Skip to content

Governance and Risk Meeting: Ep. 41 (June 27 - 2019)

Video | Audio | Discussion

References Person Text
# / 00:00:00 Richard Brown Okay, welcome everybody to the June 27th edition of the Scientific Governance and Risk Meeting at MakerDAO. My name is Richard Brown. I'm the head of Community Development. I have a few things that I like to say at the top of each call, and I'm going to continue to say them. I've posted a link in the group chat. It's a link to our discussion thread. So, every week we have these calls. These calls last for an hour, an hour-and-a-half. We talk about a lot of really interesting things. Too many interesting things to actually dig into each one of them, so we continue these discussions in our subreddit, and we also post summaries of the calls there, for people who weren't able to attend, which are wonderfully detailed and recently have included pictures as well, so if you want to get caught up, if you want to get people caught up with the things that you've said and asked, follow that thread because we post summaries, and the conversation continues.
# / 00:00:56 Richard Brown In these calls particularly though, we have tons and tons and tons to go over, but we're also extremely interested in hearing from the community and the voters, so if you have a question that you want to have answered, please type it in the chat at the side. If you have access to a microphone, please just jump in and ask questions whenever they occur to you. We're pretty good about handling interruptions. If it turns out that we're in the middle of a flow though, we'll circle back afterwards, but we love the back and forth, so please speak up if you have something to say.
# / 00:01:28 Richard Brown Traditionally in these calls, I spend the first what I usually claim to be five minutes talking about governance. It rarely ends up being that efficient. Happily, today though, everybody is spared that because I've been racking my brains to think about what I can talk about, and I probably shouldn't be looking for reasons to waste time in these calls, so here's the short answer.
# / 00:01:49 Richard Brown Governance seems to be working the way it's supposed to, which is cool. For the last three or four weeks, the voters have signaled for no change. The stability fee seems to be doing what it's supposed to do. We have a large, longer-tail conversation about whether the cadence is correct, whether the stepping is correct, whether we should ... there are 10 or 15 various tweaks that we could talk about for governance. Those conversations about how we could change governance lasted for a while, and it's probably going to be superseded by an improvement in the governance portal, which will make some of these debates moot, so I'm not going to dig back into those today.
# / 00:02:30 Richard Brown I am going to hand it off almost immediately to Cyrus Younessi, and we can start talking about some risk. Cyrus, it's all yours.
# / 00:02:41 Cyrus Younessi Sure. Same schedule today. Think we'll start off with the monetary policy stuff from Vishesh, and then we'll jump into the next phase of collateral risk. Pretty simple. Vishesh, are you on the call?
# / 00:03:01 Vishesh Choudhry Yeah. Okay.
# / 00:03:12 Vishesh Choudhry All right. Can we see everything all right? So, all in all, it's been a pretty interesting week, I think. Just to touch on the state of the peg real quick, it's been fairly steady. There's been kind of a slight drift, and a slight increase in variance over the past month or so, but nothing to be worried about, and I think this was part of the discussion around, that spread is easier to widen when the stability fee is lower, because there's a little bit less pressure, and I think that's something we can double click on later.
# / 00:03:59 Vishesh Choudhry But more importantly is, in the past 24 hours or so, there's been a huge spike in trading volume, and DAI price has actually done pretty all right in the face of that. It seems that there is some profit booking from ETH rising, which is really interesting, because actually, there's a little bit of a time delay on this graph, the time cut-off was at around 14 million. Since then, the trading volume has come up to about 16, almost 17 million, and DAI price is actually slightly higher than this now, so I'll just show that real quick.
# / 00:04:42 Vishesh Choudhry So there's been a lot of trading activity, a lot of, I think, good trading activity for DAI. I think a fair amount of selling ETH for DAI. There was some back and forth in the sense that in the past few hours, with the drop in ETH price, we had some liquidations, fairly decent sized, and so when those liquidations happen, you have some of that sale of collateral as well.
# / 00:05:15 Vishesh Choudhry So what's interesting is there's been some transaction activity in both directions, is what I'm getting at, but DAI price has been fairly strong throughout that, so I think we're going to continue to see this. DAI is not as much of a glass draw as it had been in the past, and so even in the large selling periods, the DAI price doesn't shoot up quite as dramatically as it had in the past, and then in those selling periods, it also doesn't seem to crater as much as it had in the past. So both good indicators for the stability of DAI, I think.
# / 00:05:52 Vishesh Choudhry Just the tremendous fluidity and depth of the spread that you've seen in trading activity over the past 24 hours is, I think, indicative of the large amount of trading volume, in the sense that when you have those huge spikes in trading volume, there's some slippage on protocols like Uniswap. There's a lot more depth to the order book, so the price moves in more granular increments, and so I think you tend to see this graph fill out a little bit.
# / 00:06:26 Vishesh Choudhry So that's been pretty interesting to watch, and you can see there's, at least in the past 24 hours, a larger section that's been trading above a dollar and below a dollar, but I think that's just because given all the pressures, there was more ETH for DAI selling than anything else, so that helps to push it up, but it'll level out, I'm guessing.
# / 00:06:49 Vishesh Choudhry Yeah, I think this graph, there's not much to say. The ETH price has been consistently rising. There was that sudden drop just kind of overnight, and I think what we'll see is with that sudden drop, again, the was liquidations, a little bit of a boon for DAI price, but not much change overall. What's interesting is generally when there's that profit booking happening, you'll see a commensurate decrease in DAI supply, or at least a commensurate rise in collateralization ratio, because people get scared of those liquidations, and they punch in a little bit more collateral, but also the price went down, so in nominal terms all the collateral decreased in value as well, so those kind of balance out, where in the extremely short term you would see a bit of a drop in collateralization, and then you will see it rise up as a counter effect.
# / 00:07:54 Vishesh Choudhry So, as far as the supply, it's been rising somewhat consistently since ... 17.5%, then 16.5%. It's up to 86.4 on here, but in reality it's up to about 87 million right now. So, what's really interesting about that is there were a ton of transactions, both on the minting and burning side, as well as the liquidations that were going on, so just a ton of activity in general, in the past 24 hours. And actually in the past week, there's been a fair amount of activity in general.
# / 00:08:34 Vishesh Choudhry So, about five days ago, there was two-and-a-half million that was burned and two million that was minted. Then the next day there was a million that was minted, and relatively little burned, and then in the past 24 hours, about three million-plus minted, and then about two million wiped. So even though, net, the supply has not changed dramatically quickly, it is a consistent increase. There's been a ton of action in the usage.
# / 00:09:06 David Utrobin Vishesh? A question.
# / 00:09:06 Vishesh Choudhry Yeah?
# / 00:09:09 David Utrobin I have a quick question as to something that I think would be an interesting indicator. Whenever I see large wipes, I always go on DaiStats.com and check out the burner, and I know that if the burner didn't increase by a lot of MKR, that that huge amount of wiped debt is very short-term. Do you think that that's something that's useful in any way?
# / 00:09:34 Vishesh Choudhry Yes, and maybe I can give you a preview of something interesting right at the end here, in terms of the fees that are accrued, and I think that'll help illuminate that question around accrued fees versus short-term debt getting paid back. But I actually will touch on that just now.
# / 00:09:59 Vishesh Choudhry So, the age of the debt that is getting paid back, that has been consistently cratering since just after the 19.5% increase. What we saw was the age of open debt was consistently decreasing, so a lot of new debt being created, and a fair amount of old debt being paid back. That had slowed as of the 16.5% increase, but with the past few days, what's been going on with ETH price, there hasn't been that much more old debt being paid back, but there has been a fair amount of new debt being created. So that does kind of actually touch on your question, David, which is, it has been a fair amount of new debt getting paid back, but we can actually look at the fees a little bit later on as well.
# / 00:11:00 Alex Evans Vishesh, can you go back to that graph and maybe explain the formulas behind these? Open DAI days is across all open CDP's, or sorry, across all DAI, what is the average amount of days that DAI has existed?
