Risk Parameter Update - February 2021

[jmo]

After the healthy discussion over the past week or so about the WBTC collateralFactor changes, people mentioned a few key risk factors:

• Liquidity
• Volatility
• User behavior (for example, the collateral ratios maintained by borrowers)

Over the past few months we’ve seen liquidity improve for many tokens in Compound, while volatility has also increased substantially. How do you determine if the increase in volatility is offset by the increase in liquidity? How have user-chosen collateral ratios affected any added risk? These factors interact in complex ways, and we believe the only way to understand how these myriad conditions affect risk is by running stress tests for the protocol.

We’ve gone ahead and run a set of stress tests of Compound to try to set collateralFactors that better balance the risk and capital efficiency of the protocol. We’re looking forward to doing this more frequently going forward. By setting parameters more frequently, the protocol should be able to take on more risk as we will be standing by to increase collateral requirements as market conditions change. We’ll submit a proposal for these parameters early next week, but want to leave this on the forum for a bit to get more feedback in the meantime.

Proposed Collateral Factors

	Current	Recommended
DAI	75%	80%
USDC	75%	80%
BAT	60%	65%
COMP	60%	60%
ETH	75%	75%
UNI	60%	60%
WBTC	75%	60%
ZRX	60%	65%

As we mentioned in the discussions around the WBTC collateralFactor, WBTC poses the largest risk to the protocol and collateral requirements should be increased. However, as the community just voted to change this, we’ll leave this parameter out this proposal, and include it in the next one. We’ll also be making some changes to COMP speed, as these collateralFactor changes could encourage more circular borrowing in the protocol:

	Current COMP Speed (COMP/block)	Recommended COMP Speed
DAI	0.067	0.050
USDC	0.067	0.050

Stress Test Results

Our stress tests are described here, but to recap they:

1. Deploy the Compound contracts in a test environment
2. Subsample the current distribution of borrowers to create a representative set in the test environment
3. Add synthetic price trajectories for each asset that are modelled from historical market data
4. As the prices change, positions become eligible for liquidation, and agents in the test liquidate the collateral on Compound via a slippage curve fit to real data from centralized and decentralized exchanges
5. Run simulations comprising 1-4 hundreds of times to predict potential insolvencies in Compound

We’ll focus on the main output of the simulations, which is “Net Insolvent Value” per collateral type. This is highly dependent on price volatility, which we vary across the x-axis below. A “Volatility Scalar” of 1 means that the stress tests used price paths that matched the historical volatility of each asset, a scalar of 8 corresponds to price paths that are 8 times as volatile as recent asset prices. For reference, the historically bad volatility of March 12/13, 2020 corresponds to a scalar of 7 or 8.

ZRX, DAI, USDC, and BAT have a very low risk of seeing insolvency, even under March 12 like conditions. We’ll recommend increasing the collateralFactor (and lowering the collateral requirements for borrowers) for those assets.*

ETH, COMP and UNI also carry risk, but after investigating further, that risk was mostly determined by the accounts relying on WBTC as collateral as well, so we’ll recommend those parameters remain unchanged. Accounts using multiple collateral types can create complications in data analysis, but we look at individual accounts to add color to the data.

*One thing to note. The issues that were seen with DAI price fluctuations late last year are still possible. Borrowers using DAI as collateral or borrowing DAI should be careful when approaching the collateral limits.

[Sirokko]

I very much dislike a concept of packaged deals, as this is very common way of manipulations. Change for collateral factors and change of COMP distribution speeds are different parameters and shouldn’t be packaged together inside one proposal. But rather voted separately, presenting voters opportunity to express their opinion on each change separately.

Decreasing distribution initiatives to DAI and USDC will indeed likely to decrease recursive stable coin farming, but i think it deserves bigger discussion. Certanly even after previous adjustment stable coin markets are kind of overincentivised, but such big of a cut is unjustified in my opinion. I suggest we proceed slowly and observe results. Instead of cutting it in 20% in one move by going from 0.067 to 0.050 we should consider gradual decrease instead and observe impact on markets. And maybe return to that discussion next month, when we have data on results.

