One Dollar, Many Flavors: Inferring Preferences for Stablecoins with Shai Bernstein | a16z crypto

Transcript

Uh, all right. Welcome everyone to this afternoon's A16Z crypto research seminar. I'm very happy to introduce Shy Bernstein from the Harvard Business School where he among other things co-directs crypto and web 3 center with our own Scott Kominers among others. And for those of you that think stablecoins are boring, Shy is going to tell us why actually they're very much not. So, Yep. all yours. All right. Thank you very much, Tim. So, just maybe in two sentences just quickly about myself. I mean, I joined HBS about 3 years ago. I was at Stanford for about 7 years at the Graduate School of Business for almost 8 years before that. Um, my main focus of research is always

thinking about the intersection between financing and technological innovation. Trying to think a lot about entrepreneurial ecosystems and the role of venture capital and other financing arrangements and how they influence the ability of firms to innovate. Um, the last few years I got roped into and got very excited about blockchain and started the center with Scott and um, among a few other projects, this one is on stablecoins um, together with Paul Apostolakis and Alessandro from Harvard and Slava Foss from Boston College. In fact, this is incredibly early stage work to the extent that we don't really have a

paper. This is the first time that I'm showing what we have. So, I'm going I'm very excited to share with you what we found. We don't have a model yet. This is just a bunch of facts that we were trying to come to understand better. So, I'm very much looking forward to your feedback. Um, so just kind of like to I don't I don't think I need to convince you guys too much about the importance of stable coins, but I'll do that anyway. Um so, stable coins emerged to become one of the largest um use cases of crypto.

Uh in 2022, the market peaked at uh 180 billion dollars, which declined quite significantly ever since, uh but still quite sizable. Uh settling in 2022 over 7 trillion dollars, which is more than double than the processing of Visa and MasterCard. Now, um the future or the promise of stable coins um has various aspects. One is this idea that we could generate this more efficient and cheaper and and faster uh payment system, um which might be particularly important in cross-border payments, where the transaction costs are incredibly expensive and settlement can take multiple days.

Um but the other promise of stable coins is this idea that it might enable um access to uh financial services, store value, and access to US dollar in potentially emerging markets and for populations that either don't have access to traditional banking uh services or unable to afford traditional banking services. But more realistically, the way that stable coins are being used today is is is mostly enable leverage and store value within the cryptocurrency ecosystem. Now, in order to enable stable coins to fulfill their promise to a large degree, um we really need to understand both the liquidity and run risks considerations

of stable coins, um which is naturally top priority for regulators and policy makers. So, what are we doing in this paper? We're going to try to um understand how market preferences uh shift during periods of uh significant market stress between different types of stablecoin designs. And by market preferences, I mean both traders and liquidity providers. The reason why I think it's important is we basically want to understand whether there's this there this notion of flight to safety within the stablecoin industry. And if so, what do markets deem to be safe

during periods of significant stress, whether there's one particular stablecoin design that seem to dominate the others, or is it more event-specific kind of stablecoin design that preferred by markets. All right, so just kind of like briefly remind you how stablecoins kind of are able to maintain their peg. So ultimately, it's a very simple story of supply and demand that is met by arbit- for traders. So imagine, let's say USDC is traded in a secondary market at a price above one.

Then effectively, what would happen these arbitrageurs will go to Circle, the issuers of the stablecoin, and pay cash in return for additional USDCS, which they would sell at the secondary market for profit. Circle will take this cash, deposit in the reserves, and mint additional stablecoins, basically moving the supply until the price again meets the peg. Now, if the price of a stablecoin basically goes below $1, then the reverse will happen, which effectively the arbitrageurs will buy stablecoins in the secondary market and will basically

take them back to Circle, and will ask them to redeem these assets back uh in return for dollars. So, Circle will have to basically redeem some of of these reserves, convert them to cash, and give them to the arbitrageurs, and in return will take these USDC's, burn them, lower the supply, and again, until this will this profitable arbitrage will exist until we are conflict back to the peg. Now, the idea is that that the the ability of the issuer to defend the peg depends on market belief that the the the issuer has the ability to meet this kind of redemption requests. So, I need

as a trader to believe that every time that I want, I can go to Circle, give them the digital representation of my dollar, and get my dollars back. Okay? Now, of course, it depends a lot on what exactly the issuer hold in reserves. Um as long as the market believes that these reserves are sufficient to maintain the peg, so there's no question about the ability of the issuer to actually uh provide uh to to to maintain to sustain these redemption requests, then uh the market will sustain its peg, there's no problem with that. But imagine the situation where the assets that are maintaining the reserves are actually quite volatile.

And there might be concerned that the assets may be the the value of these assets may be lower than liabilities. In this kind of situation, this might lead to initial shock to the expectations of the market about the ability of the issuer to meet this kind of redemption request, and may contain trigger this bank run like behavior well, uh where investors will basically go and try to get back their dollars, um but if reserves are insufficient, then this kind of flight can lead to complete panic and even death spiral, um and complete unraveling of the stablecoin.

Yeah. So, you think of this more of a bank run kind of model or more like the like more of a thin shim like defending the bit like currency attack? Is this Like do you think of this more like a bank run or more of a like a Yeah. So, I think that effectively, I think about that very much like a bank run when there's some expectations that them there might not be able to honor their commitments of giving me my dollars back. I'll basically base try to kind of like go ahead before everyone else to kind of like get whatever is left in these reserves. Now, um there's significant intro generality across different types of stable coins in terms of how they maintain the peg.

