Key Takeaways

Curve’s LLAMMA pools introduce soft liquidations that minimize borrower losses, improve protocol efficiency, and do not need to rely on external DEXs.
The new stablecoin, crvUSD, is pegged to a basket of fiat stablecoins through its stability pools. Peg keepers are smart contracts assigned to stability pools that aim to anchor crvUSD to its peg.
The new primitives increase the revenue potential for Curve and ultimately veCRV lockers. The CRV lock rate will indicate if investor preferences shift in favor of time-locking tokens for yield.

Curve is the largest DEX by TVL and ranks second in daily trading volume on Ethereum. It has become one of the most efficient DEXs through the creation of its novel AMM designs that provide traders and LPs with a better user experience. The protocol has performed well since the start of the year and is actively launching the major upgrades discussed in our Q1 report.

On May 17, Curve officially launched its stablecoin and lending market, marking the evolution of the protocol beyond just a DEX. Among many benefits, the expansion increases protocol revenue potential and improves the outlook for the health of CRV.

The Current State of DeFi Lending

Lending protocols began to see meaningful traction during DeFi Summer in 2020, with two lending models proving to be scalable. Aave and Compound popularized the peer-to-pool model, where suppliers provide pooled liquidity to fund overcollateralized loans for borrowers. The majority of interest paid by borrowers flows to suppliers, given their capital is required to make markets. In contrast, MakerDAO uses a peer-to-protocol model that improves protocol profitability by replacing the suppliers with the protocol itself. Borrowers still deposit collateral to open overcollateralized loans, but the debt asset must be the protocol’s stablecoin, so all interest paid by borrowers flows to the protocol.

In all overcollateralized DeFi lending mechanisms, volatile collateral poses a risk to the solvency of the protocol and must be managed. Most existing protocols use “hard liquidations'' where a borrower’s collateral is sold at a discount to repay the outstanding loan if the value of their collateral drops below a certain threshold. The threshold is often determined by a “health factor” that measures the amount of collateral relative to the outstanding debt for each borrower. The collateral value is generally decreased by a risk discount (liquidation threshold or collateral factor) based on the market cap, liquidity, and volatility of the collateral asset. Borrowers are subject to hard liquidation when their health factor falls below 1.0, meaning the loan has become undercollateralized.

Hard liquidations with tight risk parameters have effectively protected protocols from bad debt. However, the design forces the borrower to realize losses as their collateral cannot be recovered post-liquidation, even if the value of the collateral recovers. Additionally, each loan is treated individually, so arbitrageurs must execute liquidations in separate transactions. For example, if 1,000 loans have a health factor below 1.0, it takes 1,000 liquidation transactions to protect the protocol from carrying bad debt.

LLAMMA: Curve’s Lending Market

Curve’s peer-to-protocol lending model improves the borrower experience by revolutionizing the liquidation process. It utilizes the Curve’s latest AMM design, the lending-liquidating AMM algorithm (LLAMMA), to enable “soft liquidation” that minimizes borrower losses during liquidation.

Instead of experiencing a hard liquidation when a borrower's loan becomes undercollateralized, LLAMMA pools gradually convert collateral into stablecoins as the value of the collateral decreases, preserving borrower capital. The conversion happens over a range rather than at a single point and ensures the value of the remaining stablecoins is greater than the outstanding debt. Importantly, LLAMMA also “deliquidates” loans by converting the stablecoins back into collateral as the price recovers above the liquidation range.

The Liquidation Range

Curve’s lending mechanism only offers overcollateralized loans, with the max LTV dependent on the risk of the collateral. To open a loan, borrowers deposit collateral into a LLAMMA pool and receive debt denominated in Curve’s native stablecoin, crvUSD. The collateral is spread across a user-specific set of “bands," creating the “liquidation range.” The number of bands determines the length of the liquidation range and is configurable by the borrower, bounded between a total of four and ten bands. A loan with ten bands has a long liquidation range, so it experiences a gradual liquidation. A loan with only four bands has a tight liquidation range and experiences a more rapid liquidation.

The image below depicts two equivalent loans that borrowed crvUSD against 8 ETH at the same moment in time, the only difference being the number of bands. For simplicity, we use ETH as the collateral in the example. Loan A has fewer bands, so it has a more dense liquidation range that begins at a lower ETH price (1675) than Loan B (1750). Notice how the lower bound of Loan A’s liquidation range (1575) is at a higher ETH price than the lower bound of Loan B (1550), meaning it will reach full soft liquidation before Loan B. The condensed liquidation range forces the protocol to soft liquidate more collateral in a shorter period of time, so the risk parameters must be tightened. Note the image below is an oversimplified example only for illustrative purposes.

LLAMMA Price

Unlike regular AMMs, the price offered by a LLAMMA pool does not strictly rely on the balance of the assets in the pool. LLAMMA uses an external price oracle to modify the constant product invariant (x*y=k) in such a way as to always create a larger price change in the AMM than what is reflected by the oracle price. For example, if the oracle price increases/decreases by 1%, then the LLAMMA price will increase/decrease by at least 1%. The design intentionally creates arbitrage opportunities to incentivize third parties to interact with the pool, which in turn operates the liquidation process. The use of an oracle is required, and the protocol is currently leveraging a Chainlink-based solution. In the future, LLAMMA pools can be modified to utilize a different price oracle.