On the other hand, collateral markets are kind of underincentivised still, WBTC and ETH certanly could get more initiative. Previous adjustment sparked some interest in Borrowing side, which, in turn, moved supply interest into something more noticeble than zero. Currently, Borrow side of Collateral markets can boast sligtly negative interest on Borrowing of ETH and WBTC. Which is justifiable, as with current bullish expectations, going short feels a risky position for most. And yet, Borrow side of that markets is important for Compound, not only it generates Reserves for that markets, but it also provides interest for Supply side. As well as provide more balanced utilization. If we add to ETH and WBTC what we substract from DAI and USDC, not only it might help to adjust risk, borrowers are taking, but it will also likely move APY on supply side to higher numbers, bringing more liquidity for protocol. There’s also some concerns that together with big increases in valuation of ETH and WBTC we might observe actual decrease of WBTC and ETH, supplied to protocol, COMP distributions could help here too.

So on topic of COMP speed, i’d suggest consider something like that:

DAI 0.067 decrease to 0.062
USDC 0.067 decrease to 0.062
WBTC 0.01075 increase to 0.01575
ETH 0.01075 increase to 0.01575

While incentives on stable coins are likely could be cut more, i don’t think incentives for WBTC and ETH should be increased in bigger steps. Predictable distributions is strong feature of Compound protocol. And i think making small incremental adjustments on the way would bring protocol much further in the end.

Decreasing total COMP emissions per day i believe is completely different topic. And should be discussed further. Recent surge in COMP price, while brought a lot of excitement not necessarily is here to stay and not nesessarily a call for fast action.

I’m not quite convinced that both increasing of collateral factors for stable coins and WBTC collateral factor increase are coming at good time, as markets look quite overheated. At the same time i’m not convinced that WBTC liquidations is that big of a problem. ETH/WBTC pair on Sushiswap alone is over 700m of liquidity and another 300m liquidity is at Uniswap. I wasn’t a supporter of WBTC Collateral factor increase, but i wasn’t against it either.

[dabar90]

@Sirokko

collateral markets are kind of underincentivised still, WBTC and ETH certanly could get more initiative

agree, since ETH staking and very low interest rate on supply, the incentive to supply ETH needs to be strengthened. In addition to a more drastic increase in interest rates, higher COMP distribution on ETH supply is a good idea. Are there any other options?

[crypto_contrarian]

This analysis is great and really appreciate the easy to understand chart. I think more frequent quantitative analyses should be the standard of risk management for DeFi applications like Compound.

[jmo]

It is not really our intention to create a package deal, merely to change the protocol in a way that is more or less neutral to existing rewards .

Currently there is a lot of recursive borrowing in Compound that provides little value to the protocol. Recursive borrowers will for instance post DAI collateral, and then borrow USDC, having a position that is paying interest to Compound in return for liquidity mining rewards. With the current collateralFactor (cf) of .75, this means that farmers will earn COMP commensurate to 1.75x of their funds deposited. Farmers often leverage multiple times when doing this, in this case they would then borrow 75 DAI against every $100 of borrowed USDC, and so on.

	Action	Current Reward
1	Deposit 1000 DAI as collateral	1000 * 0.067 COMP / Block
2	Borrow 750 USDC	750 * 0.067 COMP / Block
3	Deposit 750 USDC	750 * 0.067 COMP / Block
4	Borrow (.75 * 750 DAI =) 562.5 USDC	562.5 USDC * 0.067 COMP / Block
5	Deposit 562.5 USDC	562.5 USDC * 0.067 COMP / Block
	…	…

You can quickly imagine that you can keep doing this over and over again to achieve farming rewards equal to L:

L = 1 + 2(cf) + 2(cf)^2 … (cf)^n

As

𝚺(cf^n), from n=0 to inf. = 1 + (cf) + (cf)^2 … (cf)^n

And when cf < 1:

𝚺(cf^n), from n=0 to inf. = 1 / (1-cf)

we have that L approaches the following as the number of recursive borrows increases:

L = 2 / (1-cf) - 1

Given cfs of .75 and .8, the maximum leverage is:

L(0.75) = 7

L(0.8) = 9

This means that if the cf is increased, the maximum leverage (and therefore farming reward) is now ~29% higher than before. Now we want changes to risk to be neutral towards rewards. This would imply a reduction of COMP rewards commensurate to the maximum leverage the cf enables.

