traders will be worse (especially for traders who already exited via selling their shares). And what's nice is that the auto-MKT-resolution also punishes the creator financially by draining their liquidity and sending it to the other traders, rather than letting them get off scot-free in the N/A case. Also, there are possible fixes to prevent gaming (eg - using the weighted average of the prob between the criteria date and the resolution time, or picking a random prob point in time from the last 24h, or something; I think the game theory works out okay on these)

This question is managed and resolved by Manifold.

#Manifold

Get

1,000

and

3.00

25 Comments

17 Holders

224 Trades

Sort by:

I think polling makes more sense than MKT, but I'm not against trying MKT for now

N/A is terrible. Mkt prior is okayish. Simply poll manifold participants and give them a cut of the fees (resolve based on 67-80% supermajority, with no payout to minority vote or market creator)

I think there's ~0% chance that Manifold won't at least try some variant of auto-MKT by Jan 1st. Whether they'll keep it is less certain, I'd buy at 75% that they keep it.

predicted YES

I'm @Milli now, sorry for messing up all links.

I think we should just try out auto-MKT resolution now (potentially plus time-randomization). If there's too much price manipulation, we can easily switch to resolving N/A in the future.

@SG My prediction is that people won't manipulate prices. It's risky, right? If two people try it and pick opposite directions then they both end up throwing away their money (I think). Also it's just kind of an uncool thing to do, contrary to the truth-seeking spirit. Eventually it will happen, of course, if it's actually profitable, but, well, I guess I'd predict it happening less than 1% of the time in the next year.

predicted YES

@dreev Hm, I think the final bettor in a MKT resolved market is always profitable, which is the biggest cause for concern. (and why I like Poisson resolution time, or a weighted post-close average instead of final %, as solutions.

@Austin Wouldn't you have no guarantee that you'll actually be the final bettor? Now I'm curious what the actual Nash equilibrium of this game is...

Market manipulation already happens in open markets, I think it's certain to happen in auto resolved markets if there's no randomness.

I don't think you can easily get away from the problem of price manipulation just before auto resolve date with a MKT auto resolution. Yes, the creator could come back and yes, other people could bet the probability back down, but fundamentally you're turning it into a race in the seconds before the auto resolve date. That doesn't seem like a recipe for rewarding rational traders, it's a recipe for rewarding the fastest clicker with the lowest latency.

predicted NO

@MattP There are fixes to that type of gaming. My favorite is to make expired markets undergo "exponential decay", so every minute there is a small chance of them auto-resolving to MKT.

@MartinRandall That's smart. Also known as a Poisson process. When the deadline hits, the *most likely* time for the market to close is always Right Now but the *expected* amount of time till it closes is always, say, an hour from now. I have code that implements the continuous version of this efficiently. And you can be 99.999% -- with as many 9's as you want -- certain that the market will actually close within some multiple of the expected amount. So you can essentially know that the market will close within, say, 6 hours of the deadline, but in the event that it does go that long without closing, all you know at that point is that it's almost surely going to close within the *next* 6 hours. (That's why it's called a memoryless distribution.) If that's too tricky or confusing, another solution is a simple quiescence rule: Keep extending the deadline until trading peters out. It's also possible that there's little incentive to try to game it by sniping at the last second because if anyone else also did so, you'd be screwed. Mechanism-design-wise, it does strike me as the kind of thing you should ignore till it actually happens.

predicted YES

I'm sold on the Poisson process! I'd propose calibrate it to a 50% chance of resolving each hour (after 7 days); simple to remember, and in 12 hours it's 1:4096 that the market is still unresolved.