# / 00:11:25 Vishesh Choudhry Yeah. Open DAI days is, on average, so denominator is the amount of open debt, and denominator for closed DAI days is the total amount of closed debt, which includes both debt repaid and debt liquidated, and then the numerator is just the total number of days that the debt that is currently open has been open, and then for the closed, it was, by the time it was paid back or liquidated, how many days had that debt been outstanding?
# / 00:11:00 Alex Evans Vishesh, can you go back to that graph and maybe explain the formulas behind these? Open DAI days is across all open CDPs, or sorry, across all DAI, what is the average amount of days that DAI has existed?
# / 00:11:25 Vishesh Choudhry Yeah. Open DAI days is, on average, so denominator is the amount of open debt, and denominator for closed DAI days is the total amount of closed debt, which includes both debt repaid and debt liquidated, and then the numerator is just the total number of days that the debt that is currently open has been open, and then for the closed, it was, by the time it was paid back or liquidated, how many days had that debt been outstanding?
# / 00:12:00 Alex Evans Interesting.
# / 00:12:00 Vishesh Choudhry Does that help your question?
# / 00:12:02 Alex Evans Yeah. So, as I understand it then, closed DAI days going up means that older debt is being paid back.
# / 00:12:08 Vishesh Choudhry Correct.
# / 00:12:09 Alex Evans Or, that the debt that's closing is older than the average closed debt up to that point.
# / 00:12:14 Vishesh Choudhry Yeah. So if I zoom out here, actually, what you'll see is there had in the past been a more dramatic increase in both, right? So as the DAI supply was growing, there was sort of a consistent amount, and it actually seems to be a trend, where there's this steady state of how old debt gets before it starts to get paid back, is how I would interpret it. That has consistently persisted, and it had slowed down recently, except in the past three weeks, I would say, there was a bit of an uptick.
# / 00:12:55 Alex Evans Yeah. It's interesting, is it only goes up, basically, which makes sense, because in the earlier days, it's a new system, and so all the debt that's being paid back is probably new, while ... presumably this doesn't go up forever, right? We haven't seen periods, except for really tiny ones, where the thing has gone down.
# / 00:13:18 Vishesh Choudhry Right. So, I would not expect the closed debt age to decrease for quite a while, and actually theoretically, and I have to double check on this, but I think it could continue to increase without ever needing to decrease, in the sense that if the DAI supply continues to grow, the average age could just level out, for debt getting paid back. It doesn't necessarily have to decrease. That would be if it hits this equilibrium, steady state. But it is interesting-
# / 00:13:57 Alex Evans Yeah, what's amazing about this, to me, is how closely it coincides with the higher-stability fee regime. Right as we hit that 19.5, it starts plummeting, and as I'm interpreting that, DAI is circulating in and out, and circulating out at a much faster rate.
# / 00:14:21 Vishesh Choudhry Yeah. Well, there's a couple of reasons for that. One, yes, higher-stability fee really, I think, lit a fire under a lot of people, and prompted a lot of repayments. Some refinancing, but I think a fair amount of repayment of old debt, which is what you really want to see out of the system. If you saw this open DAI days just continually rising, that's in some sense an accrual of risk, or at the very, very least a delaying of revenues for Maker holders, which is not good. What's interesting-
# / 00:15:05 Speaker 6 We can't ignore the drastically increasing ETH price in this as well. The timing of the 19.5 stability fee, it's correlated to the bullishness of CDP holders, so it's kind of, to me, what I saw in the increasing stability fee was one of the first leading indicators of a bull market, or rising ETH price. The stability fee was a proxy for the sentiment of the people, so those things are going to be correlated, I think.
# / 00:15:38 Vishesh Choudhry Yeah. 100%. I don't mean to claim that this is completely the stability fee.
# / 00:15:44 Speaker 6 Right. Right.
# / 00:15:44 Vishesh Choudhry But all things kind of go hand in hand, and no one item exists in a vacuum. So, one of the interesting things is the stability fee also has an impact on what happens with secondary lending, and so to some extent, you saw an increase from maybe two million DAI being supplied on secondary lending platforms to about 14 in that same sort of time span. This is a shorter time span graph, but it's still that May to June period. And then you saw the amount of DAI being borrowed on those platforms as well rise to about 11-and-a-half million from basically nothing, maybe one-to-two million.
# / 00:16:29 Vishesh Choudhry And so to some extent, 10 million-plus DAI was roughly refinanced, you could think about it, and so, out of a 90 million, or 85 million supply, that does contribute to some of the drop in the age. And I do want to touch on the fees in a bit, I'll pull that out, but also you do see this interesting pattern of a continual accrual of unpaid fees, and some periodic jumps in fees getting paid back. And there appears to be a bit of periodicity emerging, in terms of when fees get paid back, when new CDPs get opened, when old CDPs get closed out. Almost on a monthly cycle, but it's a little premature for me to be guessing trends like that.
# / 00:17:27 Vishesh Choudhry So to just kind of touch on that, is you do see almost every month, like clockwork, this dip in the percentage of draws that are happening on new CDPs versus old, so it's almost like a cycle, but I don't want to get too cute here. So what's interesting is during that same time frame of this run up, we saw far fewer draws happening on new CDPs versus old, so more of the debt that was being pulled out of Maker was being pulled out of preexisting CDPs, and presumably because the overall outstanding loan amount was continually increasing during that time, that more of those new loans were coming from secondary lending platforms.
# / 00:18:23 David Utrobin Vishesh, I have a quick question about the graph right above this one. No, no, no. Lower. One more down. Yeah. So, new draws versus existing CDPs. Are you counting that as the first draw on a new CDP, and then the second draw after that falls into the red category of existing CDPs? What's the specifics there?
# / 00:18:48 Vishesh Choudhry It's actually not. It's draws on CDPs that were opened within a time frame. I think it's within seven days or so.
# / 00:18:59 David Utrobin Got it. Okay. All right, that makes more sense. Okay.
# / 00:19:02 Vishesh Choudhry Yeah. So, collateralization has continually been rising over time. I think as the ETH price has continually been rising over time, people have perceived this greater height to fall from, and so they've kept a little bit more collateral in the system. What's interesting to note is in the past 24, 48 hours, a ton of collateral has been freed from the system. This is not even with liquidations and such, but also people have pulled out some collateral. And that tends to happen when the ETH price goes up, because presumably as the stability fee was reduced, 17, 16, around here, this 450-ish level was something that people were presumably comfortable with. As ETH price continued to rise, that number just ran up from ... you know, simple math?
# / 00:19:56 Vishesh Choudhry But then as people saw the ETH price start to rise up, they started to remove some collateral from the system. Now, I'm guessing with the drop in ETH price, you'll see the opposite trend, where people will counter the drops in the liquidations, and you'll probably see this go back up, but I don't want to call my shot too early on that.
# / 00:20:15 Vishesh Choudhry I'm going to pause for questions, or if I ran over time. I wasn't checking the clock. But I do want to pull up the ETH if I can, so just give me a moment. If anybody has any questions, I guess now would be a good time, or if Rich wants to cut me off.
# / 00:20:40 David Utrobin Did you see anything interesting with secondary lending protocols in the last week?
# / 00:20:50 Vishesh Choudhry I did. The secondary lending platforms, as I showed, the total borrow and supply volumes both have run up since the ... and they've been running up for the past week or so. The borrow has outpaced supply a little bit, and so those rates have come up a little bit. Actually, I can show that graph. [inaudible 00:21:25].
# / 00:21:27 Vishesh Choudhry So what you've seen is kind of consistently, dYdX has had a higher rate than Compound, both for ... Sorry, for borrow. For supply, it's generally been lower, so that's an interesting little tidbit. The supply rates do tend to flip-flop a little bit more than the borrow rates, but the weighted average of the amount that is being lent on those platforms, and what the rates are, has run up a little bit in the past week to almost touching the stability fee, and then at times, dYdX has run over it.
