Panel 1: Designing a DEX in 2023

Panelists: 
Max Resnick - SMG (𝕏: @MaxResnick1)
Andrea Canidio - CoW Swap (𝕏: @AndreaCanidio)
Doug Colkitt - Ambient Finance (𝕏: @0xdoug) 
Xin Wan - Uniswap Labs (𝕏: @xin__wan)
Eugene Chen - Ellipsis Labs (𝕏: @0xShitTrader)

Moderator:
Ankit Chiplunkar - Frontier Research (𝕏: @ankitchiplunkar)

We had I think 1300 1400 people apply for for this event and we could only in 200 people So every single person in here I know has a strong opinion And the goal is to let it be heard.

So don't don't be shy, please Otherwise, I'll start calling on people that I know are in here.

And it might be a little bit on the spot.

So OK, without said, Ankit, please take it away.

To start, this is the first panel that we have, Designing Decks in

I would let everyone go by order and introduce themselves and tell us what they're working on right now.

Yeah hi everyone, my name is Max Resnick, I'm the head of research at Special Mechanisms Group and we design mechanisms primarily.

We're focused on MEV and specifically on MEV on Ethereum for now.

I think you'll hear a lot more about what we think about LVR today.

I am a senior research economist at CalSWAP, Cal Protocol.

And yeah, so I also spend a lot of time thinking and designing mechanism.

And in particular, we've been working recently at Cal Protocol on thinking about how you would design an AMM or a DEX when all the trades are batched, because obviously that's what we do.

And I think we're going to discuss it a bit later.

Doug Colquitt, a founder at Ambient, or some of you might know us as Crocswap.

We're primarily focused on building an AMM where the economics are sustainable for passive LPs.

And I'm working on all things Uniswap, but most recently, Uniswap X and Uniswap V4.

We are building a fully on-chain, limit order book DEX on Solana.

I think I can just start by, maybe Eugene can start, and we can start by them.

You guys should tell us what are your spiciest opinions on AMM design?

I don't know if it's spicy in 2023, but I think it's pretty clear at this point that AMM LPs on average in expectation tend to not make money.

Why do they still do it is not super clear to me.

It seems like there is still plenty of room to innovate on the AMM design space, but at the end of the day, it does seem like kind of the goal of an AMM is to mimic to some degree what a professional liquidity provider, where they would want to place liquidity, perhaps on like a centralized limit order book exchange.

So our opinion at Ellipsis is, probably can't do it with 12 second block times, probably can't do it with 30 million gas per block.

Do we have anyone in the room who's like a passive LP on AMMs?

Like you put, you know, for profit or just for fun?

Just like only a curve, stable pools, or what?

derivatives of each other or assets where you don't have to worry about permanent loss over time causing too much loss to you because you know that they're good by nature.

Is it assets that are also traded on centralized exchanges or only assets are like not liquid on centralized exchanges?

Gotcha, okay, cool There's only 1 LP he's the only LP on all of Uniswap.

Everyone, you can go talk to him afterwards, he'll be in

I think my spiciest take might not be Dex related.

I think Dex related wise, I'm pretty mid-curve.

But I think in terms of MEV solutions, I think the best ones are always going to be kind of scenario dependent and custom to each application, rather than this 1 catch-all solution.

But I think that's probably going to cause some controversy, but that's kind of my opinion.

Yeah, I guess my spacious opinion would be I think the industry in general is going the wrong direction, moving towards professional liquidity.

I don't think you can really have decentralized markets if there are 6 trading firms that you require to provide liquidity.

That's just not decentralized, in my opinion.

I also think retail has a much lower cost of capital than professional liquidity.

I don't think Citadel is getting out of bed for anything less than 100% expected returns, whereas your average retail participant is happy with

So the big problem with AMMs is toxic flow just isn't segmented, and LPs eat all the toxic flow, But I think there are ways that you can fix that and keep liquidity fully decentralized and fully passive

Yeah, so Like as we know the problem from LP viewpoint is that can be framed in terms of asymmetric information that they may be trading against arbitrageurs that are better informed than them and therefore exploit them.

And I think that a lot of the solutions that have been proposed are to some extent missing the point, and I'll explain this by an example.

Suppose that you've inherited a piece of antique from your grandma, and you have no idea how much it's worth it.