So, and the intro generality or the design choices of different types of stable coins vary across several dimensions. One most importantly is what exactly they have they hold in reserves. Um so, um you know, fiat backed reserves are for example in the case of USDC and Tether, they basically um maintain off-chain assets um like cash and cash like securities um but then the alternative is basically for example MakerDAO through the creation of Dai, um they they maintain mostly um cryptocurrency in reserves because cryptocurrencies are particularly volatile and that may may generate

concerns about the ability to maintain the peg, they require also significant over collateralization. So, for each dollar of Dai, you actually need to contribute significantly more in uh uh in cryptocurrencies. And then finally, there's these algorithmic uh stable coins that basically generate reserves using uh completely created tokens. The ability to maintain the peg depends on the beliefs of the market that these tokens are valuable and as we've seen in the context of Terra, um these things could lead um very quickly to death spirals and unravel.

But, the composition of the reserves is only one dimension. There are other important dimensions. For example, the level of transparency. If you maintain assets on chain, then it's pretty, uh, fairly easy to understand what you have in reserves. If the assets are off chain, then you you you may need to hire auditor, depends on the frequency of auditing, and depends on the market's beliefs of the reports produced by the auditors. That would influence the level of transparency. Uh, there's also variation, obviously, in the nature of decentralization and the compliance with different types of jurisdictions.

Now, these, uh, different design choices of stablecoins impose different types of risks. So, the most important risk, of course, is credit risk in which, uh, the market may concern that the assets, the value of the assets might be lower than liabilities. Um, but there are also custody risk, for example. Let's say that you maintain assets off chain, um, then, you know, if your banking partner is subject to, uh, insolvency or financial concerns, that may, uh, influence, uh, may generate, uh, important spillovers, uh, to the ability of the stablecoin to maintain its peg.

And, of course, there's also regulatory risk. So, if, uh, in a certain jurisdiction, um, there there's a tendency to ban or apply sanctions on on these the the kind of stablecoins, then, again, it might affect the value of the stablecoins for different types of users. And, of course, there's this, uh, technological risk, uh, associated with, uh, with these, uh, smart contracts. Now, what's interesting is that we have here a situation where there's, uh, effectively a competition between different private money issuers that, uh, have, uh, quite a trojanious design choices, and they're subject to different types of risk.

It might seem as a novel problem, but in fact this is not the case. Um so there is this uh interesting quote by uh Governor Brainard that basically mentioned that the period of the 19th century, when there was active competition among issuers of private paper banknotes in the United States is now notorious for inefficiency, fraud, and instability in the payment system. Now, so in the 19th century we had fairly similar situation. Now, I'm not suggesting that this is where we are today, but definitely it's worth trying to understand what exactly um this competition between different types of uh private money issuers imply for trading and liquidity. So, let me be more specific about what we're doing.

The first question um is going to uh understand how this competition between stablecoins uh affect uh trading or the composition of stablecoins holding. Is it the case that there's flight to safety during periods of significant risk? I mean, the null hypothesis, which is a little bit of a strawman, suggests that the market may not um uh may perceive these different design choices for stablecoins perfect substitute. If that happens, of course we we should see no uh uh change in composition of um what kind of stablecoins the market holds.

But um otherwise, we might see that there's kind of flight to safety if the market perceive one type of uh stablecoin safer than the others, um which um so this would also allow us to empirically evaluate market preferences um to different types of risk. The second question that we're interested in is trying to understand if such compositional shifts exist and traders start to push and change their composition towards one type of stablecoin versus the other, uh trying to understand the behavior of liquidity We're We're to try to understand the behavior of liquidity providers.

Uh we're going to focus so obviously liquidity is really essential for the functioning the proper functioning of the markets especially during uh stress. Um we're mostly going to focus on we're entirely going to focus on decentralized exchanges. Um decentralized exchanges as was discussed yesterday provide incentives uh to liquidity providers uh in the form of uh fees. So if such if periods of significant market stresses are associated with high volume of trade we might think that this might trigger additional liquidity provision to the market.

But alternatively it might be the case that uh liquidity providers uh facing such a uh significant adverse selection that even the fee structure the the the increased revenues from fees may not be sufficient to compensate liquidity providers and the compositional shifts triggered by traders may also lead to reduction of liquidity in the market basically uh um further amplifying the the situation. And finally we want to also to understand investor heterogeneity in terms of who are the liquidity providers and how they react. So for example you may think that sophisticated deep pocket wallets may be more resilient to this kind of shocks that we are going to

explore. Um and as a result they might be able to sustain these shocks and add contrarian liquidity. Or alternatively they might be more sophisticated and they might be actually quicker to remove liquidity uh from the market triggering additional reaction by others. Okay so let me quickly tell you what we find and then I'll show you a bunch of pictures. Um So we find that during period of significant stress and I'll be more clear what I mean in in a moment. Uh we find that there is a significant volume of trading of stable to stable uh pairs.

Um, and so we we find that the market seemed to move quite dramatically to what it perceived to be safer, uh, stable coins. And these stable coins, uh um uh, what is perceived as safe depends a lot on the type of risk that trigger the particular market stress. So, it's not the case that there is, uh, one stable coin design that dominates the other. Um, and the second thing we find that despite the increase in revenue fees during this particular types of market stress, liquidity providers tend to remove liquidity very quickly, particularly from, uh, exactly those, uh, stable coins that are deemed to be less, uh,

safe. So, in a sense, the the the flight to safety of the market triggers, uh, increased run risks on these kind of other stable coins that the market perceive to be, um, less risky. And this is amplified by the removal of liquidity of exactly these assets. Um, and finally, deep pocket wallets or sophisticated wallets seem to, uh, at least, similarly to everyone else, remove liquidity. If anything, they're just doing that faster. Okay.

