Video | Audio | Discussion
| References | Person | Text |
|---|---|---|
| # / 00:00:03 | Richard Brown | Hello, everyone. Welcome to the July 18th edition of the Scientific Governance and Risk Meeting at MakerDAO. My name is Richard Brown. I'm the Head of Community Development at Maker. We have a special guest today in the form of Alex Evans from Placeholder VC. We're anxiously looking forward to hearing from him. |
| # / 00:00:22 | Richard Brown | I'm anxiously looking forward to somebody trying to get me to understand exactly what was going on in that document that he shared because it went completely over my head. So, hopefully, there'll be kind of like a preamble for that. Otherwise, I'm just going to nod sagely in the background. |
| # / 00:00:37 | Richard Brown | I also want to keep things quiet as far as governance goes because I don't have a lot to go over and that's not because there's not a lot happening in governance. There is a fantastic amount of things in governance and all of that is going to be carpet bombing these meetings over the next couple of weeks of some presentations, some request for comments, some new forum debates about ways that we're hoping the community can come together and help us to optimize this process. Because believe it or not MCD is a real thing and it's coming soon, so we have a lot of work to do. |
| # / 00:01:11 | Richard Brown | This is one of the reasons it's so hard for us to get pinned down for a date on when an MCD happens is because this is a two stage process, basically. The first stage is to make our foundation, puts all the pieces in place. Many, many, many people get all their work finished and then the work depends on the community. Largely people that show up in this room, people that are showing up in the governance forum, but we have to rely on the community to help us do a great deal of work. |
| # / 00:01:46 | Richard Brown | We need to evaluate these assets. We need to talk about the processes and the policies. We need to talk about what the voting cadence looks like. We need to get piles and piles and piles of feedback. For that to happen effectively we're going to need to formalize and internalize some of the emerging process that's coming out of the forums and that's going to happen over the course of the next few weeks. To early to talk about it now. |
| # / 00:02:09 | Richard Brown | I posted a few different links in the group chat. One of them is the discussion thread for this call. We'll continue the discussion there after this has ended. David will provide us with a summary of all the salient points. There's a link also to the thread for Alex's model. There's also a link to something that LongForWisdom is turning into a forum rockstar for us has been posting ... is this the first of the second I'm not sure. I'm hoping it's the first of many [crosstalk 00:02:40]. |
| # / 00:02:40 | David Utrobin | Its one thread. It's one thread that he just updates. |
| # / 00:02:44 | Richard Brown | Okay. So, yeah, we're debating whether it's weekly or forever. Anyway, it's a very valuable 10,000 foot view of what's happening in the forums because the forums are taking off in ways that I can only have hoped for. Lots of really interesting governance stuff happening in there. |
| # / 00:03:05 | Richard Brown | Over time, we're going to find, I believe, that governance happens in that forum basically. And then, we talk about what happened in that forum in these calls. We surface some things, we prepare the community for signals, and we do some educational materials, but largely the heavy lifting happens in those forums. So if you haven't gone, I advise you to check it out. Tons and tons of valuable content. I'm hoping that it'll have enough bandwidth to get caught up with it maybe on the weekend. |
| # / 00:03:36 | Richard Brown | I want to talk about very, very, very briefly about something I discussed in last weeks meeting and that is emergent process. We have lots of things to do. I've already touched on that a little bit, but lots of things to do, lots of people that need to do it, and we need to figure out how that works. |
| # / 00:03:54 | Richard Brown | Happily that's already been figured out largely because we're not the first distributed team on the internet that needs to coordinate. We've been doing that for 30 years now. Its called Opensource. The tools are available. And so, this is one of the things that I want to start discussing more and more about, how governance works, how these calls work, how the forums work, how the community development group in general works. How that works is Opensource. The heavy lifting has been done here. |
| # / 00:04:32 | Richard Brown | We need to start internalizing this idea of Opensource governance, which is something that we'll get into a bit more. But when it comes to the mechanics of how distributed teams of people work on the internet, especially in our space, it's all been nailed down. People talk about things in forums. They talk about things in chats. They get together in GitHub basically. For better or for worse that's what we have and then they do their work. |
| # / 00:04:57 | Richard Brown | They set up a project. They put in some issues. They comment on those. They submit a PR. And then, that works goes into GitHub. And for our purposes the distinction between slow decode and documents, a framework, a model, a Python script that approves that model is a very thin distinction. So, we're doing the same thing. We have these tools available to us. |
| # / 00:05:23 | Richard Brown | As we increase our efficiency in these calls or in the governance process, as we debate these workflows in the forums, we're going to get closer and closer to aligning with what the rest of the people in the space do. That model I'm hoping looks like this. We have a few different venues and those venues serve very clear purposes. We have a forum. A forum is where the debate happens, where input is gathered, and where documents are initially drafted and hammered out. |
| # / 00:05:55 | Richard Brown | Once those documents are cool we move them into GitHub. And then, when they're in GitHub they become and issues and then that's the final process for adding people's inputs. It's sort of like an EIP process. Once that document is finished, the PR gets accepted, it goes into GitHub proper and that's where this GitHub becomes a document, a very loose content management system for us. It's a flat file document store, it's versioned. It's public. It has an audit trail. It has all of these things we need in the governance space. |
| # / 00:06:31 | Richard Brown | And then, once that has been placed in that canonical resource anybody in the ecosystem is free to fire up Gatsby or any other tool to ingest that in their own website, in their own portals, in their own tools, or for whatever purposes they like. |
| # / 00:06:47 | Richard Brown | And so, we have this cycle of request for proposals. Our proposal comes out from some organization or group of people. Then we move into our request for comments phase. Then we move into this version control committing phase. And then, we display that data in a presentation later somewhere. We'll be digging into this a lot more over the course of the next few weeks. I just want to set the stage for that. |
| # / 00:07:07 | Richard Brown | I am going to stop talking very, very soon. If you have a question, please type that question in the group chat. David and I or I will hopefully get to it in an appropriate moment. We'll ask that question or you if you don't have a microphone. If you have a microphone, please pipe up and ask a question whenever you like. Feel free to interrupt us or click on the discussion thread for this call. If you want to have a deeper discussion, just add your question to that discussion thread and we will dig into it over the course of the next week. |
| # / 00:07:40 | Richard Brown | I am going to stop there. Alex, I think I want to potentially turn this over to you right away if you're prepared. |
| # / 00:07:47 | Alex Evans | Sure. And thanks for that, Rich. Can you hear me? |
| # / 00:07:54 | Richard Brown | Yeah, I can hear you all right. I should probably have formally ... Well, actually I think I already did introduce you, but this is Alex Evans. He's from Placeholder VC. Placeholder VC, is that what you guys call yourself or just Placeholder? |
| # / 00:08:01 | Chris Burniske | Just Placeholder. |
| # / 00:08:01 | Alex Evans | Just Placeholder. |
| # / 00:08:02 | Richard Brown | Just Placeholder. |
| # / 00:08:04 | Alex Evans | This is Alex. I have Chris Berniske as well. He's a partner here at Placeholder. Placeholder is a venture capital for cryptocurreny only and also an active MKR holder. |
| # / 00:08:14 | Richard Brown | Cool. |
| # / 00:08:15 | Alex Evans | And so, as a part of that- What's that? |
| # / 00:08:23 | Richard Brown | I was just [inaudible 00:08:23]. |
| # / 00:08:23 | Alex Evans | I'll take that. I have to do the screens. I have to share the screens. I put together some slides. Let me know when you can see my screen. |
| # / 00:08:33 | Richard Brown | That works. |
| # / 00:08:33 | David Utrobin | Okay. |
| # / 00:08:38 | Alex Evans | All right, see it? |
| # / 00:08:38 | Richard Brown | Right. |
| # / 00:08:40 | Alex Evans | As I was saying, as part of our involvement with the Maker network we put together a risk model that we put on the Placeholder.vc website and it's also now on the maker thread on the Maker forum. And so, what I'd like to do in this presentation is to describe some of the thinking that went into that model and also some of the core mechanics and assumptions that drive it. |
| # / 00:09:06 | Alex Evans | Okay. |