# / 00:22:13 Vishesh Choudhry So that is, I think, a curious tidbit, with this rise in volume that I was talking about. But what's interesting is the supply volume running up on those volumes, because when you see a fair amount more debt being drawn out of MakerDAO, and you see a fair amount of DAI for ETH trading, some people I guess are either paying back those loans or writing them off, booking the profits, ignoring the fees, and then perhaps that DAI that's getting purchased by people is being lent out on those platforms.
# / 00:22:58 Vishesh Choudhry Now, the implication for that on stability fees is curious, because if there exists this pool of loans that people have walked away from, or CDPs they don't care about, and those are recurring fees that might never get realized, those would skew the numbers, but it's really hard to filter those out.
# / 00:23:18 David Utrobin Thank you, man.
# / 00:23:21 Richard Brown All right, we're at the half-hour mark, so let's switch over to Cyrus, who's promised some intriguing slide work for the day, so I'm going to make sure ... my eyes on that. Cyrus, why don't you take it away?
# / 00:23:36 Cyrus Younessi Sure.
# / 00:23:44 Richard Brown I'm going to reuse that phrase, "slide work," constantly over the course of these next few meetings, so I hope you're prepared for that.
# / 00:23:50 Richard Brown There it is, man.
# / 00:24:14 Speaker 6 Vishesh, that was a really cool presentation. That was a lot of really cool numbers. You presented it well.
# / 00:24:22 Vishesh Choudhry Thanks.
# / 00:24:23 Cyrus Younessi Okay, are we good? You guys can see the screen, right?
# / 00:24:23 David Utrobin Yes.
# / 00:24:43 Cyrus Younessi Okay. Today we're going to talk a little bit more about tackling stability fees and expected losses. Just as a reminder where we are in the overall process, we're obviously doing the quant modeling. I think in the next couple of weeks, either next week or the week after, we'll circle back to the collateral onboarding process, and then resume with the quant stuff.
# / 00:25:23 Cyrus Younessi This is the mini outline for the quantitative stuff. ... about credit risk. Next few weeks, we'll talk a little bit about liquidation ratio, which is really just a continuation of stability fee discussions. In general, for this week and going forward, it's a bit difficult at times for me to figure out what the right level of detail is for these presentations, so if anything is too detailed, please ask for clarification, or if it's not detailed-enough, please, I encourage you guys to jump in with questions as well.
# / 00:26:19 Cyrus Younessi Okay, so quick recap from last week. We talked a bit over risk rating from our due diligence report. We still haven't seen what to do with that yet. We talked a little bit about how the risk premium as attached to a CDP is some function of its loss distribution, and expected loss being the average, or mean, of that loss distribution. That expected loss is itself a function of the probability of default of a CDP, the severity of the loss given default, and the amount of actual debt attached to a CDP.
# / 00:27:07 Cyrus Younessi And in turn, those metrics clearly have something to do with the underlying asset price and its behavior, as well as user-induced behavior, such as adding or removing collateral or adding and removing DAI. And what we are going to look at today is how to essentially think about the collateral asset, how to think about behavior, and then how to package everything up together into calculating the stability fee.
# / 00:27:50 Cyrus Younessi What is this poll here? Rich, are you doing quick polls?
# / 00:28:02 Richard Brown I'm gathering signals while you're doing your slide work.
# / 00:28:04 Cyrus Younessi Okay.
# / 00:28:04 Richard Brown It's unrelated.
# / 00:28:12 Cyrus Younessi Okay. So in general, credit risk is a pretty hard problem, mainly due to the randomness of these underlying assets, right? Anybody who claims to know what ETH is going to do, come talk to me on the side. But nonetheless, there's a lot of different approaches and ideas and models, in how to gain some understanding of the underlying asset. Probably too many to cover today, or certainly in 30 minutes.
# / 00:28:57 Cyrus Younessi But intuitively, one simple way of trying to understand an asset is by just looking at its history, right? How has a certain asset behaved in the past? And as we all know, past behavior does not indicate future behavior, but it's certainly a good starting point, and in practice, due to just the difficulties, it's often just used as a substitute, and then you can try to analyze how good it behaves over time.
# / 00:29:29 Cyrus Younessi Simple example, a company in the energy sector wants to borrow money, right? They want to issue some debt. One way to calculate the probability of default for such a company is to just look at all companies in that sector over the last 50 years, right? So you have 1,000 companies in the energy sector, and 20 of them defaulted. Well, maybe 2% is a fairly good starting point for what you expect the company that you're looking at is likely to default on its debt. And then from a similar analysis, you can also calculate its potential losses.
# / 00:30:17 Cyrus Younessi You can of course improve on this with fundamental analysis, and this kind of dives into credit ratings, where they basically compile all this data, do the due diligence, everything, and package it up into a rating. From these ratings, you can often derive the metrics that you're looking for. So, interesting question is, does this work with CDPs? Can we just look at CDPs over the past year-and-a-half, right? There's been some thousands of CDPs and some hundreds of liquidations, and just take a ratio and say roughly 5% of CDPs are likely to default.
# / 00:31:02 Cyrus Younessi The answer is probably not, simply because we just don't have enough data that spans a wide-enough range of the underlying asset's behavior either. So in the first example, we're talking about decades of data versus just a year-and-a-half, and we can certainly use it as a starting point, but probably in and of itself, not sufficient.
# / 00:31:42 Cyrus Younessi Okay, so as I've alluded to many times before, probably going to explore a variety of models. My preference is to start with an academic model, because it, in my opinion, helps build up a lot of the intuition behind this stuff. Talk a little bit about why it's not so great, see what we can save, improve it, iterate it and just kind of keep doing this over and over again. So in some ways, today's presentation does not reflect the reality, but they might, so let's see. Let's just dive right into it.
# / 00:32:24 Cyrus Younessi Okay, so first up, let's define some variables. Collateral asset price, we'll just call it A. The t subscript is of course the time, so basically the price of ETH over time. And CDP owners lock up a certain amount of collateral value, which is the price of ETH times the quantity of ETH that they lock up. We'll call that collateral value V. They draw K debt, they pay a stability fee, s, and maintain a collateralization ratio of V over K, which we'll call X. We'll also define the whole space of user behavior just by F for now. We're probably not going to talk too much about user behavior today, because I don't think we'll have time, but essentially we have to account for the fact that users, as I mentioned, can add or subtract collateral or pay down or draw more DAI.
# / 00:33:44 Cyrus Younessi This is important because CDPs are not static, so if you have a CDP that has a liquidation price of 200, I think as we all know, as we approach that 200 price point, users will tend to take action that will push their liquidation price lower. They'll either pay back debt or they'll add collateral. And this is just important for modeling. Essentially, the probability of the CDP being liquidated changes over time.
# / 00:34:31 Cyrus Younessi Collateral value V is a function of the asset price and the behavior. Just talked about this. The debt is a function of the stability fee, time and user behavior. So stability fee and time, what I mean by that is the accrued stability fees over time, and the user behavior is whether they are drawing or wiping DAI. All right, in addition to standard user behavior of trying to avoid liquidations, there are some technical edge cases as well, in particular the oracle security module, which has a one hour price feed delay, so essentially if the collateral asset just completely plummets, users are incentivized to draw as much DAI as possible, since the collateral's worthless anyway, so we need to be cognizant of this as well.
# / 00:35:43 Cyrus Younessi So, the time horizon of CDP is well known. They're open-ended, right? But it's helpful to just assume a one-year, forward-looking period. Somewhat arbitrary, but also fairly standard, and it doesn't really change analysis too much. And we know that CDPs default when they hit the liquidation ratio L, and this notation here at the end is basically saying, the first instance at which the collateralization ratio crosses the liquidation ratio is when default occurs. We'll call it T1, which is kind of arbitrary.