So, what we how we are basically evaluating that, we're basically going to focus on stable coin pair trading, um, which basically you can think about, uh, on the two the two sides of the transactions, um, um, there's a a swap of stable coins. So, effectively, it's equivalent that to the trader that holds the demand for, um, US dollar digital representation, but prefer to hold it in different form than the other.

Um, and to do that, we're going to focus on Curve, which is a decentralized exchange with an AMM that has this uh, linear invariant to to through most of the curve that is better suited to enable the trading of stable coins with stable coins and I'll talk more about that. Um if if it is in fact the biggest exchange that facilitate this kind of trade and for now I'm going to show you the results specifically from the three pool which is the most liquid pool on curve that enables the trading of USDT tether USDC of circle and die of maker down.

Okay. So um So basically let's start with the first question of whether there's this kind of flight to safety during periods of significant stress. So it turns out that it's actually very easy to reject this null hypothesis. So if we're looking at the volume of trade in the three pool um where in the red this is basically the volume of swaps and in blue this is the volume of liquidity provision. Um what you can see is that there are particular times in which there is massive increase in volume of trade.

It's actually very easy to understand what each one of these peaks what triggers each one of these peaks. For example, you can see that the collapse of Terra or the collapse of FTX or the collapse of Silicon Valley Bank led to massive volume of trades of these stable coins. Um you can also see here that the depegging of if you recall the magic internet money which led to concerns about depegging of UST also triggered this kind of volume of trend the depegging of the algorithmic stable coin of Tron also created significant volume of trade and and finally here

this is triggered by the SEC suing Paxos because of allegation that Binance USD is in fact a security. Again, triggered significant volume of trade. In fact, if you look at the 10 top days of volume, the average volume is about $2 billion while the remaining days outside of these top 10 events is roughly $100 million. So, it's 20 times higher in terms of volume of trade. Now, the question of course is what exactly happened during these events? Is this just driven by noise traders or there's some directional trades that shifts one type of stable coins towards the other following these

kind of events. So, we are mostly going to focus on these major events in which there's significant significant trading volume. Now, you could actually see what happened within the pool when you basically look at the composition of the different types of assets in the pool over time. So, the blue represent DAI, USDC, and USDT. If in in in equilibrium, if they're all pegged to the same underlying asset, the dollar, the only difference between them is the under in the the particular mechanism that enables the pegging.

So, if the market views them as complete substitutes, we should expect that the pool will be fully fully balanced. So, we'll have a third of DAI, a third of USDT, and a third of USDC. But, what you can see that there are periods in which there's significant deviations from these equal proportions of the pool. For example, over here, this is when Terra collapsed. And what you can see is that basically uh traders and liquidity providers withdrew both die and USDC um such that uh and and basically they added more USDT in return.

Okay, so the proportion of USDT increased from about 30% of the pool to almost 80% uh suggesting that markets reacted quite dramatically with respect to the Terra collapse. And you can see that these periods um the you can see the other events here as well. So this is when FTX collapsed. This is when Silicon Valley uh bank collapsed where basically um uh traders withdrew almost entirely uh the USDT from the pool dumping USDC uh um during as as there were results about the ability of USDC to sustain their peg.

Should I view the um that when the uh relative proportion is exactly a third a third a third, that means the price is equal. But if it's not a third, that means that some coin is worth more than the other even though they're both US dollars. No. So so the the AMM uh the invariant of curve basically maintain a price of one um even if you're deviating from a the equal proportion. So um and I'll come back to that towards the end. The curvature um so if we're just looking within the decks, the the the curvature is once you get to significant deviations from the balance, only then you start to see deviations from the $1 peg. Uh so this

is why and I'll talk more about that uh Curve has various mechanisms in which they're trying to push both liquidity providers and traders back to balance, but deviations from a from uh price stability, uh you need to see only uh you need to find only extreme movements. So for example, uh if we're looking at the USDT prices um uh relative to this So you see that USDT moves in this event, but there are other events in which USDT moves, but you barely notice these, uh, price effects. Uh, similarly, if you look at USDC, again, the proportions of USDC change quite a bit, but the pegging in terms of pricing happens only in, uh, when when the pool really moves away

from uh from uh a balanced proportion. Yeah. Wait, from from this graph, it looks like USDT somewhat recovered, even though this is not the case, or is this the case? Like You USDT recovered. Yes. So, so it's not UST. So, so it's not the Terra, it's the Tether, uh, stable coin, which is actually which is backed by fiat, uh, um, by cash and cash equivalent assets. Okay. So, um, So, what I want to to do next, I'm basically I'm going to show you a bunch of pictures.

Um, basically, I'm I'm going to look at different, uh, event studies that, uh, are triggered by different types of risks. So, for example, when when, uh, first I'll I'll show you, uh, events that basically, uh, there is major, um, uh, increasing uncertainty in the crypto market driven by the collapse of FTX and Terra. Then, we'll move to explore what happens with the third-party risk, uh, the collapse of Silicon Valley Bank, and then some regulatory risk, and I'll we'll basically zoom in on this picture that I showed you, and basically see how market, uh, preferences to different types of stable coins emerge.

