Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Ava Labs x CBER Ep 8: Decentralized Exchange (DEX) Aggregators and Solvers
41 Plays1 month ago
Mallesh Pai (Rice University) covers DEX Aggregators and Solvers. In more detail, DEXs are not isolated entities. Rather, a variety of "intents markets" have arisen (e.g., Uniswap X, CoW Swap) where users can express "intents" to trade and then these markets execute the intents by relying on DEX liquidity and also potentially other sources of liquidity (e.g., fillers). Intents markets are generally classified as aggregators or solvers, and this podcast explains the space of intent markets, clarifying the distinction between aggregators and solvers. The podcast particularly focuses on economic implications of intents markets. An important point is that assertions that these markets are likely to entail favorable outcomes for traders are not necessarily correct due to the underlying economic structure.
Recommended
Transcript
Introduction to the Podcast Series
00:00:06
Speaker
Hello and welcome to Crafting the Crypto Economy. I am Silvia Sanchez, Project Manager at OWL Explains by Avallabs, and today we bring you a transformative podcast series in partnership with the Crypto and Blockchain Economic Research Forum. This series features leading faculty from renowned global universities exploring various elements in the blockchain ecosystem. These episodes are a bit longer than our usual hootenannies, since we will be getting very deep.
00:00:33
Speaker
and also each episode will have its accompanying paper posted on our website for further reading. And with that, I will hand it over to our moderators Fahad Saleh and Andreas Park.
Introduction of Malesh Pai and Intent-Based Models
00:00:44
Speaker
All right, welcome everybody to another edition of the CBAR Owl Explains podcast on crafting the crypto economy.
00:00:52
Speaker
Today, we're very happy to have Malesh Pai from ah Rice University. He's an associate professor there and who is also affiliated with Consensus via a small unit called SMG. and We're very happy to have Malesh here and we're going to talk today about intent-based models.
00:01:12
Speaker
Now this sounds very abstract um but just as some background this is another edition of our podcast where we talk about topic related broadly to decentralized trading. It seems a little bit like this seems to be ah a very strong theme in our and a general podcast series but you know just for the audience obviously if you have crypto assets if we think about DeFi and the like um It's quite important that we have a very thorough understanding of how the exchange of asset works, and so this is this is what we're talking about here.
Evolution of DeFi vs Traditional Markets
00:01:44
Speaker
um Just for the audience's background, um again, in in the real or in the traditional markets, there are so many different ways how people exchange assets and trade assets. And it's kind of interesting to see how the DeFi space evolves, how it develops its own institutions and how it borrows ideas from ah existing institutions.
00:02:06
Speaker
And so in this space, I would like to you know welcome Malesh again um to have me here in his podcast. i'm I'm the host Andreas Park. ah The co-host is Fahad Saleh. And let's dive right into it.
Functioning of Intent-Based Models in DeFi
00:02:19
Speaker
So Malesh, broadly speaking, why are we talking about intent? What is the background here? Okay. So here's So when we started with DeFi, like, you know, Uniswap, V2 or V3, what you had, I mean, you had you enabled decentralized trading in the sense that you could have these passive liquidity providers, they dump a bunch of liquidity in a pool. And if you decided you liked the price, so you wanted to trade in one direction or the other, you could send a transaction.
00:02:54
Speaker
and To a first you know so first approximation, what you could do is just send a limit order. right So you could say, I'm willing to accept. I want to trade my ETH for USDC. And this is the this is the pool I want to trade with. And this is um how much i'm willing to this is how much slippage I'm willing to tolerate. like This is how much I want it to move away from the current price. The most it can move away.
00:03:22
Speaker
And that was great. And then there was this proliferation of DEXs. So the next thing that happened, as I understand it, was there were these aggregators. so And the aggregators, you didn't specify specifically what Like, you didn't say, oh, I want to trade with the Uniswap, USDC, ETH, five-bit spool. You just said, hey, you guys go figure it. I want to you know swap my ETH for USDC. You figure out the route for me. And it would go search and you know fire off the right sequence of transactions.
00:03:57
Speaker
So that's really that's really very similar to traditional markets, right? So if you're a brokerage and a brokerage, you go to your broker and say, hey, I want to trade, you know, whatever, 10,000 shares of Apple. I'll have a lot of money for that, right? And as we know, in traditional markets, there are also various different ways how we can trade. There is, say, in the US, there is not just the New York Stock Exchange where you can trade Apple, but you also have NASDAQ.
00:04:21
Speaker
and you have various different other trading platforms. And so your broker is there to really you figure out the best price, right, for you. But you go through it intermediate. So you're basically saying in the world of DEXs, we now have Uniswap, but we also have, actually in Uniswap, we have various different Uniswap. So we have version V2 and V3, both of them actually may offer trades.
00:04:44
Speaker
We have sushi swap. And so you're basically saying that there are now services that allow you to trade on multiple platforms. is that was that what yeah Do you have an example? what but Roughly speaking, yes. So that there are aggregators that don't necessarily have any liquidity themselves. Their job is to route order flow. um So an example of this would be, I believe, something like one inch.
00:05:09
Speaker
um the The next evolution, so where did intents come in?
Design and Economics of Aggregators
00:05:14
Speaker
Intents came in with the following idea, which was... Wait, Malash, before you go there, yeah just just to clarify, so is it fair for me to think about an aggregator as essentially a mechanical optimization across sources where I could otherwise sort of trade separately? Yeah, um I think that's a fair approximation. There are different different aggregators have so part of what i'm but Part of what I'm getting at is I'm trying to understand whether there is like a design space across aggregators before we go even deeper. um like they so they're they're certainly and Let me clarify because there's certainly a design space on the engineering side in the sense that like it's non-trivial to even know that an arbitrary smart contract corresponds to a liquidity pool. right
00:05:59
Speaker
so Even if it was totally mechanical, it's not like, oh, it's really trivial. I just have to like go look at all the the exchange exchanges, all the hoodie pools, and then I write a little optimization algorithm. There is sort of an engineering challenge, but if you put the engineering challenge aside, I'm trying to think more from the economic side.
00:06:17
Speaker
What is the design space for writing these aggregators? Supposing I know all the liquidity pools and I don't have to worry about kind of how to parse their smart contract logic. yeah so that's well and and that' That's an excellent question. so One of the things So already for an aggregator, there are things like but there is differentiation in like, because they have to do this fast, how many pools they're integrated with, how well they can search across these pools. Do they look to send a single order to the best pool? Do they take your order, split it up into multiple things? There's all sorts of questions around that. Do they have some of their own liquidity?
00:06:55
Speaker
Okay, but but some of those things do sound more like, I mean, so I can certainly see why even based on that, there would be differentiation, there might be some people who are much better than others, but it sounds like the differentiation there is coming along along the dimensions of things like skill, when a little bit like engineering skill even.
00:07:14
Speaker
um but So if you if you start with a mental model where everybody knows all the liquidity pools, everybody can parse them with the same sort of ability. Is there then any sort of design space left?
00:07:26
Speaker
or Or again, is it like the sort of like the simplest frictionless mental model of an aggregator? Then it's like you just take all the liquidity pools in existence and you do an optimization about given that I'm allowed to trade through any combination I want to fulfill this order, I just find the optimal solution. The way I think about it is you get this you have this sort of um smooth. So what I'm calling an aggregator is not necessarily like a single point in the space. What you have is you have the smooth.