With current COMP speeds of .067 COMP / block for DAI and USDC, lowering rewards ~25% to .050 COMP / block should offset the increased farming rewards. Given that people are not recursively borrowing a literally infinite amount of times, we rounded up a bit.

Now, as you notice, just lowering the rewards in one market is not risk neutral. The problem is that we’ve seen these rewards greatly encourage borrowing across specific collateral types, and so we’re hesitant to further encourage something like WBTC collateral, which we believe poses a systemic risk. Then that leaves other markets like ZRX and BAT, but which are also receiving needed cf updates.

We’re happy to entertain any separate proposals for how to further distribute COMP, but right now, we’re making the changes we feel best balance the interests of the protocol users, COMP holders, and community members. The changes are made with the broad brush strokes provided to us by Compound governance. We try to make the proposal as targeted as possible, and try to avoid omnibus proposals that needlessly couple unrelated changes. However in Compound, risk is tightly coupled to reward, and we are doing our best to change one with minimal impact on the other.

[Sirokko]

It’s good argument, that intention is merely to change protocol in a way, which is neutral to existing rewards. I’l take it for now. But keeping that in mind, let’s look at what you propose from broader perspective.

First of all, we all agree that there is quite a position of recursive borrowing, which debatefully not that much valuable to protocol. I believe that is a bit more complex, but i think we could agree that it’s surely not THAT valuable, considering share of COMP distribution that strategy recieves. Currently stable coin markets recieve over 70% of COMP distribution, while all others combined get less than 30%. And that is while stable coins are less than half of supply, but majority of borrow side. Additionaly good share of stable coin markets is represented by recursive borrowing.

I don’t see recursive borrowing as plain evil though. Yes, it wasn’t intended, but it’s still legit usage of protocol. It drives adoption and brings liquidity at the very least.

I’ve read Gauntlet research paper and i understand your position that Compound protocol can handle increasing collateral factor for USDC and DAI market. But while it doesn’t create sistematic risk for protocol, it allows users to take more risk.

And that is coming not at exactly calmest period for cryptocurrency markets. That is what i mean when saying that it’s potentially not the best time to allow users to take more risk.

But then who is main beneficiary of increased collateral for stable coins? COMP farmers, utilising stable coin strategies. We can pretty much safely assume that, as we do not see that stable coins are used for collateral to borrow volatile assets like ETH, or WBTC or others. Stable coin borrowing absolutely dwarfs everything else. Increasing CF for USDC and DAI is basically adding more initiative to recursive farming, like it wasn’t big enough already.

So point one: While Compound protocol could safely handle bigger CF for DAI and USDC markets should collateral factor be increased at that markets conditions, while main benefeciary of such move indeed would be recursive COMP farmers?
If we agree, that recursive farming isn’t that valuable, why push initiative basically promoting to use it more? USDC-USDC and DAI-DAI are most safe strategies who can easily take more risk from increased CF. USDC-DAI and DAI-USDC are less so due to
potential DAI volatility, but still it’s nowhere near comparable to volatile assets.

Now lets come back to your suggested solution for that.

It’s hard to estimate exact share of recursive borrowing in stable coins, as users can easily use multiply addresses, but just for a sake of simplicity let’s take it’s 60%. Like 60% of stable coin markets consists from recursive borrowing. That number doesn’t need to be accurate for purpose of that discussion, just for pure simplicity.

