About the talk
I am an applied math PhD student at Cornell University. My work is at the intersection of computer science and economics and aims to better understand complex economic systems with the goal to design systems that survive extreme events. I develop network science and machine learning tools and multi-agent models. My current work focuses on cascades in financial networks (e.g., understanding how financial systems can fail in unituitive ways) and the design of stable assets, including 'stablecoin' cryptocurrencies. Before starting my PhD, I worked in financial software developing fixed income risk models and economic classification tools. I have a wide knowledge of how pieces of the economy fit together. In an early project, I spent several months reading 10-ks and designing a business classification system. I later pioneered the firm’s network model based on this data, which motivated my PhD work. My research in undergrad focused on computational aspects of elliptic curves and modeling and signal analysis of neuronal networks. Before mathematics, I also worked on research in physics, geology, chemistry, and engineering. I am familiar with several programming languages and software packages: Python (including NumPy, SciPy, pyodbc, Pandas, NetworkX, and Google OR-Tools), Matlab, Sage (computer algebra), VBA Excel, C#/VB.net, C++, SQL Server, Bash, LaTeX, and Bloomberg. Public code repo at github.com/aklamun.View the profile
So Araya clock is moving to is a PhD student in applied math, Cornell whose work focuses, on stablecoins, defy design, applied math, complex systems and network Cascades. Do today, stock is entitled, The unreasonable design of stablecoins, Mariah. Take it away. Thanks for the insurance. So the past the past year has seen a many types of stablecoins are answering their ranks on the custodial side. We seen some of them are nice cars that are we predicted kind of before emerging as a
side. We seen you sound like a broad range of new categories, but how disable coins ficht? Well not very well at least a lot of them. A lot of them except. I'm also failed this year. And many others have encountered problems with their design in the course of logic. Cincinnati basketball a fight with this graph of a of algorithmic stablecoins. They're launched this year, with most really failing to maintain a pegged in the long run. So this talk is going to focus on this non-custodial, side of stablecoins and also be five more generally on front of like, finding what are the fundamental design
problems that need to be confronted. How do you actually designed these things? Well, and will see that. There are three main design problems. Your one called technical security, one called Economic Security, and a third called economic stability. And then we'll walk through a bit of a firework beginning to explore. It will draw from the work. The work here with the with many grapes Co operators. And I'll put these bags out publicly later so that you can do it before we really dig in. Let's go through just on
a surface level. What, what does a non-custodial stablecoin look like? What are the different parts that come into play? And it's drawing back to those different dimensions. We were seeing in the earlier before about all these different design possibilities of them custodial stable, So, one of the most important is that you have some sort of collateral value value back in the system, and this can take the form of exhaustion as collateral and dodging is collateral or implicits were really no collateral. And the idea here is making a distinction about. How does this collateral
to havenhouse circulate is, is that value with the with the health of the system itself? And then there's another category hear of a bit of a roll of risk absorption in the system for making a stablecoin. Somebody is absorbing this side of price rest in some form. And so there are few possibilities here. You could have an equity position that is observing risk, you could have agents coming into the system and deciding to individually absorb risk and then you can have the protocol itself and then there's another function here of
actually issuing the stable, and I'm sure you could have this plugged back in with the with the rest of sorts and have it be decidedly issuance of the algorithmic and then there's also sort of this like an elder is Mickey leveraging process that may be happening in many of these systems behind a bit further and some of the latest as well. And then you have sort of connecting in whitstable coin holders result of the actual and receivers of this issue is processed. And then you also have sort of needs for governance and price feeds
Governor's here, sitting like setting different hyperparameters of the system overtime and Fred feed naturally telling you, what is the, the real world value, as collateral that stuff back in the system and then behind the scenes, you have minors, who always have the ability to just answer a reorder transaction. And so from all of these different pieces, you can sort of see some areas where new risks in the tax might emerge and this is what we're going to be focusing on the rest of the top but there are certain areas where there is leveraging risks price feed and governance
manipulation wrist, extractor are coming into play. And then more generally you can just have bugs and how all these things are implemented. And these are things that are not incredibly well understood there today. So this brings us to what are the four? What are the actual fundamental design problems that we want to be tackling to understand? And one of these is technical security, and Technical security. I mean that are possible. They may be possible and these exploits are Atomic, and instantaneous implementing sort of technical
structure. And because of this atomicity, their kind of risk-free. And then there's Economic Security and this is a little bit different than technical security involved in manipulating some equilibrium over sometime. And because it has to be over sometime. If not discreet and then if you sort of have the system secure from a security perspective, you can move on to thinking, is this a, is this system actually economically stable to the incentives, leads you to the stable and actually having a cigarette. So let's start a claim. So
why is this? So it's risk-free because the outcomes of Eastside of exploit their binary for the attacker. Either attack is successful and they receive the prophet or it doesn't happen. Actually. Just texting is retarded and they only pay start of the year. The nominal cash. And there's some good examples of this, anybody that is atomic fits into this category, into this category as well as logic books. And the good news is that at least over the last couple years. Here. These are much much better than they used to be