00:07:54
Speaker
um You have this horizontal. horizon like At one end of the one end of the spectrum is like a specific V2 pool. And that's a very specific way you can ah interact with that pool. It's a specific smart contract with specific liquidity. And aggregators start getting away from that. And at the other end is really ah the the holy grail, as it were. What these intent projects are trying to do is to say, look,
00:08:23
Speaker
The user, just like in just like what Andreas was saying about the US stock markets, you never tell your brokers, or most people don't tell their broker something like, I want to buy 10,000 Apple at the New York Stock Exchange. They just say, I want to buy 10,000 Apple. Find me the best price. Or I want to sell 100 Apple shares a day for the next week. Go ahead and make this happen for me. you have some intent So the idea,
00:08:47
Speaker
like an aggregator is somewhere in the middle
Mechanics of Intent Protocols
00:08:50
Speaker
between the evolution between a single pool that you can interact with, and an abstracted experience where all you really care about is doing, you have some preferences, you state those preferences, and somebody goes ahead and ah figures out the best thing for you to do. So an aggregator is somewhere in the middle where it says, okay, you give me a specific order, I'll do some optimizations across some set of DEXs that I'm integrated with.
00:09:17
Speaker
An intent is just me saying something like, okay, I have 10,000 ETH maybe on this chain, maybe on some other chain. Again, I wish I had 10,000 ETH, but I don't. um and um ah move to move to you know I want a corresponding amount of USDC on some other chain. That could be an intent. Don't tell me the details, just go figure this out and make this happen for me.
00:09:46
Speaker
that's At a high level, that's an intent. so so Can I just interrupt for one second? What I find actually quite fascinating about this is we know we're all very deeply ah interested in the idea of decentralization and and actually having a user who has a lot of power over their assets and what they do.
00:10:05
Speaker
And in some sense, actually what we're observing here with the aggregator protocols and the like, and actually lots of other things that happened there, is that we now see the emergence of a service space around this. right Because fundamentally, an aggregator is a service, so therefore something rather centralized that figures stuff out for you. right um so Now, when we talk about intent-based trading, is is that also based on some form of service level? Or is this actually it does this sit somewhere else? Does it sit in a decentralized sense?
00:10:36
Speaker
but what Wait, Moles, before you answer that, so I think in some sense you were writing like a history context here and and we interrupted you and I think you've sort of gotten into the part that but you were going to be getting into after Agreeators. But before you do that, then could you maybe like finish the arc you were getting at and then and then address Andreas's question? Sure. So, I mean, I was almost done with that arc or maybe even done, which is that, so the next generation of these things is supposed to be these intent protocols where the intent protocol just says instead of signing a transaction, instead of signing a specific thing that will fire off a um trade with Uniswap or signing a specific trade with a specific thing with one inch who will optimize it for you. Instead, you just sign ah intent which says, this is what I'd like to do. Now, somebody go figure it out. The intent contains a bunch of covenants. Any successful fill of the intent cannot break any of these covenants.
00:11:35
Speaker
And there are a bunch of projects out there. They differ on various things. Some of them, you know they have slightly different languages in which they state their intents. There are some people who want to do intents beyond the financial space, but um where you could ask for other general ah properties. And let's let's put those aside for today. ah But the at one high level, it's just an evolution where the end user, instead of seeing instead of specifying actions, specifies what they want to achieve.
00:12:03
Speaker
and something else, a smart contract, a DeFi or or some um infrastructure along with a smart contract then is ah in in charge of figuring it out, figuring out how to fill that intent. And the cryptographic primitives mean that the one thing you can be guaranteed is that the covenants will not be broken.
00:12:26
Speaker
But what you can't guarantee or what the intent protocols promise on top of that is that we won't just find you a feasible solution to your intent. We will somehow find you an optimal solution to your intent. And that goes to Andreas's question of like, okay, what infrastructure is there on top of this? Well, we have to design some kind of microstructure to ah go ahead and fill these intents. These are often too complicated to do directly on chain programmatically.
00:12:53
Speaker
or at least the current ah scenario. So currently what we have is you post an intent and then there are these specialized entities that are that go by various names. but In the intent world, they call them solvers. um In trad-fi world, you would just call them you know liquidity providers, I guess, or um market makers. But
Impact on Decentralization
00:13:19
Speaker
effectively, they're solvers. And these solvers are in the business of turning around and so competing to fill your intent. so these are So it's a centralized service, essentially, right? So the this is so currently, I guess, it is
00:13:37
Speaker
centralized or at least has an off-chain component with plans to move that component on-chain. So, for example, UniswapX, which we talked about arm or but before the podcast, like UniswapX, the way UniswapX works currently is you sign it um you sign an intent. The intent is something like, I want to trade A for B.
00:14:01
Speaker
They won't necessarily fill it against liquidity that's available on-chain. They'll run a, I believe, Dutch option that starts from a really good price for you and works its way down and let um off-chain people fill your order. And at some point, if it doesn't work out, the reserve price is set by what they can see on-chain or what they can see on Uniswap.
00:14:27
Speaker
that's um bottom um yeah just with just a clari So was mentioning about centralization. And so I guess UniswapX here is one that opens a can of worms because, okay, so my understanding of it is there's sort of two pieces that people sometimes conflate. There's a request there's an ah RFQ at the outset, which actually helps them decide, are they going to go directly to the liquidity pools and just sort of the old process? Or are they actually going to have the Dutch auction that you were just describing? But the Dutch auction,
00:14:58
Speaker
where I think the terminology Uniswap themselves uses fillers. um Anybody can be a filler, right? um Now, incidentally, you'd have to know the order is there in order to fill it and the order is posted on Uniswap's website. So it's ah you know, in some sense, I guess you can debate about whether that's centralized or not. But I guess in principle, it is a permissionless setting. So like I guess I'm still trying to capture in my head here, what is the right mental model for me to have of solver models. And I at least my impression of them, um let's say before this conversation, and maybe still,
00:15:32
Speaker
is that I don't think of them as centralized precisely because, yeah, I guess Uniswap X is sort of the simplest example to think of, and I wouldn't describe that as a centralized ah for first ah setting.
00:15:43
Speaker
I would not describe it as a centralized setting either. ah It is definitely not, um it's not fully on-chain the way Uniswap V2 or V3 are.
00:15:59
Speaker
but we ah Ultimately, the liquid the the trades, like even if it's a a filler filling you, they are filling you on chain. They don't have to. The auction ah is not happening. that That portion is not happening on chain. wait i mean In what sense?
00:16:18
Speaker
ah yourre so My understanding was essentially, the so you have a descending price auction right in the sense that at every block, there's a different price that is adjusting. And and then there's some preferential aspect here related to kind of who won the RFQ, but let's put that but a pin in that and just say you know block by block, um there's a price where I can fill and the price is is getting ah and is getting less and less favorable for the trader and more and more favorable for the market makers. the idea Maybe I'm confusing this with other intent protocols. But but my my point was just that
00:16:56
Speaker
So, anybody can actually step in and and and and then as long as as long as they they to satisfy the constraint, they can fill it, right? And only then then, of course, you get into stuff like top of block, right? Because like you want to fill it, I want to fill it. Well, in principle, we I mean, there's this thing again about the r the guy who won the RF2 having preference treatment, but if you set that to the side, let's say we're facing the same sort of constraint, then it just becomes you know who's going to pay more to get top of block?
00:17:23
Speaker
No, if that's fair. um No, you're right. um Maybe I'm... Could you describe the model that you were thinking of? So maybe maybe you're thinking of a different protocol, but still like just to help the the you know the viewers and and us even understand what is the sort of space of these things. So what's ah what's an alternative?
00:17:38
Speaker
ah The space of these things, an alternative version would be, for example, that the um auction the the the auction or whatever competitive process you have to fill the um to fill these intents. So you sign an intent. There's a bunch of solvers who compete to fill that intent. That competition portion happens off-chain.