| # / 00:09:08 | Alex Evans | I'm going to try to do that in four steps. First, I'm going to give a little bit of context and background. Then I'm going to give a brief overview of the model and some of the assumptions behind it. Then I'm going to provide an example of how to estimate the parameters in the model. And then, I'm going to give a example of how to apply the estimates from step three in the risk management example. |
| # / 00:09:29 | Alex Evans | Okay. |
| # / 00:09:31 | Alex Evans | Some context in motivation here. Now that I think about it, too, I don't think there's a need for a lot of context. I think Cyrus, Vishesh, and some of the members on this call have done a pretty good job of setting the framework on what risk is, how we should think about risk, how we should model risk, what are some potential frameworks that we can use to formalize our understanding about risks. I don't want to repeat any of that. |
| # / 00:09:54 | Alex Evans | I think everybody here at least is aligned as we are and believes that the risks of Maker are significant, they are important, and that as community members we need to think systematically and rigorously about risks and debate the best ways to manage risk. |
| # / 00:10:12 | Alex Evans | And so, there's ways in both the academic literate and the industry that we think systematically about risk and model it. Cyrus has done a good job presenting a good number of those models in the last couple calls. Maybe one thing that I can do here is give a little bit of color. |
| # / 00:10:30 | Alex Evans | I apologize for that noise that you hear in the background. If you can't hear me let me know. |
| # / 00:10:35 | Alex Evans | But why we chose a ratings-based model. There's really three reasons why we went with this approach. The first one is it just gels really well with the way that smart contract based financial applications work and architected and that they are ... they're build on Ethereum that's a state machine and so they think about the world in terms of states and transactions representing transitions from one state to another. What these models allow us to do is they incorporate this idea of state and state transition very naturally and intuitively with very few assumptions. |
| # / 00:11:10 | Alex Evans | The second one is these models were designed historically for the types of instruments that had state dependent pay off. When we come to MKR we really care about let's say how long a loan stays open or whether a loan is bitten or is liked or ends up in some other state. And so, that is a perfect application of that kind of framework because it's built for those kinds of derivatives and financial instruments that have these state dependent pay offs. |
| # / 00:11:36 | Alex Evans | And then finally, these models are really adaptable and modular, and it's easy to change their assumptions without really messing too much up. You can play with it and create new assumptions of your own and see how the conclusions of the model change depending on how you alter the assumptions. Okay? |
| # / 00:11:55 | Alex Evans | So without further ado getting into what those assumptions are. Can you see everything on the screen? |
| # / 00:12:08 | Richard Brown | Yeah, it looks good. |
| # / 00:12:10 | Alex Evans | Yeah, I think there's that. Okay, maybe that's better. |
| # / 00:12:14 | Alex Evans | There are ... by the way, as I said in the last one, each of these assumptions, I'm listing some assumptions here, everyone of these assumptions can be changed and can be altered. In fact, I think it's a good exercise to try and go in and tweak some of these. We can have really any unit of analysis here. We can use individual broad transactions. We can have a discrete state space, continuous state space. This state space's a bigger one. A smaller one actually in the next slide. I think I'll contract the state space a little bit. It can be Markov, non Markov, homogenous, nonhomogeneous. It can be continuous time, discrete time, whatever. And so, that's really one of the core intents of the model. |
| # / 00:12:47 | Alex Evans | The way that this model thinks about Maker is ... and I'll try to explain it as simply as possible. If you don't understand something it's my fault not yours. Please stop me. The way this model thinks about Maker is it thinks about Maker as a collection of loans primarily. And the way that we describe the global behavior of Maker is by looking at the emergent properties of that pool of loans. |
| # / 00:13:09 | Alex Evans | The first thing that we need to do is that we need to find a model to explain and predict how loans will behave across time. Okay, so the first thing that we need in order to do that is we need to define what a loan is. And so, that's what I mean here by unit of analysis. Here a loan is defined as an individual broad transaction. Whatever somebody goes to the Maker contract and draws another DAI we call that a new look. |
| # / 00:13:32 | Alex Evans | Now there are other ways to model this. We do talk about some of the advantages and disadvantages of each of them and the actual paper that we have on Placeholder.vc or you could use CDPs or you could use DAI and we'll get into those trade offs. You can go ahead and read about those there, but just remember that these are individual broad transaction. |
| # / 00:13:48 | Alex Evans | Now, just as in if you ever use a CDP you know that when you create a new CDP or a draw some amount of DAI you're automatically in a safe state. And so, you're CDP is totally safe. |
| # / 00:14:02 | Alex Evans | And then, at any given point in time the loan could be in one of four states. It could be safe where it started. It could be unsafe. In other words, it could be liable to being liquidated because it's below the 150% collateralization threshold. And then, it could be wiped or it could be bitten. |
| # / 00:14:18 | Alex Evans | And so, these two states, wiped and bitten, are what we call in this model absorbent states. In other words, if the loan is wiped or it's bitten it can't go back to safe or unsafe. It stays in those states forever. |
| # / 00:14:29 | Alex Evans | You don't need to really worry about these Markov and continuous time assumptions here. They're discussed in the paper as well, but just important to remember that they exist and are important assumptions. Okay? |
| # / 00:14:40 | Alex Evans | Here in the top right I have a diagram of the state space and the kinds of transitions that are allowed between different states. You can go from safe to unsafe. You can't go from safe to bitten. You have to go to unsafe first. But you can go from safe to wiped. From unsafe you can do to pretty much any state. And then, once you get to wiped then you see there's no arrow leaving those states. That's what those arrows mean. |
| # / 00:15:06 | Alex Evans | Up here what these are aren't transition probability as some people that have seen discreet time Markov chains would be familiar with. They're what are called transition rates. What these rates allow us to do is they incorporate a notion of probability, the probability that you transition from one state to another, but they also incorporate a notion of time. |
| # / 00:15:23 | Alex Evans | And the reason for that is we actually really care as MKR holders and as community members about the amount of time that a loan spends in a given state, right? If a loan closes having been open for a month that's very different to it closing after it being open for a year. Presumably in the later case it grew significantly more stability fees. Hopefully, that's clear. |
| # / 00:15:45 | Alex Evans | Here right below we describe how these different transition rates are calculated. First, they're organized in this matrix called the generator matrix. It's just it can be a way to organize the information. That's the best way to think about it right now. Here we see, for example, rate from safe to unsafe as this QSU. From safe to wiped is the QSW. And the way these are defined is the diagonal elements ... oh, and by the way, wiped and bitten are all zeros. There's no notion of transitioning from wiped to unsafe for example so that's zero. |
| # / 00:16:18 | Alex Evans | The diagonal elements are just the sum of the elements in the road corresponding to that diagonal element. The transition rate out of the safe state, which is what this first one, one entry is here, is just the sum of transition rates to safe, unsafe, wiped, and bitten. The transition rate out of safe is the transition to you can only transition to three states, so the sum of those transition rates is that transition rate you get here. It just convention to do the negative of it, but really you ignore the minus sign. That's the transition rate out of the state. |
| # / 00:16:50 | Alex Evans | These off diagonal elements incorporate this notion of probability and so you take the transition rate out of the state and multiply it by the probability of going to let's say the unsafe state. And then you get the transition rate too from the safe to the unsafe state. |
| # / 00:17:04 | Alex Evans | Is that clear? Does that make sense? |
| # / 00:17:06 | David Utrobin | Yeah. I have a quick question. From the state change from unsafe to bitten that is qub in the flowchart on the top, but in the matrix it's qsb. |
| # / 00:17:20 | Alex Evans | You're paying attention. |
| # / 00:17:20 | David Utrobin | Or qwb, yeah. |
| # / 00:17:23 | Alex Evans | Yeah, yeah, so that's qub, right. Thanks. |
| # / 00:17:28 | David Utrobin | Okay, cool. |
| # / 00:17:29 | Alex Evans | Yeah. That's just a typo. |
| # / 00:17:32 | Alex Evans | What's interesting is that from the unsafe state you could wipe it as well and you could bite it as well. Actually, these transitions do occur in practice. You do see loans going actually doing this oscillation a lot. That's just a side note. |
| # / 00:17:44 | Cyrus Younessi | Are you dealing with safe to bitten at all? I mean, I think you said you were going to ignore it, but I also noticed you put a spot in in for the matrix. |
