Resolves to whatever easy-enough resolve-to-mkt mechanism @jack likes the most.
Pick your favorite schelling point!
Related questions
🏅 Top traders
# | Name | Total profit |
---|---|---|
1 | Ṁ34 | |
2 | Ṁ28 | |
3 | Ṁ27 | |
4 | Ṁ18 | |
5 | Ṁ2 |
@Sinclair the trick is apparently to make the market really long duration so everyone forgets about it.
@Sinclair Perhaps @acc could use its HFT connection to the server to be the last whale on all these markets, then users wouldn't bother trying.
good thing there was no manipulation, because if there was I would have misresolved.
The algo I actually did just now: resolve to last prob regardless of what it is.
The algo I was supposed to: if the market looks non-manipulated, resolve to last prob, otherwise do some crazy math or in some other way (poll?) resolve it to something else
("@jack's favorite mechanism" is underspecified because jack likes many different resolve to mkt resolution methods but after messaging him he thinks this resolution is ok)
it's a bit unsatisfying. feels like the acausal magic worked even though I casted it wrong >.<
@Sinclair It was definitely manipulated, I spent the whole market betting towards 50% instead of my beliefs.
@jonsimon resolve to market just means "i will resolve this market to prob based on the market price at close"
Markets are supposed to be individualistic. If you call something a "market" you should expect that people will try to profit. The norm of a market is that there is no unfair price for an apple, nothing dishonorable about asking $2 or $200 - it is just that other people may not buy it.
Betting down on Joe Biden 9/15 is not a lie, any more than betting Twitter above its fair market price is. The incorrect bettor deserves to lose money, but not honor.
I'm thinking auto-resolving is... not half bad? We could say it's 49% bad?
The experiments from Martin and Jack have been fascinating and the biggest thing I've learned is that prediction markets (all prediction mechanisms?) are not reliable when the true probability is near 0 or 1. Even real-money prediction markets with ground-truth resolution have this problem, like we saw on PredictIt in 2020: Trump lost the election but the market price refused to fully reflect that, due to fanatics and trolls. I was very confused about how that could happen until these experiments. Now it makes sense to me. Near the extreme probabilities, there's a big asymmetry favoring the trolls. It costs them peanuts to keep the market price skewed.
Auto-resolving exacerbates that problem but the problem is more fundamental.
Here's one where auto-resolving works like a charm, I think: https://manifold.markets/dreev/will-at-least-1-out-100-bernoulli-t
@dreev ok but predictit is bad. 2020 dem primary had probabilities that added up to much higher than 100% (and you can't singlehandedly arb this back to normalcy because $850 limit).
I agree it's kind of bad, but the question is what's the least bad way to resolve hard questions? I used to do more polls, then I realized it was a bunch of vote for a pretty small number of votes, and very error-prone too.
Empirically, the vast majority of markets in https://manifold.markets/group/selfresolving I'd classify as working well - if I had to label whether the market ended up having a "good" vs "bad" outcome, among the resolved markets I count 2 bad and 7 good, which gives me 2/9=22%. Betting down to 22%!
@jack My verdict on the resolved markets:
1. Flipping a coin. https://manifold.markets/jack/what-is-the-probability-that-a-coin So bad. It's obviously lower than 48%. Suppose my prior is that about half of coins in the world don't have a head on one side, so my prior is 25%. Do you think I should update that prior in any way based on this market?
2. Trading fees. https://manifold.markets/MarketManagerBot/poll-should-manifold-direct-some-po So bad. The correct answer is NO, as seen from current reality where there are no fees. Only 6 bettors, because it's a resolve-to-mkt-or-creator-whim on a market where the creator is betting. Also it was improperly resolved, as it resolved to YES instead of round(MKT).
3. What is love (songs). https://manifold.markets/MarketManagerBot/presupposing-that-there-are-these-t . This is DPM. DPM is going away. DQ.