# / 00:36:36 Cyrus Younessi Any questions so far? Okay. Cool.
# / 00:36:47 Cyrus Younessi Okay, a few more. A CDP is liquidated with probability Q, and so this is the probability that that T1 threshold is reached before our one-year time horizon, so this is definitely something critical that we're interested in. And we know that this probability depends on the evolution of the collateralization ratio, which in turn depends on the value of the collateral and the amount of DAI debt, which in turn depends on this stuff we talked of before, asset pricing, user behavior.
# / 00:37:29 Cyrus Younessi At liquidation, fraction of the debt that is recovered from the collateral sale is w. Somewhat arbitrary variables. This represents loss given default, or one minus loss given default. And as we all know, this represents the slippage adjustment for when you liquidate the collateral. The less slippage there is, the higher amount of the debt that you recover. And then of course we have this amount offset by the liquidation penalty as well, or I should say, added back in. r is the risk-free rate, from traditional world. Plays a role. And our credit spread, which is essentially what we're solving for, is defined by s minus r, and we'll ignore DSR adjustments for now. And the expected loss of a single CDP, which is what ultimately we're trying to solve for here, is the product of these three variables.
# / 00:38:59 Cyrus Younessi Probably all good. Any questions so far on any of these pre-model specifications? Cool.
# / 00:39:11 Cyrus Younessi Okay. So we already know we want to solve for probabilities, losses and exposure amounts, and we need to develop some intuition around the asset price and user behavior. Standard approach to modeling an asset price is to apply what's known as a stochastic process, which we'll talk about in a second, and it just helps us understand the asset in a variety of ways. It's not perfect, but it's fairly standard and it has quite a history behind it.
# / 00:39:54 Cyrus Younessi And then to model user behavior, we can analyze historical CDP stats and then go from there. Vishesh and Alex have done a lot of great work in this area, and I hope, I think, we'll be able to explore that a little bit later as well. And in some instances, you can just look at user behavior correctly, and maybe even bypass collateral assets.
# / 00:40:26 Cyrus Younessi So the way I like to think about it is, you can look at the CDPs, loans, by themselves, or you can try to pick the CDP apart and look a little bit deeper at its underlying assets. I think there are advantages and disadvantages to both, and in some cases we can even do a hybrid. To clarify that a bit, if you think about that corporate debt example I used a few slides ago, it's not always easy or even possible to actually model the underlying assets that is behind debt. Sometimes you can only just look at it from the debt level, and try to model how the debt behaves over time, or historically, or in some cases, such as our CDPs with ETH, you can actually model the underlying asset, which I think gives us an advantage. But there could very easily come a time where the collateral that is put into CDPs is just a representation, right? It's not something that can be modeled directly, or easily. So it's interesting to have different approaches in mind.
# / 00:41:59 Cyrus Younessi Okay, so we're going to talk a little bit about that stochastic model for the ETH price that I was talking about. And I'm going to impose a bunch of assumptions right now, and then relax them slowly, because it's a lot to do it all at once. So I'm going to ignore user behavior for now. The ability for users to add or remove collateral at will at any time, and to add or subtract DAI at any time, makes things really complicated. It causes a lot of headache. Just adds a lot of complexity, which is what makes MakerDAO such an awesome protocol, right? It's extremely flexible, but it becomes a little bit difficult to analyze.
# / 00:42:51 Cyrus Younessi Going to assume no slippage for collateral amount. So essentially, as soon as liquidation is triggered, you can recover everything. Constant risk-free rate. Constant volatility. And volatility here is a historical volatility, based off of the trading history. Returns are normally distributed. I'll get into it. That is for next slide. No early liquidations, which seems like a silly assumption, and asset price cannot gap up and down.
# / 00:43:31 Cyrus Younessi Okay, so essentially the reason why I've imposed all these seemingly random assumptions is if you go through all this, then you can create a nice, little model called the Merton model, which is almost a 50-year-old model by now, and this is actually part of the pioneering stages of modern options pricing theory. Basically, if you assume that the collateral asset behaves a certain way, you can do interesting stuff with it.
# / 00:44:10 Cyrus Younessi So, this equation up here at the top is known as a stochastic differential equation, and essentially what it's saying is that an asset price can be reduced to ... In every small time increment, it will move a little bit in accordance with its historical mean, plus some random component that is scaled by the volatility. And that random component is quite literally just ... almost think of it as an RNG. Essentially, you can simulate thousands of different price paths per collateral asset, where each price path is a potential likely simulation for the evolution of this asset price.
# / 00:45:12 Cyrus Younessi So you can say Ether will tend to go up over time, according to some expected return or some mean, but then of course there's this volatility component, which will cause it to veer in all sorts of ways. And so what we can do is we can calculate these sample paths, and then let's say we compute thousands of them. And then we can merely just calculate the number of them which end up below our liquidation threshold.
# / 00:45:53 Alex Evans So, Cyrus, are these the returns that you were making assumptions about their distribution? Is this a return of the underlying asset?
# / 00:46:05 Cyrus Younessi Yes. You mean these?
# / 00:46:08 Alex Evans Yeah, so returns are normally distributed. So this would mean the logs of the returns are normally distributed, so the returns themselves would be log normally distributed, right?
# / 00:46:20 Cyrus Younessi Yeah. Yeah, yeah. Yeah, exactly.
# / 00:46:25 Alex Evans Got it.
# / 00:46:34 Cyrus Younessi Okay, so a few things we can glean from this kind of model. One is, as I mentioned, we can calculate the frequency of the paths which end up below our threshold, which will help us determine the probability distribution, and actually at the same time, we can just calculate how much loss each particular path is associated with, and essentially knock out two birds with one stone here. Kind of, those two metrics come together.
# / 00:47:11 Cyrus Younessi So let's keep going for a bit, and hopefully it becomes more clear, but if not, don't worry about it, because this is fairly theoretical as well. So, first assumption that we can relax is no early liquidations. That's kind of an easy one to get rid of, and here we can kind of see the same style of model, but essentially as soon as the paths breach the threshold, liquidation is triggered. And that's more in line with the model that we're used to. This green one should not have an arrow too that.
# / 00:48:07 Cyrus Younessi And let's get rid of this asset price cannot gap up or down as well. And so we've modified our process a little bit by adding a jump variable, which essentially means that the asset price can gap up or down at any random moment, and this Y variable here, this dY, is essentially what's known as a Poisson process, which essentially means that you know a event will occur with a certain frequency, and then the goal is to try to calculate what the frequency of that event occurring is, and with what size.
# / 00:49:02 Cyrus Younessi And so one interesting question is how to actually calculate the frequency of these big jumps up or down. This is something we've been doing a little bit of research on, essentially poring through historical data for markets and trying to say, "Okay, how many times has ETH had a sudden crash? What was the magnitude of the moves?" and so forth. And then going from there. So that's some interesting research we've been doing on the side. I'm happy to chat about it on the chat.
# / 00:49:47 Cyrus Younessi Interestingly enough, we haven't actually used our risk rating for anything yet. And essentially, even without it, with just the trading data, we could have actually calculated everything up to the stability fee, but that's not necessarily very reflective of reality, and the easiest example for that would be the ETH versus some fiat-backed stablecoin, which everybody knows has a chance of having a sudden plummet, right?
# / 00:50:30 Cyrus Younessi So if we wanted to incorporate some likelihood of an asset collapsing, we wouldn't be able to glean that from the historical data, and this is where we could incorporate our qualitative information for collateral asset. So we can essentially use that to say we believe that the intensity of the frequency with which these jumps are likely to happen is higher, lower, based on the risk characteristics of the organization behind it.
# / 00:51:18 Cyrus Younessi Okay, so from the rest of this list, not going to be able to cover everything today. So this first one, behavior. Super-important, which we will get to, hopefully next call. Slippage as well, when it comes to calculating the liquidation ratio and evaluating the liquidity of the markets as well. This third one is also fairly easy to relax. Essentially requires using its own interest rate stochastic model. And these last two are a little bit contentious, so we'll talk about that another time as well.