| # / 00:17:54 | Alex Evans | Yeah, so there is a spot for it, safe to bitten, technically, and we note this in the paper. Also, a very, very good question. This should be zero. |
| # / 00:18:03 | Alex Evans | Now, the reason I keep it here is because if you look at smaller time intervals ... sorry, larger time intervals. It might show up as like something goes from safe to bitten, but actually it went through unsafe. You're just not noting it because you're time interval was too big. I keep it there for that reason. |
| # / 00:18:19 | Alex Evans | That's a really, really good question. Anything else? Okay. |
| # / 00:18:28 | Alex Evans | How do you we use this model? One of the ways and I didn't mean to scare you with all of these formulas here, so just ignore them. I want to note just one thing with this slide is every single one of these formulas that represent either global or local state you can compute with just these six assumptions. Actually, just as Cyrus just noted, it's actually five. You only need to compute these five rates and then anything here you can compute directly. |
| # / 00:18:54 | Alex Evans | With individual loans you know all these Qs, right? Because you have them for the matrix. You know the average time that a loan stays open is this thing here. I derive all these formulas. The proofs of how they all are derived are all in the paper if you're interested. |
| # / 00:19:09 | David Utrobin | Awesome. |
| # / 00:19:09 | Alex Evans | The only thing to note is the proximity of them all. Very few assumptions you can say with the average time a loan stays open. What is the probability that it it ends of bitten. The ps are these things here. |
| # / 00:19:18 | Alex Evans | And then, with additional assumptions about the birth process, i.e. the rate at which new loans are created, and the expected size of each loan, you can now start to project things like at any arbitrary point in time T how many loans should I expect to open? Or in the very, very long-run equilibrium of the system at what level does the number of loans stabilize? Or what is the total liquidation fees between now and some arbitrary point T, like in a year from now? How many stability fees do I get between year two and year five? You can get that in this formula here. |
| # / 00:19:54 | Alex Evans | Each of these formulas, is basically these five assumptions, these five rates that we had on the previous slide and just these to extra parameters. You can get really, really far. If you get good estimates of these parameters, you can make pretty good prediction about other parameters in the system whether they be local pertaining to individual loans and the risks associated with them or global parameters that have to do with the overall state of the system. Okay? |
| # / 00:20:16 | Alex Evans | How do we actually go about making these estimations? I won't get to into a lot of detail on the statistics here, one, because of the time constraint and, two, because it might be boring. There is a note in the model that includes some work that Vishesh has done with a Python code that estimates the empirical transition matrix from on chain data. |
| # / 00:20:37 | Alex Evans | Here is another approach that is included in the [Arc 00:20:39] code that accompanies the model on Placeholder.vc and, basically, tries to use covariances to estimate the conditional transition rates based on different values of the covariances. |
| # / 00:20:51 | Alex Evans | Here what we've done ... this O you might not recognize. This O is just your refusing the state space for safe and unsafe and just calling it open. Everything that's safe or unsafe we kind of just treat as one and we call it open and so then we just have three states open, wiped, and bitten. It just simplifies things. It's not necessary that you do this, but it's just a way to convey the insight a little bit more clearly. |
| # / 00:21:15 | Alex Evans | And so, here we just need to estimate two transition rates and then everything else kind of falls out from those. We have this open to bitten transitions rate and we estimate that condition on the value of this covariate for the collateralization ratio. This is a factor variable that is coded zero if the collateralization ratio is above 280%, it's coded one if it's be 250 and 280, and then it's coded 2 if it's below 250. The assumption being that as the collateralization ratio increases the property rate at which loans are bitten also increases. |
| # / 00:21:49 | Alex Evans | Similarly, we go and estimate this open to wipe rate and that's estimated based on this ability fee. Again, the hypothesis is as this ability fee goes up more people are like, "Screw this. I don't want to keep my CDP open," and so they wipe it. Okay? |
| # / 00:22:04 | Alex Evans | And these, you can think of this sort of as a regression table. It's a little bit different. We go into some of the intricacies in the actual paper, but the idea here is, for example, the stability fee you see this 72.4 here, 72.39. If we were to increase the stability fee by 100%, not multiplicatively but at 100% to this stability fee by one, we would see an increase, the model predicts an increase in the rate at which loans are wiped by 72 point .... Sorry. By a factor of 72.4, which means that all of those would pretty much be wiped out of the system within day or something like that. |
| # / 00:22:43 | Alex Evans | And then, similarly, open to bitten you see for instance the rate of transition to the bitten state when the collateralization ratio is below 250%. That is 14.3 times higher, so significantly higher. Okay? |
| # / 00:23:01 | Alex Evans | So question on this? |
| # / 00:23:03 | David Utrobin | I have a quick question about the very bottom, the negative two times log likelihood. What is that? |
| # / 00:23:13 | Alex Evans | That will get us beyond the scope here. You and I can sync offline about that, but it'll just tell you the type of fit that you're getting here. This is ... you're doing a partial likelihood basically across these ones. You can think of it almost ... it's not quite an R-squared because you need to do the standard errors for these. But anyway, it's reasonably good. There's some ... Here you also see 95% confidence intervals. I'm not getting into some of the stuff here that could be fun. |
| # / 00:23:39 | Alex Evans | There's more discussion of this in the paper and how this all works. These are 95% confidence intervals and then you can also pull standard errors for each one of these and do various things to understand the goodness of it, but I won't get into that here. If you don't mind, David. |
| # / 00:23:56 | David Utrobin | No. |
| # / 00:23:56 | Alex Evans | All right. But good question. |
| # / 00:23:59 | Alex Evans | Okay. What can we do with this? One application of these is in risk management. What's great is Cyrus has already done all the hard work of explaining all this. Cyrus has explained expected losses, unexpected losses, and so forth. |
| # / 00:24:14 | Alex Evans | Here we focus on what's called the conditional loss distributions. So additional values of the covariant. For a given pool of loans what percentage of them do we expect to, for instance, be bitten is what we're trying to do here. And this is what's called scenario stress testing sometimes. Let's say for above 280 this is above this is this green line so the probability will be a really small probably the loss is small. Probability of largish losses all the way to very large losses is extremely small. You have this peak here what this is around 17 and that's probably that 1% standard deviation where you get most of your liquidations happening. |
| # / 00:24:50 | Alex Evans | And then, on the other hand, IF the collateralization ratio is below 250 percent you see that the probability of small losses is pretty small, but these quite large losses are pretty probable. You can see that a bit more clearly here. The probability that your loss will be contained at the 60% level is quite small, but the probability that they are quite large, they're contained at the 85, that they exceed that 85, is quite high. |
| # / 00:25:21 | Alex Evans | That is that. That's all I had to talk about today. I want to say thanks. Vishesh did a lot of really great work here, not just on the data, but a lot of the ideas that went into the model, so a huge thank to him. |
| # / 00:25:33 | Alex Evans | This is just one attempt. I think it's just the first and most imperfect attempt at a community risk model. I really hope and believe that these will get better over time. I will be available to answer questions on these on the Maker forum or reach out directly on Alex H. Evans on Twitter. I think I'm also Alex H. Evans on the Maker chat, so you can reach out to me there as well. I'll try to make time to answer questions. |
| # / 00:25:58 | Alex Evans | One other thing here is we at Placeholder will be putting together a little bit of a more formal risk discussion or risk committee. It will include several ... it will be a smaller sort of forum. We'll share some of the results from that as we start to do that probably around the launch of MCD. |
| # / 00:26:15 | Alex Evans | And yeah, that's all from me today. Thank you very much. |
| # / 00:26:18 | David Utrobin | Really quickly, Alex. I saw that Matteo Leibowitz had a question in the chat about how you define a loss. Is it defined as bite or is it defined as liquidation? |
| # / 00:26:29 | Alex Evans | Yeah, this is just bite here. Now, in order to get the actual quote/unquote loss ... also great question. In order to get the actual quote/unquote loss you have to have an additional assumption about what amount, what percentage of CDPs fit in below what level. You need to have a distribution over what's called the recovery rate. Of the CDPS that are bitten how much are below 100%? Then also what is the recovery rate should they be below 100%? Then you can actually talk about dollar amounts or how much the MKR holders truly lose. That's a really good question. |