Now, if someone shows up, and this person is a super expert in antiques, you know, he's gonna screw you Okay, he's gonna like Tell you a price that is just convenient to him on the other side if someone shows up and you know This person doesn't know anything about antiques is just tell you a price that correspond to his own, how much he likes it, then OK.

Statistically, you may get a good price or a bad price, but it's non-systematic.

Now, when you see this way, you immediately have like, I'm sure you can think of a solution to the problem, because if there are multiple experts out there, you can play them against each other and still get a good price, even if you do not know anything.

And so I think, to some extent, a lot of the discussion about AM design has been about, how do I know the correct price?

For example, feeding the price of an Oracle through an AMM?

Or how do I distinguish between the informed guys and the uninformed guy?

The informed guy wears fancy dresses and speaks with a French accent.

And the guy who is not informed is more scrappy or similar things.

But the idea of, in fact, we like the toxic flow, provided that we can play the informed players against each other, has somewhat escaped.

But this is a bit like the work that we've been doing recently.

I think what other people have said is that toxic flow is the problem, that long block times accentuate it, which I think is true.

But I think maybe 1 thing that isn't discussed is that the proposer and their role in the system on chain, where there's 1 person who gets to write to the chain, who gets to decide all the order of all the transactions, is contributing to the problem as well.

And it makes it really difficult to pit all of the training firms against each other and give users good prices.

So we have a paper that's about how it's difficult to hold auctions quickly on chain, which is kind of what you need to do if you want to solve this problem, is you need to hold auctions quickly.

We know that we can capture some of the LVR.

Like if you go look at the PBS auction, the PBS auction is redirecting a ton of LVR to the proposer of the block because they're like the only person who can do it.

So I think if we could figure out somehow how to run an auction in a market system where the proposer wasn't a monopolist, then we could push more revenue back to the LPs.

I heard a lot about toxic, non-toxic flow, how to differentiate how traditional finance does something on it.

Does anyone want to take on how do some of these firms differentiate between toxic, non-toxic order flow?

Do you think there is a venue that we can even do something like that on chain?

In the theoretical literature, the difference will be between informed and uninformed traders.

And informed traders will be someone Who is paying attention to the prices everywhere, say the Binance price if that's where the price discovery is done.

And as soon as you observe a deviation from your local DEX from that price, we'll trade so to exploit this.

So the difference is that of information and instead non toxic flow will be people that kind of want to trade at a reasonable price So for example a cow swap, you know, we receive intention to trade We aggregate them in a batch and then we there is a mechanism to look for the best execution.

Now we claim that we have a non-toxic flow precisely because the batch imposes a delay, and it's also unclear then from the trader point of view how the batch will be executed.

There is a kind of a competition that will determine afterward.

So it's very unlikely that an arbitrageur can exploit differences in batches in the various prices through our batch.

So We think that our flow is non-toxic, but instead arbitrageur that go directly on a DEX and compete to have their transaction included earlier in the block, precisely to be the 1 that rebalances, will be the 1 that are the toxic.

Yeah, I mean, yeah, like, I mean, to first approximation, toxic flow right now is just the price of Binance.

Like, here's the, like, the reality is price discovery happens at Binance.

And if you look at Statarb on-chain, it's pretty much just check the price against Binance and go there, right?

So like there is arbitrage doors need to exist.

So it's not necessarily getting rid of toxic flow, just making it pay enough.

So if the price of Binance gaps 50 bips and the Statarb MevBot is paying 5 bips to trade on the AMM pool.

So it's not necessarily about getting rid of toxic flow.

It's just making toxic flow pay to the LPs a fair share of the actual economic rents.

And right now, those rents go to Ethereum.

So it's great if you're an Ethereum holder, juices the burn, raises staking yields, but sucks for LPs.

So like a lot of what we focused on is, okay, like how can we move, how can we move that layer, keep it decentralized, but probably Ethereum itself isn't sufficient to do that.

So like we're really excited right now about coprocessors.

We think you can do a lot of really complex stuff in coprocessors, off-chain.

Oh, yeah, coprocessor is basically some decentralized way to do compute off-chain and then post the proof that the compute is accurate.

So ZK would be an example but you know depending on what the security risks are even something like the enclave might be fine right and like you can pump in into SGX or something and you can make reasonable assertions and if it's not fully secure maybe someone gets to swap they hack SGX swap at 1 basis point instead of 50 basis points, and that's fine occasionally.