00:18:05
Speaker
But you then settle on chain in the sense that whoever wins the auction gets goes ahead, takes that signed intent, dumps it on chain and settles it. And the whole thing processes only when all the covenants are met. So this gets into like a related aspect, here yeah a really good question, which is that, and it's related I think to what Andreas was asking earlier also about like the service aspect.
00:18:27
Speaker
awesome So i I mean, the natural thing that I could think about traders caring about is price, right? um And when and it is something when it's only price that matters, it actually seems kind of odd that you would do things off chain to begin with, because you know, why do you need exactly like in UniswapX's case?
Computational Challenges in DeFi Protocols
00:18:44
Speaker
And to your point, like in UniswapX, I think you described it as, you know, there's not like a lot of contingent things that you throw in there. You just kind of say, I want to trade, you know, ETH for USDC. I don't want to price worse than this. And that's kind of it, right? So in that setting,
00:19:04
Speaker
it you know I can see why they would want to do all this stuff on chain. But I guess there's a whole there's a whole like space there potentially where it might be a little more complicated and then you might have some value of doing things often. So could you maybe talk a little bit about the sorts of things that a user might actually afford in the order other than just like best price execution and How that relates to kind of how the solver would be set up like maybe that means they have maybe it means practically speaking. They probably have to do stuff off chain. So some of these languages like some of these intent languages are super general. So I can state something like here is a list of tokens that I have. Here is a list of tokens that I want.
00:19:46
Speaker
And here is a utility function I have over, I mean, roughly speaking, here's a utility function I have over a budget set. find Find the best thing so you can specify a fully combinatorial order.
00:19:59
Speaker
um And some I don't know, you know, I don't know that people. ah You have a protocol in mind, I'm not actually very curious, like, what protocol are you describing? Or is this more of a um I think we mentioned some of them in the paper, and they are ah they are indeed ah working towards very general languages. So you can specify ah not just limit orders, but multiple token orders, ah if this, then that, like either trade it into this, or if you don't get a price better than this, trade it into, take my ETH and put it into USDC, but if you don't get this good a price, put it into DAI, something like that. ah
00:20:42
Speaker
there's There's a bunch of others. Does that create constraints, or I was going to say constraints, does that create a sort of cost structure where you would want to start doing things off-chain? Meaning like, I got to imagine the more computational work you got to do to figure out what you want to do, you don't probably want to do that on-chain. Exactly. um So some of it is just you can't install um you can't solve the assignment problem on ah on-chain, even a single like at least for a single pair trading, something like a Dutch auction is feasible on chain. ah As we showed in an earlier paper, an ascending like a single one round sealed bid auction is not actually possible on chain for large orders, um at at least with current chain architectures.
00:21:30
Speaker
um So it's partly that, partly it's like other stuff, right? Like I might want, I as a solver might want to take Yu Fahad and Yu Andreas' intents together. I might have preferences that I can fill both of them together because I can cross them. They're crossing their, you know, they're trading in opposite directions, crossing the spread. So there's also a combinatorial problem on the filler side.
00:21:53
Speaker
that might need to be solved, not just ah optimization of a single person's intent. So all of this is somehow, um all of this is moving portions of the architecture either off-chain or moving them to specialized side chains, things like that, depends on the protocol.
Target Audience for DeFi Protocols
00:22:12
Speaker
can i interrupt for one moment and maybe take ah a step back um and so take the perspective maybe more of a really boring old-fashioned finance guy and so the first thing and that comes to my mind as I listen to this discussion is who is this actually for right so um now if I I mean you know I'm I'm poor I don't have it much I don't have much crypto if I wanted to trade my 0.05 ETH obviously it's fine if I go just to Uniswap right so that's fine um And then at the same time, they they're they're really large guys that have very complicated trading needs as you describe them now, and and they need a specialized service. ah now i'm kind of trying to figure and Now, for one inch, for instance, one inch, again, is actually quite a simple web mask. It looks exactly like Uniswap. as you As you say, here's how much I want to trade, what can you give me? And and you get kind of ah you can get a price even if it is something that is reasonably large that you want to trade.
00:23:09
Speaker
so um I'm trying to figure out what the what you describe. is What market is this actually for? Is the intent here? Sorry, no pun intended. um Is the intent that this is something which is a like a generalist service which will be available to anybody and so that even I as a you know poor little 0.05 each guy can can make but get a better outcome of my trading or is this something which is more very special like trading needs? It's meant for everyone. Now your next question might be well who has these kind of complex combinatorial preferences and I don't know the answer to that but some of that is that
00:23:49
Speaker
There's been this theme for the last year or two in crypto that we're building like broader infrastructures to move global finance on chain. And this is, I guess, one direction where ah if you build it, you don't have the ability. there there are a few Actually, there are a few ah markets, there are a few startups even on TradFi. Actually, one is based in Toronto, Andreas, that tries to do combinatorial market clearing.
00:24:17
Speaker
But if ah but if i if I'm just trying to think of, so, so but again, you're you're describing a protocol, right? So um a protocol is always something very complicated, complicated right? um I mean, you you could also just simply say, you know, I'm a really clever broker, right? So so sort of let me take a step back. um In the traditional world of finance, we have, um you know, routing technology providers, right? So we call them vendors.
00:24:46
Speaker
they provide you know for existing markets the technology so that a bank for instance can easily access multiple markets given a certain set of constraints right so like for instance if i'm a large yeah large fund and i want to trade a particular large amount of a particular stock I use some form of algorithm. um Now these algorithms are developed usually, I mean it could be um developed by my broker, but oftentimes they're actually developed directly by a tech vendor because they have the technology and then actually my broker goes and uses just that technology as it's there.
00:25:24
Speaker
So I'm i'm wondering if if this is, you know, what you know if this is what what you're describing is really something that is a tool that a tech vendor would use or because, I mean, would you describe something really general? but what Why is this an advantage over just having a specialized firm that that provides a service for me as ah as somebody who needs to do the trades? I think the question comes down to where, like,
00:25:51
Speaker
The specialized firms in TradFi are, so firstly, they, okay, so a couple of things that I can think of. One is they're specialized firms that at the end of the day, various regulations mean that, yeah you know, there are multiple markets with liquidity, but that's like a finite number of markets in the crypto, thats sorry, in the TradFi world. So if you are a bank in the US and you have some complicated need, you still have 10 menus plus OTC to execute your trade.
00:26:25
Speaker
um I guess part of it is that those tools are just not available right now. ah so this is you can have if If you have some large complicated trade that you want to do in the crypto world, you can um optimize it yourself and fire off order by order by order to execute it the way you want. um This is just ah this is just A technology, they my understanding is these intent protocols want to run them as ongoing services, including having ah including as a two-sided marketplace where the solvers are also market makers, liquidity providers. They will come in with their inventory. They will fill some portion of that, and only some portion of the intents will go and hit you know a Uniswap v2 or something like that.
00:27:15
Speaker
um So they're running it like that, but you they could, I'm sure there are protocols that also are just providing, that want to provide an API and ah like a bunch of code that someone could use. So now if I come back to the stratify world then, um and then also thinking about, again, for who this is for. So one complication that people face often, and that's that's really a big concern in markets, is that you don't really want to tell anybody what you're doing.
00:27:46
Speaker
Right? Because you expose yourself to a lot of risk, right? So, the moment you basically say, hey, i you know, I want to trade, you know, a large quantity, um you you have to, even even though it is illegal to do, um ah but you always worry that your broker will front-run you, right? Because you never know if, you know, their desk is not maybe sitting right next to somebody who can look over their shoulder or listens to a phone call, right?
