Will Gurkenglas convince me that multiplying the net winnings of the market creator by (L_tot-100)/L_tot would not effectively fix the "free money for market creators" "exploit" currently extant in the free daily market system?
4
20
149
resolved May 22
Resolved
NO
(see description) Note that this is based on my judgment - if there is some extremely convoluted way (especially involving coordination with other actors) for the market creator to still pocket a significant portion of the daily $100 I would likely still consider my solution worthwhile. Similarly, if due to some minor quirk of the fee structure or similar, market creators can still net some small daily amount (say $5 or under) I would still consider the solution worthwhile. For a YES resolution, Gurkenglas would have to articulate a hole in the solution large enough and reliable enough for me to think the exploit would be worth continuing to use. May 21, 11:36am: (in the question title, L_tot is total question liquidity at time of resolution) May 21, 11:45am: also, regarding the proposed solution - this would only apply to daily free markets. Market creators would not have their net winnings penalized on markets they ante'd up for themselves.
Get Ṁ1,000 play money

🏅 Top traders

#NameTotal profit
1Ṁ225
Sort by:
Alright - final resolution: * I am convinced by the experiments Jack and I ran that multiplying the market creator's net winnings (aka "profit") by (L_tot-100)/L_tot will successfully prevent easy mana farming by creators from the free daily markets * I am also persuaded by the critique that this does make betting on one's own free daily markets quite unattractive in all cases (not just mana farming), which may be an unwanted externality that makes this solution unwise overall Thanks for provoking some really interesting exploration of the system @Gurkenglas! I definitely learned a lot.
bought Ṁ50 of NO
Related note - it seems that part of the $2-3 residual profit remaining after applying my rule is actually attainable even in a non-free market. I generated $1 (or at least, an increase in the displayed ledger of $1 - not sure what's going on before the rounding) from nothing here: https://manifold.markets/MattP/test-case-money-for-nothing-clicks
Can't say I fully understand exactly why (liquidity is confusing), but it looks like @Jack and I have both found experimentally (with free daily markets) that if you apply my originally proposed rule to the amount displayed as "profit" in the interface, you come away about $2-3 ahead when all is said and done due to some quirk of the fees/creator earnings structure; nothing close to the $100 people are currently able to farm. These tests have made me fairly confident that applying my rule would vastly reduce the incentive to "exploit" the daily markets for income. If no new information/arguments arise before the question closes, I'm planning to resolve this as NO per the criteria I set out in the original description (aka net winnings $5 or below being insignificant enough to not be worthwhile using the exploit IMO).
Addendum to the above - as Jack discusses below, this (almost) zeroing out of net winnings works whether you inject liquidity before or after buying shares, which matches our experimental results (Jack injected liquidity after and I injected liquidity before, we both applied the profit scaling factor and would've nearly zeroed out our net take).
predicted YES
I repeated the same test and this time used a free daily market and kept track of my balances: https://manifold.markets/Jack2/test-7a8d0ec91708 End result was I gained M$20 under the current system, which means I would have gained 400/500 of that = M$15 under your rule. It's kind of surprising to me that this ended up with more profit than what you tried. I didn't try to optimize it at all, I just literally did what @Gurkenglas suggested which I assume was some arbitrarily chosen numbers. I suspect that by choosing the bets more carefully you could do better.
bought Ṁ100 of YES
Oh wait, I did the calculation wrong. Under your rule, the net payout of the bet M$63 is multiplied by 400/500 -> $50. Then the net balance change would be a $2 gain. So that's actually quite similar to what you saw.
@jack Curiouser and curiouser. Yeah, I'm pretty confused at this point. Before your most recent comment I was thinking I hadn't had my liquidity returned to me as you had - most likely due to injecting the liquidity before I bought the shares, while y'all have been injecting liquidity after you bought the shares (right?). What I was thinking was that if you inject liquidity before you buy the shares, you're then consuming that liquidity when you buy the market up to 99% (and getting more YES shares for your money). Then the winnings from that extra liquidity is scaled back per my rule, and it ends up in a wash. Whereas, if you inject liquidity *after* buying the shares, you just get that back (but not as winnings) so you are able to skirt around the effect of the rule. But, then you commented your results were actually similar to mine. So I'm left thinking I'm maybe mostly right?
predicted YES
Yes, I injected liquidity at the end - I don't know if that's a better strategy than doing it before. I think it's better but I was starting with 50% probability which is probably suboptimal. Even a m400 bet only captured m63 of profit, so there was apparently a substantial amount left on the table. I still suspect there exists a way to make the payouts significantly bigger.