So we think there's a lot of room to do computation off-chain, but still in a decentralized way, still in a fully trustless way or near trustless, and move that logic off-chain, and then have basically the chain and the AMM be the settlement layer.

I think 1 of the ways that they deal with this in traditional finance is actually payment for order flow.

So Citadel wants to distinguish between the toxic flow and the non-toxic flow, so they buy all of the non-toxic flow from Robinhood, and then they give them pretty good prices because they're not toxic.

You know, it's debatable whether those prices are very good.

But I think that's something that's very clearly coming down the pipe in crypto, and The question is whether the people who are able to pay for the toxic flow will end up being LPs somehow, in which case they'll be able to kind of price discriminate, give good prices to uninformed traders because they know they're not getting screwed in your example of the antiques assessor, right?

And then give much worse prices to the toxic flow.

I think what's going to happen, though, is that the citadels of our world are going to be able to pay more than the LPs for the non-toxic flow.

And then what's going to be left over is only the toxic waste and that's what's going to be for the LPs.

So even though I think private order flow has the potential to alleviate this, it might actually make make it worse depending on who kind of wins the auctions for the order flow.

I think Uniswap X also launched recently, which is also very similar to doing some compute off-chain.

Jin, you want to take some of this question as well?

So I guess I can first talk about the lever question a little bit, and then we can talk about Uniswap X.

So I think with the thinking around returning lever to LPs, in general, Uniswap has always taken a market-oriented approach to solving pretty much anything.

And specifically for designing pools with custom functions and encourage people to innovate is this thing called hooks, which basically allows each pool to have its own custom logic and more or less arbitrary logic that can be executed while the pool is executing certain functions.

And people have been discussing on Twitter, oh, what we can do with these hooks and those ideas that my panelists just mentioned have all kind of come up before, which I think are like promising ways to solve this problem.

Like you can auction out the ability to, for example, trade in a pool in a block.

And that essentially just passes on some of the kind of willingness to arbitrage from the arbitrager to the LPs.

It's unclear to me at this point whether the details are going to actually work out.

But I think these are promising research directions.

And I think Uniswap X is just tackling a separate problem, which is the routing problem.

When swappers are swapping, especially with Inswap v4, you will see a ton of pools.

And it will be almost impossible for you to find the best routes yourself without having this beefy machine and complex math and all that.

So Inswap X is really a product with the swappers in mind.

And say, again, outsource routing and execution to the market and say, Whoever gives me best execution can have the trade.

So we're kind of saying these are 2 orthogonal problems that can be solved better if they're decoupled, instead of putting them together.

So it looks like a lot of these newer versions of AMM designs are trying to split into 2 types of users, toxic and non-toxic type of users.

Does anyone in the audience have any opinions on this direction that AMM Design is taking?

They pass it on to other exchanges, to like for instance the one-inch RFQ system.

If there's something that one-inch market makers don't want to fill, it might go on to Uniswap or Dodo or some other entity.

But Uniswap is unique in that when there's some flow that the market makers don't want, it would then transition on towards the Uniswap liquidity providers.

And it just seems to me like you guys are the only ones that are sort of like at a conflict with yourselves, so to speak.

And I was just curious, like, I don't know, it just seems like a really socially complicated, like, you know, governance, like, I'm curious, like, how you guys are handling that from a variety of different angles.

I think it's a bit of a nuanced topic, I would say.

I would agree with you on the fact that it is a hard trade-off to balance, because We do have 2 type of users.

But sometimes, they are kind of in this position where you have to make a decision on where you want to be on the spectrum.

But It's not true that all the flow has to be, all the residual flow has to be routed to Uniswap.

For example, if I'm operating a Stata shop, I'm very well positioned to route everything to any exchange.

In fact, it might be cheaper to just hedge your position in Binance as stock side.

I can win the order first, and if I don't want to internalize, I can just route it to Binance.

So it really depends on the situation, I would say, and not exactly clear.

I fully support the Uniswap front end routing to Ambient, personally.

I think there's a distinction to make here between not necessarily routing all of the bad flow to your LPs versus making sure that they never get any of

because all of the good flow would get filled by the RFQ system.

Yeah, under this sort of RFQ system, you turn the passive LPs essentially into this liquidity of last resort where the RFQ liquidity providers who are probably going to be these professional firms, they're going to take a look at the flow that is coming in, decide if they want to fill it at a better price than the passive LPs will fill it at.