# / 00:52:18 Cyrus Younessi So, the next steps after this, what we've done so far is single asset for a single CDP, and we want to evolve this to single asset with multiple CDPs. So right now we have thousands of CDPs, each with their own debt, their own collateral, their own liquidation price, and we can't do a full simulation one by one for CDP. It's just not feasible. So what we would have to do is take distribution of CDPs and model them as a whole. And that's been a lot of the work that we've been doing.
# / 00:53:04 Cyrus Younessi And then furthermore, we don't want to ideally have just one liquidation ratio bucket, but you might want to have CDPs with a 150% liquidation ratio that have a certain stability fee, but another set of CDPs that have a 200% liquidation ratio that might have a lower stability fee, because they've committed to keeping a higher collateral reserve, so these are all things we can do with the model that I've just described. Of course this would require modeling the actual collateralization ratio as opposed to the asset, which is just a few small tweaks. Still a lot of research to do, in terms of sharpening our corners, but we're getting there for sure.
# / 00:54:06 Cyrus Younessi And so essentially, everything we've done so far is to calculate that expected loss, which, how much do you expect to lose on a particular CDP, or a group of CDPs, right? And big question is to how to get from this to the actual stability fee calculation. So, the easiest way to think about that is to ... let's first take a quick minor step back, and let's look at a hypothetical risk-free CDP. So imagine a CDP that takes out 100 DAI debt, stability fee of 5%, right? They'll pay back 105 in a year.
# / 00:54:56 Cyrus Younessi And let's assume that that stability fee payback is completely independent of the collateral asset price. That no matter what fee, it goes to zero. You're still getting back 105 DAI. Well, if that's the case then that CDP would not have a stability fee larger than the risk-free rate, because there's no risk to it, right?
# / 00:55:28 Cyrus Younessi I know it's the turn of the hour, Rich. I'll be done in a couple of slides. Now, of course stability fees do depend on the underlying asset, and we know that below a certain price, which is 100% collateralization ratio, there are losses that start to accrue, thinking of this from the perspective of the payoff structure of the CDP.
# / 00:56:02 Cyrus Younessi Oh, this slide didn't come out good. So, essentially what I'm trying to get at here is that ... let me get the ... so, we have this risk-free CDP, and we have a CDP here that looks more like our actual CDPs, and that difference here is this chart here in the middle, which, if anyone has picked up on the clues, this whole presentation is very reminiscent of standard options theory. And essentially, the payoff structure for a CDP resembles, behaviorally, an implicit put option on the underlying assets. There's no actual put option here, but it's a very helpful heuristic to help us model the expected loss.
# / 00:57:14 Cyrus Younessi The expected loss is the price of a implicit put option on ETH, where the strike price is equal to the amount of DAI debt that's been borrowed. So another way to explain this is, right now as a CDP owner, you may ... let's say you were the counterparty to the CDP owner. You have that downside risk, where if the collateral value drops below the amount, then you would start to accrue losses, and those losses, which is what we've been trying to calculate so far, resembles that of a put option.
# / 00:58:19 Cyrus Younessi And so, just a hypothetical example, collateralized one ETH, borrowed 200 DAI, and assume that your expected loss, or that the price of the put option with a strike of 200 is $10. Then what you are trying to do is find the yield of this risky CDP, which is a combination of the risk-free and that implicit put, and that is actually how to get your stability fee.
# / 00:59:04 Cyrus Younessi Any questions?
# / 00:59:04 Cyrus Younessi Okay. Last slide. Was this all worth it? Maybe. Definitely a good amount of complexity here, and a good amount of assumptions. In particular, the log normal distribution of the asset price, which is what underpins the stochastic model, underprices the tails, and that's how you end up with outsized moves that you don't expect. It's potentially not great for assets that don't have a lot of historical data, and in many ways, looking at empirical models would be better in some regards, because I think it could have a better tail distribution, but as I've mentioned, it doesn't necessarily factor in the right qualitative risk characteristics, so I think some ideal blend could be combining the empirical with some sensible, qualitative risk adjustment, and then, interestingly enough, there is also a thing called "expert models," where you can just defer to somebody who claims to know what they're talking about, but we won't be doing that one, unless you guys want Rich to just determine all the risk parameters by himself.
# / 01:00:54 Cyrus Younessi That's it.
# / 01:00:55 Richard Brown All right. Thank you, Cyrus Younessi. There's a lot to digest here. If people have questions, they want to absorb some of this stuff, surface it in the chat at the side. I guess this is the slight segue to talk about signaling and debate, and these deeper discussions, because this is a path that MakerDAO is going to be heading down more and more as the world and crypto gets more complicated, particularly in the governance calls and particularly as we begin to shift focus more towards risk.
# / 01:01:34 Richard Brown Right now, the deep discussion venue that we have, unfortunately, is Reddit, and Reddit, and this may come as a surprise to people, but it's not designed for deep and thoughtful discussion. So what we're going to do, what we have done, is we're going to release very soon a forum for the MakerDAO ecosystem, and in that forum we have numerous top-level categories. Those top-level categories will be, and, slash, are, community, governance and risk, and X, Y and Z as well.
# / 01:02:11 Richard Brown And this is going to be where we begin to really get serious about debates, long tail debates, so we'll be able to take some of the really amazing manifestos that have been posted into Reddit, and were subsequently ignored, because there was just too much for people with a Reddit attention span to absorb, myself included, transitioning those into an official forum, sort of along the lines of EthResear.ch, where we have a proposal, a manifesto, a slide desk, algorithms from Cyrus's team, and we can spend days, weeks and potentially months debating the finer points of those things.
# / 01:02:55 Richard Brown The target for that launch is July first, the week ... It's called the first week of July, for our commencing, but that's coming, and so I wanted to put that out there, that if you don't have a question right now, you haven't missed the boat. We recognize the need for a long-term, very detailed debating mechanism, with some light reputation management and polling built in, and that's been built. Last tweaks are being applied, and it will be released to the community, first week of July.
# / 01:03:30 Richard Brown Cyrus, did you have anything else that you wanted to leave us with, or can we move to the weekly narrative now? I know that you'd threatened to drop a bomb on us, and I'm trying to give you an out of whether you still want to do that.
# / 01:03:47 Cyrus Younessi I did what now?
# / 01:03:48 Richard Brown You did what now? Hey, Cyrus, do you want to talk about the debt ceiling? How's that?
# / 01:03:55 Cyrus Younessi Sure.
# / 01:03:57 Richard Brown So let's lay the groundwork for it, and see if we can start a debate, but I also want to make sure that we give Matthew enough time to have his discussion, so must be cognizant of the time. It's ten after the hour.
# / 01:04:11 Cyrus Younessi So, who's going to go first?
# / 01:04:12 Richard Brown You're going to drop bombs on us, and then we're going to talk to Matthew.
# / 01:04:18 Cyrus Younessi Yeah. I was just noticing the debt ceiling was starting to creep up a little bit. I thought it would be interesting to discuss. As a quick reminder, or refresher, there is obviously some risks involved with not increasing the debt ceiling.
# / 01:04:48 Richard Brown All right, so let's bring this thing ... I'm going to try to put some energy into this convo right now, so here's the idea.
# / 01:04:56 Cyrus Younessi I'm exhausted from the presentation, yeah?
# / 01:05:00 Richard Brown Yeah, that slide work takes a lot out of you, I can tell. So here's the idea. We've constantly been debating the fact that we have a limited number of levers in this ecosystem. The lever that's been exposed right now is the stability fee. There was a lot of back and forth, a lot of very considered discussions about whether we should be playing with more levers than that.
