[POLL] Yes/No markets: how often would you prefer to bet "up or down", vs "on a value"?
9
102Ṁ124resolved Jun 10
Resolved as
82%1H
6H
1D
1W
1M
ALL
Complement to https://manifold.markets/Austin/poll-numeric-markets-how-often-woul
Today, all betting on Yes/No markets is "up or down"; all betting in Numeric markets is "on a value". We may want to support both mechanisms, but I'm curious to see what our default should be.
To participate, comment with a % representing how often you'd rather bet "higher or lower than current probability", rather than "a normal distribution around X".
Eg "80%" means "8 out of 10 times, I'd rather bet on 'higher than 17%' rather than 'normal distribution around 20%'"
This market resolves to the average % submitted across all responses.
This question is managed and resolved by Manifold.
Get 
1,000 to start trading!
1,000🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ11 | |
| 2 | Ṁ3 | |
| 3 | Ṁ2 | |
| 4 | Ṁ1 | |
| 5 | Ṁ1 |
People are also trading
Sort by:
I'd be willing to bet that if this market were rephrased we'd get a noticeably different outcome. There's the part about normal distribution which doesn't seem to make sense to myself and the other commenters. And what Mathias just said about "I have no idea what betting 'on a value' for a yes/no market means"
Here's my proposal: you should be able to make trades of the form "buy until the price reaches a specified amount". E.g. if the current price is 50, I can say "buy up to 60" - this buys yes until the price reaches 60. Or "buy down to 45" buys no until the price reaches 45.
I also think the interface should let you put in either a M$ figure or a probability figure, and show you what the other number is, i.e. today I can start by inputting M$10 and it'll show me what the end probability is, I want to be able to do the reverse too where I input the end probability and it shows me how much M$ that costs.
Some other nice things about this:
This can be seen as similar to limit orders, and is nice because maybe they can be presented in the same interface: you specify a) a max M$ amount, b) a max price, and c) a duration your order is open ("right now" executes the trade now, "day" or whatever = limit order). The system should then execute your trade against the AMM at the best possible price - e.g. when current price is 50 and you input an order to buy up to 60, your trades should execute continuously over the interval 50-60, not all at 60.
It's also a more robust way to input trades. Currently, if I put in a buy yes, and someone else buys M$1000 YES right as I'm clicking the "submit bet" button, moving the price from 30 to 90, I can end up buying at a much higher price than I intended, which is really bad for me. My proposal cleanly prevents this issue.
Another nice thing is if I already have no shares, and say "buy up to 60", this can automatically start selling no shares first, then buying yes shares. In comparison, today you have to manually make two separate trades, or you just buy yes and each pair of 1 yes+1 no is converted into M$1 - but this doesn't behave quite the same as selling your no and is slightly worse for you I think, at least in part because buys cost fees while sales do not.
@jack this make sense to me! I would compare it to an autobid — keep bidding until the price/probability hits a certain level.
Why would someone ever bid though?
Either you have the money to move the market directly to your probability (adjusting the bet and looking at the probability change), or you don't have the money (in which case neither a large initial bid or an autobid will move the market to that probability, you simply don't have enough M$ for that).
@MathiasFoster Yeah, autobidding is a great analogy.
I'm not sure I understand your question, but I think maybe you're talking about setting bids for later. Here, I'm still talking about trades that execute right now (i.e. market orders) - setting bids for later is a separate thing you can easily add on top of this.
Also, buying yes is the same as bidding the probability up. And the reason to do it is to make a profit - you buy YES because you think it's underpriced. And if you don't have enough money to move the market to your target probability, that's a good thing for you as a trader - it means you can invest however much you're willing to invest at a cheap price.
If I say "buy M$10 of YES" that means keep buying YES from the market until I've spent M$10. The interface also shows me what the price range is: probability 50% -> 58% means the first share I buy is at 0.50, and the last share I buy is at 0.58 - every time I buy a YES share the price goes up a bit (I'm approximating for clarity, the math is actually continuous).
If I say "I want to buy all YES shares available on the market right now at below a price of 0.58" that's basically just another way of framing the same thing (except with some other nice advantages I mentioned above).
I think to many people, saying "buy M$10 of yes" maybe feels more intuitive, which is why I think it might be better to frame it as an order where you put in the max M$ you want to buy, and optionally put in the max price you're willing to buy at. (And if you put in the price limit first, it should autofill the amount it'll cost for you)
Created followup market and poll here: https://manifold.markets/jack/poll-should-you-be-able-to-input-a
Here's a good example of why betting yes/no can be confusing when the market is used to represent a numeric outcome - it's not intuitive what a bet on yes means: https://manifold.markets/misha/on-august-1st-which-fraction-of-joe#wTZUtnhNZlyhQ1bSVZSk
100%... I'd rather bet to a specific probability than either of the options you mentioned, but "high/low" is much closer to that than "normal distribution around X%". The latter doesn't really have any useful interpretation to me.
