Next Generation AMMs - Eliminating LVR by Anna George | Devcon SEA

[Music] hello um hi I'm Anna from cow wait one second ah perfect now we have the right slide and um today we're here to talk about amm designs why they're currently broken how we can fix them and what other positive side effects it will have if we fix amm who here is currently depositing liquidity on Unis swap yeah you might um find out and I think in the previous talk we actually had the question who of you is hatching against lvr and not so many people were raising their hand so um we are talk about here to talk about today to let you know why you should rethink about about how you want to provide liquidity in the future um so let's start why do we want to change the system it seems to be working fine amm have been around for a few years at this point they have a ton of liquidity so they seem to be okay however research has shown in the last couple of years that actually liquidity pools leak a lot of value and this phenomenon we are calling loss versus rebalancing or in short lvr or lever and it essentially describes this adverse selection risk that you take by potentially taking the wrong side of a trade So In traditional Finance this happens if you trade against an informed Trader but on blockchain this problem is accelerated it essentially happens on every single block and so you basically are guaranteed to trade at an information disadvantage so I don't call it adverse selection risk I call it adverse selection guarantee and to describe it with help of an example today most of the price finding unfortunately is still taking place on centralized exchanges so um if you think if you see that the East price is dropping from $3,100 to 3,000 then you can now go to binance and buy one East for $3,000 on Unis swap the pool will show still show you the outdated price so you can sell your one e for $3,100 so the arbiture is making making a $100 profit and then the liquidity providers on Unis swop they sell at a loss you can measure this loss by looking at how your liquidity position is comparing is performing compared to a rebalancing portfolio if you look at concrete numbers research or basically what we can see is that there's a lower bound of at least 30% of um the volume that has been settled on Unis V2 post merge is coming um from arbiture Traders so arbiture volume isn't equal your lvr loss but you can be assured that Arbiters will only trade if they can make a profit if you look at concrete numbers research suggests that between 5 to 7% of liquidity providers liquidity is lost annually due to IV this equals an US dollar an estimated amount lower bound of $500 million and this is on ethereum alone and from everybody who was raising their hands that they are providing liquidity to Unis swap part of this money is coming from you so in other words providing liquidity on chain currently is not the happiest thing for liquidity providers so we need to look into how we can solve this problem how can we solve lvr and in order to understand how we can solve it we first need to also understand how the current systems are operating so if we look at the traditional amm Market most of the amm are using some type of constant function amm the reason is that essentially on blockchain currenty you have a lot of different transaction in a single block and then they get executed sequentially so you are first second third or wherever in the block one trade gets executed after the other however you get different prices right like the first Trader might get a different price than the second trade so now this turns out to be a bit problematic because at the same time you also want to then fulfill s to satisfy par Independence which means that there shouldn't be an incentive for for you to split up your trade if you trade $1,000 worth of a token there shouldn't been an incentive for you to instead settle 10 trades of $100 each or maybe to have like a very simple way of describing it on the constant function your price is determined deterministically whereas the actual price might be might be somewhere outside so you basically have these two constraints on traditional amm right that you're processing TR sequentially and that you have to fulfill path Independence now we were going back to the drawing board and like see is there like a different way of how we can actually fulfill this or make this less problematic so we looked into batching a solution and if you collect all the transactions of a block into a single batch and you execute them actually at a uniform clearing price then the order of the transactions doesn't matter anymore because anyway all of the transactions within this batch received the same clearing price and then it also doesn't matter anymore if users want to split their trades into multiple because in any case the batch is reaggregating all those trades so as I mentioned I'm from cow and cow has been looking into batch auction mechanisms for a long time so we have a live mechanism for over 3 years and so we decided okay let's try it out let's build an amm on top of this existing batching mechanism so really brief overview of how it works all transactions are collected in a batch over a certain period of time and then there's an auction where multiple external parties who we call sers are competing to finding the best possible price for the orders that are in this badge and they also have to comply to some rules one is this uniform clearing price essentially all Traders with the same assets should get the same execution price and then there's another Criterium ebo that requires that also the minimum price met is at least equal or better to existing other onchain prices having this competition guarantees then that even if somebody is trying like to take advantage of the a low like of a price discrepancy there's an there's an auction there's this competition between different parties who will all Drive the price up and then lastly there's this consolidation of all transactions in the batch you can net them out against each other and only the leftover amount is executed then against the amm and by having this additional noises of like additional transactions you can generate additional Surplus that is then all distributed this captured lvr and the additional Surplus that is captured is distributed to the liquidity providers so in other words the AVR is captured and redistributed in this way the amm doesn't suffer from it um just to show show you briefly how the price functions differ on the left you have the constant function