| # / 00:27:05 | Cyrus Younessi | Alex, where you guys able to do any research on potentially calculating those transition probabilities? And if so, can you talk a little bit about the challenges associated with that? Because I imagine it's probably not an entirely straightforward thing to do. |
| # / 00:27:25 | Alex Evans | Right. There's been some background discussion on this. One, by the way on GitHub we shared with the GitHub page both the codes were run all of these in generating all the results of the estimation and graphs and also the data. If you look at the data that Vishesh was kind enough to contribute it also includes the collateralization ratios at bite of every loan that was bitten. And so, if somebody were to do that exercise and try to do it well that would be where I would start because it'll tell you what the collateralization ratio of that bitten loan was. |
| # / 00:28:02 | Alex Evans | And so, that is one way to do it. The only issue is you really haven't had uncollateralized loans and so all of that is kind of guestimation. You could try to project based on that distribution, but we have a whole bunch more of like what is the probability it ends up below certain number. That's probably how I would do it. It's just we haven't. We've been quite conservative in terms of collateralization ratios and the liquidation system has worked sufficiently well. We haven't seen great example of that happening and so it really is an exercise in conjecture. |
| # / 00:28:35 | Cyrus Younessi | Yeah. |
| # / 00:28:36 | Richard Brown | What's this- Sorry, go ahead. |
| # / 00:28:40 | David Utrobin | I'll just say, yeah, that's a great point on the guestimation side. I mean, I think right now all risk models are in a state of trying to deal with the lack of meaningful historical data. |
| # / 00:28:53 | Cyrus Younessi | Actually, I think one of the, well I don't know if it was an intentional design towards it or anything, but something that we should be looking for in design choice is, what is the easiest way or what is the easiest model that'll help us abstract some of those necessary guestimations, as you like to put them, and put them into the model? That's what I'm trying to get at. |
| # / 00:29:19 | Cyrus Younessi | If we wanted to input some sort of manual or expert adjustment into this model do you think that's a ... is that doable? Is that something we can say that based on our understanding of the underlying collateral or would you be more comfortable just using this purely for collateral types that have like years and years of historical data? Do you think this would work equally well for ERC20 tokens or would you say this is better for security tokens and longer historical assets. |
| # / 00:29:59 | Alex Evans | Well, the more historical data the more you can extrapolate and the more significance you're going to get in our statistics. The other one is though for security tokens that may be a little bit different and we haven't seen them used in DeFi as much. While with other tokens we could for instance look at Compound loans and how those have behaved over time and whether there've been liquidations there and how that system has behaved in response to it. We can see some examples. |
| # / 00:30:25 | Alex Evans | Now, the data's really, really recent. For example, if you look at a collateralization ratio one of the reasons why it's a factor as opposed to just continuous collateralization ratio variables is really each one of those numbers have your sample size is like two, right? |
| # / 00:30:40 | David Utrobin | Right. |
| # / 00:30:41 | Alex Evans | You can't really do much. And so, you split it into just three big buckets of collateralization ratio that's like pretty well resolution, but it's about as good as you can do in order to actually get some significance and then talk about some results. |
| # / 00:30:54 | Cyrus Younessi | Yeah. Actually, something else that I was talking with Vishesh about the other day was that this transition probability's. Do you think they would be similar across different collateral types or? I mean, certain collateral types they'd be completely different, right? Certain like binary type of assets where one day they're just going to collapse to zero or whatever. But for a particular set of standard assets do you think that those transition probabilities will expect to be the same across different types? |
| # / 00:31:34 | Alex Evans | The hypothesis here is that those transition probabilities and rates depend on two things. One, they depend on the bites pass of the collateral. You've done a really good job explaining this. I'm just taking this directly from your presentation. One is the bites pass of the collateral. The second is the borrower behavior associated with borrowing against that collateral. |
| # / 00:31:57 | Alex Evans | Do you behave differently when you're borrowing against the house versus when you're borrowing against ETH? The hypothesis would be, yes, it's a different type of borrower with a different kind of mindset and different types of behaviors associated with that type of lending. |
| # / 00:32:13 | Alex Evans | And then secondly, the price pass of those two assets are very, very different. One is more volatile than the other and acts quite differently. |
| # / 00:32:20 | Cyrus Younessi | Right, okay. |
| # / 00:32:20 | Alex Evans | And so, the constant through the transition rates would be different numerically. Their estimation shouldn't be that different because really the system cares the state. It says, what state are we in right now and what states will we be in in the future? All you need to do as an analysist is step back and say, okay, what's the probability that we'd get to that state? |
| # / 00:32:45 | Cyrus Younessi | Okay. And so, this is where- |
| # / 00:32:48 | Vishesh Choudhry | [crosstalk 00:32:48]. |
| # / 00:32:48 | Cyrus Younessi | Sorry, was somebody saying something? |
| # / 00:32:49 | Vishesh Choudhry | Yeah, sorry. I was just saying, can I ask a follow up to that or rather a clarification, right? To what extent is the relative rate between different states determined by behavior versus on an aggregate the amount of liquidation you would expect a function of asset price volatility, right? How do you separate- |
| # / 00:33:12 | Alex Evans | Hang on, Vishesh. Say that again because I didn't understand what you were saying. |
| # / 00:33:16 | Vishesh Choudhry | How do you separate the effects of behavior of CDP owners? For example, rebalancing when asset prices go up or shoring up when they see liquidations, how do you separate behavior versus asset price in terms of this is something that happened with ETH or this is something that CDP owners tend to do? |
| # / 00:33:36 | Vishesh Choudhry | And so, that's where I started to think variance, yeah. I mean, that's where I stared to think about the relative transitions between one state verus another versus the overall aggregate amount of liquidation for example that you would expect, right? |
| # / 00:33:54 | Vishesh Choudhry | I would say that the total amount of liquidation you would expect would primarily be a function of that asset. But how much people shore up capital in their CDPs versus taking it out when asset prices go up versus down I would say would be more standardized across different assets as more of a behavior function. |
| # / 00:34:12 | Alex Evans | That's interesting, yeah. So sorry, was there a question there? |
| # / 00:34:23 | Vishesh Choudhry | I guess it was more of a clarification than a question. |
| # / 00:34:25 | Alex Evans | Okay. |
| # / 00:34:29 | Richard Brown | All right, there's some questions in the forum, Alex, from Patrick [inaudible 00:34:33]. |
| # / 00:34:32 | Alex Evans | I'll go in and respond to them. A good number of them are in the paper. And so... |
| # / 00:34:37 | Richard Brown | Okay. |
| # / 00:34:38 | Chris Burniske | May I ask a question? |
| # / 00:34:39 | Alex Evans | You may ask a question, Chris. |
| # / 00:34:42 | Chris Burniske | We talked some in the process about how this would be applied valuation. I'm curious if you can expand on how this can be applied to valuation. |
| # / 00:34:47 | Alex Evans | I knew you were going to ask that. |
| # / 00:34:52 | Chris Burniske | I'm curious if you can expand on how it can be applied valuation? Evaluation of MKR. |
| # / 00:34:58 | Alex Evans | You had to do this. Okay, so there is a brief note at the end of the paper taking about valuation of MKR. One of the ideas that I mentioned very briefly at the beginning of this presentation is that this type of model is very good. |
| # / 00:35:08 | Alex Evans | You can actually see in the original paper that it sort of derives from it references this idea of derivatives with state dependent pay offs. That should immediately ring a bell because it sounds a lot like MKR. It's a type of ... |
| # / 00:35:21 | Alex Evans | You can think of it almost like a portfolio default swap written on the existing loans where it collects premiums from that pool. And then should there be an issue it provides a residual form of capital to capitalize any undercollateralized loans, okay. |
| # / 00:35:35 | Alex Evans | That idea you would have to have a way to figure out from actually traded loans with interest rates that are free floating unlike Maker to strip out the risk premium and then apply it to take the empirical transition matrix that we discussed in the presentation and turn it into risk neutral matrix. And then based on that matrix go back and compute what the true risk neutral price of MKR is. Now, that's a theoretical exercise as I say in there because we don't have access to those free floating loans. |