and there are some good ways to begin addressing such history program analysis, and formal methods to specify. And some good examples of this about to come up and stable coins. In particular. There is the origin dollar flash loan. Explict there also a bunch of lending platform examples recently. So how does this compare with economic security? So is you remember, Economic Security was supposed to meet them this manipulation of equilibria over sometimes. And so really just sort of like, taking advantage of, you know, the technical
structure. There, may not be an exploit, that's possible instantaneously. But if you explode during its properties of this, these markets that come up over time as all my people in this market, then you may be able to sort of explain the technical structure in lieu of exploiting the delivery. And this is not risk-free for the attacker because there are tangible upfront costs. You have to actually do this. This Market manipulation before you can process it and it's not too, so you can't say they sound like they're transaction gets reverted if your back was not successful in the
end and this means that the there's a real possibility of attack is not successful and you can also see this estimate the market because you're All of the market to see. Can you actually perform this time in Appalachian prophet? I thought these things are much less tight and they include things like a governance extractable value, but workout gebre are the tax. Also include Market manipulation exploit and fundamentally to address these sorts of problems. We really need economic model with Avi systems and
agents work. Until we can't just do a program analysis because the program analysis and also have to build into assuming some sort of actual economic models. How the markets work. And so, will talk to you a couple examples, just to illustrate this one illustration, which is not a clear exploit. What does nicely illustrates? That was possible was in November 2020, compound and compound takes its its price of died from the or the dentist. I'm at least from from coinbase and in coinbase, there was a either manipulation or some sort of Market phenomena that occur
where the price of died spikes to a dollar Thirty instead of a dollar over just a 15-minute time. And because of this sort of thing, recorded in the compound price feed. I triggered a very large amount of liquidation you can count and why this could have been an exploiter why you could design and explain to take advantage of this, is that the Liquidation in compound. If you're performing those liquidations, you get an Arbitrage profit. And so if you can actually manipulate the stock market liquidation, you can potentially And this is essentially what then occurred
later. Just passed your night. Where was it could exploit in Venus. We're essentially the same sort of thing was happening in exploiting their their market for a squatter. So let's move on a little bit of talking about sort of like how we're helping to fill in this area of Economic Security. So we begin by by by formulating, some of these economic theory Express on Market manipulation attacks around liquidations that can also lead to mine, are destructible value. This is coming from her instability for the
blockchain paper, back in 2019 and a variant the best way to record in. I actually on black Thursday and using mental manipulation. And then we also formulate some potentially tractable, what we call for working models to be able to, to model out to you, and this is coming from our stablecoins. 2.0. And then we also formulated this new area of government governments extractable value what we call G-Eazy and just separate into two main types. One type is
actually the ability for Governors to make a short-term decisions that may be lead to short-term profits for them. But maybe not going to call and then also potentially outright malicious. Tavern expects such as run pulls are really indirect. And so we begin addressing the sort of area by formulating some the first models for remodeling is infected and these are based on capital structure models where we incorporate this idea of a price of Anarchy. So another words that you're making a
truly decentralized system that does not have the protections of the legal system with Angela price of Anarchy at the governor's incentives, need to be aligned with out. That's a, that's that outside legal system to play a role. And one of the problems that can arise here is that honest incentives are around for the flight. Attendant may not be enough to make the best decisions for the protocol and this leads us to rent in possibility conjecture about how some of these systems are designed today. And this is
so you consider a bank. If it's bank is unsecured, if it's Equity Value City, value of its stock is less than 2 times its assets under management or deposit deposit into your bank. And it's quite obvious that no purchase no deposit because it's their responsibility. If they want to participate to be boosting up. The bank's Equity value is just too enormous to make sense. And this is coming also from our stablecoins, 2.0 Paper. We formulate diesel models.
And then we also come across a potential solution to this in one of our recent blog post. And this is formulating what we call an optimistic approval mechanism B potentially bypass this impossibility. And this the second is only ideas that gives users the option to start coming from our blog post, you change the side of the system so that if the governor's are planning an exploit and actually perform it, then rational users will choose to veto this or this Anthony possibility to and The effect will be that the tactician performed and the
system is only worse off for the Governor's because less people want to participate because they see malicious governor. So these are the ideas of security and the rest of the talk. I really want to focus on this last topic, our work on economic stability. And this is kind of asking a question. Then after, you know, that the system is secure in terms of a house designed and how the incentives are set up. And in terms of time, you exploit the system and steel collateral. So I'm taking this for granted are the incentives actually set up so that
the issuance process in particular. Does it read the actual stable outcomes for the stable, How How likely is it that the stable going to be able to retain its intent to pick? And one of our main results here is around characterizing deleveraging spiral skin died. This is coming from two of our papers instability for the blockchain and well stability. Last. Tell the poor getting into these, you see leveraging. Spirals. Let's talk a little bit about how these leverage, they stablecoins like. I work. So it's parallel to a CDL structure.
