By 2022-04-01 Manifold representative will admit that reward function used as of 2022-02-28 results in silly results for nearly impossible/certain events.
11
39
resolved Mar 3
Resolved
YES
Given odds of troll market or missclick are noticeable AND return is abyssal for markets at 97% or 10% and someone predicting obviously true event AND it is necessary to lock up money for some time there is no incentive to predict that unlikely/certain events like https://manifold.markets/GustavoLacerda/will-russia-invade-kiev-by-end-of-2 or https://manifold.markets/colorednoise/will-a-nuclear-bomb-detonate-in-a-p or https://manifold.markets/AYev/will-ukrainian-troops-enter-in-mosc or https://manifold.markets/ACXBot/will-there-be-a-major-flareup-worse or Events that already happened sitting at 93% (Kiev invasion by Russia) are embarrassing for a prediction market. It is barely profitable that chance for exploding nuclear weapon over populated area is below 4% ---------------------------------------------------- Resolves as yes if by 2022-04-10 comment will be posted here demonstrating that condition from the title was fulfilled. (according London time zone if near edge) No idea how to solve it but "Will aliens land before March 2022?" at 1% and being unable to profit from pushing it lower and risking that it would mistakenly resolve as yes is extra-silly. Mar 2, 11:09am: loans are a weird case. Is it changing specifically reward function that I mentioned in the title and condition?
Get Ṁ200 play money

Related questions

🏅 Top traders

#NameTotal profit
1Ṁ8
2Ṁ2
3Ṁ1
4Ṁ0
5Ṁ0
5 Comments
26 Trades
Sort by:
sold Ṁ1 of NO
> It's designed to counteract this part: Note that I also mentioned some associated issues. I have not considered something like that, but if reward function can be constructed in way that includes also whether funds are deducted immediately from balance, or only at completion.
bought Ṁ1 of NO
It's designed to counteract this part: > AND it is necessary to lock up money for some time But it doesn't change the payout for correcting markets at 90+% and the risks of troll market resolutions or miss clicks still exists. So, up to you whether that counts I guess.
bought Ṁ1 of NO
loans are a weird case. Is it changing specifically reward function that I mentioned in the title and condition?
bought Ṁ20 of YES
James implemented loans and said that "one use case is to calibrate long term markets without locking up capital" which implies that long-term markets were poorly calibrated.
bought Ṁ1 of NO
It's plausible that adding some decimal points to the UI without actually changing the math would be sufficient to make these feel less silly. The implied probability with 250 Yes and 6400 No (the approximate maximum of No on the "aliens before march 2022" market, reached around the 13th) is 250^2/(250^2+6400^2) = 0.15%, but the UI rounds that up to 1%. Also relevant is that the starting ante (presumably 50 M$) is a significant portion of the yes in this case, so the market creator is partially to blame for even the implied odds being that high.

Related questions