Um, so, in fact, I mean, in my research, I kind of like I tend to run regressions and various econometrics, but this case, I mean, the pictures kind of like are sufficient to tell the story. Um, so, for example, when when Terra collapsed, uh, you see that up to the event, uh, of Terra, the pool was fairly balanced. So, you This is basically the percentages here represent the proportion of of USDT, USDC, and Dai in the pool. But, when Terra collapsed and there was significant uncertainty in the market, you see that basically traders and liquidity providers, I'm not I'm not distinguishing between them yet, started to withdraw

USDC and Dai from the market and the proportion of USDT increased dramatically from about 30% to almost 85-90%. Suggesting that, you know, Tether was There were a lot of concerns about the quality of the assets of Tether, and which also was particularly not transparent about the quality of these assets. And during periods of significant uncertainty, the market was concerned about the ability of Tether to maintain the peg and sustain redemptions. And as a result, it seems like the flight to safety here was basically towards either Dai or USDC, both much more transparent

stablecoins that it was much easier to evaluate the extent to which they might be able to sustain um redemptions if needed. So, you see that basically traders tend to shift the type of still they're interested in holding stablecoins, but the type of stablecoins, the mechanism in which it's sustained peg is entirely different. Now, if you are interested just in understanding the traders' behavior, the liquidity takers, you can see basically similar story when the You can see significant decline of significant decline in the amount of USDC and Dai available in the pool,

which is exchanged by uh USDT. Okay, so the proportion the the amount this is here in in millions of dollars, you see the increase of of USDT. Question. So, um you could imagine sort of shifts in demand being caused by things other than just sort of risk attitudes, right? So, like one concrete example would be like if the price of ETH is plunging and everybody has these over-collateralized loans on Maker DAO, all of a sudden they really need DAI desperately so they can sort of close out their loans. Right?

It's not so much that they think the other stable coins are bad, it's just the Maker only accepts DAI. So, is there going to be any effort to try to tease apart these different things or Yeah. Yeah. Yeah. Yeah. So, it could be that um also the yield you might get on different types of stable coins in other instruments are kind of different. Uh that also may shape your preferences. Um what we kind of like did so far is basically we tried to to look at the yields um for example on Aave how they react to these kind of events and whether they can explain these kind of shifts. If anything, we find we find um um we find the opposite. So, first of

all, the magnitudes here are are are enormous in terms of you can see kind of almost a billion dollars of USDC and USDT are basically being taken out of the pool. Um so, we can also So, one thing one way that we could explore directly what you're saying is basically see where this money is going. Uh so, do we see that, you know, if there are suddenly very attractive yield somewhere because of these changes, we might see that these uh these stable coins are being directed to various DeFi protocols. So, we can actually trace that. Um the the second thing is that when we were just looking at yields in Aave, we didn't we didn't see that this can really explain. If anything, it was going the other direction. But, there's still kind of like much more that we

need to do uh in terms of tracing uh for what what were the purposes in which uh the you know what these wallets basically did with these assets after withdrawal. Yeah, cuz yield optimization doesn't really seem like a good fit for looking at these sudden price shocks, right? I mean you know, when Terra's crashing, I don't think too many people are worried about like getting 5% instead of 4% on a lending protocol, right? Whereas this idea of like you actually need a specific stablecoin cuz it's the only one you can use to close out certain positions, right? That is an in-the-moment kind of effect, right? So Yeah.

Yeah. Yeah, interesting. Um Yeah, so so that would be in in the context of Maker DAO. So we could definitely uh, uh, uh, at that in particular. The explanation for USDC might be uh, need to think about you know how that would work there. something you could do is like after the swap does the stablecoins that they swapped into, do they do anything with them? Or do they just hold them? Exactly. give you some indication, right? Exactly. Yeah. Um But the interesting thing is that the the the this this story uh, you you basically see a very similar picture that emerges uh, when when FTX collapsed. Again, one thing that is not clear to me is is

first of all is why USDC and DAI kind of like are traded in this like, you know, almost perfectly together within the pool um during this time period. I guess there were some uh, some stories that you know, DAI was kind of like almost entirely backed by USDC. So then maybe that was kind of like led them to be traded so similarly. But in any case, you kind of see again that when uh, FTX collapsed, again we see major shift in the composition of the pool um where uh, um traders are basically taking out um, uh, USDC and DAI in return for basically dumping into the pool um,

uh USDT. And and then and again you see kind of like this swap activity and how it affects the quantities. Um Now, the another story is basically looking at regulatory um activity. So, that's kind of like a different types of risk. Um So, for example, Tornado Cash, um which is a decentralized mixer, uh was banned by the US Office of OFAC, um and arguing that it had played a central role in laundering more than 7 trillion dollars uh 7 billion dollars. Um Following the ban, Circle started to freeze and blacklist addresses that were associated with with Tornado Cash mixer, and that

created the surge of concern about privacy, the autonomy um of being able to use USDC. Uh it raised concerns about what are the next steps of um the US Office of OFAC. Um Subsequently, interestingly, uh Tether uh responded to the US regulations or the US sanctions by claiming that, "Well, we First of all, we are not supporting American customers or conduct business in the US. Therefore, we view this as a uh we might voluntarily choose to do that, but we feel no obligation to comply."

Um What this triggered, as you might suspect, is um a significant withdrawal of USDT out of the pool. This kind of concerns about additional sanctioning of USDT um led to a replacement of USDT with Tether. So, again, kind of like regulatory arbitrage, where the where in this case Well, it's a little bit harder to see, but you but there's this kind of like replacement of different types of digital representation of the underlying US dollar, but now moving to assets that are less subject to this uh regulatory regime of um in which uh the USDC was forced to blacklist

wallets associated with Tornado Cash. In fact, you see very very similar story also when the SEC sued Paxos. Um So, this is very recent. So, when the SEC announced that it intends to sue the the stablecoin issuer Paxos um because it was alleged that the Binance USD is in fact an unregistered security because of various privileges Binance provided to holders of BUSD, Paxos in response halted minting the new BUSD tokens.

Um at the same time it raised concern about okay, what does SEC do next? And again, you see this regulatory arbitrage where there was again in the within the three pool, there was this substitution where there was withdrawal of USDT, which is less likely to be subject to the US regulatory changes. So, again, this kind of like regulatory risk seem to shift what kind of stablecoins traders were interested in terms of holding. And then, the final thing, yeah.