And in this sort of structure. There's a portfolio of a landmass split this portfolio. And now it's their losses that are incurred their born first, by the junior truck and the senior tranches intended to be protected. And this is parallel for these leverage. They stablecoins, like die, in that. That unitranche is this risk absorber sort of role in the stable, going to CPS and I and disable coin. Holders are the two senior transfer intended to be protected from from Price fluctuation. It also has this edition of ADA
leveraging process, which tries to make this the city of like structure and two are a Perpetual instrument. And so, the idea is that if there are enough losses that I'm card and you try to downsize the actual Supply, a size of the system and supply of the guy in the system so that there's always going to be protecting what's left of the oven in your truck. And so in this your papers, we've been remodeling the price on a Max of the system and is kind of clear but maybe it's worth making explicit here in the original die.
System is supplies at Raymond. In this Lover's Market. Sort of what do you have? The CD PS? You are the fulfilling this junior Tron. Should I risk of a sort of role? They're coming in providing collateral in inside Heath and deciding to leverage in that heat. And so the guy they actually create is through their leverage decision and how much how much do want to borrow against my wishes importantly, a very risky decision. And so, to be conserving to be conservative, as they may not borrow up to the soul women, which is commonly what happened. And this leads
to, and I'm dodging his prices of died in the system actually because you don't have his guarantee that the supply is actually equal to demand at the dollar price. And so you may actually have supply problem and at different prices than a dollar at certain times. And it's also means that the traditional financial leverage models, don't really apply in this context because they don't have them for the feedback attacks built-in. And so we've developed some models that that helps you protect your eyes the system and they lead to what we call deleveraging spirals, which are essentially
short, squeeze effects, also amplify collateral Toronto and justifying for the stable and unstable regions for the systems and important. So it's basically set up a system. There's a collateral constraint and also to freeze freeze over this a little bit, but it is made formal. There is a decision then by the the stablecoin, if you decide to Sable, supplied by this, a b, c d p Speculator and it takes into account this potential liquidation effects that can occur from. Because if the value of their collateral Falls to to quickly, they may be liquidated in the system or they
may fall into a relief liquidate themselves to be leveraged and in doing so they can affect the actual market for. And if weeds to this busy leveraging sort of effect. So basically if there is a Liquidation in the system, it means that some of the collateral is being used to have to buy back the supply and reduce it. Which has an upward effect on price because now the demand is not really in line with the way with Supply and it is because you have to be buying this guy off at the market using, as collateral and
the price of appreciating because you need to be fed and balances Supply Second liquidation, because the price of a, of a tire and it is the second leading liquidation occurs. You actually need more of the of the collateral actually reducing supplier, which has been affected and of increasing, press further. We would have needed to bring in a spiral and is actually what what happened to a large degree to dye on black Thursday, which was one of the biggest So this led to
some realizations bit that you may not have a stable region. If if you don't have service like positive expectations around to use that is collateral types and it's seeming contradiction that are you really want to be centralized stablecoin. We can only make it fully stabilized by adding. I'm clearly a desk and these are currently custodi and patching. This has been a major topics in and there's been several attempts to do. So one is maker which Incorporated their tag stability, module or PSM. And the idea here is to set up was basically a
primary market for staple crying where prefer die or you can come in with a USB C. And you can mention you die by depositing that usdc or you could take a guy and redeem it for a USB C. And this has the effect of adding his most recent Just this past May cryptocurrency prices. We did not see the same sort of effect, Cindy leveraging, spirals occur because of this psms. But we did see the volume of TSM. I really growing. And now it's actually 60 per-cent of the year that died supplies, actually, backed by a
b u s d c. There's also Rye witches been Incorporated in negative rights to reshape the incentives in the settings. And if we ask you some questions around, like, what is equilibrium participation, is there expected to be negative rates for long. And also they definitely dressed for what's the weather going to look like? And then there's also stablecoins like a liquidity crisis. Well initially, there were a lot of redemptions and liquidity at the start of the crisis potentially because it's a new stable, and this was its
first. Its first and Redemption is a feature of the other look what he design. But then via the liquidity. I was really able to absorb the shock over time. But this had the effect really of postponing this, the shock and really smooth and the effects, but still having this sort of like price appreciation short squeeze like a factory. And then we also have a kind of a generalization. There's something she has to mechanism which is kind of our next to our next topic which is going to go into algorithmic stable. And this is something we're working on around
Reserve back to primary Market. So, with that in mind, the rest of the talk was so focused on thinking about algorithmic, stablecoins, and how would you design? This is coming from a paper that we was not released yet, but will be released very soon. So before we tackle, that let's run through a little bit of what actually back the currency tag. So there are two main sources of value that actually backed this. One of them. If you have some sort of thing, I miss is a tangible
and then you also have some degree of economics and potentially saying there's a reason to be holding. These are these currencies and it is to you need the currency to unlock access to this particular economy. Economy is a particular size. Then there's some real value to you are coming into the holder. And then, if you, if you have a $1 Target for the server currency Peg, if there's sort of enough of this a new sources of values stand, the pain can be suspect. But now if you have a shock to one of these sources of value from sincerity,
that it actually shifts how much value you have. And now the pegs match we break. And I want to stress that this is like a highly simplified so that you can read more about it at that precise modeling in this paper. Put out what actually that's algorithm you really start off with. No, native usage. There's no real economic data usage in the first place. But some of them often tried to be under backpacks at the beginning. And so, you really have the economic usage for this reason, the tag off in Brick.