@Austin Reasonable! I'd recommend the continuous version if you think it's worth implementing at all. Keep me posted and I can write up the algorithm I'd recommend. I have a lot of experience with this stuff -- it's what tagtime.com uses.

predicted NO

@MartinRandall OK Martin, +1 for suggesting the Poisson process. Haven't heard anyone talk about those since grad school, fun trip down memory lane. My $0.02 is to calibrate it at a significantly lower chance - say 20% per hour. Why? We're not in any particular rush to auto-resolve I wouldn't think - and the lower the chance, the lower the chance anyone tries to game it.

predicted YES

Yeah to be clear, we're not intending to really use the actual prediction here as the mechanism by which to decide. As with most of our polling markets, the intention is to allow users to have a place to leave comments on our different proposals!

predicted NO

Voting yes, betting no.

predicted YES

Mainly I'm in favor of automatic resolution. If I had to choose I would pick MKT over N/A for two reasons: * N/A eliminates all winnings, so people are less inclined to do research / participate. In the worst case market creators who are losing M$ "accidentally" let the market auto resolve with less face loss than fraudulent manual resolution * Markets resolved as MKT display the sentient at that time clearly and are more valuable for the future If the issue is a deal breaker for some (rather "no auto resolution than X") I would pass the decision to the market creators by setting a default that can be changed (see 6. below).

predicted YES

It seems to me we have a binary market to decide on a question that has 3 fundamental answers: 1. No automatic resolution 2. MKT automatic resolution 3. N/A automatic resolution Market participants who are in favor of 2. can clearly state that by buying "YES", but then it gets complicated: The majority could be in favor of 3. but buy "YES" to not risk 1. It could also be that there is no absolute majority for any of the answers so people are buying "NO" by default. And this is not even considering the Keynesian beauty contest. I feel creating another market would spread the discussion over even more markets than there are already on the topic so I'm against it.

predicted YES

1. Yes, makes sense! 2. I'd keep the close date in the header unchanged, until AFTER the market closes; then have the date switch to showing "time until auto-resolve". This keeps our UI less complex, always showing the traders what's the most relevant. Especially with 4., people can find out already 3. I think the EDIT button should only appear next to wherever the date is shown; so if we go with 2., then only in the INFO panel. 4. Great! 5. I guess that's reasonable, to set a "infinity" option or something like that. Not sure about this, I could go either way 6. I _think_ that's okay, to make any of MKT/N/A/YES/NO the default resolution. But I don't feel strongly, could also see the case for "only MKT" or "only N/A" to keep things simple. If I were personally implementing this, I would keep it as "only MKT", with a distant second preference for "only N/A", to keep things simple, then add other options if people feel very strongly. 7. I think existing markets that have closed should have 7 days starting from when the feature was implemented. (or maybe 14, to give a bit extra time). Existing markets not yet closed should get the default "resolve in 7 days".

@Austin +1 to all of this

predicted YES

@Austin @7.: How should market creators be informed that their markets had an auto resolution date added?

predicted YES

From @MilliOnaire: After the discussion yesterday I thought I'd specify what I plan to build in regards to https://manifold.markets/Austin/will-manifold-set-abandoned-questio, before I build stuff that has no future. 1. Automatically set a date for automatic resolution as MKT for 1 week after market close (in the future possibly liquidity removal) 2. Display the auto-resolve-date in the header of the market (see screenshot for first draft) 3. The "EDIT" button next to the dates opens two date pickers, for close and auto resolve 4. Mark automatic resolutions in Market info/Stats I'd like some feedback on this questions: 5. Should the market creator be able to completely remove the automatic resolution? Dates far in the future are functionally the same but look silly. 6. Should the market creator be able to set other automatic resolutions than MKT? * James argued for N/A * Market creators might want to do YES/NO for things that have a default solution ("Will there be a nuclear war?", "Will I resolve this market manually?"). 7. What should be done for existing markets? Should automatic resolutions be added to existing markets or just newly created ones?

I'm very pro auto-resolution but am still agnostic on resolving MKT versus N/A. Price manipulation for MKT resolution seems like a very big problem. It's true that randomization could help, but that makes the final mechanism a lot more complicated.

@SG N/A is obviously superior.

Related questions

Related questions