| # / 00:36:07 | Alex Evans | One of the things I that's interesting that this type of modeling allows us to do and some of the models that Cyrus has been presenting over recent weeks allows us to is that actually the valuation exercise and the risk management exercise kind of converge into one thing. We're trying to think about how do we maximize our risk adjusted return as MKR holders and it turns out that the actual answer to that is similar to the answer to, what is the best way to manage risk in the system? |
| # / 00:36:35 | Cyrus Younessi | Wait. Sorry. Could you reexplain the information you're taking from the free floating loans, like the traditional? Can you expound upon that a bit? I didn't quite understand that part. |
| # / 00:36:48 | Alex Evans | Right, so in the [JLT 00:36:51] paper he describes this technique where you go out and find traded prices for let's say different corporate bounds or corporate liabilities and different [crosstalk 00:37:02]. |
| # / 00:37:01 | Cyrus Younessi | Right, right, right. |
| # / 00:37:02 | Alex Evans | You go back and then based on those in the actual transition probabilities you do this recursive procedure where ... It's tricky. I won't go into the details of this. He describes it super well in the paper. And then, you pull out these risk premium. |
| # / 00:37:14 | Alex Evans | You have this spectrum premium that you multiply the empirical transition matrix by and you get basically what's called a risk neutral matrix. And then risk neutral matrix is basically the same thing as the natural transition matrix but with risk neutral probabilities. |
| # / 00:37:29 | Alex Evans | And then, you take, for example, under the risk neutral measure all your discounted stability fees and liquidation fees less whatever you're going to have to make up. Then you have a value for MKR. |
| # / 00:37:41 | Alex Evans | Now, that's a theoretical exercise. You would need loans that are extremely similar to CDPs which don't exist publicly in traded form right now. You have to have, let's say, in single collateral DAI, it would have to be 150% collateralized with ETH, they would have to have a keeper reward like a liquidation penalty of 10%. They would have to reward keepers with 3%. All these things, right. |
| # / 00:38:04 | Alex Evans | So Dharma loans and Compound loans are kind of similar, but they're not similar enough where you can make that kind of risk control argument to say well under the ways that the market is pricing risks in this scenario if we apply that same risk premium here this is the value that we get. We can't really do that. |
| # / 00:38:20 | Cyrus Younessi | Okay. |
| # / 00:38:27 | Richard Brown | All right, thanks for that, Alex. There's probably a lot more questions. I'm not sure what your schedule looks like. But we need to get into, Visheshalytics for the next 15 minutes. But if you want to hang out, maybe we can ask some questions. If not, we can bombard you in the forums. But thank you for that. That was enormously valuable. |
| # / 00:38:45 | Cyrus Younessi | Thanks, Alex. That was great. |
| # / 00:38:46 | Alex Evans | Thank you. |
| # / 00:38:51 | Richard Brown | All right, Vishesh, are you ready to take it away? |
| # / 00:38:54 | Vishesh Choudhry | Sure. All right, so just to bring you back to some numbers after all that theoretical goodness. As far as the DAI price, I mean, so right now volumes trading at like 5.6 million. It's been considerably higher than that over the past couple of days. The spread is kind of waffled around 98, 99 cents for the last few days. I'll just touch on that. |
| # / 00:39:32 | Vishesh Choudhry | I think the big picture is things are doing all right. I think quite a few people expected things to be going a little bit better especially with the stability fee increase. It's interesting for us to dive into a little bit more of the like holistic measure and what's been going on instead of just looking at like price stability in the silo because things can get a little bit oversimplified and actually by extension a little bit confusing if you just try to look at stability fee price. |
| # / 00:40:02 | Vishesh Choudhry | Yeah, just to touch on, I mean, this has kind of been the trend over the past month or so as things were hovering around a dollar, 98 cents, and 99 cents. Then there's a bit of this dip just at the beginning of July and a slight trend down with the 17.5% stability fee increase and the 18.5% stability fee increase. |
| # / 00:40:24 | Vishesh Choudhry | It had started to look like things might be recovering then ETH decided to start to make some insane price movements and so there is this sort of double overreaction. I do kind of want to touch on this because this has been a topic in the chats for quit a while is, how did we get into this scenario and is it actually a problem or is it a temporary state of being? |
| # / 00:40:49 | Vishesh Choudhry | And so, with the 17.5% increase ... Sorry, with the 17.5% decrease, there this kind of initially this narrative that at 19.5% the state of the system's too expensive. DAI was trading above a dollar at times. And so the stability fee really needed to be decreased. |
| # / 00:41:10 | Vishesh Choudhry | That was probably a bit premature, right? With that initial increase to 19.5% there was some temporary push up of the price and then a bit of a strong overreaction with DAI price trading over a dollar for just barely a week. |
| # / 00:41:27 | Vishesh Choudhry | And so, then it was to successive decreasing. That's when you started to see, as I had mentioned at the time, more kind of variability. The price was a little bit looser though it wasn't necessarily lower. And so, you'd see a widened spread. |
| # / 00:41:43 | Vishesh Choudhry | I should show the volatility of this, but at the end of the day that started to drift down with the ETH price movement. And so, it wasn't necessarily that the 16.5% stability fee caused that decrease, but rather increased the sensitivity of the system to what was going on with the ETH price. |
| # / 00:42:04 | Vishesh Choudhry | So when there was sort of this sharp drop in DAI price for a few days there is again a strong reaction by the community, "Okay, now we need to increase the stability fee to 17.5%," and then again to 18.5% shortly thereafter because in the span of four days it did not look like there had been an improvement. |
| # / 00:42:30 | Vishesh Choudhry | And so, this was again maybe sort of that same strong double reaction pattern that I guess is now emerging, which I would caution against, right? If you make changes thoughtfully the first then you don't have to sort of scramble to unwind on the second time. That's one sort of governance feedback that I will give. |
| # / 00:42:52 | Vishesh Choudhry | The other piece is with the 20.5% stability fee increase a lot of people were kind of confused that the DAI price didn't immediately shoot up over a dollar and stay there and they were a little bit confused as to what was going on. Again, the stability fee does not inherently fix the DAI price. It doesn't inherently bring the DAI price down. It just improves the sensitivity of the system to what's going on with leveraging behavior and by extension ETH price. |
| # / 00:43:24 | Vishesh Choudhry | This increasing the stability fee to 20.5% is not a inherently just going to bring the value price up because what you effectively need is people to buy back that DAI, which means you need people to be willing to unwind their leverage. The problem is some people given where the price of ETH has gone may be unwilling to give up on their positions at this time or they may have been liquidated out and thus they don't have to buy any DAI because they've already been liquidated. If those liquidations happen at sort of effectively lower DAI prices then you'll have kind of missed out on some that upward price pressure. |
| # / 00:44:12 | Vishesh Choudhry | A little bit of a complex process with a lot of moving pieces, but at the end of the day I think the takeaway is even with the 20.5% stability fee it's going to take a little bit of time for this curve to slowly get pushed up with that buying behavior. |
| # / 00:44:31 | Vishesh Choudhry | Again, I would caution against these sort of strong overreactions. Just because things seem fine at the 19.5% stability fee because ETH has not been stressed in that time doesn't mean that it's necessarily time to lower the stability fee. And vice versa if it was increased to 17 and 18.5% it doesn't necessarily mean that price is going to recover tomorrow. You need actual movement in the trading ploy. |
| # / 00:45:00 | David Utrobin | Vishesh, really quickly. There is a question in chat about this. Is Maker governance then forced to take directional bets on ETH price? |
| # / 00:45:10 | Vishesh Choudhry | Forced to, no. I don't think the preemptive stability fee changes approach, which is basically what you're getting at and what's been discussed a lot, is necessarily scientific is what I'll say. You could certainly choose to try to manage the system by guessing what's going to happen with the ETH price, but if you just try to manage to the level of confidence that you have ...so what I'll say is if you think that the system does not have robust enough transaction going, does not have robust enough holistic health in terms of level of leverage, amount of sticky demand, amount of oversupply, then you should be more weary of having a lower stability fee. |
| # / 00:46:04 | Vishesh Choudhry | I think the problem is not so much that you have to guess what ETH is going to do. I think the problem is you have to be mindful of how much you're opening up the system to fluctuations in ETH price based on things like the level of organic demand, right? If you have a greater degree of oversupply what that's effectively saying is ... demand is a non monolithic concept, right? Demand is basically a curve of stickiness. And so, if you have a lot of DAI that's been draw out on fairly transient demand in terms of people wanting short term leverage, you really should just count that as oversupply. |