one of the traditional amm that sort of like deterministically defining where you where your track gets executed and then now with the new approach by batching you can now think of a price function where at the actual price point sits outside so with every trade that you execute you can move up this price function this is why we are calling it a function maximizing amm and looking concretely into the return of your investment on the traditional amm it is sort of like a rebalancing strategy just that you have the loss that lvr incurs and then on top as a bonus you get the fees that the amm collects on our new approach the function maximizing amm it also operates in a way like a rebalancing strategy but you do not incur the loss from loss versus rebalancing and on top of that you get additional Surplus so in other words batching combined with this new price function removes ivr and can finally make liquidity providers happy this sounds great in theory um how does it actually perform so as I mentioned we did do an implementation of top of cow protocol and we call it cow amm it's been live for a little over over 3 months and it has around 20 million in TVA at the moment it's deployed across ethereum arbitrum and nosis chain and has um essentially 29 pools that have more than 10K and tvl the most interesting data point up front it currently outperforms Unis swap V2 pools in 50% of the cases you might wonder why do we compare against Unis swap V2 the reason is that Unis swap V2 is currently performing best for Passive liquidity providers so we wanted to compare km against whoever is performing the best um let's look into the methodology we using we looked into growth of 10K so essentially if you deposit $1,000 US dollar into a pool on cow amm versus the equivalent un2 at the same moment in time how does this strategy perform and the reason why we like this methodology is because it can be negative because at the end of the day as a liquidity provider what you care about is the total assets that you hold in your balance in your wallet we also look at AP comparisons um just because this is what users like to see they're really used to it um but in a way it's somewhat misleading because it can be very positive you can sometimes have moments which shows oh amazing I have 100% 200% APR but it doesn't tell you so much we use it for essentially like short-term price volatility okay let's look concretely at the pools the interesting part here is I already said that km outperforms univ V2 pools in 50% of the cases but the interesting part here is we also looked at tvl and so Unis swap has larger tvl then C amm across all the pools we compared and the interesting component here is that in those cases where unisab tvi is only between three to nine times higher than on c amm c amm already outperforms Unis Swap and only on pools where the tvl of Unis swap is between 30 to to 80 times higher cm is still underperforming and this is kind of expected because the more the like the more TV you have the more like the more overproportional it also attracts trading volume so it will also generate more fees so we should expect that as tvi is growing on cow amm so should also increase its performance then we were actually curious to see if there's currently Surplus left on the tables because we literally like K was just launched three months ago and it also of course relies on good solver support and unfortunately so far only three solvers integrated C natively so we looked into token paare usdc westpool um and looked at different performance periods I think this one particular shows for one week or a little longer actually and um looked into um how many trades could have been settled against cow amm but didn't and what is the Surplus that was left on the table and it turns out that unfortunately 96% of the Surplus is still not captured um and then we've Rend the analysis across different time periods and also across different token pools and turns out that this captured 4% is even an upper bound so sometimes even more surpluses left on the table so we're currently working together with solers that it do natively integrate CM because then also the performance will significantly go up um loss ver rebalancing ivr is particularly problematic in times of high volatility right like the more there's a price discrepancy between a centralized exchange compared to a DEX the more value can be extracted so we looked into those High volatile times and there's actually a bunch of examples I'm only going to show two now that are both pretty recent so this the save token got listed on upit and within a time span of 15 minutes it had a huge price increase to over of over 80% so we looked into this time interval and there was a bunch of rebalancing transactions on km a bunch of them were juicy one particularly juicy one um generated over 10% in Surplus if you compare this with how it would have performed on Unis swap there here to remind you how does Unis fop capture lvr not at all the only thing they have to kind of circumvent the losses of lvr is by taking a fee but the fee is fixed and the fee is fixed to 0.3% at least on the equivalent pool and so if you compare the profitability CM essentially did 30 times better than Unis swap um then to give one second example just to show this is not a rare occasion last week the cow token got listed on binance also led to a really big price jump and um there essentially even 15% Surplus was captured in one single transaction again price jumps like the price movement obviously happened over time interval of a few minutes not in one single block so there's a bunch of like individual rebalancing transaction on C amm and I just picked here one particularly juicy one um and again if you compare this to a 0.3% fee on Unis swap then on this particular trade um CM outperformed Unis swap 50 by 50 times so what's the lesson here that's essentially if we remove RR and we add Surplus into the mix we can actually really maximize the liquid the returns for liquidity providers now if we remove the problem of lvr we can finally also think of C amm as an investment strategy why is this the case it's because in a way it is performing like an index fund because it automatically rebalances the reason why Unis swap wouldn't work as a good Index Fund strategy or rebalancing strategy is again because you have the loss due to ivr which is now fixed and then it essentially even outperforms an index fund because instead of having to pay fees to to pool managers that are doing the rebalancing for you it does