So on balance, you know, ignoring anything that v4 might change, it's probably going to reduce the amount of non-toxic flow that hits the the passive LPs.

There's definitely a trade-off between the swapper welfare and the LP welfare although it's definitely not this like purely zero-sum game because right now there's like a ton of loss external to the system.

So I do support the general framing that you know you kind of want to reduce this loss to arbitrage.

I think it's super super clear right now that AMM LPs pay like way too much because you see some of the folks in this room here bidding like 10 ether to be able to build a block where they get to make the first trades and that is getting just like lost from the AMM system entirely just going to the Ethereum holders.

Who in this room would consider themselves to be toxic flow?

Would any of you want to take, like, be makers on only the orders that Citadel and Jump and Jane Street and everybody who just raised their hand didn't want?

So we've worked recently with Balancer at CalSwap on an integration.

It's still experimental, so it touches on certain pools, not all of them.

But essentially, the idea is that our solvers, that are the 1 that executes the batch, so the orders in the batch, can access those pools at 80% discount on the fees.

So this is to say that the logic there is rather that even some of those orders may be matched by private liquidity, some of those because we have a batch some of these orders may be matched peer to peer, so within the batch with coincidence of wants.

And what is left, however, is still considered non-toxic, so much so that Balancer is willing to give a very large discount, or it's experimenting with this idea of giving large discount to the flow coming from Cowswap.

So yeah, so in this sense, I would tend to agree with the idea that what is left doesn't need to be toxic necessarily, but definitely we, I'm going to say, there is no, we act in the interest of our traders, period.

We do not have the LP side, so this is a kind of an Initiative that we started that seems beneficial to both of us, but ultimately we are working in the interest of the traders.

I Think like a lot of LP leaking value is due to price discovery happening on off-chain venues like by Nance or other faster exchanges What is this panel's opinion on how can price discovery, especially for the most liquid assets, can happen on chain?

Maybe Eugene you can you can take this question.

Yeah so at least at Ellipsis our opinion is there's all of these really really sophisticated liquidity providers that exist in traditional finance.

They've been playing on these electronic markets.

They've invested tens of billions of dollars in R&D and infrastructure into getting really, really good at providing liquidity.

And sort of the result of this competition, and there are a bunch of zero-sum or negative-sum aspects of it, but the end result really is that on these centralized exchanges there are really, really competitive prices, really, really deep liquidity, because the way you compete as a liquidity provider is by providing better prices or by providing at the same price with more size.

So in contrast with the AMM approach, which is sort of taking this very prescriptive view on what liquidity ought to look like, with a limit order book, you sort of let these liquidity providers put the liquidity wherever they want to put it in a way where you're, again, allowing them to compete on price.

And the results of this game, which might be zero-sum or negative-sum between them, is better prices for the non-toxic flow for the swappers.

And in this world, yeah, there's probably very little role for the passive AMM LP, and I would argue that that's okay.

I don't think there's any reason that passive capital necessarily should be able to make a return or be providing better prices than a Citadel or a jump trading, who again have spent so much resources into getting really, really good at this particular task.

And if we're able to put all of this on chain, then maybe you're still never gonna be quite as efficient as Binance, but you can approximate something that's more likely to facilitate some of the price discovery happening on chain.

In a tradified way, market makers compete.

The problem is they don't necessarily compete on price.

A lot of times, to be honest, they compete on who gets first position in the queue.

If you ask any HFT market maker, would you rather have the best alphas in the world or would you rather always get first queue position on new level formation?

9 out of 10 of them are gonna tell you we just want the best queue position.

So the reality is like Traditional markets and order books or even professional liquidity always turns into this PVP game.

So if Citadel and Jane Street are competing for order flow, even if that order flow is non-toxic, it becomes toxic because they're both bidding against it, and Therefore, they don't bid as aggressively as they could, because if Citadel outbids Jane Street, they're implicitly eating Jane Street's alphas, taking the other side of that.

Anytime you introduce a PVP dynamic, you introduce this adverse selection problem, which inherently raises the cost of liquidity.

1 thing that I think is really interesting about AMMs is you basically take away the PVP game.