| # / 00:46:46 | Vishesh Choudhry | And so, when you have that oversupply when people are ready to effectively dump that DAI at the drop of a hat then you should be more weary about having a lower stability fee because, one, you're either going to exacerbate that problem or, two, just be mindful of then if ETH goes and decides to make price movement your sort of flapping in the wind. You're subject to whatever those market movements are to a pretty significant degree. |
| # / 00:47:15 | Vishesh Choudhry | But if you're oversupply is really low, right, and you have strong transaction volume, you have a high percentage of transaction to trading volume, you have not as much outstanding leverage position that's ready to be closed out, then you can be a little bit more confident. And so, it's again not so much about guessing what ETH price is going to do, but being mindful of how much you're exposed to it at any given point in time. |
| # / 00:47:45 | Vishesh Choudhry | And so, this is part of why DAI's price will always fluctuate, right? This is a topic that's been discussed for quite a while is DAI is never going to sit exactly at a dollar. The goal of management of the system as I see it is to try to manage the system health so that when it does fluctuate it's fluctuating within acceptable ranges. |
| # / 00:48:11 | Vishesh Choudhry | And so, again it's not ... If you try to move the stability fee every week to try to counteract every price pressure, you're going to be chasing your tail. I think you're just going to propagate more degree of variability and thus likely more error into the system, but I think if you try to keep the system more stable and less subject to those fluctuations and facilitating less oversupply or less of that sort of transient demand at any given point in time then you will I think experience more stability. |
| # / 00:48:46 | Vishesh Choudhry | It's just it's kind of a management choice and a level of awareness rather than trying to preempt data because that's really hard and I don't think you're going to do that successfully. |
| # / 00:49:00 | David Utrobin | [inaudible 00:49:00]. Thanks. |
| # / 00:49:01 | Vishesh Choudhry | Yeah. So to talk about some of those behaviors. This is just the ETH price. For context this really started to tank around July 9th, 10th, which if you look at what was going on with some of the trading volumes there is a decent spike then and then a little bit of a dip and then a huge spike a couple days later. |
| # / 00:49:28 | Vishesh Choudhry | This huge spike, although the ETH price was just sort of continuing to go down, was unfortunately probably lot of people levering back up, some other people selling. We'll go over to the supply metrics to sort of highlight that. This is shortly thereafter like that same timeframe. |
| # / 00:49:51 | Vishesh Choudhry | What we saw was overall ... I'll zoom back out here just real quick. Overall this supply had been consistently growing for quite while around that like 90 million, 92 million level. There was a lot of these problems with DAI price. The stability fee was consistently increased. During that period supply came down to like 82 million and then the stability fee was lowered as we just talked about to 16.5% and supply sort of started to grow pretty consistently topping out 91, 92 million. That's when we started to have conversations about the debt sealing. |
| # / 00:50:32 | Vishesh Choudhry | I think during this time period there was an accumulation of a lot of that transient demand, a lot of that leveraging behavior, and what you should effectively call over supply. And so, that's where you started to accumulate some of this risk which you're not starting to see the effects of. |
| # / 00:50:53 | Vishesh Choudhry | A lot of that transient demand kind of went away. There was refinancing. There was some liquidation. There was some closing of leverage positions. And then, unfortunately, I think it had ETH skyrocketed you would have actually seen a similar effect in terms of DAI stability and DAI price. That's again another reason it's hard to try to make bets based on what happens with ETH price because strong movements in both directions can actually end up having similar effects. |
| # / 00:51:25 | Vishesh Choudhry | And so, had the ETH price shot up you would have seen a lot of folks cashing out potentially, but you may have also seen folk leveraging up and so it can go either way unfortunately. |
| # / 00:51:41 | Vishesh Choudhry | Yeah, with the 20.5% increase that supply had started to come down but not because I think purely of the stability fee. This is where correlations start to become really difficult because the stability fee was increased do to instability in DAI and Instability in DAI was due to ETH price. If you try to sit there and just make the correlation of stability fee was increased and then supply went down you might draw the wrong conclusion. |
| # / 00:52:09 | Vishesh Choudhry | So the supply went down primarily due to those exogenous market factors. There is probably four or five million DAI that was wiped during that time, but a fair portion of it was also do to what was going on with secondary lending platforms and a fair amount of refinancing to Compound, for example. |
| # / 00:52:28 | Vishesh Choudhry | It's a combination of first your supply was starting to go down with some liquidations. Then you started to have some people refinancing. You did have some I think closing out positions, just sort of giving up, but I think more so those people were probably liquidated out. And so, the remainder probably just are sitting with their positions. |
| # / 00:52:55 | Vishesh Choudhry | I think this funny enough ties into what Alex was presenting around those transition states and is part of the reason. Now I'm glad that model was out there and hopefully people understand this concept a little bit better. As this stability fee changes you see often a significant impact in what's going on with the circulation of debt. |
| # / 00:53:18 | Vishesh Choudhry | And so, that's where the age of debt is pretty decently impacted by what's going on with the profitability of loans, the extent that refinancing is going on, the liquidations are going on, et cetera. With the successive stability fee increases earlier in May the ... I'll just go for context here. The age of debt had kind of consistently been rising accumulating a lot of ... and this tracks with the sort of unpaid stability fees graph, so consistently growing and outpacing the extent to which loans were being liquidated or paid back. |
| # / 00:53:59 | Vishesh Choudhry | Then successive stability fee increases, that age starts to decrease fairly significantly. And then, with the 16.5%, 17 and then 16.5% drops that sort of started to stagnate. And then you started to see with a lot of refinancings and the success of secondary lending platforms in this sort of May, June timeframe, you did see some decrease I the age of open debt. A lot of that was new debt being created, yes, because the supply was also growing during that time, but also with some slight uptick in the age of closed debt was due to refinancing to secondary lending platforms, so a combination of factors. |
| # / 00:54:49 | Vishesh Choudhry | What's really interesting now is these two lines have kind of converged, so you've actually got a really interesting slightly decreasing steady state. I say slightly decreasing because the stability fees have now been increased, but only for the last couple of weeks. |
| # / 00:55:07 | Vishesh Choudhry | Honestly there is far more liquidation going on in that time frame, so the age of debt will be a little bit quieter during time periods like that. And so, these two lines have kind of converged where roughly the age that debt is open and the age that debt is closed is roughly at the same level and slightly decreasing. |
| # / 00:55:32 | Vishesh Choudhry | To kind of touch on what that looks like in terms of draws, we saw a spike in draws around the end of June and we saw a fairly heavy rising amount of draws in July. But a lot of that was on existing CDPs, and so this percentage of draws on new versus old CDPs has been relatively constant but slightly decreasing. |
| # / 00:55:58 | Vishesh Choudhry | The affect of that on collateralization ratio is pretty interesting. Actually, next time I want to graph this against the ETH price so you can see the impact, but basically as the ETH price had ... and we went over this last week. The ETH price, when it goes down, you see a decrease in collateralization ratio but then also commensurate increase as this counter behavior. |
| # / 00:56:26 | Vishesh Choudhry | What's really interesting is right around this end of July time period you have a fair amount of collateral being added after the dip. And so, this is exactly what we're talking about, sort of that countereffect. A lot of people got scared from those liquidations and that drop in ETH price and so have actually added collateral into the system bring up the collateralization ratio a bit. |
| # / 00:56:57 | Vishesh Choudhry | As far as secondary lending platforms ... actually, I'll just jump over real quick. This is something that we've kind of touched on as transaction volume. The quote/unquote demand and the extent to which DAI is moving around, why it's moving around is an important question. I've been asked this question from quite a few people. How is the transaction volume of DAI versus the trading volume of DAI? |