00:28:09
Speaker
um and and could ah could trade a ahead of you. Now, obviously, there's mechanisms, compliance mechanisms in the traditional world to prevent this. But from what I hear here, a lot of it sounds about like you're auctioning off orders and the like, and in doing this, you know you can put all kinds of constraints in, but by the time you've put them all in and you said what your constraints are, you know you already may have lost out on ah on an opportunity.
00:28:34
Speaker
so this is Again, this is a design parameter. There are two parts to this. One, as you know well know, in the TradFi world, there's also like you know there are things, there are these papers like Sunshine Trading, which say, okay, if you have a really, really large order, but you're not, you're doing it for some programmatic reason, you actually want to announce it loud to the market so the market doesn't think, why is someone dumping? So there's that. um Here, I think ah there are intent protocols that are thinking about things like privacy.
00:29:03
Speaker
okay so can I take if I have enough order flow I can take it cross some of it with itself there's this so you don't necessarily have to go order by order and try and you know, announce, okay, Fahad wants to trade this. That's, I think that is the problem, or that's the problem a lot of people have with the basic ah RFQ system, which is an RFQ system is literally announcing Fahad wants to do this. Does anyone want to announce the price? And then of course, who's going to come in? it's ah It's the person who wants to pick off Fahad the most. Sorry, Fahad. I'm not saying that's what happens, but that is of course a concern. So there are separately, there are
00:29:45
Speaker
better I guess I'd say the larger space is trying to figure out what is the correct set of market designs and market microstructures that will work. We are, for better or worse, for now, unconstrained by regulation.
00:30:00
Speaker
and um That means you know at some point, the regulators are going to come in, and they already are, and ah clean up bad actors. But I think that also gives some freedom to experiment and you know just see if we were to reinvent this from scratch with all the technology we have now, would some other market microstructure other than the ones we have in TradFi work better?
TradFi vs DeFi Market Regulations
00:30:20
Speaker
And that's an interesting side question.
00:30:22
Speaker
well but i mean so Just for the record, just ah since you mentioned this, right so even when it comes to the trading of orders or handling of customer orders, there's very different philosophies among different regulators in the world. right so In the US, there's a very firm system, very firm set of rules of where orders have to go. In Europe, it's ah it's a very different setup. right so ah I mean, I don't want to go, bor I'm going to bore the audience for a moment, right? So in the US, you basically, there is a concept of protected markets. So you always have to go to the market which has the best current price, right? Whereas in Europe, that rule doesn't exist. It is at best implicit to what's called the best execution practice, right? Where you should get the best price for your customer, which kind of would mean you should get to the best price venue.
00:31:10
Speaker
But the the way the its implement is very loose, right so you can basically say, well, I'm going to send everything to the London and Stock Exchange. you know That could be your announcement of your best execution practice. So what I'm trying to say with that is like the regulations of the US are very specific and they're historically driven and nothing in the world says that that that even is the best way to do it.
00:31:32
Speaker
So and in talking to traders, I should note that even the very specific regulations in the US have interesting loopholes. So for example, one of the things they'll do is they'll get their order flow by microwave feed, and they'll get the NPBO sort of their store is coming through a slow line. So they have like a 200 millisecond or whatever x millisecond time differential.
00:31:56
Speaker
um where they can say, look, we filled at better than the nbbo that we saw, but they just have a slow feed on it. So they actually have 200 milliseconds of play. And we know that those things make a huge difference in tragedy. So even there, there's, there's you know, there's stuff that maybe ah hasn't been caught on, or or just accepted. It's arbitrage, eh?
00:32:21
Speaker
Exactly. Physics arbitrage. I mean, we're at the point where we're doing physics arbitrage. Anyway, maybe and we're we're I think we're getting quite off track with our far region. Yeah, sorry. We are going off track. I was going to say, I think this is a good point to to what to to get into, but maybe with a little bit more of the crypto concrete context. What I mean by that is that ah So for instance i think it's at some point andreas asked about you know like who are these markets for and your answer was something to the effect of there for everybody and if if you think about that from like a business perspective it's like hey i'm gonna i have a i have a new business and i want everybody to buy my stuff like that that
00:32:55
Speaker
That is a good posture to begin with until you think about things like, well, maybe marketing goods to this segment means that it'll be sort of costly to market to another segment because producing both of the goods that these groups like or somehow, you know, it's it's very, good I don't have an economy to scale across them, etc. So the real question then becomes like,
00:33:14
Speaker
what exactly does the market want? So you also sort of alluded to, you know, intents can be very general. You talked about like, I'll give you a utility function. And i'll so essentially, I could like just specify like, almost an econ paper style optimization problem. um but But, but there's a real question about like, is this actually what people are doing, right? So like, I think I think what might be useful here is if you could talk about kind of like, what is the sort of cross section of intense protocols, at least amongst the bigger ones. And like what are the differentiations in terms of what they allow their customers to do? And then maybe even to the extent there are differentiations, does that lead to sort of specialization where like this particular solver model goes after this part of the market and this one goes after a different one? Or is it really like for all the intents that most people are you know asking for? um a
00:34:03
Speaker
one one and One protocol could basically take care of it if it's you know run efficiently enough and there's not really any sort of sense of specialization that matters here. It's a great question. um So the two big the two big things that I would and And this also gets to this sort of conceptual question of what is an intent protocol. So the two big intent protocol, because that's that's you know there's a fine line between pure aggregator and then eventually ending up at an intent protocol. The two that I would call intent protocols are UniswapX, which ah is big. And the other, which I think of as kind of an intent protocol, but also doing sort of limit orders, is Cowswap. So Cowswap is um
00:34:50
Speaker
Why am I calling it an intent market? ah It's you state a swap and there are these solvers. I think they call them solvers. um And these solvers get a certain amount of time. They can choose. like but but Before before for you go on, I think so. I think we we should describe COW swap. But actually, if you could like even take a bit of a high level point, like what does COW stand for? Like what are they going for here? And part of their idea was, look,
00:35:18
Speaker
if you Fahad want to buy ETH for USDC and you Andreas want to sell, there is no and and there is there there are gains from trade, then we don't necessarily have to have both of you hit the Uniswap pool. We can cross your orders against each other. So their idea was based around that, they would do these periodic batch auctions where they'd have a bunch of flow the solvers would try and cross up as much of the flow with itself as possible and then either fill the remainder. So if there was, if Faha did like some excess demand,
00:35:53
Speaker
um If aha had some excess demand ah for it, the solver has an option to either send it to an on-chain source, fill it from their own inventory. So that that was a coincidence of, that was cow swap. But they they started with coincidence of wants and then really leaned into the cow thing. So if you look at their website, they have cows and all that fun stuff.
00:36:17
Speaker
um The more general ah intent protocols, i'm not um the the more general stuff I described is coming.
Comparison of CowSwap and UniswapX
00:36:26
Speaker
I don't know if they have achieved, like I don't know how much of it is fully in production yet. I'm sure when I say this now, I'm going to get flamed on Twitter whenever this hits.
00:36:38
Speaker
ah the interwebs, but there are multiple, the the more general intent protocols. I'm not aware of how far they're into production. But actually, so I think it's it's useful then to think about COW swap versus UniswapX already as, let's say, a way to think about the extent to which i there can be differentiation. So because we talked earlier about like, again, you can have very, in theory at least in theory, you can have very sophisticated intents that you articulate. Essentially, anything that is sort of within ah the state of the blockchain that the blockchain understands, you could you could basically, I suppose, ask as an intent, I want
00:37:14
Speaker
my you know my order to satisfy these criteria yeah and that could be quite rich, right particularly because we can have oracles feeding and all sorts of stuff. um ah But at least when I think about cows swap, and I don't know that much about cows swap, so maybe maybe ah a maybe I haven't thought about it enough, but when I think about it, I compare it to Uniswap X,
00:37:33
Speaker
I don't, it doesn't seem to me that the reason people are using one versus the other is because they have a different intent, like the space of things. It's more maybe like, I feel like this batch option thing is going to give me a better execution than if I were to try UniswapX or something like that. And so there it feels like the heterogeneity is more coming from the solver side, like, okay, it's the same with the solver. Both are very limited languages. Both are very limited languages in the sense that currently they both just enable pair flow, pair tricks.
