What Drives Liquidity?

Transcript

[Music] Okay, everyone. Uh, my name is Shin. I'm the author with the underscore on the slides. And, uh, today I'm going to discuss the factors that we think drives liquidity across different decentralized exchanges. Um well specifically here I'm just using a sample of unisoft protocol but uh you would expect that LPS behave sort of consistently across different dexes because nothing stops them from being on a different one. Uh and this is joint work uh I'm at UNISOP Labs and this joint work with Columbia University

and Circle. Um yeah my opinions only and not that of my employer. So um the role of liquidity is uh as everyone understand to be very important but uh there is always this um perception that maybe liquidity indexes behave a bit differently than liquidity elsewhere. And it's unclear first whether that's true and second if it's true uh what explains it because you can say oh it's a different group of people who's providing liquidity or you can say you know regulatory they have different

constraints or they require different type of return or they just you know think differently build differently um so the goal here is to not uh make a normative statement so I'm not here to say whether people are doing the correct thing or the optimal thing. I'm also not here to predict uh what happens in the future. I'm simply here to take what happened in the past and try to explain it and say, you know, these things happened and I think it's statistically correlated with these other factors. Maybe some of them are causal, some of them are not, but uh that's sort of the goal of the study.

So the first question to ask is uh what are the key uh onchain offchain factors that drive liquidity on dexes? And when I say drive, I mean that uh a change in that factor today would likely uh lead to some predicted change in the future uh in future liquidity on the dexes. So some of the examples are gas prices, token pair returns, volatility, the amount of fee you can earn in the pool and uh the sharpness of your counterparty.

Now once you identify the factors, the second question you can ask is how do these factors affect liquidity? Um so specifically you can mechanically break down uh liquidity into TVL and concentration and I can show you how we do it. Um and we can try to ask which of these are most affected by these factors that we we identify. So the TLDDR here is gas prices and returns uh mostly affect liquidity concentration.

Um volatility, the amount of fee you earn and the sharpness of your counterparty uh act through both channels. Uh and the final question is uh how does external liquidity from other dexes or liquidity that is not public uh affect onchain liquidity? And uh we found that competitor share which are which is defined to be the amount of liquidity in other dexes um to negatively affect liquidity uh through the channel of concentration and internalization by aggregators. Meaning um if an aggregator choose to uh take the

trading flow in their own book or send it to some uh liquidity sources outside of unis swap that actually has no statistically significant impact on market depths. So this is just summarizing what I have just said. Uh the first thing is we should think about liquidity as TVL plus concentration. Uh second thing is we identify volatility fees and informed trading to uh affect the total depth of liquidity and then con uh competition between dexes uh creates liquidity fragmentation. Finally uh aggregators ne

don't necessarily uh kill dexes. They can coexist. Okay. Uh, I'm skipping the literature review because I used this for academic presentation before. Um, I'm also skipping unis swap uh intro because I would assume people know it. Um, this is just to show what concentration looks like. Um, okay. So, this is the data source.

Um, we used data from Uniswap V3, uh, going from May of 2021 to July of 2024. So that's about three years of data. And we had all of these pools in our data sample. So across four different chains uh, in each one across seven different pools, sorry, seven different pairs. And some of these pairs have different fe tiers. Now, uh let's talk about the outcome metric. So, in a in a simple setting, you can think of our analysis as a regression. And in a regression, you

have a y and you have a bunch of x's. So, this effective spread is one of the y's. And what do I mean by it? I mean that if you simply take a fixed amount of ETH and you try to simulate a transaction in a liquidity pool uh the amount of uh price movement that your trade would incur um or rather sorry it's basically the same thing but the difference between the quoted price and um the um the quoted price for a sell versus a buy of the same fixed amount uh gives you

this uh spread and we get this quote quoted price through a quot contract. The second Y um and again we're not touching the axis yet. The second Y in the regression uh is uh TVL and that's pretty obvious. So we'll run two regressions, one with the first Y, one with the second Y. Uh, and here I'm just showing what the Y's look like as a time series. So on the left side you have uh four chains of one pool and keep in mind that we have a ton of pools and this is just for one pool and you can observe uh the change in spread on the left side uh with log

scale and the um the TVL on the right side with log scale as well. And this is um taking sort of the same data from before and instead of treating as a time series I'm just showing it as a histogram to give you an idea of the distribution. And you can see that for the same fixedsiz trade um the liquidity on mainet is much deeper. And so there's an inverse correl uh relationship between spread and depth. Uh the deeper the liquidity, the smaller the spread

for a fixed size trade. And here you can see that for the fixed size um the spread on mainet is much smaller, meaning the liquidity on mainet is much deeper. Okay. So those are the two axis h sorry the two y's. Um and and remember that when we set out we said the first goal is to identify the factors that drives liquidity. The second goal is to understand how those factors drive liquidity. So here I'm going to introduce some intermediate wise. So you can think of them as Y's that we suspect are uh representative of the channels

through which uh our factors affect the final Y. uh one of them is this thing we call a counterfactual V2 spread. So what this is is um you take a v3 pool and you look at the TVL in the pool. You say okay let me withdraw imaginarily not like not in reality imaginarily you withdraw all the assets in the pool and you rebalance it such that it's the same pair but it's a v2 pool you create this imaginary new pool and you simulate the same fixedsiz transaction in this new pool. Um what this does is it isolates

uh the effect of TVL in that V3 pool and the difference tells you how concentrated your V3 pool is compared to a imaginary V2 pool of the same TVL. Um so uh let's just take one pool as example. Let's say we have a V3 pool that's extremely concentrated. Um what that means is for a size that's within the TVL of that pool. Uh it's not going to create any price impact or if you compute that uh first Y which is spread you will see it spread is tiny.