# / 01:05:21 Richard Brown I think that we've seen that the measured discussion was correct. We fixed things as a community with the stability fee, but the market conditions have changed. Inventory levels presumably have recovered to healthy levels. Secondary lending platforms have arisen to take some of the pressure off of us. We are no longer in an extended bear market. Well, maybe it just started in the last hour again, so who knows? But up until today, we weren't in an extended bear market, so maybe it's time to actually start looking at the fact that raising or tweaking the debt ceiling might actually be a valid option at this point, because we're not adding insult to injury, and we're not trying to disguise the fact or overcompensate for supply and demand issues. We actually have a healthy ecosystem that could potentially demand a higher debt ceiling.
# / 01:06:15 Richard Brown And that's as far as my understanding goes, so I can jazz up the convo a bit, but I'm not clever-enough to know whether it's a good idea or not, and I think that's where the analysts and the community needs to jump in with some suggestions, or some ideas, or some perspective on whether ... Is it smart to raise the debt ceiling, or should we leave things the way they are? That's the question.
# / 01:06:40 Cyrus Younessi Yeah. Let's hear from ... Am I on mute? No. Let's hear from Matthew and Vishesh on that, because I'm sure they have some good thoughts on that as well.
# / 01:06:47 Richard Brown Yeah. Vishesh, go. Tell us what's happening. What would Vishesh do?.
# / 01:06:55 Vishesh Choudhry All right. I hate that meme. But obviously I think the thing to think about is we know we don't want to hit the debt ceiling. I think that's fairly agreed-upon. We can debate that if not, but I think the question is, what is the likelihood that we hit the debt ceiling? What is the downside of having the debt ceiling be slightly higher if we're not going to hit it? Which I think is relatively low. And in what scenario should we worry about the debt ceiling?
# / 01:07:32 Vishesh Choudhry And so I know that's a lot of questions. I'm basically unpacking your question, Rich, into four or five. But I think some of the answers are pretty clear, which is, okay, let's assume we don't want to hit it. Let's assume there's relatively low risk in increasing the debt ceiling if we are not going to hit it. Now, the one caveat is, if the debt ceiling is increased and somebody goes ahead and aggressively votes down a stability fee, that's your risk scenario.
# / 01:08:06 Vishesh Choudhry So given that both of those things are under the control of this governance process, presumably there's some higher hive mind intelligence here that realizes, "Don't do both of those things." Because in the scenario that stability fee is artificially depressed or inflated from external market factors, and those external market factors change, like, oh, I don't know, a sudden drop in ETH price, then you might be worried about people deciding to lever up, DAI price potentially taking a hit, and us deciding, "Hey, we need to react and make a change to the stability fee." So that kind of reactionary behavior in response to market forces is riskier when there's a higher debt ceiling, because then you-
# / 01:08:58 Richard Brown That's interesting. Sorry.
# / 01:08:58 Vishesh Choudhry [crosstalk 01:08:58] 20 million more ... overnight.
# / 01:09:02 Richard Brown So if we did expose two levers, it's going to be even harder for us to figure out which one we should be pulling, and also disincentivizing the community or making sure that the voters don't pick the easiest, short-term win in that model.
# / 01:09:20 Vishesh Choudhry Yeah. You're trading one risk for another, and the risk that you are reducing is the risk that you hit the debt ceiling, and you have to work harder to unwind some of the negative effects that get bubbled up from doing it, so the risk that you are gaining is the risk that the governance process lets a fox into the henhouse, reduces the stability fee and somehow somebody decides to bully the situation and overnight mints a bunch of DAI, either crashing the DAI price or creating a whole bunch of other negative effects.
# / 01:09:53 Cyrus Younessi Yeah, you want to walk this tightrope where you don't have the risk of hitting the debt ceiling too quickly, but you don't leave more open room that you really need to. So you want to just keep a safe distance from the debt ceiling.
# / 01:10:15 Richard Brown There's an order of operations-
# / 01:10:16 Vishesh Choudhry And the weighting changes, right?
# / 01:10:19 Cyrus Younessi The what?
# / 01:10:20 Richard Brown I was just saying-
# / 01:10:21 Vishesh Choudhry The weighting changes, so ... delays are funny.
# / 01:10:28 Richard Brown Sorry, Vishesh.
# / 01:10:28 Vishesh Choudhry Go for it, Rich.
# / 01:10:29 Richard Brown All right, well, I'm just saying that potentially there's an order of operations here, where you make the system healthy with the primary lever, which is the stability fee, and once you have established, or if you have some kind of a monitoring system that can tell you if the inventory levels are healthy due to the stability fees being appropriate, and that matching user behavior, secondary lending platforms, whatever, then it's safe to raise and lower the debt ceiling, but the debt ceiling is not the primary risk parameter to tweak, right? Is that the idea?
# / 01:11:04 Vishesh Choudhry So, I think there's two caveats. One is, to the extent that there is a 14-million float on secondary lending platforms in terms of DAI supply, that reduces the fear that you should have of hitting the debt ceiling. The more excess supply sits on secondary lending platforms, the less worrisome it should be that you're nearing the debt ceiling, because that's kind of your buffer range.
# / 01:11:31 Vishesh Choudhry As that comes down, yeah, you have the chance of grinding bone against bone and getting some pain out of that, but I don't think we're quite there. I do think it's definitely worth starting to raise that conversation and starting to be aware that that may be something we want to do soon, because even at the current stability fees, the supply has been increasing.
# / 01:11:55 Vishesh Choudhry Now, this supply on those secondary lending platforms has also been increasing, so it may be a little bit misleading to just look at the DAI supply. But, that being said, I think it's a good conversation. I think it's worth raising. It's something we should think about soon, and on the flip side, what I'm saying is if you do decide to increase the debt ceiling, then you've got to be a hell of a lot more careful about decreasing the stability fee, because even if it's all quiet for a week or two, or a month even, and nobody prints a massive amount of DAI, if the stability fee is lowered, it just increases your susceptibility to that kind of an event occurring.
# / 01:12:35 Vishesh Choudhry And so that's what I just mean. Is, increase your weight on how cautious you are about decreasing the stability fee if you increase the debt ceiling.
# / 01:12:44 Matthew Rabinowitz You need to leave enough of a gap in there so that one rogue actor can't decide to mint a bunch of DAI and crash the place. You want to leave enough. You don't want to have so much of a buffer where you make the debt ceiling 500 million and the current outstanding DAI is 80, and if you lower the stability fee, you're just taking unnecessary risk.
# / 01:13:07 Vishesh Choudhry Right. And so the stability fee I actually look at as kind of like a speed adjustment, so when the stability fee is higher, the supply moves slower, in terms of in response to demand, and when the stability fee is lower, it moves faster. Now, it's also not just demand, but also moves faster and slower in response to ETH price changes, so the one caveat I would say is it doesn't necessarily have to be this malicious, rogue actor or whatever. It could just be ETH, and ETH could be our rogue actor, and make significant price changes, drop to 260, whatever, I don't care, and cause a bunch of different people to act on behalf of market forces, which is basically the same effect as a rogue actor.
# / 01:13:57 David Hoffman Is it possible to write into the code that if DAI ever touches the debt ceiling, then there is an automatic plus-5% stability fee implemented? Which would allow you to lower the buffer between the outstanding DAI and the debt ceiling. You would let it get closer a little bit, right?
# / 01:14:19 David Utrobin But wouldn't it depend on the state of the peg? That's also something [crosstalk 01:14:24] that-
# / 01:14:24 Matthew Rabinowitz [crosstalk 01:14:24]. I mean, you're talking about existing code and single collateral DAI that we're not going to change, A, and we're about to launch multi collateral DAI, that-
# / 01:14:33 Male Yeah, that's what I'm-
# / 01:14:33 David Hoffman I'm just talking theoretically.
# / 01:14:33 Matthew Rabinowitz No, no, I get it, but the point in general, and I'm about to do a little presentation of this, is that in single collateral DAI, the fundamental lever to control the supply and demand was the stability fee, right? In multi collateral DAI, to some extent that's going to shift away, and it should shift away. So changing the underlying risk premium for a specific collateral shouldn't be the method to control supply and demand.