| # / 00:57:22 | Vishesh Choudhry | And so, you saw sort this graphs on trading volume, right? So a huge peak in June 26th and like roughly July 14, 15 to the tune of like 14 million and 8 and 9 million, but if you look at the total transaction volume on chain that's more to the order of 140 million. And so, this is, and I know Matthew (Rabinowitz) had asked me this question on one of the chats, was their record. Just shy of a record, so November 2018 156 million was the record, but pretty darn close. |
| # / 00:57:59 | Vishesh Choudhry | And so, you've seen this consistent rise in the transaction volume of DAI since basically the beginning of April. A lot of those secondary lending platforms had not really taken off in full force until May/June, so this is not purely due to smart contract activity and secondarily lending platforms although that definitely contributes. |
| # / 00:58:23 | Vishesh Choudhry | I think the demand for DAI had been growing a bit and justifies part of the increase in supply, but I do think there was a fair amount of oversupply there in the last few weeks. The total transaction volume of DAI is partly attributable to that demand and then partly attributable to DAI's usage in smart contracts, secondary lending platforms, refinancing DAI, and things like that. |
| # / 00:58:52 | Vishesh Choudhry | So you can see the amount that DAI was traded. Call it 14 million in the last few days. The amount that DAI was burned and minted, a total of like six, seven million, but the total of transaction volume was roughly in the 140s. And then, the amount of DAI that was moved over from Maker to Compound was like roughly five million. So, even taking those kinds of transfers into account, there's still a fair amount of activity over the past month. |
| # / 00:59:28 | Vishesh Choudhry | Bites. We saw with those dips in ETH prices there's a fair amount of liquidations, but again nothing quite record setting. It was just a 30% price drop. But with the way the collateralization ratio has been going, this was something Alex touched on, is that liquidation percentage is hugely driven by what this collateralization ratio graph looks like over time. |
| # / 00:59:58 | Vishesh Choudhry | And even though things have kind of dipped and gone up, et cetera, just compared to the past it's been huge, almost double what it had been in the end of 2018 beginning of 2019. That huge buffer of collateralization ratio is going to protect against a lot of those liquidations and the potential undercollateralized liquidations, et cetera. |
| # / 01:00:20 | Vishesh Choudhry | I know times going over so I'm just going to make one last point here, which is the borrow on secondary lending platforms has shot up with all this ETH price volatility. Also partly due to things like the Instadapp and things like that, but also do do the stability fee. |
| # / 01:00:38 | Vishesh Choudhry | And so, there's a lot of discussion around what's the natural role of these secondary lending platforms in this ecosystem. You see the excess supply as has consistently gone down over the past few weeks and the utilization rate, I mean, dYdX is lower, but Compound utilization rate is at like 93%. This makes total sense because to the extent that you can get cheaper DAI on secondary lending platforms there's no reason why people wouldn't do that. |
| # / 01:01:13 | Vishesh Choudhry | What's really interesting is as that excess supply dissipates then you can actually say that the supply levels are closer to the organic demand levels and you can be more confident that you have less oversupply. And so, then when you start to talk about what the rates should be and whether the stability fee should be decreased, et cetera, I think these are really good indicators to look at. |
| # / 01:01:38 | Vishesh Choudhry | I really want to caution against a lot of the, and I've been doing it form months, caution against the like over simplistic thinking of DAI is down, increase stability fee DAI is up, increase stability fee. I think if we get of of that kind of narrative and we actually wait more than a week before making changes, it will ultimately manage the system better. |
| # / 01:02:00 | Vishesh Choudhry | Okay, so- |
| # / 01:02:01 | David Utrobin | There was a question in the chat. Is there a healthy amount of oversupply? |
| # / 01:02:09 | Vishesh Choudhry | Healthy? There's not a significant amount of oversupply now I think. Had there been or like is it good for there to be oversupply? There was oversupply. Is it good? Obviously, not. Healthy is complex. I think having a buffer on secondary lending platforms is a good thing in that it does help protect the demand function from a stability fee that is potentially to high. |
| # / 01:02:42 | Vishesh Choudhry | There should be some buffer, but I don't ... it's sort of like if you ask your doctor, "Should I eat more salt?" The doctor is going to say you get enough salt in your diet. The excess supply, the oversupply, is going to be high enough to give you like enough of a healthy buffer I think anyways that you don't need to think about like artificially increasing it or anything. I think you generally want to manage it down and then there will always be some amount that exists. |
| # / 01:03:12 | Vishesh Choudhry | If the excess supply on the dYdX and Compound went to zero tomorrow, then I would say stability fee is probably too high. You have no oversupply and it's just bone on bone. You probably should decrease the stability fee because there's enough demand out there that justifies a higher supply. |
| # / 01:03:36 | Vishesh Choudhry | I will stop sharing, but if there are more questions I'll be happy to answer. |
| # / 01:03:49 | Vishesh Choudhry | Yeah, so- |
| # / 01:03:49 | Richard Brown | All right, there's no question incoming. |
| # / 01:03:51 | Richard Brown | Sorry, Vishesh. Were you going to say something? |
| # / 01:03:54 | Vishesh Choudhry | No, just that Matthew Leibowitz, Matteo Leibowitz's question or his comment is right is what I'm saying. You want some excess on those secondary platforms. But again, I don't think you have to try to increase that. |
| # / 01:04:13 | Richard Brown | All right, it's seven minutes after the hour. That was a nice presentation from both Alex and Vishesh. If it turns out that there are more questions or people want to hang out and just talk about risk, we can do that for like the next half hour or so. But for people that want to drop out, I just want to say thanks to Alex again for that presentation. Thank you Vishesh. |
| # / 01:04:39 | Richard Brown | Check out the forums. Follow us . We've transitioning away from Reddit. As I said at the top of the meeting governance happens in the forums, so please join us there. There's a lot of really smart people saying really smart things and it's worth catching up on. Please join us. There will be a summary of this call and Alex's slides will be posted there as well sometime today. |
| # / 01:05:02 | Richard Brown | All right, I think ... so I wonder ... Yeah, let's just hang out for a couple of minutes and if there's an awkward silence we'll put an end to this thing. If there's questions, we can chit chat. All right, thanks everybody. |
| # / 01:05:17 | David Utrobin | I have a question for you, Vishesh. Do you see a scenario where secondary lending platforms basically half the effective DAI supply that would have been if they didn't exist at all? I'm trying to think of the right way to formulate the question because I look at Compound and it's a $20 million DAI market, right? That's a quarter of the total DAI supply. I don't know. I'm trying to figure out how to correctly ask what I'm trying to think about. Do you kind of get what I'm trying to say? |
| # / 01:05:59 | Vishesh Choudhry | Are you worried that like secondary lending platforms are contributing to the supply? |
| # / 01:06:03 | David Utrobin | No, that secondary lending platforms are ... you know what? Let me think about it a little more because I don't really know how to articulate it. |
| # / 01:06:17 | Vishesh Choudhry | Yeah, so the one obvious like criticism or concern that I've heard is that secondary lending platforms are taking the market share away from Maker somehow. People using those platforms instead of opening CDPs and so forcing Maker into having higher stability fees, which again as you can tell from my description there I don't get how the argument follows, I would say. |
| # / 01:06:49 | Vishesh Choudhry | If that's the concern I don't don't see a logical chain that brings you to that conclusion that they're taking market share aware from Maker. I mean, to the extent that the rates are cheaper, and I think this is going to be an interesting point, is to the extent that the rates are cheaper those platforms people will and should and I think are now finally waking up and realizing it go to those platforms to obtain the same asset cheaper. |
| # / 01:07:17 | Vishesh Choudhry | But the question would be, why does that supply exist in those platforms? |
| # / 01:07:22 | David Utrobin | Right. |
| # / 01:07:23 | Vishesh Choudhry | I think there are two different reasons why that supply exists. One is people who sort of capitulated there's CPs and sold their DAI and have no intent in reclaiming that DAI at least for quite a while or to the extent that people are holding onto DAI for some reason and have not decided to refinance. |
| # / 01:07:49 | Vishesh Choudhry | But I don't expect that to be significant portion of the supply on those platforms. I think it will primarily be purchased DAI that is being supplied, but the borrow volume is fairly significant. It's like 18 million, but as of four days ago it was 11 million. And then, as of seven days ago it was roughly that and it hasn't changed for quite a while. It was 7 million in June 19th. It's kind of been slowly and steadily increasing except for in the last four of five days which is again partly due to UI improvements, partly due to stability fee. But a dYdX borrow volume, for example, has actually gone down a little bit, but [crosstalk 01:08:51]. |
