This is a market inspired by this recent blog post. I am trying to capture the scenario from the section "No, order is not preserved" as closely as possible.
Suppose there’s a conditional prediction market for two coins. After a week of bidding, the markets will close, whichever coin had contracts trading for more money will be flipped and $1 paid to contract-holders for head. The other market is cancelled.
Suppose you’re sure that coin A, has a bias of 60%. If you flip it lots of times, 60% of the flips will be heads. But you’re convinced coin B, is a trick coin. You think there’s a 59% chance it always lands heads, and a 41% chance it always lands tails. You’re just not sure which.
... people might figure out if it’s an always-heads coin or an always-tails coin
There are two markets in this set, representing two different options our hypothetical futarchy could choose. Of the two, only the market with the higher final price will resolve YES or NO. The market with the lower final price will resolve N/A. If both prices are exactly the same (according to the precise prices from the API) I will flip a coin with FairlyRandom to choose one to resolve.
The rules for resolving the markets involve randomness, and are as follows:
24 hours before the market closes, I will ask @FairlyRandom to (publicly) generate a number from 1 to 100. This represents the chance that coin B is an always-heads coin - if the number is 1-59 (59% chance), then B is an always-heads coin and if the number is 60-100 (41% chance) then B is an always-tails coin.
Then, once the market is closed, I will do the following.
If the A market is to resolve I will ignore the previous number and ask FairlyRandom for another number 1-100. If the number is 1-60 I resolve Coin A to YES, otherwise I resolve it NO. I resolve coin B N/A.
If the B market is to resolve, I resolve it YES or NO according to whether it is an always-heads coin or not, respectively. I will resolve coin A N/A.
1,000🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ2,849 | |
| 2 | Ṁ1,793 | |
| 3 | Ṁ238 | |
| 4 | Ṁ67 | |
| 5 | Ṁ31 |
Derivative market links:
Followup experiments based on Dynomight's followup blog (which is responding directly to this market)
@Tumbles I don't believe these markets prove the validity of the flaw so not going to bet. We can make a different market about some general consensus in the future on whether this is a real flaw.
At the close, the prices will reflect all the information so far and as such "work".
Making a scenario with random close time doesn't disprove this as argued by blogpost author. You just need to perform unnecessary extra calculations at the end to get the true probability.
This market is also flawed. As you are guaranteed that coin B either resolves N/A or YES. And A has 60% or is N/A. I'm confused how can such a market prove anything? It's true if basic logic axioms are true not necessarily the blogpost.
@UnconditionalProbability Until it resolves it will in fact be P(X | X doesn't resolve N/A). Over time however it will become more and more clear that some X in all worlds doesn't resolve N/A and therefore that flaw will disappear at resolution.
@patrik I think even Robin believes it is a flaw because he addressed it in his paper. I think he believes it's not a problem in practice if you set up the market to minimize it. This market was designed to maximize it of course.
Side note: does anyone think using the close price to make the policy decision is a good idea? That means the last millisecond is super important and the rest of the trading window is irrelevant.
@travis I thought futarchy is about making decisions based of values at close so I still don't see this as a flaw. Maybe I misunderstood futarchy then.
@patrik Yeah it's interesting, I think sometimes it gets proposed that futarchy be based on the average price over a time window (this Hanson post gives me that impression). But I think if you do this, then something like the example in this market (which is not really flawed in its current form because everything works out ok in the end) actually might become flawed in the sense that it could make a wrong decision given all the information available.
@BoltonBailey Interesting. Btw I'm not saying the flaw is not real just that it's not "fundamental flaw" or that it makes "the whole concept faulty" as argued by the blog...
Yeah, I feel like every trade should have an equal effect on the decision. Using close price means only the last trade matters. So why leave the market open for more than 1 second (or 1 trade)?
Most of Robin's examples use the choice that has the highest price for the longest time (I'm simplifying, there are other variations/complexities). I think if that rule had determined which coin would be activated in this market, the bias in the coin B price would've been considerably reduced. However the decision might not reflect information that becomes available part of the way through.
And I agree the flaw doesn't seem very important because I think it would be rare that important information drops in the middle of the trading period.
I'm more concerned about the idea that as the price of a proposal drops, the probability of na/void approaches 100% and the price will reflect increasingly bizarre possibilities.
@NivlacM Perhaps we learned that "N/A resolutions which allow withdrawals halfway through can cause negative balances and futarchy is therefore bad for cost-of-capital reasons".
@NivlacM This empirically confirmed Dynomight's claims about markets measuring correlation instead of causation (unless other measures are taken to avoid this).
@BoltonBailey One possible fix for the cost of capital problem is to denominate the bets in shares of another prediction market.
So first you would make a market for "Which coin will be flipped?". Then the coin A conditional market would be be priced in "Coin A will be flipped" shares. If the market is at 25%, I could then spend 1 "Coin A will be flipped" share to buy 4 "Coin A lands heads" shares. If coin A is flipped and does land heads, then my 4 "Coin A lands heads" shares pay out 4 "Coin A will be flipped" shares, which each pay out 1 mana.
If coin A is not flipped, then we don't need to reverse transactions on the conditional market. Instead, all payouts have been multiplied by 0.
This setup also allows someone to invest in both the coin A and coin B markets using the same capital. They can spend 1 mana to get 1 share of "Coin A will be flipped" and 1 share of "Coin B will be flipped". I can then trade in both conditional markets using the same mana.
(These kinds of share-denominated-markets are not supported on manifold.)
@UnconditionalProbability But if there are ways to prevent the problem doesn't that disprove the conclusions of the essay?
@patrik No. I was responding to Bolton Baily's capital-related concern, not Dynomight's essay.
The capital consideration is orthogonal to the correlation-causation one.
Dynomight's essay also mentions that fancy market structures that fix his concerns exist but hove downsides and are almost never used. It would take more than just stating a clever structure to disprove that claim, and it is arguably the most important one wrt futarchy.
One possible fix for the cost of capital problem is to denominate the bets in shares of another prediction market.
Yes I agree. I would conceptualize this as making a multiple choice market with four alternatives, one for each combination of {Heads, Tails} and {Coin A chosen, Coin B chosen}. This way of looking at things is nice because we can see that (as on manifold) the total price of all four markets should consistently stay at 1 and have the market maintain this invariant.
Ok, coin B has resolved at 98.8% and coin A at 61%, so coin B resolves YES and coin A should resolve N/A.
@mods it is telling me that "negative payouts are too large" to resolve the coin A market N/A, is there anything I can do about this?
@BoltonBailey your random number is: 39
Salt: cx50qqc7ou5, round: 5268328 (signature 8fdb51f4821f48e35bb5e3dccd95a22c5279bbb2c30b41d7cee23266b9e080e73b2b48176e1e5a80ac8af0707b7640471955af90506148e8f8497ef494e9cc6fccdc06b16134766897cf460415013985e9a27ffe1d7811f621cd15e70c8f2683)