I In all these graphs, like, you know, USDC and Dai are really closely like correlated and USDC is not. Is that Is it always the case like even on normal trading days that US DC and Dai are very correlated then? Yeah. Yeah. So, we we need to to further check that, but I think that the correlation of USDT, USDC, and Dai increased dramatically over the last few months. One speculation is because they kind of like heavily used USDC as as a collateral. Although, I checked this morning and that was not the case. So, I don't really have a good answer for you right now in terms of how come these two are such so highly correlated uh the the three pool, but we we need to

check also whether there are is highly correlated also outside. Um So finally uh um while there are benefits from um having these kind of off-chain assets in reserve that are either cash or cash-like securities because of the increased stability, you also have to depend on a banking partner that will store these assets for you off-chain. Uh the Silicon Valley Bank collapse was kind of like a really nice uh you know, not nice, but it was kind of like a reminder of like of the third-party risk uh in which stablecoin might be dependent um this kind of stablecoins might be dependent on the traditional finance

financial system. Uh when Silicon Valley Bank uh experienced a a traditional bank run, um Circle announced that in fact um about $3.3 billion of its reserves is in fact stored out of the $40 billion is in fact stored in reserves um at uh Silicon Valley Bank, and there were significant concerns about the ability of Circle to um get access to these assets. And naturally, once you're subject to this kind of third-party risk, immediate uh uh as you you would expect, you see that within the three pool, there was immediate withdrawal of USDT as a

replacement um of USDC. While the amount of USDT almost got to zero, then you also started to see that people were starting to pull out DAI um uh just to substitute to substitute USDC. And then, um around March 12, the the US government announced that all depositors will get their money back, so kind of like the uncertainty uh um subsided, but again, you can kind of like see how the third-party risk led to this massive, reallocation and this kind of flight to safety of of stablecoin traders.

Okay? So, just to summarize, so I'm going to um, why do I think so? So, now I want to talk about the liquidity providers, but but basically the point of going through these event studies is just to illustrate that first of all, there's this kind of flight to safety that various types of events that are driven by aggregate market risk, regulatory risk, or counterparty risk are leading traders to change the type of stablecoins they want to maintain. So, the mechanisms through which stablecoins maintain their peg um, expose them to different types of risks.

And again, we don't find that there's a single dominant stablecoin design that holds all these different types of risks. So, every time we find that the market seems to kind of like jump to different types of stablecoin depending on the nature of risk. What's also quite important is that once there's a run to one type of asset, it's exposed the other stablecoins that are deemed as less safe to increased run risk. Okay? So, so that kind of like generate this kind of negative externalities in the market.

So, so and and basically this notion of competition between stablecoins during this period of stress in may impose this additional increased run risk. Now, you might also argue, well, um, this nature of flight to safety is just a feature of this emerging technology and maybe this is just early days in stablecoins and that therefore this might disappear. A counterargument may suggest that flight to safety may persist as long as there are differences in regulatory risk, reserves quality risk, and counterparty risks.

As long as kind of risks maintain, then we will continue to see this kind of flight to to safety across different types of stable coins. Okay, so let's let's talk about liquidity provision. So, yeah. In terms of being early and like a technology risk, do you think that a How likely do you think it is that that eventually some of these will just collapse and the market will consolidate around like stable coin? Because if there are different types of risk, but they show up sort of stochastically, you might expect like eventually one of them might collapse and then or two of them collapse and then you're left with Yeah.

the strongest of Yeah, so I think that definitely it seems like that there's consensus among regulators in the US and Europe, you know, both against algorithmic stable coins towards, you know, fiat-backed types of stable coins. Um Um so it's it's still conflict to be to be seen how regulations across different jurisdiction and to what extent the different jurisdiction will actually collaborate with each other to ensure that there is conflict one standard of stable coins. As long as there's no collaboration across jurisdictions and different jurisdiction enforce different um standards, then we might still see this kind of flight flight across different jurisdiction. We might still see

variations in quality of reserves and and and this kind of risk may may continue to persist. Okay, so I'm not going to spend a lot of time about Yeah. So so you you concluded the last slide talking about the different types of risks. Do any of your examples, I'm curious what you mean by counterparty risks in particular separately from what you would have from reserves quality risk? Like your circle example, that's also a counterparty risk or just reserve quality risk or Oh, so you mean you You the Silicon Valley Bank example?

Oh, that's the one you meant. Yeah, yeah, yeah. So, so basically, when you have off-chain assets, you have to rely on traditional banking uh in order to you know, custody your assets and and and therefore you're subject to risk that these banks may suffer some significant shocks. So, that's what I mean by I guess the custody eventually translates into reserve quality risk, right? Exactly. Exactly. So, it's kind of like it's a subset of uh so Yeah. I mean, the reserve quality risk Ultimately, everything boils down to the to the credit risk, right? The bottom line is can you uh sustain redemption requests um and the counterparty risk um

and and also the aggregate market risk may also influence the volatility of the assets that you have in reserves and might kind of like expose you to ultimately different types of uh credit risk. Or the transparency the lack of transparency is also a sub is going to um increase concerned about credit uh credit risk. Right, I guess the distinction I had in mind, you could imagine a counterparty risk which is not that You could imagine having counterparty risk even when they're fully collateralized and they just don't like you, right? Versus right, so there's a distinction of like can you not redeem because they literally don't have it or can you not redeem because they're sort of I don't you know, they're sort of censoring you in particular.