00:38:03
Speaker
but yeah also So there's no particular reason. Yeah. But so then the differentiation there is more like, OK, so we're trying to get best execution. But um we just have different approaches to get best execution. But although although I wonder, actually, is there a temporal dimension that differentiates these two in the sense? Yeah. So like could you talk about that a little bit? Because maybe that's a little bit in the background in terms of? Yeah. So cow my understanding of Cowswap is um so solvers submit Solvers submit um solutions to the existing flow. that get scored on something like um ah They get scored on ah a cow score, ah which is like, how well are they filling all these orders? And then the best one wins that order, ah wins that collection of orders, gets to fill it the way they said they would.
00:38:53
Speaker
and gets some get some for it, some of it in cow tokens, I believe. um you But I believe an order on cow swap can stay open for up to 30 minutes. So if you fahad state a particular order, um it can take it can just stay on that system for a while before it gets filled. it does They don't have like UniswapX, because of the Dutch auction component, has a point at which it just pulls the plug and goes on you know goes and just fills you on chain or just says this order is unfillable. right but but so the I think one of the points here though is that
00:39:35
Speaker
um There is price preference, but there's also a time preference, right? So for instance, if I say, you know, I just want to, I want to trade subject at being at least as good, but, but I also value, or I don't even care about the time. Like I don't mind if you fill me an hour from now versus me, like, no, I want it on the next block. This seems to introduce a ah ah design space for the, for the solver model, right? Like for instance, you describe batch auctions, right? like if you try to do a batch auction in the sense that like in every block you want to cross all the trades, there's not that much liquidity. Well, this isn't going to work out so well probably.
00:40:09
Speaker
um ah and so Actually, could you talk about that? Also, like so and if you think about something like doing a batch auction in this context, how do you think about how frequently you actually want to try to cross these trades and how does that relate to kind of the i guess the ah trader base that you're facing?
00:40:30
Speaker
Yeah, so I will say before I say that, let me say one thing which I think you've hit upon, which is really important, which is this um ah time priority. So there, as we know from TradFi, like there are people like me who are uninformed and just wake up one morning and say, Hey, I want to buy Apple today looks really good to me. And then there are people who ah are doing sensible portfolio management and have things like, okay, I want to do, I want to set up a slow recurring time weighted order.
00:41:02
Speaker
or a slow recurring price-weighted order. um And they do this to minimize the amount they hit, like they're actually taking from the order books to have low price impact. They don't have a time preference. They have some rebalance preference. um It's very hard. So and another reason for things like intents is, when i remember, when I sign a particular transaction,
00:41:26
Speaker
um it's very the transaction gets by and any current chain, the transaction gets red and it doesn't action and then the transaction concludes and it doesn't get touched again.
00:41:38
Speaker
So it's very hard, it can be done, but it's very tricky to takes to take a standard transaction and have it do something over the course of multiple blocks. but keep um you can you can if You can achieve directional average trading by saying, I'm going to be a liquidity provider and I just want to so that I can essentially be filled in the reverse direction to I'm providing liquidity.
00:42:02
Speaker
But it ah otherwise, it's hard for me to do things like that on chain right now. So one of the things that Intents allow us to do is to state things like time preferences. um Here's what I want the solver to do. I wanted to program, I wanted to fill as close to time-weighted average. I wanted to dump 10,000 ETH for me over the next week. That is something potentially that a solver language also allows that would be hard to implement in a current blockchain just directly with a single transaction. Okay, so one way, so tell me if I'm thinking about this correctly now. So you alluded to one differentiation across, you alluded to a couple differentiations.
00:42:43
Speaker
ah across potential traders. but I think the simplest one is the is one I was alluding to that a lot of people think of, which is, in some sense, informed versus uninformed traders. right Like uninformed traders, like, just fill me at this price, I don't know, within the next day or whatever, right? I don't know, up up until block number X. Next minute, really. I just want to see that little you know confetti drop on my Robinhood screen. Yeah, and but but in some sense, then like the market could move massively against you. And they're like, well, whatever. you know I got failed.
00:43:10
Speaker
And then there are informed traders who are very aware of how the market is moving. But even within the sense, so then later on, I mean, you then got into this other point, which I think might be kind of worth digging into. In the set of informed traders, people have different trading strategies. They might have different risk preferences. There are a whole bunch of heterogeneity there. So I guess one question is, to what extent do we see these heterogeneity is already manifesting ah like because otherwise I guess you you alluded to the fact that there's uniswap x and cow protocol that are kind of the big ones. But at least intuitively to me if there's a lot of traders doing a lot of different trading strategies with a lot of different objectives kind of let's say at the level of risk preferences and so on.
00:43:48
Speaker
why wouldn't there be more, like like the coincidence of wants things doesn't seem to and necessarily line up particularly well for sophisticated trading strategies or Uniswap extra setup. Why don't we see more of going after like particular kinds of trading strategies or they're just not that well developed at this point? I think part of it is like the really sophisticated people are just, they're never going to just tell you what they want. they They're going to... and Isn't there a role for like privacy elements to be incorporated into these things? There is a role, I mean, there is a role for privacy elements. And again, the optimal privacy is do it yourself, right? Just fire the transaction you want at the Uniswap pool you want when you want to. So like, I i don't think professional trading units are directly using CowSwap or UniswapX. I could be wrong. I think the
00:44:38
Speaker
the professional trading units are actually filling on cow swap and UniswapX. They're the solvers and the fillers on the other side. um And they're using that they're using these orders to achieve the execution they want or filling it or at least filling it when they see it either as a direction they want to move in at a price they like or they just see an order that's profitable. But so here here this is more of like Maybe my naivete is about like traditional and my market microstructure. Maybe Andreas can can weigh in here. and when you When you give that answer, the way what I first started thinking is that, then why aren't a lot of these sorts of players themselves becoming solvers? On the one hand, they can like do their own off-chain stuff. and but like
00:45:20
Speaker
you know maybe they get flows on top of it that they are able to make, that they're able to monetize. They are. these these solvers the There are two kinds of solvers really. and at this if maybe i should so Yesterday, there was a bit of a back and forth about this, but Danning Sui, who you guys have had on at CBR and is a wonderful um data scientist at Flashbots. She put together a thread where she actually looked at how much who's filling Uniswap X orders.
00:45:49
Speaker
Okay.
Creating Competitive DeFi Markets
00:45:50
Speaker
And um it's like, you can see like, you know, for small units for PEX orders, there's also a lot of smaller moment. what we call just um search about then small mom and pop operations like one or two people um who have a good compute and a basement somewhere or that's my mental model of them but as you get to larger and larger orders like ah her finding if I'm stating it correctly was that orders over like a million dollars in size really have only two bidders and there's those two bidders are also the um
00:46:26
Speaker
Also, the largest market makers in the space also happen to be the largest block builders in the space. So, um ah you know, as you get to a certain size of Uniswap X order, um you you start to get like only sophisticated market makers can fill that order or can have the size to fill that order and they are the only ones who do and and in the data.