Then you say if we increase CF, than those participants, can safely increase their leverage and potentially could get up to 29% more in COMP rewards for their capital invested. That if all of them do that, which might or might not happen. But then
where do that bigger rewards coming from? Is protocol paying more? Nope, protocol distribute exactly same amounts. Ok, so bigger rewards for recursive farmers coming because they increase their share in stable coins markets. So basically, having
same capital invested, they will potentially be able to increase their share in USDC and DAI markets and thus dilute share of other 40% participants which utilise other strategies. So they will get more per capital, than pure non-recursive usage like
Supply ETH - borrow USDC for example. But that is worst-case a potential outcome. So, potentially, 40% of non-recursive part of stable coin markets would get less of COMP distribution than before, which at maximum might be up to 29% less.

And then your suggested solution is to cut COMP distribution to USDC and DAI markets, which guarantee decrease of 25% in rewards for every participant of USDC and DAI markets, unless you are specifically a recursive farmer, in which case you can make it up for yourself by doing more leverage and then your rewards per capital invested will stay same.

So the only loosers here are users, who are not recursive-farmers. Also such change imply that we actually decreasing total COMP distribution per day. As we decrease distribution speed for USDC and DAI markets and not putting that distribution anywhere else. This is certanly not neutral change to existing rewards.

I’d say in that case the treatment is worse than desease. Not only we introduce CF change wich directly benefit specific strategy, but we go further, and then cut rewards for everybody else, who isn’t using that strategy. I wonder how that go well with statement that recursive farming isn’t that much valuable for protocol. Why then we introduce initiatives stimulating that activity.

I would say if we go with increasing CF fo USDC and DAI we should let market decide about decreasing/increasing rewards proportions between recursive/pure borrowing. And not instead execute literally the worst possible outcome for other non-recursive users.

I have no comments on your proposed CF on ZRX and BAT markets, these are relatively small markets and proposed change is rather conservative as well. I don’t see problems with that change for now.

And on topic of decreasing recursive borrowing my position is same. I believe we should allocate bigger portion to other markets. In current conditions Collateral markets (WBTC, ETH) at the expense of USDC, DAI markets. Decreasing distribution on stable coins always should have decreasing effect on recursive borrowing as you only can leverage till it’s profitable enough, lower rewards lead to less sustainable leverage. Increasing rewards to Collateral markets at same time lead to bigger utilisation and make other patterns of usage of Compound more rewarding as well, even if slightly.

In the end, i would like to mention that i absolutely like job, Gauntlet is doing for Compound protocol, as well as involvment with community. But i believe this particular changes as they just were drafted certanly need considerations.

[jmo]

@Sirokko

Not only we introduce CF change wich directly benefit specific strategy

To make one thing clear- raising collateral factors should help all users

@Sirokko

I believe we should allocate bigger portion to other markets.

We’re looking forward to changing the compSpeeds in future proposals as well

[Sirokko]

@jmo

To make one thing clear- raising collateral factors should help all users

That’s not true. Raising collateral factor for USDC and DAI benefits exclusevly specific portion of users who use USDC and/or DAI as collateral. And since 80% of borrowing side consists of stable coins, we can safely assume that primary beneficiaries are recursive farmers. (since if you using stable coin collateral most likely you still borrow stable coins with it, and that have no economic sense aside of COMP farming) How raising collateral for USDC and DAI should help a user, who isn’t using DAI or USDC as collateral? For example, users, who supply ETH or WBTC and borrow USDC or DAI? Maybe i’m missing something here?

Decreasing distribution speeds of COMP for DAI and USDC markets, decrease rewards for ALL users of stable coins markets on the other hand.

And that was the main thing i wanted to highlight in my previous post.

[getty]

I want to preface my comments that I wish I had more time to do a thoughtful, independent analysis of each asset.

TLDR: I would vote no on this proposal if it were made as is.

Why:

• I don’t like package deals. If you want to adjust multiple collateral factors at once, I think it is possible (unlikely) to do that. If you want to change multiple comp speed variables at once, I think it is possible (likely) to do that. Adjust multiple collateral factors and comp speeds in the same proposal is setting a bad precedent. If you had ample evidence for these changes, perhaps I would change my mind.

• I don’t see a reason detailed in this proposal to decrease the comp speed for Dai and USDC.