Yeah. So, those seem distinct to me. Yeah, so so So, for some reason they decide to ban or to ignore your request um Right, which I don't think is there in any of your examples, but in principle Yeah, yeah, at least empirically we don't have it, but potentially I mean, you can imagine that you know, if there were you know, centralized protocols that at some point they will freeze any redemption request or al- al- I mean, to some degree I mean, this is kind of like these uh there are many example that you know, in finance you can think about that you know, banks provide uh differential treatment to customers or those who are of like they do with them frequently business and and and when you freeze assets

because there's conflict bank run on the stable coin again you might conflict if it's not operated through smart contracts there might be concerns that you might there there might be kind of this kind of preferential treatment which is interesting we we haven't observed yet this kind of example but this is kind of interesting concern um for example the way that circle works they conflict they don't have the the market they don't operate with you know permissionless set of arbitrageurs they select a few and then you know you might have problems with one of the arbitrageur might get into difficulties in terms of their relationship with circle so that's

conflict might be definitely an issue um interesting point okay so very quickly about AMM so we talked about it yesterday so I'm not going to talk about too much but for example the constant product AMM you just need to maintain the same level of liquidity as as the pair as trade traded pair is traded against the against the um against the curve so imagine just a quick example so imagine the situation where we start with balanced pool that the the price of B is effectively is one A so they're fully balanced um now let's assume that there's a some

price shock and the price basically drops to price of B drops to half A so just there's this kind of aggregate market shock that leads to to decline of the price of B basically what will happen in order to get to the price of half A what we're going to see is that in form traders will will basically will will sell eight B coins in return for six A coins and generate profit along the way and move the the the the decks to to this kind of point where the price is going to be half A. Okay, so um Now, people typically tend to talk about

I mean, we talked yesterday about lever. You know, other way that people are thinking about potential losses for for liquidity providers is this notion of impermanent loss, which is you're comparing the pool value in this new point relative to a buy and hold strategy. Um so, what happens? The pool value is basically in the new point we're going to have 14 types of asset A and 28 types of asset B, which are now worth half of asset A. So, the the value of the pool is going to be 28 A.

But relative to the main problem is that in a buy and hold strategy, you basically you didn't change the quantities. So, notice that basically when you're trading along the curve, um the the quantities are shifting that the liquidity providers now hold more assets that are worth less. Okay? And and that and as a result, if you were just hold applying the buy and hold strategy, you would still have 20 A and 20 B. Um in on the on the curve, now you have 14 A and 28 of B, which is now worth less.

So, if you think about that in the context of stable coin, what happens here is that if there are concerns about the price of the other stable coin, basically traders will come in and take out the high-quality assets and replace it with the lower-quality assets and as a result, there's this kind of adverse selection problem that informed traders are causing losses to the liquidity providers because they're draining the good stable coins and dumping the bad stable coins. In fact, if the price further collapses to 0.2a, then you can see that the new point on the curve is is is you see even more extreme example where where the um

um the traders continue to drain uh the pool from the high quality assets that maintain the peg and now the uh the liquidity providers is left with 45 uh coin B type of assets which are kind of um uh are not sustaining their peg and only nine assets that sustain the peg. So basically the when there are informed traders and this is kind of like a general results also from more traditional finance um if traders are informed um then passive liquidity providers may suffer significant losses because particularly in the example of stable coins, informed traders will come in and exchange um

uh high quality stable coins with stable coins that there are concerns about about their um uh about their peg. So so they kind of like face this kind of significant adverse selection and as a result um the the the the mechanism will try to compensate liquidity providers um uh such that uh they will get a significant share of the fees. Hopefully the fees will um uh will be sufficiently high to uh outweigh the potential losses that liquidity providers are facing when trading against informed traders.

Now um Curve in particular has various interesting mechanisms in which they're trying to maintain the balance of the pool. One thing is the invariant. So as I mentioned earlier, uh the invariant of Curve has significant linearity that effectively almost mechanically, if as long as you're trading against the the the curve in these prices, remember from yesterday's talk, the price is basically the derivative of the curve. So, the derivative is going to be equal to one in this linear part. So, basically you're going to sustain price stability even if you're deviating from the balance of the pool, and the prices might start to change only in extreme

cases. The second mechanism that Curve applies to try to maintain the the the pool balance is that they enable one-sided liquidity provision. In fact, they will give you bonuses if you provide liquidity on to the least demanded coin. And finally, there is also mechanisms of dynamic fees. When you're deviating from a balanced pool, then Curve will increase the transaction fees.

Again, hoping that this might increase the revenue for liquidity providers, and as a result, may may encourage additional liquidity provision. Now, the question ultimately is whether these kind of mechanisms on Curves are sufficiently to offset adverse selection losses that are associated with these contract major events. Yeah. Do the fee increases trade in both directions, or just trades that go even farther out of whack? As long as you're as long as you're farther away from balance.

But so, suppose a trade Suppose there's a trade that comes in that's going to make it more balanced. Do they pay the higher fee? Um I think I think as far as I remember, it's just kind of like on distance from the balance. Um I'm not sure To be honest, it's very hard to understand from Curve documentation what's going on there. So, I'm kind of like this is based on like if you read the Curve V2 white paper, it literally end with this kind of like equation. That's it. That and Yeah, so that's and this is kind of like the representation of this equation. But other than that, they don't provide any documentation. So, I'm kind of like but it's true that the actual transaction

fees tend to fluctuate quite a bit depending on when you depending on the distance from from balance, but you know, this is kind of like as far as the um Okay, so so now so the the question here is kind of like empirically are these mechanisms that try to ensure the balance of the of the pool is sufficient to compensate liquidity providers? And the answer is kind of like quite strikingly no. So, um this is simply the volume of trade during this particular period. So, you would think that during this particular when the volume increase, this is during these particular