sold Ṁ30 of YES
I dug into the liquidity payout math - details in the comments on https://manifold.markets/Jack2/test-7a8d0ec91708. The amount of injected liquidity lost in this example is m63 * 400/500, because m63 is the amount of YES shares taken from the initial liquidity pool which is the losses incurred by the initial liquidity provider, and 400/500 is the proportion of those losses that you as the second liquidity provider take. Let L_tot be the total amount of liquidity injected (500 in my example), and P be the trading profits before being reduced by your formula (63 in my example). P is exactly the amount lost by the liquidity pool. Of the amount you inject, you lose P*(L_tot-100)/L_tot - that's the proportion of the liquidity you provided, which happens to be the same as the multiplier you proposed. So this exactly cancels out the amount of profits you get! Then what is left is some tiny amount which I think is coming from the author commission. So I think it happens to work at eliminating most of the profits - but the more important issue in my mind is that this mechanism would make the author lose money on most legitimate self-trading, so it creates more problems than it solves.
predicted YES
Oh, I meant to add: my calculation was for betting first then injecting liquidity. Now I actually think doing it in the other order would be better.
sold Ṁ43 of YES
If you inject liquidity first then bet: First you put in L_tot - 100. You buy YES from 0% to 100%. Your net profits in the interface are basically L_tot (i.e. you extract all the liquidity). Then the adjustment makes the actual payout L_tot - 100. But that's the amount you put in yourself! That's why it ends up canceling out to some tiny amount, which is related to fees/commission.
@jack so you're saying I inadvertently stumbled on a fairly simple multiplier that works to almost eliminate profits in both of the simple cases (injecting liquidity pre and post share purchase) despite the path to that point being different? Fascinating.
predicted YES
@MattP Seems to be - very cool how the math worked out :) Although I think the way liquidity payouts work today is not ideal and I'd suggest changing it to more closely pay your liquidity back out (if we are planning to stick with this general CFFM liquidity approach), and in that case the patch I suggested earlier would be appropriate.
bought Ṁ490 of YES
Alright, tested it out for myself on this market: https://manifold.markets/MattP/test-dont-bet-on-this My starting total balance: $388 Starting probability of market: 1% (for cheapest YES shares) Liquidity injected (prior to buying shares): $200 YES purchased (in a few steps): $100 Market total liquidity (prior to resolving): $301 Gross payout: $399 Net profit/winnings (in interface): $299 My total balance after resolving YES: $490 My total profit under current rules: 490-388 = $102 (not sure where the $2 came from but whatever) Compare to how it would be under my proposed rule, with everything the same except the net profit/winnings (displayed in the interface as "Profit") scaled per my rule: Adjusted net profit/winnings (under my proposed rule): (301-100)/301)*299 = $200 Balance after resolving YES under my proposed rules): $391 Total profit under my originally proposed scheme: $3 So.... actually, I think this test just convinced me (experimentally) that my rule actually *would* work to eliminate the single-actor "mana harvesting" issue. Not remotely what I expected, given that I bought this question's probability up to 98% right before doing data review making this comment... but it is what it is. I still give @Gurkenglas (or anyone else) until the close date of this market to change my mind. Gonna be pretty difficult at this point though, given the test I just ran.
Any formula that only applies to the market creator can still exploit if you have one friend to collude with. The friend take the free M98 and this formula doesn’t apply to them.
predicted YES
@Undox agreed, but that's not the problem I was attempting to solve. Any time there's regular "free stuff" it will be pretty doable to collude with others to exploit it. Much harder problem to solve than plugging a no-collusion exploit.
bought Ṁ100 of YES
Leaning heavy yes now, because I think I might've just been a dumb. If the *net* winnings are being multiplied by any positive value (which my factor would be), than you always recover your costs and get at least something on top. My solution works if you don't inject any liquidity (scalar is near zero), but injecting any liquidity lets you keep a significant portion of your winnings. The only reason I'm not 100% convinced yet is that I'm still not quite sure what happens to the injected liquidity. Seems that it might not be recovered, in which case I'd be correct in my original assertion and the proposed solution would actually plug the "hole", so to speak.
predicted YES
If the way I currently suspect it works is correct, I think fixing this issue would require the *gross* winnings (which is to say, the payout from the shares) to be modified by a factor, rather than the *net* winnings as I originally suggested.
predicted YES
(or just modifying net winnings but replacing the 100 with the total author injected liquidity, as @jack suggested)
predicted NO
I'll run Gurkenglas' suggested experiment tonight at my next free market and report back at that point.
- **dreams**
I see now that you can neither delete nor edit comments haha
bought Ṁ20 of YES
I tried this here: https://manifold.markets/jack/can-an-author-profit-despite-the-ma However, I'm confused about what happened to the liquidity I injected. Did I just burn m$400? lol
predicted NO
@jack hmm, so what actually happened?
predicted YES
@MattP I think it did work. I got confused about the liquidity, I believe it did get returned to me as expected.
bought Ṁ240 of NO
@jack When I ran a similar test, I didn't get the liquidity back because it was included in my winnings already. Still not 100% sure how liquidity "return" works - I suspect if I hadn't bought up all of the liquidity in my example it might've been returned to me, though. Maybe the key is to inject liquidity after buying the shares, as you did (rather than before buying the shares).