I will resolve YES if N/A resolution is worse than never having made the market and NO if N/A resolution is the same or better. Ignoring opportunity cost and inflation.
Aug 25, 8:58pm: the initial description of this market may be incorrect, I recently learned that initial liquidity worked differently to my model of it. Resolution criteria remain the same.
Sep 26, 11:41pm: Attempted to restore resolution criteria after corruption.
Related questions
🏅 Top traders
# | Name | Total profit |
---|---|---|
1 | Ṁ68 | |
2 | Ṁ33 | |
3 | Ṁ29 | |
4 | Ṁ21 | |
5 | Ṁ18 |
I made a market that's sort of an opposite of this one. Losing UTBs is really the only penalty that currently exists for resolving a market N/A, so if Manifold wants to make N/A resolutions less discouraged instead of more discouraged, this is probably the most likely way they would do it.
@MartinRandall
I disagree, if we have :
P(A | B) = 0.2
P(B) = 0.05
P(A) = 0.1
P(A and B) = P(A | B) P(B) = 0.20.05 = 0.01
P(not A and B) = P(not A | B) P(B) = 0.80.05 = 0.04
P(not B) = 0.95
It means we have more market at extreme values, and it is bad because it means you should take more risk in them : People should bet more to hope the same expected return, with a risk of losing everything.
So the probability of the market risk to be screwed toward average value.
The risk of N/A for P(A | B) decrease you expected return, but at least it doesn’t increase your risk.
Given that
1) Increasing variance of return seem bad.
2) You get 2% of loan each day, which greatly decrease the cost of blocking mana for long bet or bet with a great chance to end-up being N/A
3) Anyway, a lot of people have a lot of mana in reserve.
I think conditional bet is better.
@MartinRandall
Also I think :
- Conditional probability is often easier, more natural, to think about.
- In the long run, arbitrage could be made easier, or even fixed by bot.
Also, sometime you just care about the conditional probability.
For example, I created this market today : https://manifold.markets/DenisBaudouin/will-i-like-guix
This is conditional on me actually testing Guix. But I don’t care if people think I will effectively test Guix or not, and I think it would be weird to bet on that (the probability of an average Archlinux liking Guix seem to be somewhat useful information, me personally doing what I intend to do in this case, not so much).
@DenisBaudouin A lot to cover in this discussion.
It means we have more market at extreme values, and it is bad because it means you should take more risk in them : People should bet more to hope the same expected return, with a risk of losing everything.
So the probability of the market risk to be screwed toward average value.
Well I previously made this market, so I understand the perspective:
/MartinRandall/will-manifold-limit-markets-to-199
But I think people should spread their NO 1% bets out over many markets to reduce that variance, and having more P(A and B) and fewer P(A given B) markets can help that.
Also, I think it's pretty clear that, all else equal, low-probability conditional markets skew more towards 50%, or towards whatever market participants want to signal, detached from base reality. Whereas the "interest rate" skew of markets with low probability is smaller and relatively well understood.
Conditional probability is often easier, more natural, to think about.
I don't think it is. Or rather, it feels easier and more natural, but that's often a trap.
https://dynomight.net/prediction-market-causation/
For example, I created this market today : https://manifold.markets/DenisBaudouin/will-i-like-guix
I'm going to assume that you think you have a ~95% chance of trying Guix. To do the same thing, I might create two markets:
Will I try Guix?
Will I try and like Guix?
On the first market I would buy up to ~95% and place a limit order YES at ~95%. Now someone who wants to make a pure bet that P(like given try) is high can bet YES on market 2 with a small NO bet on market 1 as a hedge.
@MartinRandall While I strongly agree with https://dynomight.net/prediction-market-causation/, that doesn't mean that the conditional probability isn't easier to think about. I think about P(X|Y) as: in worlds where Y is satisfied, what is the probability of X - this reflects correlation regardless of whether it's causal, and imo this is often a lot easier to reason about and trade on than P(X&Y) and P(Y) separately.
In the example you gave, I see having to think about the two markets and hedging between them as a major source of friction compared to a single conditional market.
@jack If I know that Guix will suck out Denis's eyeballs, then it's easy to bet NO on P(X&Y), and reveal that information for profit, but hard to bet on P(X given Y).
If I'm bothering to get on P(X given Y) then I must have an opinion on P(Y) anyway, which makes betting on P(X&Y) easy. I can hedge if I want to, but I don't have to.
@MartinRandall Yeah, I definitely agree that is an issue. If P(Y) exists you can still bet on that and reveal your info, but it's true that betting no on P(X&Y) is likely better. However, I still think reasoning about conditionals has a ton of advantages, and so I think it might be good if the underlying markets are P(X), P(Y), P(X&Y), etc but Manifold automatically displays the implied conditionals and also lets you make trades based on inputting your beliefs on the relationships between the variables (I think this shares some similarities with latent variables https://www.lesswrong.com/posts/HygyWpDcwsekqmfnS/will-manifold-markets-metaculus-have-built-in-support-for)