| # / 01:08:50 | David Utrobin | So about dYdX borrow volume, is that because the people that actually use the supply DAI on dYdX are using it specifically for dYdX instruments and not just pure borrowing like on Compound? Are you able to like pure borrow on dYdX? |
| # / 01:09:09 | Vishesh Choudhry | I think you're going to start to see emerging use cases for different platforms. Things are going to become more differentiated as they go. I think about a month ago a lot of people would have asked the question of like, "Why would I use one platform versus the other? What's the difference?" But I think those are going to becoming increasingly differentiated. |
| # / 01:09:30 | Vishesh Choudhry | And so, dYdX is what you see is you see generally higher rates. Although those have actually come down fairly significantly, but they're definitely more unpredictable rates. You've seen the utilization come down pretty ... I'm sorry, I'm looking at a graph here at [Sunfair 01:09:55]. But like pretty negatively correlated with Compounds, Compound utilization. |
| # / 01:10:03 | Vishesh Choudhry | And so, what's interesting is Compound may have taken some of the market share of DAI borrowing from dYdX. I think part of the reason is education in UI, so UI's not to be underestimated. Definitely I think there were some people who were borrowing from dYdX at a higher rate and then realized they could get a lower rate on Compound and decided to switch. That I'm sure is part of the impact. |
| # / 01:10:31 | Vishesh Choudhry | But also I think dYdX is in large part going to be used for like leveraged long positions and for derivative structures. And so, I think there were a fair amount of people who'd kind of given up on long positions and probably thus disproportionately affect the borrow on dYdX versus other platforms. Because I think dYdX has it's brand and it's going to be more used more expensive short term plays or at least more unpredictable shorter term plays that probably are for a greater degree of leverage, like going directly 3X or something. |
| # / 01:11:13 | Vishesh Choudhry | With dYdX UIs has the easiest method of going directly to a 3X leverage long position. Compound I think is going to be basically the cheaper version of Maker or at least how a lot of people are going to think about it right now. |
| # / 01:11:30 | Vishesh Choudhry | So to the extent that supply exists on Compound it is going to get eaten up, but I don't think that's necessarily bad is kind of my point. I think it actually probably an economic positive because what was happening with that supply if it wasn't being borrowed on Compound? It's not like people are drawing DAI out now to just go lend it on Compound because that makes no economic sense. If people are that will go down over time as people smarten up, but I don't think that's a reasonable explanation at this point. |
| # / 01:12:03 | David Utrobin | Right. That was a really good answer. |
| # / 01:12:08 | David Utrobin | There's a comment that I think is interesting by Sibil in the chat. He says, "I'm concerned that the new 22.5%, which is the winning governance poll right now, is going to put on even more delayed pressure and we're going to find ourselves in a supply crunch. What if it goes up to 22.5 and DAI supply goes down another 10 million from 85 to 75?" |
| # / 01:12:31 | Vishesh Choudhry | This is the hard part, yeah. This is the hard part between like calls versus e-synchronized chats and things like that is we're moving so quickly that like some of this I haven't had a chance to convey this on the calls since then, but like it's definitely been discussed in chats that this idea of .... okay, stability fee has increased to 19.5%. DAI is going to moon, yeah. |
| # / 01:13:01 | Vishesh Choudhry | I don't think that's an reasonable guess, no, but if there is not oversupply there's obviously going to be a supply crunch, obviously going to be a liquidity crunch on DAI, and so the price is very likely to move up. The problem is if that supply starts to decrease. This is where I talk about quote/unquote bone on bone. If that supply starts to decrease because demand starts to decrease and people want to start using USDC or something else, that's when I think you should be concerned. |
| # / 01:13:39 | Vishesh Choudhry | And so, you don't want the stability fee to be 50% because then you're self selecting to where only people who are going insanely levered and are really into going levered on DAI would use the system. I'm not sure that self selection selects for a very stable demand function. |
| # / 01:14:02 | Vishesh Choudhry | And so, I am basically cautioning against increasing the stability fee yet again and I made the point prior that going from 17 to 18.5% happened too quickly because going from 19 to 16.5 happened too quickly. We're just like hyper overreacting in both directions and we're going to get into a crazy loop. |
| # / 01:14:32 | Vishesh Choudhry | Because then if you create a more unstable demand function where people are holding that DAI or are putting it into leverage positions but are not necessarily buying it or trading it, maybe they're borrowing it on Compound or who knows what, then I think you could get into a scenario where DAI price is further depressed. And then, people are like, "Oh, we have to increase the stability fee again." That's like a weird sort of counter scenario to get into where you create your own demons. |
| # / 01:15:05 | Vishesh Choudhry | This is very hard to scientifically show, but like to what extent is the current state of DAI or you know for the past few weeks a function of the reaction of what had happened to the 19.5% increase? And then, it starts to become a really complex psychological equation of had it been increased only to 19% instead of 19.5 would people have been less overreactionary to dropping it to 16.5 creating less of a need to 20.5 or 22 for that matter? |
| # / 01:15:39 | Vishesh Choudhry | And so, it's like you have to really visualize how these things play out because you could create your problems. And so, I think the 20.5% increase was even maybe a little bit premature. Definitely now increasing to 22.5 would be premature. |
| # / 01:15:59 | Vishesh Choudhry | Yeah, I mean, Psybull here. This is a problem of governance. People could just decide whatever and not listen to anything I'm saying here, but data doesn't necessarily have to drive how people make decisions. I'm just suggesting what might not be supported by that data in the event that Maker holders want to make scientific decisions. |
| # / 01:16:30 | David Utrobin | Makes sense. Thanks. |
| # / 01:16:35 | David Utrobin | All right, I'm going to be hopping off the call. I'll see everybody later. |
| # / 01:16:42 | Vishesh Choudhry | All right. |
| # / 01:16:44 | David Utrobin | Maybe I'll hop off after I give you this question actually. |
| # / 01:16:53 | Vishesh Choudhry | I think my prime fear is Rich paraphrasing me. |
| # / 01:16:57 | David Utrobin | Yeah, right. |
| # / 01:16:58 | David Utrobin | Kieva asks, "Will peg get stronger as volume increases?" |
| # / 01:17:02 | Vishesh Choudhry | At the 20.5% stability fee you decrease the profitability of leverage transient demand on DAI. It's more likely than not then if there is trading volume on DAI for ETH that that would be buying volume. And so, logically, yeah, it's more likely to improve the DAI price. |
| # / 01:17:26 | Vishesh Choudhry | Oh, okay, so just long term you're asking? Yeah, I mean, obviously as transaction volume increases long term then you have a stronger demand function. That can drive up your supply more sustainably than if you have lower transaction volumes. So yeah, I mean, I think that's a pretty intuitive conclusion. |
| # / 01:17:55 | David Utrobin | Does increased liquidity also increase the time delay of the effects of a stability fee adjustment? |
| # / 01:18:04 | Vishesh Choudhry | Increased liquidity should ... did you mean to say increase? |
| # / 01:18:10 | David Utrobin | Yeah, increased liquidity. Like if the DAI market was way bigger, would the stability fee have a larger time delay or about the same? Does the liquidity of the market have any like real modifying effect? |
| # / 01:18:25 | Vishesh Choudhry | Well, presumably if you have more liquid order books you would also have more trading volume. |
| # / 01:18:31 | David Utrobin | Right. |
| # / 01:18:31 | Vishesh Choudhry | And so, if there is more trading volume on DAI and you have a higher stability fee, then again presumably a bigger percentage of that trading volume is going to be purchasing. And so, again unless the underlying asset is doing something crazy at the time. |
| # / 01:18:48 | Vishesh Choudhry | But what's important I think is as your transaction volume grows, as your order books become more liquid, more of those market impacts are transient and I think get processed by the system faster. And so, they will pass. And then, I think if they pass quicker voters will be less likely to react to them or feel the need to react to them and so the system will be stable for that reason. |
| # / 01:19:13 | David Utrobin | That's interesting. Okay, cool. |
| # / 01:19:18 | David Utrobin | All right, bye for real. Take care everybody. |
| # / 01:19:22 | Vishesh Choudhry | Okay. |
| # / 01:19:30 | Richard Brown | I think things are naturally winding down a bit, so maybe we can stick a fork it it there. Thanks everybody for your involvement. Thanks everybody that stuck around for the additional Q&A session. |
| # / 01:19:42 | Richard Brown | Final reminder, please check out the forums and get engaged. Lots of interesting things are happening and this is the opportunity to have your say. |
| # / 01:19:49 | Richard Brown | All right, thanks everybody. |