4. Community guidelines. https://manifold.markets/jack/how-clear-do-you-think-manifolds-co . So bad. In the comments you mention how it failed to converge to a high percentage despite most comments being positive. I think the Biden and Sky markets give a hint about that.
5. Market manipulation. https://manifold.markets/Duncan/is-a-market-that-always-resolves-to This is not a resolve-to-MKT, it's a change-my-mind. DQ.
6. AI vs Pandemics. https://manifold.markets/JamesGrugett/is-ai-a-greater-existential-risk-to . So bad. Heavy market manipulation. Final answer is flat out wrong.
7. UBI. https://manifold.markets/ManifoldMarkets/universal-basic-income-should-exist . This is a resolve-to-mkt-or-creator-whim market where the market creator wasn't betting. I would predict it to settle based on bettors beliefs about the market creator's beliefs about whether UBI should exist. I think it provides less evidence for whether UBI should exist than just having a chat and then asking the market creator.
8. Plastic straws. https://manifold.markets/ManifoldMarkets/plastic-straws-should-be-banned This is a resolve-to-mkt-or-creator-whim market where the market creator wasn't betting. It resolved to NO instead of 4% based on creator whim.
9. AI doom. https://manifold.markets/JiaobeiMandos/i-have-been-entrusted-with-an-ai-in . So bad. Heavy market manipulation. Final answer is flat out wrong.
So, 9 markets, no happy endings.
@MartinRandall My bar here is "did the market do badly or not". I am considering a subjectively reasonable result as "not bad". Like I've said, if we can answer a complicated/ambiguous/subjective question within say 10-20%, I consider that pretty good! On the other hand, answering "Was Biden president on 9/15" at 90-95% is pretty lousy, I'd mark that as bad.
1. The title and description were pretty different and there's a pretty reasonable argument Yev raised about what "heads" normally refers to. (I will rerun this experiment with a more accurate title). I think it's a reasonable answer given that 90% of the time, "heads" = obverse side of coin. OK
2. I interpret the question as "if there are fees, should some of them go to liquidity" because if there are no fees, it becomes both yes and no (paradox) and so n/a is the only reasonable answer. And my answer to that conditional question is yes. OK
3. Sure, DQ, let's just consider binary resolve-to-mkt. N/A.
4. I think there's a difference between asking "is it clear" and "how clear is it"? The poll clearly indicated that most people thought it was clear, and 65% answer to "how clear is it" is consistent with that. OK.
5. Yeah, this is not resolve-to-mkt. N/A.
6. Agree that the market price graph looks pretty wrong. BAD
7. Disagree, it's resolve-to-mkt with creator leeway to override market failures or to extremize to yes/no if there's a clear consensus (I think that's a reasonable rule given that we know markets are bad at betting all the way to 0% or 100%). In my view it's clearly mostly resolve-to-mkt, not mostly resolve-to-creator-whim. And the answer looks reasonable to me. OK.
8. Just like above. OK
9. Clear market failure. BAD
That's where my 2 bad score came from. Not sure which 9 I was counting, but I agree removing the free response ones makes sense, so that leaves me with 2 bad, 5 ok.
And we can add another one that just resolved:
10. https://manifold.markets/MartinRandall/is-the-sky-blue-resolves-to-mkt. Seems like a reasonable way to average out different interpretations of the question. OK.
That leaves a total of 2 bad, 6 ok. 2/8 = 25% is my score.
@jack Out of 7 and 8, only one of the two resolved to MKT, so I don't agree that they are "mostly resolve-to-mkt". Maybe 50/50. Markets with a probability of resolving to market can do well, depending on other factors.
@MartinRandall Yeah I'm being broad here in what counts as resolve-to-mkt, e.g. I think https://manifold.markets/jack/did-hans-niemann-cheat-against-magn is close enough.
@jack You should adjust your rating based on the Biden market being a disaster. And I think you should weight the ratings by amount of mana and attention spent.
@MartinRandall Good point. The Biden market was intentionally an experiment with a market design with no protection against a failure that we predicted, but it was certainly an even bigger failure that I had predicted, so still evidence to update up here.