# / 01:15:02 Vishesh Choudhry Yeah, and I'll echo. Probably there are some Redditors who are droning right now. The changing the stability fee doesn't necessarily solve your problem, right? So even if you hit the debt ceiling, you jack up the stability fee. It doesn't necessarily mean that it's going to solve the underlying friction that you run into from hitting the debt ceiling. So let's say the stability fee was 14, 13%. You hit that debt ceiling because somebody minted a bunch of DAI. People can continue to source DAI, theoretically, from secondary lending platforms, to the extent that that supply exists, but if you increase the stability fee, you may also reduce the supply that exists on those secondary lending platforms, and you may not necessarily reduce the DAI supply. It sort of depends on what the rates are.
# / 01:15:57 Vishesh Choudhry And so it could solve the problem. It doesn't necessarily solve the problem. And this is where I think some people have voiced some frustration, that the stability fee is a bit of a blind instrument. Totally true. The stability fee works well when the relative rates on other platforms are two to three percent lower, let's say, or there's a decent buffer range. If they started to run up to be equal to the stability fee for some reason, or even higher, then increasing the stability fee might not solve your problem at all, and so it's a really complex kind of problem, and so I would caution away from thinking that if you hit the debt ceiling, okay, just jack up the stability fee and that will fix everything.
# / 01:16:46 Matthew Rabinowitz Yeah, and the danger, or the scary part right now, is all the thinking we've constructed thus far under single collateral DAI, and how we use the stability fee. It's somewhat changing, right? When you go into this multi collateral stuff that we're rotating into, and it's a little segue, this whole learning process goes on, it's painfully humbling, and I want to say thank you very much to Vishesh, Cyrus, Alex and other guys in the Maker Chat and Reddit, for some of this stuff.
# / 01:17:14 Matthew Rabinowitz From the conversation we had last week, took some of the commentary that folks had and digested, what were my failings? What were the pieces that I were basically missing on that? And let me just show this. Kind of reconstructed back to what I thought was where we started, which is in single collateral DAI, where we only have Ethereum, and we know our liquidation, we know our collateralization, we know our debt ceiling, it's changeable and the risk premium is what's changing dozens and dozens of times, or will be or has been.
# / 01:17:46 Matthew Rabinowitz And that's what we are using as our fundamental, quote, unquote, collateral package. And, here's my fourth-grader drawing again, as that relates, where you've got a risk premium, and that interest rate is what we're using to balance back and forth, and how we're trying to swing back and forth to try and stabilize the price, it's not the world's best graphic, but I hope the message kind of comes across.
# / 01:18:08 Matthew Rabinowitz But the idea behind it, when we switch over to a multi collateral area, where we're going to have, insert your name of collateral, the liquidation ratio will be static for that package, so will the collateralization ratio, and so will the debt ceiling, for the most part. But the risk premium for it, it's going to get set by a risk team member. It'll be modified, maybe, weekly, but its condition's based on the risk of the collateral package as a whole. It shouldn't have anything to do with the price, as it relates to the price of DAI. It should only be risk-driven, and risk-driven to the point where this collateral package as an aggregate, as compared to the whole, should be viewed in effect as being riskless. In effect, an insurance wrapper around a collateral, of whatever that type is.
# / 01:19:00 Matthew Rabinowitz If it's Ether with X liquidation ratio and Y collateralization ratio, those are going to be different, and we set the risk premiums accordingly, but we really shouldn't have to change them that frequently. Conditions on the ground that drive the risk, and Cyrus's comment about putting and energy company, and putting their debt as a function of collateral, conditions on the ground of oil prices go to $5 a barrel. Okay, there's a high chance that company might default. The risk premium should get adjusted accordingly. But that still doesn't change anything to do with ... it shouldn't have anything to do with the price of DAI yet. It's purely derisking a collateral package to the point where it's riskless.
# / 01:19:44 Matthew Rabinowitz So the corollary to this, it's kind of in an MCD world, where we're in effect using the DSR and weighting it against the riskless offset of this litany of riskless collateral packages, some of them mint DAI, some of them burned DAI, and the DSR is what's used to mop up the riskless DAI that was minted off of them, which is kind of a little bit of a toe into what we discussed last week about, if DAI being minted isn't riskless, then you inject risk into the system because you can't really tell if the DSR is using a riskless tool to mop up DAI that was coming from a risky source.
# / 01:20:31 Matthew Rabinowitz To keep them in balance, the risk premium has to be set by risk team people that know that. This will tee into the end of this conversation about, what are we actually going to be voting on six months, in a year, in three years? And how will we be able to handle the quantity of risk premium voting? And will be even be voting on that, or will we be just voting on the recommendations from the risk teams?
# / 01:20:59 Matthew Rabinowitz So it's because of each collateral package, when they're dialed in or tuned and they are basically riskless, using the DSR is warranted, because the riskless natures offset each other, and for all the electrical engineers out there, or engineers, this is my poor man's starting circuit concept, where at the end of the day, you have your DSR, which in effect absorbs whatever the collateral that comes out, but your real objective is such that this piece over here stays with whatever energy is out there, but there's no real current moving back and forth. There's no DAI or electrons that are moving. You want it to be stable.
# / 01:21:42 Matthew Rabinowitz So that goes, really, to the question for the Maker Foundation and the community as a whole, is how will we handle voting on risk premiums? One? Two? Forty? Or will we not vote on them at all, and outsource it to a risk team that provides a recommendation? Or will we be taking multiple risk teams and taking some type of median or average, or weighted average, based on some system I don't know, and almost like a shard of the decision making, and then taking some type of computed version, saying that, "Here are the 400 collaterals we've got hanging out here. Here are the recommended slight modifications this week," and then either vote for it, or you veto it, or ... that's just an outstanding question. I don't know the answer to that one.
# / 01:22:32 Richard Brown So are these rhetoricals, or are these things you're asking right now? Because I can [crosstalk 01:22:36]-
# / 01:22:35 Matthew Rabinowitz Oh, we don't need to know it today. It's more of a question of legitimately in six months, how can we ... the biggest concern that I think all of us should have is those risk premiums that are going to be assigned to a given collateral package that Cyrus and team will compute and spend valuable time figuring out what it should be, how do we maintain them? And-
# / 01:22:57 Richard Brown Okay, let me start answering some of these theoreticals, or some of these hypotheticals with some theoreticals. The idea is that we don't know, but it's not like we're unprepared. We've been discussing precisely that, basically, internally for the last eight to twelve months, because there's no surprise that complexity is [inaudible 01:23:24] governance process that we've been experimenting with for the last 41 weeks is to determine exactly what that picture looks like, and to get the community educated and engaged enough to help us to discover the answer to those questions.
# / 01:23:37 Richard Brown So that's a bunch of euphemisms and unknowns, I understand, but here's the idea, is that we need the community to tell us what their appetite is for engagement. We need to figure out exactly how involved people are. We need to figure out where these discussions happen, and then we need to figure out whether we need to optimize or whether we don't, and all those things, we can and have come up with elaborate plans to anticipate and/or guess about how that might turn out, and to talk about how we'd like to have that turn out, but it's ultimately not up to us.
# / 01:24:13 Richard Brown So the idea is, and I touched on this briefly earlier, that we have a forum coming. That forum is going to be the place where we figure out the answers to these questions, and so we'll take things like your most recent weekly narrative, Matthew, which I loved, and we'll give it a pride of place in the forum, so in the governance forum, we'll have, "This is the questions about how we address packages, or tranches of risk parameters, and how often do we bite these things off? What's a reasonable amount of work that we can expect from the community, versus the friction involved with constant maintenance through the voting portal?"
