Will Manifold significantly improve the reliability of extreme market probabilities before the end of 2024?
19
แน€2.0k
2025
8%
chance

The reliability of probabilities on Manifold decreases as it moves further from 50%. A market being at 1% rather than 0.1% indicates pretty much nothing about the actual probability of the event; it's just not worth tying up mana for such a tiny return on investment.

This is a huge problem for markets on unlikely but high impact events such as nuclear war; their probabilities are often meaningless.

Feel free to suggest solutions in the comments.

Get แน€600 play money
Sort by:

if there's general agreement that the status quo means mana's being left on the table, there might be nice alignment if this were improved through an @acc-like "house bot"

one way that it could be done without risking a large amount of mana on an autonomous bot would be to have a side web app where trusted users give their odds for these types of markets (possibly just binning them into 0.1%/0.3%/1%, but it could allow the user to enter a specific percent too like metaculus). if a market is unambiguously considered far off, the bot could kelly betting to push the odds in the right direction

an initial list of candidate markets can already be generated with something like https://outsidetheasylum.blog/manifold-search/?liquidityWeight=0&closeTimeWeight=1&creatorVisibility=true&p=0.5%3A5%2C95%3A99.5&l=500%3A but this could be improved (or it could consider markets that have been manually added to a given group)

the bot could pay out some percent of its profits back to the trusted user group to keep them incentivized. you'd have to think about frontrunning, but maybe that could be mitigated by randomization (if folks are holding YES/NO waiting for the bot to pull the trigger, that actually adds to the amount of mana the bot can push towards this problem)

@AndyMartin This sounds very easy to exploit.

@SG really? I'd agree it's an ugly hack on top of the wider market system and it might not be worth doing on those grounds alone, but it seems like there are a lot of mitigation options and there's not actually that much incentive to exploit it because the movement in odds/price is so small

it would need to assume mods will reverse false resolutions, but that seems safe

if you only allow folks with established track records into the user group and say "using alts == manifold ban", I don't think anyone would actually do this -- they would need to frontrun 100s of independent markets to make a significant profit, which would be very easy to detect (it's possible they could use 100s of independent alts, but at that point it's probably a better use of their time to do other things with them...)

beyond requiring votes from many users and being very sensitive to a few ">1%" votes, it could also inspect the current holders and look for frontrunning right before it makes a purchase

there's maybe a concern that this would incentivize early betters to cash out, allowing the price to swing back, but this is arguably a good thing as it lets these users put their mana to more useful markets (and lets you all make more profit on the market by betting it back down again)

predicts NO

@IsaacKing yeah I'm following/funding that, but I don't think it's quite the same (although @EvanDaniel please correct if I've missed something)

this idea can only be done by "the house" (or maybe an account like @NinthCause ?) because so much mana is needed

in general, there's no need for the specific "users binning things" scheme mentioned above, just a reliable way to identify liquid markets that are "hung" but still far from 0 or 100

this is pretty simple already with queries like https://outsidetheasylum.blog/manifold-search/?liquidityWeight=0&closeTimeWeight=1&creatorVisibility=true&p=0.5%3A5%2C95%3A99.5&l=500%3A but someone spending a few hours on it can likely do much better

beyond marginally improving the extreme probabilities issue, it has a couple unrelated positive follow-on effects:

I haven't followed it as closely since creating https://manifold.markets/AndyMartin/will-manifold-use-a-loan-rate-that , but I think significantly raising the loan rate for users (and bots?) with strong track records is worth exploring

I've put a M$5000 limit order on NO, as a reward for anyone who can make this resolve YES.

How would tying loan rates to predicted probabilities affect this? Currently we get 4% of the amount bet, regardless of what probability the market was at after we bet. If this percentage was lower for bets close to 50% and increased as you get further away from that, would this achieve this goal?

predicts NO

@BrunoParga I think it would help (I've suggested the same thing), but it would exacerbate problems around people defaulting on the loans and leaving the site.