And then you also have a question of what are the actual assets that are backing the system in some cases. This is potentially a senior care system and the value of these are endogenous so-called Equity. The problem with use this, at their value is quite circular for the health of the system. And so you may have a side and we're right. When you need these assets, the most they decrease in value because the system is in distress minute warning. And then you have another category of Reserve back to stable coins, where there's some portfolio.
So with this in mind, one useful way to categorize. Be stablecoins is first. Ask a question as a user. Coming into the system. They're paying $8 for a new stablecoin if it's actually a system is working. And where does this? What is this dollar actually? Go do it. One case, it goes to the park, with the pockets of the stakeholders. In another case. It goes to. The prison part way to use some Reserve is backing the system. And a third case, there is it's going to 100% Reserve back and decide.
There's some of no value working but system. And this is what sort of many bases type systems are essentially doing in another one. You have some small reserved parking, but it's inherently smaller than sort of this up. It's a hundred percent Reserve backing. And so, we end up seeing a lot of problems with with, with, with many of these systems because There's actually heading out to a lot of the assets that could be used for stability to the two stakeholders in the system. And a couple other dimensions that are worth considering also a one is again the composition of
the reserve. And another is sort of like, what is the, what is the protocol actually maintain liquidity? And this is kind of a final point of discussion here, which is around. How do you design algorithmic primary Market until here? This is essentially saying, how does the protocol actually maintain its liquidity and it makes its primary Market? Which is responsible for making and redeeming. And just primary Market has a Redemption trailer which is the price of redemption as a function of the system State. How are these Redemption trailer? So I designed what do they look like?
And I could look like I'm here plotting price versus the amount of Redemption on a dollar have various degrees of a curvature. And in the basis case you eventually have flat at zero. And so we seeing if that doesn't work. I don't really have time to go through all of his case study but u s, e, c, u s, d. T essentially have to fuck her about a dollar but it has other issues are there was recently an issue with Faye and you can understand it. Really in terms of implicitly having a bad shape to the production which
then led to put of like life Redemption and the inability to handle those Redemption. And then you have seen her reach their system where the problem is really that look up the value of this if I could really change depending on the health of the system. And so you have sort of like liquidity at a dollar up to a point. But beyond this point, I like what he may be exhausted and he may be left with zero. And the recent example of this is sort of the Titan and iron the iron steak. And then we have some work, which I also won't have
too much time to go into this year, for the market, in a more fundamental way with desirable properties. Some of the problems with the current space is that they're very at Hawk and really end up using governments to it to make quick fixes. And can I see? And that's what it motivates us actually like a desirable properties. Formulate a way to achieve those properties. And this is in our upcoming paper, designing our primary automated my computer. And that's what it's like in a way
where there's liquidity up to a point but there's our circuit breaker protection and she is a nice sort of like bounded loss for participants of the protocol. And so this is like some of the main properties here that we want bound and loss for the protocol and redeemers. And also importantly that sufficiently to get a boat launch. So that's really brings us to be at the end of the of the talk of the important takeaway really is bit. Like these are feeding the problems and models that might be going forward. And
if you're working in the Deep by space, you should spend carefully about which type of problem, you're tackling and what sort of models you really need to tackle it. And thank you for. Thank you for listening and feel free to ask me questions.
Buy this talk
Buy this video
Our other topics
With ConferenceCast.tv, you get access to our library of the world's best conference talks.