00:46:51
Speaker
ah She then went ahead and stated that maybe it would be better if like one of the things Uniswap X and its design does not do is when Fahad says he wants to buy a million dollars of ETH, ah a given filler has only a take-it-or-leave-it offer. A filler can't say, oh, I'll i'll take 500,000 off that and find the next 500,000 elsewhere. So she said maybe it would help ah if Uniswap X, the design, allowed splitting orders. It currently doesn't. um And that's sort of the kind of thing that you know these newer intent protocols, um the next generation, also want to allow. ah That got into a bit of back and forth on Twitter on whether weather
00:47:31
Speaker
Uniswap was a X was achieving good price execution and I haven't really dug through the data on that. so yeah so and so I'm going to just bring this back again to the to the question of the who's the customer and who this is really for. right um so If you go back to traditional finance, the the way how brokerages handle the concept of best execution means finding the best possible outcome is is ah is is actually rather loose, right? So um the regulation is a little loose. it It doesn't actually tell you what you have to do, right? It basically loosely describes given the circumstances and the time of the order, do the best possible thing that can be done and it is never enforced on an order by order basis.
00:48:16
Speaker
So this is actually kind of critical here because in a way what you're trying to do is you're trying to find an algorithm or a concept such that can be done. So a few observations here. Number one, um you know, your intensor is a very, very complex problem, right? Potentially you have basically arbitrary, ah you know, usage of it, which is it's great for an economist, right? We say we have a mechanism, you know, people say what they want and then we give them the best that we can Google them but they what they want.
00:48:46
Speaker
ah In practice, ah people don't know what they want, right? Or rather, they cannot actually fully express it because they actually don't know the full parameter space, they don't know the full, they don't have a complete contingent plan of what actually is possible for them, right? And so therefore, this this is what the broker does, essentially, is they try to read the customer's mind in the sense of say, well, we want to do the best possible thing for them given the circumstances.
00:49:10
Speaker
and Then they built their the processes based on essentially a process. right so There's an execution committee that says, this is the general rules that we should follow. and and so there's really not a yeah there There is not really a ah precise policy or precise algorithm that you can use for you know complex market situations and complex orders, not even complex orders, but you know complex complex scenarios scenarios that could play out. ye but I just want to i want to agree with that, but I want to say you've got to understand where decentralized finance is coming from, which is you're correct that users can barely state their preferences, um but
00:50:01
Speaker
currently like to trade on a stand, like on a single decks, you have to know which pool you want to hit and then sign a transaction for that specific pool, right? So the level of, but you can't just like, now aggregate, or you need to know that aggregators exist and go to aggregators, which now a lot of people do. ah But there isn't this notion of a broker, or it's, there is somewhat, but The level of abstraction that we have achieved in TranFi where for you, the sort of retail user to just know that you want to just trade stocks and you don't need to know anything about the plumbing underneath, we are not there yet. And I think the intent, like one way to think about the intent market is about these intent protocols is they want to get to the stage where they can provide similar levels of abstraction to users, where users can just say, this is what I want. um And then then the rest is details.
00:51:00
Speaker
Yeah, I know that essentially, I mean, we fully agree on this one, right? So I think actually the big power of a place like Robin Hood, or in Canada, Wealthsimple, I mean, they're really not comparable, but let's just let's just say it as such, is the specification of the usage, right? And so my biggest pet peeve with DeFi is, it's just Super complex to use the learning curve is steep, right? And so what you're really looking for is and I think this is actually the way how how this will all pan out is I don't think that people will actually go and search for a Market and nobody wants to sign up with anything. You want to have a wallet and i had a piece of software on your phone and say oh well i actually want to change this token for that token and just want to press a button and then in the back office is all of the things that are happening for somebody to drop in the really sophisticated guys that we are i mean that you know the sophisticated traders
00:51:53
Speaker
You know they they have this they solve their own problem but that actually doesn't that will not apply to what you know hopefully the market of defi should be right. Sure um and and yeah and and indeed we have all the major wallets have for instance their own sort features that don't even require you to as long as you have a wallet and you have some funding in it.
00:52:13
Speaker
you can do swaps within your wallet without even going to Uniswap. On the flip side, Uniswap has a wallet now. So there is um there is this there is this understanding, I think, in the crypto world that the learning curve is currently too steep and ah intents a one thought process on how to start plugging those um plugging those learning curves. I'm going to just take a step further. right What you want to have is you want to have a wallet like MetaMask and you want to basically dictate with Siri into the wallet and have an AI which basically then translates this into an intent. right that's kind of This is what we're looking at the front end there, right? That's really what you want. That's what I think ah
00:53:00
Speaker
or because otherwise I mean and we are only talking about the finance side right right now but there's all this other stuff right you want to understand gas tokens and you want to understand what chain you're on and like try explaining any of this to your mother or your grandmother or your grandfather whatever well I have a hard time doing this and I'm I'm I'm into this problem yeah so it is very tricky um It is very tricky to explain this stuff, and we definitely need levels of abstraction that make stuff usable for um normies, as we call them in the crypto world, which are people who are not steeped in the traditions of DeFi, but instead just want to do stuff.
00:53:44
Speaker
and Yeah, so if I go back now, let's go back to actually the kind of problem that you're trying to solve with your paper um So I always wonder when
Challenges in Designing Intent Protocols
00:53:53
Speaker
I I mean when when you when you think of trading um and if you're treating the papers in in the in this space um There's often a question that we ask of you know, we have liquidity providers. We have ah liquidity Demanders and we have in your case also some form of service providers, right? Which is the solver modeler that you have so what what exactly is ah So whose whose incentives are you or whoses whose welfare are we trying or utility do we try to optimize here? Do we have a welfare function here where everybody, what we're trying to look at, what is the what would be the setup such that everybody is best off? Or do we say, for instance, say, well, our liquidity providers have to make net zero profit or our liquidity takers or the solvers have no profit function or ever have a constant cost function or the like.
00:54:44
Speaker
So how how is this set up? so know That's great. I think that's that's sort of where... that's sort of where our paper comes in, because as we were surveying with my co-authors, and to be clear, this is is joint work with Tarun Chitra, Shethij Kolkarni, and Theo Tiamandis, we were talking a lot about these intent protocols, and I think the thing that bothered us was there was this notion that
00:55:15
Speaker
um The goal, and I don't want to call out any specific protocol, it's it's sort of pervasive across the space, I think, which is, thats as long as you set up a good marketplace and you have end users, then competition among these solvers is going to achieve optimal outcomes for ah the end user. that's That's sort of, if you read every one of these um papers, that's always sort of the implicit statement, which is that we're going to write the tech, and then the magic of the market or the magic of a second price auction or an optimally designed auction is going to do the rest of the work to get a collection of solvers to solve all these hard problems for you um to translate your intent into optimal execution.
00:56:07
Speaker
for free or for close to free because they're finding the optimal answer for you. That was that was sort of the common myth that I think a lot of people believed. And that was where we just wanted to write our paper to say, you know what, there is, and and these are things we know from TradFi and we know from information economics in general, like solvers are doing things. They're doing things that are costly. You are revealing private information to them.
00:56:33
Speaker
um you're not necessarily just going to solve that problem by saying, hey, run a second price auction among solvers. Okay, so ah the point of the paper was just, well,
00:56:48
Speaker
how should we, and it's in a very simplified model, I don't think we nailed it, but we we just wanted to say it's almost, the paper is not a negative result, it's just a think more carefully result in the sense of you really want to think more carefully about what is your solver objective function, what are your solvers fixed costs, how are they, because that's going to determine things like solver entry, as we're seeing um you know There just aren't that many solvers above a certain size, for example, in Uniswap X right now. um And then you don't have, you know when you only have two people in an auction, it's not necessarily as efficient or as competitive as five or seven people in an auction. This is also a very complicated search space. There are thousands of DEXs plus um centralized exchanges plus other sources of own liquidity. So this complex the complexity of the search space itself means that
00:57:46
Speaker
three or four or five ah use like three or four or five bidders which might be efficient in a single um you know in a single item auction or you know you'll your standard sort of Theorems will tell you that the equilibrium is pretty close to the ah best price. So the true underlying price may not happen ah in in this space. So that that was the sort of punchline of the paper. Can I actually come back to something that you said earlier? Because this is also, again, this is out of curiosity and I've not been in part of the auction literature for a while. right But you mentioned before that when when orders get very large, there's usually only one bidder or so right that can do it. And it would be useful maybe to break up orders.