episodes, then the revenues from fees increase quite dramatically. So, you would think that well, this should increase liquidity provision, but unless there's significant losses associated with with this kind of adverse selection. And what you can see that every time there's this kind of like a significant period, you see a massive removal of liquidity. So, the the value locked in the pool shrinks quite dramatically, suggesting that whatever compensation Curve offers to liquidity providers is not is is quite far from being sufficient to entice liquidity providers to come in and and

provide additional liquidity. And in fact, you can see this very clearly. For example, if we look at the collapse of FTX, um in the blue line, you see the swapping activity, and the red line, you see uh liquidity provision. Um and you see uh that the there's a significant decline in uh liquidity provision um um around the time that FTX collapsed. So, exactly the the during times that you would think that liquidity is most needed to lower transaction fee or lower slippage, um in fact, we see that markets tend to almost um to to contract by significant decline in

liquidity. Um you kind of like see similar uh reaction also when we're looking uh for die decomposition. The reason why I'm I'm I'm showing you that for every coin separately is because you you can provide uh you can provide a one-sided liquidity for one of the tokens. You don't have to provide an equal proportion uh of uh of liquidity in in curve. And finally, you can you also see that um uh there is also removal of liquidity also when uh for USDT. Even though there's there's almost simultaneously, uh you see that people are um trying to uh avoid um

avoid um they're basically dumping USDT um in the market, but then at the same time, they're also eliminating liquidity of USDT uh in in the market. One one interesting aspect, I mean, again, if you look at the uh Terra Luna crash, you kind of like find very similar results. One kind of like interesting exception is that you see that um for USDC, when uh Luna crashed, there was an initial spike in liquidity provision at least temporary for a few days, but then as kind of like losses accumulated, uh liquidity was removed nevertheless from the market.

So, but but kind of like it it doesn't matter how we look at that. You kind of like see that liquidity provision essentially disappears uh during this kind of uh events. And this is also you see also similar decline in liquidity also during Silicon Valley Bank and and across uh basically all all of the uh all all other cases. The the last point that uh so so basically what we find is that despite the increased revenue from trading volume, despite the increased fee structure due to deviation from balance, and despite the increased incentives to add liquidity to least demanded uh stable coins, um we we we see across the board

removal of liquidity during periods of significant market stress. Um so, it seems like these kind of incentives are not sufficient. Uh this is um quite interesting. There's a quite interesting distinction from the operation of uh you know, market making during traditional in traditional finance, where basically the spread is going to react to market forces. Uh liquidity providers in in uh order limit books will basically increase the spread when there's when they're facing more informed trading.

But in curve, these parameters are basically set you know, through the DAO or what So, they're basically just set. It's not these parameters of the nature of the incentives are being set exogenously and not as reaction to market forces. Definitely not in the few days right after the shock happens. Um and the lack of um market forces shaping the incentives, uh we see that basically these parameters are not set, you know, empirically it seems like the parameters are not set sufficiently to offset concerns about adverse selection and informed trading uh during these these periods.

Uh the last thing that I want to show you and I'll just mention that very quickly. Sorry, yeah. One thing, I mean, in fairness, right? I mean, the bid-ask spread may well be infinite at equilibrium, right? By market makers if the volatility is like sufficiently high, right? I mean, which is kind of the same as all the liquidity being withdrawn, right? Yeah. Yeah, no, in it's not clear that there is a set of, I mean, of course, let's say let's imagine a world in which you dramatically increase transaction fees, that might also kind of like, you know, as we heard yesterday, that might generate significant other consequences uh for arbitrageurs and and so on. So, it's not clear that there's a there's a clear mechanism that would actually solve this problem

or there's a set of parameters that could offset um because unlike, you know, for stablecoins, unlike maybe ETH or stablecoins, uh the the outcomes are incredibly binary. So, you're not concerned that, you know, the price will drop by um you know, 10 or 20%, you're concerned that the price will drop by 100%. So, in such scenario, it's not really clear that you can design sufficient incentives to offset these kind of concerns. So, I I completely agree which, you know, it's it's an interesting question given the magnitudes of these events whether you can actually design parameters that would offset these the these kind of risks.

Yeah. Actually, those pools look like they're actually designed to keep this balance like when the fluctuations are low, but actually like aggravate things when the fluctuations are high because it's like linear in the middle of the function, but it but suddenly the derivative Yeah. That's a really good point. So, basically what happens is that because of lack of curvature, it makes it very cheap to withdraw to drain the pool effectively from the good assets. If there was more curvature, then the price changes would make it more expensive to to remove assets and maybe that might limit the extent to which you know, we see this sharp reactions in

the pool. So, there's this trade-off. On the one hand, you want to maintain price stability, but on the other hand, yeah, there's absolutely this kind of consideration which kind of like brings this question of, you know, is this what is your I mean, maybe rest what we talked about during lunch, which is kind of like what exactly is the objective function? Are you trying to minimize slippage and transaction costs or are you trying to minimize run risks during periods of significant stress where you maybe you'd like to increase the cost of removing assets from from the pool.