If the question is "will X event happen?", I can answer "I think there's a 75% chance X event happens" and that's something that means something. The answer "I think the chance of that happening is a normal distribution of probability around 75%" is not something I think 99.99% of people would have any idea what they mean by - not to mention is enormously difficult to determine calibration for.
(and to be clear, if the dichotomy is "bet on a specific probability vs bet on up/down", I choose bet on a specific probability and my answer would be "0%" - I'm not remotely convinced that betting on a normal distribution around a probability is the same thing as that, though, so I pick the interpretation that's clearer to me)
To address the "YES/NO markets and numeric markets should be the same" topic: I think there is a big fundamental difference between numeric and YES/NO markets. The difference between 0% and 1% on a YES/NO market is infinite, while the difference between 0 and 0.01 on a numeric market ranging from 0 to 1 is exactly the same as the difference between 0.50 and 0.51. You can think of this as the shape of the scoring rule/function. YES/NO questions have a logistic shape or something like that, while numeric questions with a linear scale (which is the only option Manifold has) are linear. Thinking of Metaculus for comparison, there's a reason why Metaculus has binary and numeric markets and they have different scoring rules.
More simply, the key point is that probabilities do not work on a linear scale.
Also, I have made some YES/NO markets where I mapped a numeric range to 0%-100%, which can kind of work as an approximation, but it works best near the middle of the range (where a logistic function is closest to linear). As you get close to the extremes, it breaks down in the sense that the market probability stops being anywhere near the expected value of the numeric result.
@jack Hm, two things I suspect (and am still trying to prove!)
1. Straight up mapping 0 to 100% to a numeric range (eg 5 to 25) does have the property that betting up the estimate from 15 to 16 is much cheaper than moving from 23 to 24, and moving to 25 is infinitely expensive. But I don't view this as fatal, I think the return on betting is still aligned? Which is to say that if you're correct that the expected value is actually 24, you net money by moving from 23 to 24. The fact that it's more expensive just reflects the fact that the liquidity provider initialized under Uniswap is less willing to sell you those shares.
2. One way to understand what's going on is that when the liquidity provider/market creator sets a cutoff from (5, 30), what they're actually doing is setting up a Uniswap v3 range order from (5, 30) within a hypothetical total range of (-inf, inf). A regular yes/no market in fact has distributed liquidity across this total range (in odds space, not prob space). If another liquidity provider was then willing to set up a wider range order, eg provide liquidity from (-10, 60), then they would smoothen out the cost curve so that it is more linear in terms of actual payout.
I believe each separate range v3 range order is basically the same as someone offering to provide liquidity for a separate mini-market that follows the same resolution criteria, and then the algorithm chooses where to buy the cheapest shares across all such mini-markets
"As you get close to the extremes, it breaks down in the sense that the market probability stops being anywhere near the expected value of the numeric result." -> I realized that I was wrong about this, I got confused partly by fees and rounding when I went to empirically test this in a market.
Yeah, you're right, the market still rewards betting towards the correct EV. The cost to do so increases away from the initial probability setting for the liquidity (towards infinity near 0% and 100%), which doesn't seem ideal but isn't necessarily a dealbreaker.
10% - I think about markets by figuring out my probability estimate, and then trade to bring it partway towards that. In practice, I choose my bet size primarily by tweaking it up and down until I get the probability I want (which is very annoying). The exception is in high-liquidity markets where it takes a very large bet to move the probability to where I would put it - but this is a rare exception in my experience. I'm strongly in favor of supporting both.
However, I'm not sure what you mean by "a normal distribution around X" because that's not really how I think about predictions on YES/NO markets.
@jack yeah, I am also confused by the intent there. You can't have a single person have a normal distribution of probability estimates for a single event occurring or not occurring. I mean, I guess you sort of could if you squint but it's extremely unclear what that actually means in the real world.
@MartinRandall Numeric markets and YES/NO markets should actually be the same thing; I'm betting heavily that we'll combine the underlying mechanisms haha https://manifold.markets/Austin/will-manifold-change-numeric-market
@Austin in a true yes/no market that will only resolve yes or no or NA, what would be the difference between be up
@Austin I don't think you're correct about the numerical type being actually the same as a yes/no market. You could try to use it in a similar way, and maybe you'll be successful, but it's a fundamentally different (and more complicated) thing IMO.