@jmo

We’ll also be making some changes to COMP speed, as these collateralFactor changes could encourage more circular borrowing in the protocol:

Perhaps you meant to say discourage? Either way, I disagree.

• I read through your February Market Risk Assessment. While the document is very well written and the first 14 pages do a great job outlining the protocol, I do not think the paper’s evidence is sufficient for justifying these collateral changes. While I am generally a proponent for increasing the collateral factors, I also believe sufficient evidence needs to be provided. My main concern for the review is what volume numbers you are using for each asset. Throughout the paper, $100m is used as the ether volume, but I believe that wildly underestimates today’s liquidity. The paper also cites volume stats from 2019 & 2020, which is well out dated by now. The only significant historical stat worth looking at is volatility. Estimated volume numbers should be as current as possible.

• I think the analysis also significantly discounts the state of liquidators and their sophistication. Liquidators are no longer just users. Whole protocols/dapps are being built to be liquidators.

With the limited research I have done per-asset, I think the CF can be increased for DAI, USDC, and ETH. If I had to include a fourth, it would be UNI.

[rleshner]

First, I’d like to applaud @jmo, @tarun and team for beginning the process of methodical, consistent changes to risk and incentive parameters. This is a welcome improvement; thank you for stepping up

Process
I agree with @Sirokko and @getty that packaging multiple sets of changes together makes it more difficult to:

• Analyze the impact of changes / experiments
• Exclude changes that have public opposition

As the community gets into the groove of governance, it makes more sense to start slow; a proposal with specific risk changes, then a proposal with specific incentive changes (or vice versa). If they can be separated, they should be, in my opinion. This goes for a category of changes, too: if we’re modifying risk parameters, why not start with one, and measure the impact of it, before changing many?

Risk
Over the past few days, the risk of the protocol has increased significantly; there is now $7.9B of assets supplied, and $3.50B borrowed from Compound. One month ago, we had $4.5B supplied. Take a step back – these are staggering numbers and velocity.

Prior to increasing risk (by increasing CFs), I believe as a community we should evaluated every additional safeguard possible, to decrease the risk of the protocol. This morning I laid out one strategy for improving the liquidation system, and there are many other ideas/approaches. Optimizing CFs (and capital efficiency) is a tool for growth; now is not a time to put the petal to the metal, in my opinion–we already have extreme velocity. We should add better navigation systems, cameras, and tires to this vehicle, first.

Incentives
In Proposal 35, your team laid out a great framework to think about, and model incentives. Prior to changing incentives, let’s revisit your framework and scientific method; deliberately experiment with incentives, and measure the results.

The idea of modifying the COMP distribution is sound; we could modify one stablecoin, and one collateral market (taking @Sirokko’s idea of re-allocating it to collateral), and measure the results. Decrease DAI, increase WBTC, for instance (or, USDC / ETH, etc). How will this change ETH vs WBTC, or DAI vs USDC? This is great data to arm ourselves and the community with.

Conclusion
This is a great direction to set the protocol on–by getting the methodology correct up front, we’ll have a recurring system that works well to calibrate the protocol.

[jmo]

We’re happy to start slow, so we’ll break these parameters changes up into a couple of proposals as you mention. Given gas costs and also the long lead time for protocol changes, I hope that we can build confidence to bundle changes in the future.

Tenatively, we’ll suggest the following schedule:

• Tomorrow - Update the ZRX and BAT collateral factors. I know it feels like the “pedal [is] to the metal” but for these markets, the pedal very much is not.
• Next Week - Update the rewards for the USDC and another market
• Later Next Week - Update the WBTC collateralFactor as this is a case where the pedal may in fact be through the metal

[jmo]

@getty

Throughout the paper, $100m is used as the ether volume

As mentioned in my original post, we use live market data for these values. We’re actually working on sharing some of these risk factors (like liquidity/volume) via a Dashboard, which I will post about later today.

[jmo]

We’ve published the first proposal here:

Compound

(Ethereum Transaction Hash (Txhash) Details | Etherscan)

Thanks again for all of the feedback!