it automatically and then on top of that so you don't have to pay fees and on top of that you even gain additional fees from the Surplus we call this new product dtfs or decentralized traded funds and it has all the financial upsides that I just mentioned and and it can also already support multi tokens because you can just create two different pools with like a 50/50 or id20 um weight and then you can take those newly created IP tokens and put them in a new to in a new pool and this way you can already create your multi token asset pool we will also uh build on top of balancer V3 so then there's going to be a more native more straightforward integr of multi tokens but you can already do it today and you have full control over your assets naturally and can access them anytime and then there's this last really really nice benefit also that comes from integrating on top of car protocol which is that you can actually leverage your yield bearing tokens because if you look at the unisa pools they're all against West right but a lot of people want to hold rep stake ease whatever any interest bearing token and you can use them and put them into an amm one because like you have this really nice rebalancing strategy but the highlight of kocol is that its underlying Sol mechanism can easily swap in and out of this token so if what somebody is interested in buying W and not rep St East they can still take advantage of the liquidity that's in the pool that you created so instead of keeping all your idle ass assets all your ID yield bearing tokens in just your wallet you can now think about hey why don't I deposit them in an amm and earn additional yield on top so rebalancing on km plus generating Surplus and leveraging interest bearing tokens is essentially a little bit like providing liquidity on LSD the summary is um by removing so ivr today is the reason why liquidity provision doesn't work well we can fix this now with a function maximizing amm and the current implementation of it C amm is already outperforming univ 2 in 50% of the cases and the performance will increase as there's going to be more server support and as the tvi will grow and as of today dysfunction maximizing amms can already unlock the potential of passive investment strategies so I'm inviting you to try it out thank you thank you Anna I am definitely interested to try it out so let's take some questions first one if C amm volumes come from the left over of the batch meaning that the noise Trader volumes don't hit the amm aren't LP getting very little trading fees due to low volumes and how do you solve that I would say that at the moment yes probably if you compare the fees obviously that they receive compared to units of volume it's lower but the real benefit is so essentially there's studies showing that the fees that Unis V3 is making are not making up for the losses that avvr is creating so just by comparing this cm is already performing better right because km is removing ivr by pushing the prices to the real price where the price should be and then even if the Surplus is SL might be slightly lower than the fees the Surplus is on top of this lvr removal and on Unis fob it's only trying to make up for the lvr all right next question how much of the car operation happens on offchain well the basically at the I mean we are working on a decentralization white paper because the goal is in the future to have everything fully on chain but the matchmaking currently is taking place off CH okay right so the next one if it outperforms univ V2 on 50% of the traits what happened for the 50% remaining does it perform the same or Worse um I think I showed actually the performance on the slide and it's currently performing worse but it's like by 0 point something perc it depends on the pool depends on how much tvl it has and depends on how much volume we be we are able to capture on those pools so there's no this is the fixed percentage across the pools but yeah it is SL slightly underperforming at the moment okay does c lping Works similar to Unis swap V fee no because Unis V3 is leveraging concentrated liquidity and this is not the audience we're targeting we are targeting passive liquidity providers um actually if you look in the performance of univ V3 everybody who's like trying to do passive like who's not super active and is not an informed Trader and like a specialist and knows what they're doing are actually suffering even worse like they're their positioning return is much worse than on univ V2 so our Target group is the passive um liquidity to provider who wants to go who wants to get a return on the investment and not leak value and so you know it's different definitely how can you come up with lower bound on percentage of swaps that are Arbitrage um let me try to understand the question are you come up with lower bound of percentage of swaps that are aage uh do you mean the I I assume that what is meant is essentially the the slide where I showed this so this is not research that we did internally um but there's a bunch of research cers that I think I quoted in the source of the of the presentation so if you're interested you can look into it okay okay so the next one would be in the example you gave how amm performs outperforms uni V2 in 50% okay that that has been answered so we're going to the second one can you call it an index fund when a token holding strongly depends on the price isn't Hing strictly necessary um so as if you like basically if you look at an ETF strategy what you do is you invest into it right and then you have like a company who's doing re balancing for you actually much less frequently than it happens on an amm so um depend on the price so in other words for this part like if if you're like an informed Trader who's providing concentrated liquidity on univ 3 then probably you do hatching if you're a normal passive liquidity provider who essentially just wants to earn some interest on their hard earn tokens without being a professional at what they're doing then no you don't need it because overall you're not losing money but you're making money on top okay so we have time for one last questions do c amm LPS outperform the hle strategy yes we did look into it and um in maybe so the problems we only have data periods of three months but the short answer is yes it out performs it um maybe if you look at like if a newly created pool was just set up in this a couple of days or that maybe there might be cases where it doesn't outperform it but over time yes it does okay thank you very much Anna thank