Chaos Labs - E-Mode Methodology

[ChaosLabs]

Executive Summary

E-mode allows the Aave community to increase capital efficiency on related assets. In this post, we discuss this relationship, how to test for it, and what it means for E-mode risk parameters.

We argue there are two key properties we must require for E-mode assets:

1. Pairwise prices must be historically mean reverting - the sampled mean exchange rate of any two tokens is stable. If the exchange between two assets doesn’t mean-revert (e.g., it drifts up), users will experience liquidations, and bad debt might be accrued.
2. Pairwise prices must mean-revert fast - from the moment prices diverge, they must take less than 24 hours to converge back to the mean. This way, Aave has some confidence it won’t be left holding debt for long.

Furthermore, major price deviations between E-mode assets may lead to bad debt, either from missed or adverse liquidations (as we explain below). We set our LT/LTV values sufficiently low to mitigate bad debt and preserve a good user experience for retail traders.

Testing Methodology

We consider USDC, USDT, DAI, and LUSD for the stablecoins class and stETH, cbETH, and rETH for the ETH liquid staking derivatives class.

We measure for mean-reversion and mean-reversion speeds using 1 year of Chainlink data, including the latest collapse of SVB and subsequent depeg of USDC. Our analysis is based on rigorous statistical tests that are common in the financial engineering literature. We describe these and how we used them in our full-length paper found here.

Setting Risk Parameters

Fig 1: DAI/USDC exchange rate during the USDC de-peg.

LT: As a conservative measure, we set LT/LTV ratios according to the most significant deviation we’ve observed for any two assets. For DAI/USDC, we saw a deviation of ~3.6% after the USDC de-peg this weekend. As we discuss in our full-length paper, we can set the liquidation threshold such that we minimize the accrual of bad debt from adverse liquidations or potential attacks from arbitrageurs:

LTV: We can minimize the chances that retail traders are liquidated by increasing the gap between LT and LTV. This way, even during major depeg events, retail traders will not be harmed by liquidations (and we further minimize the chances of accruing bad debt):

For USDC/DAI, we get an LTV of 93%. If we included USDT, we would have to decrease LT/LTV to 0.87 and 0.76, respectively, which we explain below.

Results

We find that all stablecoins are mean-reverting at 99% confidence, whereas there is insufficient data to assert the same for ETH LSDs. Furthermore, we find that LUSD has historically taken too long to mean-revert and is therefore not considered.

Fig 2: Results from our ADF and Hurst Exponent tests on mean-reversion for stablecoins. We find that all pairs considered mean-revert within a 99% confidence interval.

Fig 3: Historical time to mean reversion for stablecoins. We measure how long prices take to converge to 0.1% of the long-term mean once they have diverged by 0.1% from the long-term mean.

Fig 4: Stablecoin mean-reversion results. We find that there is insufficient data to assert that stablecoins mean-revert at a 99% confidence interval, likely due to their brief lifespans and the underlying redemption mechanisms (which differ from stablecoins).

A Note on Bad Debt

Bad debt can be accrued in two ways:

1. Missed liquidations: The prices of borrowed and collateral assets diverge and don’t converge. In this case, if liquidations were not profitable as the prices diverged, Aave is left with a debt that cannot be repaid with the current collateral.
2. Adverse liquidations: Adverse liquidations occur as prices converge: liquidators are rewarded with all the collateral but only repay part of the outstanding debt. This likely occurs when prices are converging (because there is more buy-side liquidity on the collateral asset).

We have accounted for both risks in our methodology: we recommend choosing assets that have a history of mean-reverting quickly (1), and we set the LT sufficiently low such that debt can be repaid at deflated prices (2).

[sakulstra]

Do you have any data on current v3 stable eMode usage?
With only supporting USDC, DAI in eMode, how much of the current usage will be covered? - does this eMode make sense? - i think not.

Assuming this is a true summarization of usecases:

Gauntlet Aave V3 E-mode Methodology

I’ve seen the following usecases for eMode on stables. 1.) Shorting/longing the Euro. Some markets (polygon) offer stable-eMode for both EURO and USD denominated stablecoins. This was a popular way to bet against the falling EURO in H2 2022. 2.) Shorting USDT Or more generally, people use eMode to short stables which, they view, are more likely to depeg (to the downside). 3.) Arbitrage (as mentioned earlier) If we were to turn stablecoin eMode into ‘USDC eMode’ (DAI, FRAX, USDC) we would …

1.) Shorting/longing the Euro.
Some markets (polygon) offer stable-eMode for both EURO and USD denominated stablecoins. This was a popular way to bet against the falling EURO in H2 2022.
2.) Shorting USDT
Or more generally, people use eMode to short stables which, they view, are more likely to depeg (to the downside).
3.) Arbitrage (as mentioned earlier)

With what you(and gauntlet) propose seems like a useless eMode eliminating all use-cases.
If that’s true, does it perhaps make sense to just not have an stable eMode?

---

Alternatively I think it could make sense to explore:

This would require us to set a very low LT/LTV to mitigate the risk of bad debt, which is not our goal with E-mode.

The goal with eMode is to have a more attractive, but not inherently less save liquidations i’d say. This doesn’t mean it must be 96.5% LT - it just needs to be noticeable more attractive than non eMode.

Let’s assume non-eMode LT should never exceed 80%¹ - in this case a 90% LTV, and 93% LT, 1.5% Bonus(random numbers) on eMode could still be attractive, but allow including more stable assets.

¹On a side note I think it would make sense to actually to embrace a rule like that, as 86% outside eMode is imo quite risky in pools where volatile low cap coins are listed.

[ChaosLabs]

Thanks for the comment and feedback, @sakulstra.

Based on the feedback provided in various E-Mode threads (in this post and here), the community should decide on its stance regarding the use of a stablecoin E-Mode, which is subject to the constraint that a single asset can only be part of one E-Mode configuration.

To simplify matters, we view two possible options for the configuration of stablecoin E-Modes:

1. Aggressive parameters with a limited number of assets
   1. If the community chooses this approach, our recommendation would be for an E-Mode for highly correlated assets, excluding USDT.
2. Conservative parameters and a wider range of assets
   1. If the community chooses this approach, we will follow up with recommendations for a more inclusive configuration on V3 Ethereum as well as updated parameters for all other V3 deployments. As shown in the post above, the parameters would be significantly more conservative, making it possible to include other stablecoins in the future or increase the LTs if market conditions allow.

Chaos Labs recommendation:

Following internal discussions and considering community feedback, Chaos Labs recommends Conservative parameters and a wider range of assets as the most viable solution. The main reasons for this recommendation are:

1. Inclusivity: this option is more inclusive, particularly for USDT, which has a market capitalization that is twice that of USDC and DAI combined. By including USDT as a borrowable asset, E-Mode will be more accessible to a wider user base.
2. Conservative LTs: More conservative LTs will better accommodate more significant short-term de-peg events, providing users with greater stability and security.
3. User experience: Increasing the LTs is always possible, but decreasing them can have a significant negative impact on user experience and liquidations. Therefore, starting with a lower LT is a more prudent approach.