# / 01:24:54 Richard Brown There's all kinds of things we need to figure out. So the answer is, we've anticipated the stuff. We had a lot of ideas. I think the community also has a lot of ideas, and the next step is to start formalizing them, and figuring out which ones work, and starting to put those into place. You've raised some really interesting points that we've been debating for a long time now.
# / 01:25:16 Richard Brown So, I realize that was a bit of a non-answer. So, you raised a lot of great questions, and we agree that those are great questions. We have some ideas, but the solution to finding the answers to those great ideas is we need to improve our signaling methods. We need to release the next version of the portal, which allows us to do multiple polls at once, instead of being completely dominated by a weekly stability fee poll, and we need to have long, deep discussions in a semi-formalized manner, and that's where the forum comes in.
# / 01:25:51 Richard Brown Did that address any of your concerns, Matthew?
# / 01:25:58 Matthew Rabinowitz Yeah. No, no, definitely. I think the greatest concern of the entire project is how to maintain the, call it tranches, packages, whatever the mnemonic will be, that those risk premiums are priced consistent with the risks compared to the system, and that they maintain the exponential nature of the risk that they actually have. If we start having too much apathy towards that, and just a little bit of ignoring the conditions on the ground, and I'm not picking a crypto token. I mean, the price of oil, the price of real estate, you name it. We start getting lax on that, and the whole project can implode on itself, and that's what is very dangerous.
# / 01:26:40 Richard Brown Yeah, we agree, and this is something that we've ... I mentioned that we've been having these discussions. So, we talk about governance all the time inside MakerDAO, and we get to talk about it publicly for an hour on Thursdays, but one of the most often-requested, floated or suggested future plans for governance is that the reality of the situation is that not everybody in our ecosystem is either willing, able or has the time to be deeply engaged and weigh all the evidence, so people are keen for delegates, or some kind of a delegation system, or some kind of mechanism where we identify people that are trusted-enough to make some of these high-level decisions.
# / 01:27:30 Richard Brown We already have a model for that. That model is the risk teams. That they get together, they surface, they collect the data, they create the models, they present their frameworks and then the community ratifies those. And potentially, as that ecosystem, or as that model, expands and gets more refined, then we can start applying that to other things, so maybe there's a utility token tranche, the administration team or something. There's all kinds of options available to us, but very reluctant to have the Foundation airdrop that onto the ecosystem. We want the community to tell us what they want to do, and how they want to self-organize and work. We're anxious [crosstalk 01:28:13]-
# / 01:28:13 David Utrobin So it sounds like the scale governance ... Sorry, Rich. So it sounds like the scale governance, in the end, Maker holders are going to have to rely on expert models, where you defer to the risk teams, right? That's a little bit what Cyrus was talking about in his last slide.
# / 01:28:28 Richard Brown Yeah, that's basically ... well, I shouldn't say "basically." That is literally what the whole system is designed to do. So if you go back and, I recommend that people do this, is read the Foundation governance risk framework that Steven published for us, or you can just skip to the third one, but that's what it's all about. We do scientific governance and risk. This entire framework is designed for that first iteration of gradual decentralization, is that we have risk teams who are deep into the science and the history of this world. They present models and ideas and datasets, and then the neophytes and the amateurs like me are in a position to read all that stuff and figure out whether we have confidence in those systems, and then we vote them in or we vote them out.
# / 01:29:20 Richard Brown That same model applies to basically everything that governance is going to be doing in the future, but it's an evolutionary process, and I think that we've cautioned ... I hope that we've been good about this message, is that I can't really iterate strongly enough how interested, how focused, MakerDAO is, the Maker Foundation is, on decentralization. That's our core principle. In order to do that responsibly and effectively, we need to do that in steps while we understand how the process works, and the first process, the first step, in that was the interim risk teams, and the decentralized risk teams, which are coming on board soon. We're going to iterate and evolve that process, and then keep on identifying skilled groups of stakeholders, and figuring out whether we need to split out the different groups, or have different venues for deeper discussion, or we identify some other class of actors in the system which we want to empower to start surfacing signals for us as well.
# / 01:30:20 Richard Brown The governance ecosystem that we have right now is the bootstrapping phase. It's going to get significantly more complex in the coming months, I guarantee. But that complexity will be driven by the community, and we need to have people in this call and in the forums to tell us what that's going to look like.
# / 01:30:43 Richard Brown Having said that, I need to jump out because big things are afoot in the MakerDAO ecosystem. If you want to know what those big things are, I suggest you go to blog.makerdao.com, and you will figure out what is up. But I do need to bounce and deal with some of the resultant buzz. Actually, I'm going to, Cyrus, make you a host, if you want to continue?
# / 01:31:10 Cyrus Younessi Sure.
# / 01:31:11 Richard Brown And then just make sure you stop the recording before you shut down the call. Because the last time, I discovered that there was three-and-a-half hours of black screen that I had to edit out, and I don't want to do it again.
# / 01:31:24 Richard Brown All right, thanks everybody for joining. Great discussions again, as usual. We'll be posting summaries and recaps and videos and audio to Twitter, the subreddit and the discussion thread. Please continue the debate there. All right. And that is it. Talk to everybody later.
# / 01:32:09 Cyrus Younessi Anyone have anything they want to talk about?
# / 01:32:15 Akiva Dubrofsky Yeah. So, I have a quick question.
# / 01:32:17 Cyrus Younessi Sure.
# / 01:32:19 Akiva Dubrofsky Once you start looking at the fundamentals of an asset, doesn't that get rid of the assumption of normal distribution of returns?
# / 01:32:32 Cyrus Younessi I mean, looking at the fundamentals of an asset doesn't necessarily have much to do with what its returns look like. Yeah, no asset has normal returns, right? It's just a completely made-up assumption. It's a made-up assumption, to help the models work, but that's it really.
# / 01:33:00 Cyrus Younessi That didn't ... okay. Can I answer your question in a different way?
# / 01:33:09 Akiva Dubrofsky Well, I mean, if you're just looking at equities for example, their returns for the past 10 years haven't been normally distributed.
# / 01:33:17 Cyrus Younessi Right.
# / 01:33:19 Akiva Dubrofsky So once you start looking at the fundamentals, let's say interests are low and there's QE, so them are the fundamentals, and you say, "Well, I don't think the returns will be normally distributed," so why aren't I seeing that from my models?
# / 01:33:31 Cyrus Younessi Because in some ways it's the best we can do, right? I mean, all derivatives are priced with a certain set of assumptions which in practice are extremely difficult to unwind. To relax that assumption would imply that you have some better distribution of what the asset returns look like.
# / 01:34:05 Vishesh Choudhry I also think if we're thinking about, what are the inputs to modeling a potential price path for an asset, it's one thing to be able to say, "I know what the fundamentals are, and so I can tell you how this asset is going to do over a one-year, two-year, whatever, period." If you can do that, you don't need to look at any kind of simulation of a normal price path, or log normal price path. Now, it's not right, it's not going to be perfect, to be able to say, "Hey, I've got this estimation of log normal price path, and I'm going to assume that that's the case," right? We're not doing that either.
# / 01:34:50 Vishesh Choudhry What you're saying is, "Under various potential assumptions that could produce various log normal price paths, this is what the expected losses would be," and if you're modeling all of the reasonable potential outcomes and you're taking an average of what all the potential loss values would be under different circumstances, what you're really saying is, "Given a historical asset volatility, given certain assumptions about the probability of a price crash, et cetera, what am I going to expect in terms of my losses at the end of the day across all the different potential outcomes?" And then the way you're arriving at those outcomes is by saying within each of those iterations in that simulation, you've produced a log normal price path, because there's no better intermediate alternative. Does that make sense?
# / 01:35:46 Akiva Dubrofsky Yeah, that makes sense.
# / 01:36:36 Cyrus Younessi Okay. Well, I have to jump off myself, so I'm going to stop the recording. Anybody want to take over as host, or not?