[Pinotio.com]

Hey @jmo , thanks for the analysis plot, a few questions:

1. How is volatility parameterised. Are you taking it as Gaussian or Power Law or something else?

2. Have you done a simple convexity analysis? One example:
   a. Calculate collateral levels with a 20% drop in WBTC/ETH prices within a time gap delta_t.
   b. Calculate collateral levels with a 40% drop in WBTC/ETH prices within a time period delta_t
   c. Check whether collateral is convex up or down comparing a. with b. (you can also check convexity with time). If it is convex down, the platform is under-collateralised. If it is convex up, it is buffering and may be sufficiently collateralised.

3. Have you included second order effects of continuously falling WBTC/ETH prices as liquidations occur?

4. Have you looked at second order effects of DAI (which itself will face collateralisation challenges in a downmarket)?

Generally, it strikes me that - at current collateralisation levels - a lot of trust in the liquidation mechanisms is required. I’ve laid out some high level ideas and suggestions here:

Are DeFi Lending Platforms like Compound and Aave market neutral?

[jthrack]

I’m deeply saddened to see this fail because not enough votes were cast. While I definitely support keeping proposals small and simple, I wonder whether the small number of votes was because the economics of the change were too small or affected too small a group, such that voters didn’t want to bother wasting time or gas to vote.
Maybe worth pointing out again that everyone can vote for free at comp.vote, and also highlight the site whenever we link a new proposal. I definitely forget to use it sometimes, even though it saves a few dollars for each vote.

[jmo]

We’re going to propose again in the next day or two. I think as gas prices increase, this will happen more often, so we’ll just try to reach out directly to a few token holders to see if we can push things over the line for this proposal.

[jmo]

@Pinotio.com

How is volatility parameterised. Are you taking it as Gaussian or Power Law or something else?

Our price paths in simulation use Geometric brownian motion, using historical data to set the volatility.

@Pinotio.com

Have you included second order effects of continuously falling WBTC/ETH prices as liquidations occur?

We then add in a “price impact” from liquidations that we train from data on Coinbase Pro and Messari:

(This picture is from an old market snapshot but should give you an idea)

[Pinotio.com]

Thanks @jmo ,

Re Point 1. I guess I’m just saying what you know here, but geometric Gaussian won’t cover price jumps or power law type behaviour - the specific behaviours that would likely give liquidations most issues?

Re Point 3. Thanks for this plot showing the fits. Is it possible to share a point where you highlight 2017 and 2020 crash data among all of the other data points.

Curious to get your answers to Qs 2 & 4 (even if you could give me a sense of what kind of convexity you see in the models that would be worthwhile).

Adding one more question - 5:

Have you built a list of assumptions covering what is not in the model but could put the liquidation mechanism at risk? Could you share that list?

[jmo]

@Pinotio.com

geometric Gaussian won’t cover price jumps or power law type behaviour

We’ve used jump diffusion in the past but it’s not super useful. Collateral factor of 80%? Add in some 25% jumps, yeah liquidators come and force insolvency immediately. Highly volatile GBM price paths cover a wide variety of scenarios. Jumps just don’t clearly make the model more accurate, though is a fun corner case to look into. It’s my understanding that power law price movement is used to approximate the exact sort of liquidation spiral we explicitly simulate, why did you want to look into it?

Regarding future model features, we plan to share more info on that over time. Right now, we want to use these models to make Compound better, and our focus is on pushing parameters changes that do so. We’re more than happy to answer questions from community members that help them understand why the model is valid, but at the same time encourage everyone to do their own analysis as well. You should feel free to do any of the complexity analysis you mention above, I’d be delighted to read it.

[monet-supply]

How do we feel about raising stablecoin reserve factors?

DAI is at 15%, and with currently high rates and utilization, reserves are building up at a rate of $30M+ per year. USDC and USDT are both gaining reserves at a steady clip, but I don’t see the harm in laying on the gas by increasing the reserve factors for these assets.

If USDC and DAI have their collateral factors raised, increasing the reserve factor could also play a role in reducing returns from circular leveraged positions.