00:58:27
Speaker
ah Just out of curiosity, um um my my recollection is that ah One problem with if we if we break up an order into smaller units, we have essentially multi-unit auction and multi-unit auctions have all kinds of problems with manipulation, right? So what's the trade-off space here? have you Have you thought of that? Is that because that that discussion is kind of critical? I should clarify that this was, that that was a statement that was, so that was Danning's suggestion. We haven't analyzed the counterfactual because we don't have data. um But exactly that's the problem, which is
00:59:02
Speaker
the unit For instance, the Uniswap design is incredibly simple to the point where it can run fully on-chain. And if you start moving to ah combinatorial half-fills and so on, at some point maybe it can't run on a chain. Or maybe it runs um it needs a sidecar so other Uniswap affiliated or ah there's there's an associated collection of projects that are trying that that are trying to do portions of the auction, portions of this auction off-chain.
00:59:32
Speaker
ah in order to in order to do more complicated stuff that they couldn't do currently on-chain. An example of that is Sorella. So Sorella is a startup um that is going, I believe, to have a Uni V4 hook. And part of their design involves um Well, I don't know enough about the design to comment publicly on it, but I believe there is an off-chain auction component. And I'm just going to just um kick something in here just speak for the audience that is maybe not fully aware of all of the different versions that we have in Uniswap. Uniswap version V4 has something called a hook, and the hook is really just the condition that you can put in. right So you can make conditional bids and the like. um it is And it allows you to have exactly one hook.
01:00:23
Speaker
right so i mean you can maybe fold multiple conditions into one, right? But just ah to clarify that people understand what that means. Yeah, sorry. i yeah it's it's ah It basically makes so UniV2 and UniV3 had very um opinionated statements on how liquidity, but how orders were filled against existing liquidity providers. And Univi 4 lets you design um various statements. So um you can you can do all sorts of stuff. The simplest stuff is stuff like dynamic fees. You can look at, for instance, the volumes in the last block potentially or something like that and say, okay, if the volumes were high, I charge higher fees or look this pool will charge higher fees or lower fees.
01:01:08
Speaker
But something like Sorela is basically allowing an off-chain option component to my understanding on top of Unity as well. Can I bring this back to to your work um on this? um so at At a very high level, I think part of what you were saying is that there's sort of a mental model that comes from very simple models in economics that I think some of the people maybe behind these protocols are thinking of. So, for instance, if I am a margin 80 in the sense that everybody's like, think of a producer market for simplicity, everybody's cost of production is the same.
01:01:40
Speaker
And there's free entry. Now you're probably going to get price fighting and then ultimately the price will just be off. But trying competition will bring us down to marginal cost. Exactly. And that's sort of, let's call it the mental model maybe that people want others to have. And when you were talking about your paper, I think I heard two sort of pieces there that let's say interfere with the intuition of that very simple frictionless model. One,
01:02:04
Speaker
Was this point about entry like maybe yeah, maybe people can enter but there's some cost of entry and so on and so you don't actually get uh, you know, uh, the sort of of full entries like are you know, um, uh, the other piece was heterogeneities, right? Like if so, um, if one person has the lowest cost relative to everybody else, even if there's, you know, an uncountable infinity of them or whatever,
01:02:29
Speaker
um You're not going to get that person to... to to You're not going to get that person's cost. Yeah, exactly, because he he has no incentive to do it. So could you talk a little bit more about sort of what specifically in the context of your analysis and and more maybe practically thinking about... but We don't have to name particular protocols, but I mean like just in broad strokes, given what we have a sense of the world of of of these solver models, what are the specific details um that break the the perfect competition intuition?
01:02:59
Speaker
um and kind know How do they play into your results? like what of i understanding is like ah What is more important and how are they important? and how do they There's interplay, of course, also about heterogeneous types and entries. Could you maybe dig into that? so Let me say a few things. some of them so One is not from my paper, but it's something I've noticed when looking through certain off-chain auction data.
01:03:22
Speaker
um And this is about heterogeneity. Different solvers have different strategies. This isn't as Andreas pointed out way at the beginning. This is an incredibly high dimensional
Barriers to Entry for Solvers
01:03:34
Speaker
space. So you can of course abstract it, especially if you're an economist and just say everyone has in their head the current liquidity of every possible thing.
01:03:43
Speaker
But there are maybe 500 DEXs. There are maybe 30 other places that you could get flash loans from ah to enable other strategies. And if you just think about that combinatorial collection of things that you could do to fill a particular order,
01:04:03
Speaker
optimally, that search space is massive and you have maybe a few seconds to actually try and fill it. So you, Fahad, might have one strategy that you use that works well for you. I might have a different strategy that works well for me. Andreas might have a third strategy. All three of us might be efficient, but we might be, like, we're constrained efficient in the sense that we have made our approximations that work well for us, which means on certain kinds of orders, my strategy might perform really well. Certain kinds, your strategy might perform really well.
01:04:33
Speaker
bigger bot programs like bigger programs um Bigger trading houses have multiple bots running simultaneously with different strategies. Just because there's no one bot that's large enough to search the entire space,
01:04:47
Speaker
um So one person uses a, I'll fill this order and optimally background it strategy. Another person says, can I find a flash loan strategy? These are, um I'm just. instead So the mental model here that I have in in what I'm hearing is essentially um maybe based on the particular order and the particular state, meaning the real world, like the economic state of the world at any point in time.
01:05:11
Speaker
Oh, somebody might be differentially better than everybody else who's at least trying in this space. And so you kind of have the classical ah breaking of there's a very sort of clean way or very simple labor. Yeah. So, I mean, basically, the best guy can extract some reds. Right. and like So and how much that is is sort of you now have to trade off between the number of people you have. It's hard to put this in a mathematical model. We try because it's it's sort of how rich the space is determines how many people you need doing different things ah to get a good fill. But so then the question is barriers. I think then that brings it back to the barriers to entry idea, right? Like, so for instance, if there's
01:05:57
Speaker
infinitely many players, then maybe this doesn't matter as much. um And you're not going to get infinitely many players because each of these players need to justify the costs of whatever infrastructure they have, plus whatever human capital they spend ah improving their bot.
01:06:16
Speaker
There's also running costs. so For example, if i'm there are two so there are two kinds of solvers. There are ones who fill purely based on on-chain liquidity. so they are like um or not two kinds, but as a mental simplification, you could think of two kinds. of ones Some of them don't bring any of their own or to any size, they don't bring their own capital. Instead, they're just searching for good routes for you. they are They're just trying to fill your intent with all the stuff that's available on chain the best possible way.
01:06:49
Speaker
Other solvers will also keep an inventory of their own. So their bot will have some amount of various tokens. Sometimes they might choose to fill you from inventory. They might have connections to centralized exchanges. So they can quickly buy yoto the tokens that you want on a centralized exchange and then fill you on chain.