And 1 reason I think AMMs actually can beat professional liquidity and can provide better prices is without the PVP games, You can provide really, really aggressive pricing than you can when you don't have these professional liquidity providers competing against each other and holding back because they don't want to eat each other's alphas.

In fact, there was a paper doing exact this analysis.

So they take, I think, central limit order book data from Canadian Stock Exchange or somewhere.

And they sort of compute the counterfactual liquidity, assuming the priority is Parata instead of price-time priority.

And then you do see that better risk sharing among LPs would marginally improve liquidity given some, you know, assumption on risk preference.

The other question I would have for Eugene is, as far as I know, the reason that we have this thesis that AMM liquidity providers are kind of picked against is because, you know, long block times and latency compared to centralized exchanges, which, you know, it would be reduced somewhat if block times are shorter.

But fundamentally, you know, like even Solana block times are like 100 years compared to Binance, right?

So does building central limit order book on a blockchain really solve it for professional market makers if that's a dynamic that's always going to play out?

I think the actual answer is we will just have to see in practice as the participants in the market become more and more sophisticated.

Long block times, definitely still a problem for Solana, although not to the same degree as they are on Ethereum.

I think the main difference here between the active liquidity and the passive liquidity is the passive liquidity is effectively paying the arbitrageur some amount of money to move the price to the quote-unquote correct price, which maybe we will just agree for now is whatever the price is on Binance.

The liquidity is following some curve, so there's some amount of some size trade that needs to happen, and then there's some loss inherent in there.

With a active liquidity provider, kind of the hope is they can move their prices before someone is going to toxically pick them off, essentially like moving the prices for them.

And there's a bunch of nuance here as well.

For example, like you're playing these top-of-block games, but there's I think a lot of room for innovation here in the same way that there's a lot of room for experimentation with like the microstructure of the AMM.

I think 1 of the most exciting things about working in the DeFi space is we have this opportunity to really play around and experiment with different microstructures and have conversations like this 1.

Whereas if you're the NYSE and you want to implement even a new order type, this is a process that's going to take 3 years.

it out tomorrow and see what happens in a live market.

I'll just add, actually, we actually just completed some research on this, Block Times, and we're actually working with Blocksroute on this about an instant inclusion type hook so we'd see, OK, if you can actually guarantee sequencing when things arrive, how much less is the toxic order flow?

And it's pretty substantial when you go from 12 seconds down to, I think we found, like, diminishing returns are, you pretty much hit diminishing returns at like 100 milliseconds, 50 milliseconds, so I think that's like kind of the sweet spot, at least based on like the ETHUSDC pools and the order flow.

So I want to ask actually Doug on what you just, I was surprised that what you said about that AMMs could, by not having toxic order flow, could actually potentially give gradually better, but does MEV, as the guy who's against MEV auctions, the existence of MEV auctions, does that make it a PVP setup, in which it puts it back to users to getting the worst price, or not?

Yeah, I mean, that's definitely the case if you do have an auction, right?

You can play around with it, though, and you can say, okay, maybe instead of an auction every block, you auction off the right for the next whatever, 12 hours or something like that.

That's why retail order flow, you don't compete on every single order flow coming through, right?

Citadel comes up with a contract because they don't want to compete on every single order that comes in from Robin Hood with Jane Street or whatever, right?

They bid, but they bid at like at a larger granularity where it's less of an issue.

So like the cool thing again, going back, coprocessors are really cool because now everything's on chain.

We can make assertions about, okay, are you — you know, you've held this position.

Have you met, you know, certain criteria?

Again, it's on chain, so we can all agree, like, what the results are.

So, I think we'll have to experiment in terms of how do we do those MEV auctions, but yeah, I do believe kind of auctioning off larger granularities reduces the adverse selection problem.

And follow up on that is that you all kind of say, well, you know, really price discovery happening on Binance, which is, we all know what Binance is doing right now or how it's doing.

But from my perspective, if DeFi keeps, you know, being tagged behind CeFi, it's always going to be like a sidecar it's not going to be the real thing It's just like where the arbitrage happen and a bit other do you guys think price discovery?

Can happen on defy and is it important or not?

Like I think it's important, but you might disagree

The big thing that's really missing right now, to be honest, is PERPs.

You just have so much more liquidity and volume on PERPs on centralized exchanges, and that's where price discovery happens.

See all Frontier Research transcripts on Youtube

MEVday Paris: Designing a DEX in 2023