Um but if you create this imaginary V2 pool and simulate the same transaction, it will have a bigger spread. And the difference between the the two tells you that the depth in that V3 pool is mostly coming from concentration instead of TVL because the same TVL on a V2 pool would do much worse. And now just to formalize it uh we're going to define this uh quotient to be the ratio of the actual V3 pool spread versus the spread we compute on this imaginary V2 pool. And this tells us as I just kind of walks through an example how well concentrated this V2 V3 pool

is. Okay. Okay, so here's an example of a well concentrated pool um which is uh we uh WBTC 30 bips pool on polygon and you can see that the V3 spread uh consistently stays under uh the imaginary V2 counterfactual spread and now here's an example of uh what we call a poorly concentr poorly concentrated pool because uh a lot of the times uh concentration gives you no additional benefits in terms of depth. Uh you can do equally as well by just putting that uh money in a v2 pool.

Okay. So now uh I just defined you know three ys. The fir the first two are spread and uh well I forgot the notation but you kind of remember there are three Y's and I'm going to regress them on all the factors that I think uh is going to affect LP's behavior specifically their gas prices uh return on the pair of the past. So notice that there is a subscript t+1 for all the y's and there is a subscript t for all the x's. So I'm I'm looking at the correlation between future liquidity depth and current um factor values. So uh gas price log

returns volatility fee revenue markout uh and there are some statistical control that I'm adding to the regression. Okay. Um after we do those three analysis, we also want to say is our result robust against some slight changes in experiment design. So we're going to uh kind of change the trade size, change the horizon, and change change the horizon of two of the factors.

Okay. Uh so now I'm going to overwhelm you with a very big table. uh just be mentally prepared. Um but before I show you the the number uh the horizontally uh is what we have defined to be the y's. So the the outcome metrics that we're interested in explaining uh and uh vertically are the factors. So let's look at the first one. The first Y is just the V3 spread. And this shows um that in fact log prices of this period has no stat statistically significant uh impact on depth in the next period. Um but return uh does and

negative is good or sorry negative is intuitive because uh if you have a high return this period uh there will be more liquidity in the next period which means less spread. Uh so that that's kind of the direction we're going. Uh similarly, volatility is intuitive because more volatility, less liquidity. Um higher revenue, more liquidity. That's easy to understand. And then markout, um more markout, meaning you know you you make more statistically from your counterparty, more liquidity.

Um so now I'm going to try to break this down, right? the the V3 is sort of the combined outcome and now I'm showing u the counterfactual V2 spread which is the effect of TVL and the spread quotient which is the effect of concentration. So the first column should be the combined effect of the second and third column. And now it's it's a little interesting because you can see that gas prices actually has a significant effect and it affects um concentration when gas price is high in this period uh concentration will reduce in the next period uh

significantly. Um and then for for return uh it's also working mostly on concentration and sometimes you can think of concentration as you know active liquidity is pulling out uh and passive liquidity is staying in the pool. Uh volatility and fee revenue on the other hand affects both channels in the same direction. Uh markout little bit uh unintuitive here but markout has a uh weird effect on this quotient or maybe um concentration and uh let's just think about this for a minute. uh markout is your the sharpness of your counterparty and you you would

expect sorry markout is like the inverse of that because the science is reversed. But if you would expect that um the more uh the smarter your counterparty is uh the higher um the selection effect is for liquidity in the pool. And you would expect liquidity to be uh maybe more concentrated.

Uh but what this shows is actually the opposite that when the traders or the takers of liquidity in the pool is smart, um the active liquidity kind of leaves the pool and we're left with a pool with relatively less concentration. Um, which I guess also makes sense. Um, but it's just kind of interesting that it's going in two directions. Um, okay. And I'm going to skip because of time. Uh, this paper is online. So if you're interested in, you know, any specific number, you can just go online and search for this paper. What drives liquidity?

Um, and then quickly want to mention two other things we look at. one is the market share of competitors and uh the second is the internalization ratio. Um and both of them uh we do it through kind of a similar setup except we just add those uh two factors into the regression. And okay, maybe one thing to highlight is uh we show that at least for the final outcome that we're interested in uh there's no statistical uh statistically significant uh number for this factor which means uh even if

you have you know the the likes of um 1 in and cow swap where they take trader uh order flow first and they uh try to internalize it or send it to people who has the ability to internalize it. Uh before it sent to Dexus, uh the existence or the prevalence of uh those players uh didn't statistically affect the depth of liquidity on chain. But but again I just want to remind you that this is empirical study. So it doesn't say what should happen or what is optimal. It simply describes what has happened and this is what we observed so far.

Uh okay. So quickly summarize um liquidity we think of it as a combination of TVL and concentration and then we identify the key drivers to be gas prices liquid uh volatility return fee revenue and markout. The last three are kind of you know different ways of expressing the same thing. Um and then there are two channels that we uh uh that we test all these factors through. Um and we found that you know some subset affect specific channels uh uniquely. Uh finally we tested external liquidity um and their effects on liquidity within unis swap protocol and we found that uh competition intuitively

reduces concentration within unis swap and uh internalization so far does not harm overall market depths. Uh that is it. If you want the paper, you can um scan. Thanks. Do I have time or Oh, just on time. Okay. So, if if you want to ask me, I'm I'm going to be outside.