01:07:10
Speaker
um um all of these, this you know this increase increases the cost of entry because you have to be maintaining, like you have the cost of capital plus whatever risk you're carrying on your filler book. So you start becoming more and more like a traditional trad-fi market. But so can I ask you this quantitatively, how big is approximately, not price numbers, ah each of these? Because for instance, like to your point, when you were saying that first thing I thought is like that second group,
01:07:37
Speaker
It's going to be hard to enter and compete, like what what am I going to be able to start up a hedge fund and so I'll have an inventory and stuff like that. That's not really plausible, I think. ah so So there's going to be a certain sense in which that's not going to be competitive. um But the the first set seems like OK, if I don't, if not keeping any inventory. um yeah So first question is, is that first set actually occupying?
01:07:59
Speaker
large amount of the volume. Are they actually competitive relative to the second set? Because I don't see what their differential advantage is. I guess, in theory, they could be smarter about routing or whatever, but I don't see why a sophisticated financial institution couldn't get that expertise. The first set is big. The first set is big. The first set is big. I don't know if it's big in terms of total profit, it's definitely big in terms of number of entrants. So there are 20, 30.
01:08:42
Speaker
Yeah, I think it's just. Two things, one, because the search space is so large, they find little niches where they can continue to maintain um some advantage. um The market size of those niches might not be big enough for a big shop that has, even like even the big shop is not huge, right? So the biggest shops in the space are um orders of magnitude smaller than like a Citadel.
01:09:10
Speaker
The large ah jump was one of the larger hedge funds that was trading in the space and they stopped. ah So jump capital stop to my understanding so the largest on-chain ah people are winter mute SCP and a few others there are three or four of that size or that order of magnitude and But they're not massive firms. They are not you know, somemark some market some certain optimizations They just you know, they have a list of
01:09:44
Speaker
going ah across and getting it. I'm not saying these small searcher shops will maintain that edge forever, but we're nowhere close to equilibrium. and that's you know you can you can run If you find a profitable strategy, you can probably run it for a few years and make a nice amount of money until it becomes big enough for a larger firm to find copy. What's the primary barrier to entry to these searcher shops that you're describing?
01:10:10
Speaker
I don't think there's any other than programming skill and the ability. so if you look at and but
01:10:20
Speaker
so It's an interesting thing. like If you meet some of the original searchers, so the original one ah and the solver shops are also, in some cases, outgrowts of searcher shops who were just doing various kinds of MEV extraction ah back starting from the early days of DeFi 2018-2019. The thing is,
01:10:39
Speaker
As you do strategies and you fill them on chain, people see what you're doing. People try and back reverse engineer how you're doing it. They improve on you. So it's a constant race. you ah there's this I'll try and find it. There was this lovely story about ah ah a particular searcher and you know for the first month he made like you know, just random person ah in their basement or whatever, ah not even super high quality computing made um an insane amount of money, like and not insane insane, but of the order of a million dollars in the first month, then suddenly sees that the third month, they've gone down to zero.
01:11:17
Speaker
And then they're like, okay, we have to get back. We can't just you know party. We have to start finding the next thing, the next thing. And at some point it hits, you know goes out of their skill range or the kinds of things that they're comfortable with. And they either drop out of the space, go on to do something else.
01:11:32
Speaker
um So that that's the life cycle for these strategies is not very large because it gets discovered and gets ah the alpha gets eaten away. But the search space is also increasing. So it's it's a cat and mouse game between those two things.
01:12:14
Speaker
They do? and Yeah, and and that's exactly what that is. To my understanding that's happening, there are ah some ah like many of the OG shots and sometimes searches make a lot of money and just aren't interested in the game anymore. So they made some money now that that the alpha from that strategy is gone and they say,
01:12:32
Speaker
ah i'm gonna say um' I'm going to sit and enjoy for a few years or I'm just going to move on to a different type of move to a different space. So some of the original searches have moved on to other things.
01:12:44
Speaker
um ah it's it's it's definite I don't think there's sustainable long-term advantage, but they like what is long what is short term here? Is it short term like your bot could like your collection of strategies and skill sets could be relevant for four or five years, maybe? ah There are definitely shops that have been around for two or three years and have not been competed away. I have to feel i mean i kind of feel that when if these markets were bigger,
01:13:12
Speaker
Then the incentive, for example, to take away the the brainchild of this search strategy is going to be much more significant. And so I do wonder i do i wonder whether, when you scale these markets up, zooming you know DeFi gets orders of magnitude larger, whether we're going to see more centralization owing to that fact, because then it becomes worthwhile to sort of ah do those sorts of things.
01:13:36
Speaker
yeah And that's exactly I think that is what we are seeing right now. So we're seeing it in other context where, ah for example, ah in a different like we see this a lot in builder markets where we went from. ah On Ethereum, we had ah we started this thing called proposal builder separation, which is um validators don't have to propose their own block, they can just sell the right to propose the block in an auction that's run um sort of off-chain. And we went slowly, if you look at the concentration, back in
01:14:14
Speaker
When they started, so back in ah early 2023, late 2022, there were maybe 10 competitive builders splitting it up. And the two largest builders right now are also the two largest market makers. Is it our thing to Beaver build?
01:14:30
Speaker
R-Sync Beaver Build, two of the three largest builders R-Sync and Beaver Build are in particular Winter Mutant SCP in that order. And Titan is the third other big builder. And all the other builders have fallen away and we've had this sort of centralization of the builder space. So similar things will probably happen and this is You know, these are specialized activities. So specialized activities as we know from, you know, there are
01:15:05
Speaker
gains from scale. There are information rents as you learn to do things better than others. So this idea, I guess, we just wanted to write this paper, pulling it back to that, to point out that this idea that you're going to get competitive markets where billions of dollars will be traded and solvers are going to make a penny, sub pennies in profits is just not going to happen.
Conclusion and Reflections on Market Structures
01:15:28
Speaker
ah They are going to make rents and we need to understand the trade-off between these rents. ah you know sort of We need to design markets so that they're making an appropriate amount of rents and not more than that. Yeah, I mean, forcing a competitive market is generally a difficult problem, as we know, right? So because every every person who operates in a competitive market wants to become a monopolist. and but I mean, this is kind of in the nature of the beast, right? um so I think this is actually ah probably this is actually one of the biggest messages that you know people in the in the DeFi and blockchain space have learned, and but sometimes still need to learn. is that The idea that we can have a competitive market and it just magically appears.
01:16:12
Speaker
as As lovely as it is when it's there for the consumer, it' it's just unfortunately something that is an ongoing battle. um And that's probably a pretty good note to kind of slowly transition to the end of this podcast.
01:16:28
Speaker
sort of Just to add, like we know some things about how to make markets competitive, right which is to make them simple. We make barriers to entry low. We make the underlying product that we're selling undifferentiated. like If we're selling widgets,
01:16:41
Speaker
ah you know if if it like if if we ask if we're trying to like um buy widgets, we can have a competitive auction because there'll be a lot of widget makers if you're trying to buy like super specialized software, you'll only get like Microsoft and volunteer or something selling to you. So we know how to do this, we just need to think about these design questions carefully, which is why we wrote that paper. Yeah, no, this is great. And I think this is probably the best message here that that we can pass on to
01:17:12
Speaker
the audience as a whole that you know getting widgets, and is is there's there is ah there's value in this. right Designing markets for widgets or widgets or rather designing markets in such a way that people offer widgets or commoditizable services. I would probably call it something which is less economy, CEC. It's probably a good thing.
01:17:35
Speaker
so Yeah. Well, with that, um let me come to the conclusion of this podcast. ah Again, Malesh, many thanks for this. I think we had a very wide ranging discussion. I hope that the audience enjoyed it for hard too. So we're we're very grateful to have had you and and to be able to pick your brain on these very complex problems. And thank you so much.
01:17:59
Speaker
We hope you enjoyed this podcast. Thank you for listening. As a reminder, you can find additional materials on owlbexplanes.com and can stay updated by following us on social media. That's all for today.