Yeah. My take from what you've said so far is sort of a little bit more optimistic in that I mean essentially you're trading here nickels for dimes, right? Two nickels for for for a dime. And of course, when there's some threat about oh maybe dimes are going to be withdrawn from US currency, so everyone shifts to the nickels. So, you see these massive quantity changes, but and despite the decreased liquidity and so forth, nickels are still trading two for two for a dime, right? The prices are way more stable, right? So, you have massive quantity shifts and only tiny tiny price shifts and it's the price

shifts we care about. So, like isn't this really actually showing hey, stable coins are actually pretty stable? Yeah. So, so these are um we we we focus purposely on the three most, you know, stable stablecoins out there. Uh if you venture a little bit outside uh to stablecoins 5 to 30, then then you see that actually the price impacts become dramatically much more uh volatile. And and secondly, um you know, the extent to which, you know, these these stablecoins survived, you know, they were depegged for some time and ultimately recovered. I mean,

I I think I think yeah, so on the one hand it's kind of like illustrating that um you know, market forces were able to kind of like converge back to the peg, which is it's it's a really important point. Uh on the other hand, you know, it's kind of like these mechanisms and market forces seem to nevertheless um embed significant volatility in the quantities. You're right that the quantity fluctuations are going to translate into prices only in the extreme cases, but still um uh these kind of like you you see that the quantity uh fluctuations are pretty volatile uh over time. Now, you know, to the extent so it kind of

like illustrates at at the least, you know, and I'll conclude here because we're out of time. So, um at the least what we kind of like find that there's significant flight to safety. Um and there's the flight to safety is driven by different types of risk. So, it's not triggered by one type of risk um all the time. And this kind of like this imposes run risks on the least preferred stablecoins that is amplified by removal of liquidity. So, these are kind of like forces that go towards um breaking the peg. Uh the fact that it didn't ultimately or the peg recovered, I think is is kind of like is comforting, but it's not

suggesting that these kind of forces are not important if we're thinking about ultimately stability. Um Uh right now the the you know you can also think about, you know, well, this occasional volatility or depegging could be sufficient to prevent from the use of these stablecoin users' money, right? So, with with when we use money, we don't ask questions, well, is there a kind of like uh particular features that are risky or not? We just use it. And so, as long as we're not there, that might be also a a concern.

But, this is we can continue this conversation. Um Finally um you know, um the result that I didn't get to show you is that we find that sophisticated and deep-pocket wallets uh do not seem to differentiate uh differentially provide liquidity. If anything, they seem to run first by removing liquidity from the market. Um so, super early, looking forward to talk more to you. We're still trying to figure out how to um you know, continue uh move forward with this project. So, thank you very much for your comment.

And now, about the very last point. How So, you made it sound like it's not su- I would have expected them to really be the first ones to flee. And you But, you what you said suggested that it that was uh not very noticeable. Um so, this is what the this is the results where we still kind of like uh we we we still need to make more progress. But, what we did so far, for example, in the case of FTX, we look at the value of the wallets at the time of the collapse, and we uh said, "Okay, let's see the reaction of the top quartile in terms of value of the wallets relative to the others." You can see that in terms of Here, we we we see the liquidity uh removal.

And you you actually see not with what you said, you see that those that top 25% the top quartile based on the value of the wallet, they seem to remove liquidity first. Um and that seems to be consistent also for die and and and USDC. Um and and and in other way we we're looking at these wallets is based on the level of sophistication. We look at, you know, how complicated the transactions that they tend to execute on curve. And again, we see that they seem to react first uh and withdraw. In some cases, they seem to kind of like react similarly, but if anything, they seem to be on average move much much faster. Uh the other thing, you know, that was suggested to me earlier today is also to see whether exactly those that remove

liquidity um are also those that swap um uh at the same time. So, if you want to avoid USDT, you're not only going to avoid by swapping to USDC, you're also going to remove the liquidity of the market. So, this this kind of like going to be uh almost kind of like a double impact on on on the stability of USDT. And then finally, we we actually know who are the wallets to provide liquidity. We can see how concentrated are the liquidity providers and whether the liquidity removal is driven by kind of like the big the the the big whales so so to speak or driven by more smaller liquidity providers, but which is not

consistent with what we find here. Do you have any sense of the geographic region of these wallets or, you know, any sense from the data or anecdotally how those trends vary across different jurisdictions? Uh no. No. In fact, if you have any ideas how I can identify uh the geographic location uh of these wallets, that would be super interesting. I guess like something like Chainalysis comes to mind, but Yeah.

But that would be interesting. Yeah, time of day. Yeah. Time of day, that's right. Yeah. So, what what hypothesis do you have in mind? So, Well, I'm thinking of um some of the promises you laid out at the beginning. Yeah. Yeah. Yeah. there's a lot of rhetoric and some anecdotal evidence of stablecoins being highly adopted in places like African continent for these reasons. So, it could be interesting to see a lot more fine-grained like between um stablecoins and like derivatives Right. different points in the market. So, yeah. So, time of day of trading could be an interesting way of looking at that. In general, I think that it's really interesting to see where it's adopted and how it's being used. Um If you have exchanges or different

different countries have different exchanges that are popular like central exchanges that are popular with them. If you could link like central like sexes to what's trading on the decks. Right. Right. Um Cool. I have two quick comments. Can everybody hear me or not? Yeah, we can hear you, Sam. Cool. Um the first is on the USDC die pegging. Um there is a die has a thing called the PSM, the peg stability module, that allows a one-to-one trade. It is capped. I think it's up to about $850 million at this point. Um so, that is probably why you see that that that it's um because there's a risk-free arbitrage created if it

deviates in any other market up to $850 million of liquidity, but that's pretty deep even for a big crisis. Um and then the other thing would be at all of the times of all of these crises, you USDC did back about uh 45% of die. That has decreased over time, um but but at at the time of all of those events, it it was back. Yeah. Yeah. Interesting. Um and then Tim, to your comment on on curve fees, um I think the fees are based on the the price at the end of the trade rather than the beginning. Um so, trades towards the center or towards rebalance um have lower fees because the last price

Yeah, makes sense. Thanks again. Yeah. Thanks a lot, Sam. Awesome. Thank you so much.