Collateral Conversion via Soft Liquidation

LLAMMA pools need to soft liquidate collateral to remain solvent when the market price of the collateral decreases, so they incentivize arbitrageurs to swap crvUSD for the pool’s collateral. During collateral price declines, the LLAMMA price decreases more rapidly than the oracle price. Therefore, Curve’s lending pool sells collateral for less than the current market price. Arbitrageurs take advantage of this opportunity by buying collateral from the pool with crvUSD and instantly selling the collateral on the market, capturing the spread between the market price and the LLAMMA price. The arbitrage transaction converts the collateral into crvUSD, soft liquidating borrowers that have collateral in the active band.

When the market price of the collateral later increases, the LLAMMA deliquidates borrowers by incentivizing arbitrageurs to sell collateral back into the pool. As the oracle price increases, the LLAMMA price increases at a faster rate. Therefore, the lending pool is buying collateral above the current market price. Arbitrageurs now buy collateral from the market and sell it to the LLAMMA pool, capturing the spread between the LLAMMA price and the market price. The arbitrage transaction converts the LLAMMA’s crvUSD back into collateral, returning capital to borrowers.

The animation below walks through the lifecycle of the soft liquidation process for a crvUSD loan against ETH collateral. When the AMM price is above the borrower’s liquidation range [1], the LLAMMA pool holds the collateral in ETH. When the AMM price moves into the borrower’s liquidation range [2], soft liquidation is initiated and arbitrageurs are incentivized to convert collateral into crvUSD. If the AMM price moves beyond the lower bound of the liquidation range [3], the user is fully liquidated with all collateral held in crvUSD. The LLAMMA pool ensures that the amount of crvUSD is greater than the outstanding debt at full liquidation, maintaining the overcollateralized nature of the loan. As the collateral price recovers, the crvUSD is converted back into ETH [4], but notice that a small portion of the borrower’s capital is lost.

While arbitrageurs play a critical role in maintaining the health of the protocol, each transaction erodes borrower collateral. In other words, borrowers must pay a small price for the potential ability to recover their collateral after being soft liquidated. The length of the liquidation range, determined by the number of bands, impacts the frequency and severity of collateral erosion. Loans with fewer bands have a tighter liquidation range, so they are less likely to be in soft liquidation for extended periods of time. Therefore, decreasing the number of bands decreases the time exposure to collateral erosion. However, a dense liquidation range holds a larger percentage of the borrower’s collateral within a single band. Thus, tighter liquidation ranges experience more erosion on a percentage basis when in soft liquidation.

To quantify the impact of the collateral erosion, Curve Founder Michael Egorov has been tracking the change in health of his loan. He opened an 89% LTV loan with a ten band liquidation range to test how it would react to soft liquidation. Note this is a very aggressive loan that was meant to push the limits of the protocol. After eight days in soft liquidation, the collateral price recovered above the liquidation range and he regained 98.6% of his collateral. The borrower did not need to worry about hard liquidation and only paid 1.4% of the collateral to do so.

If the liquidation process fails to function properly, hard liquidations become a backstop to protect the protocol from insolvency risk. Curve calculates the health of a user’s loan, similar to traditional lending protocols, and allows for hard liquidations if the health factor falls below a specified risk threshold. Additionally, hard liquidations can occur if a loan is in soft liquidation for an extended period and the borrower’s capital is eroded below the risk threshold. For this reason, it remains important to monitor the health of LLAMMA loans, especially those with a high LTV. At its best, Curve’s lending model significantly improves DeFi liquidations. At its worst, it reverts to the battle-tested hard liquidation model used by the largest DeFi lending protocols today.

Soft Liquidation Efficiencies

Since the collateral is held in an AMM rather than a simple pool, liquidations are processed on the pool level instead of the individual loan level. One soft liquidation transaction updates the balance of the LLAMMA pool, so all loans in the active band are updated simultaneously. This is a significant efficiency improvement over hard liquidations that monitor the health of each loan separately.

Curve’s lending model is consistent with its approach to DeFi. For example, Curve concentrated liquidity pools (Curve V2) rebalance the entire liquidity pool with one rebalance transaction. That is a fundamental design difference to Uniswap V3, which relies on LPs to rebalance their positions individually. Curve continually takes a holistic approach to its mechanism design, resulting in a more efficient protocol.

Further, hard liquidations can cause large market sell-offs and cascading liquidations. LLAMMA improves this by gradually selling out of the collateral, avoiding sharp and sudden market impact. This allows Curve to loosen its risk parameters when determining the appropriate max LTV for a listed collateral. Liquidators have to sell the collateral onchain to book profits, so deep DEX liquidity is an important variable to determine an asset’s max LTV in a hard liquidation model. Less liquid assets must have higher liquidation bonuses to ensure profit. This parameter can be loosened with a gradual soft liquidation process. Instead of $10M of collateral being liquidated in one transaction, it is spread evenly over the borrower's liquidation range.

Current State

The team relentlessly tested the contracts before officially recognizing them and opening the UI to users. The contracts were deployed a total of four times before finalization, and Curve Founder Michael Egorov personally tested the final deployment with around $1M of his own capital. Critics cited the multiple deployments as a sign of trouble, but in reality the final step of the process was testing in production. Curve contracts are immutable, meaning they cannot be upgraded after deployment, so additional testing on mainnet is necessary.

On May 17, the official deployment was recognized, and the frontend went live. The first listed collateral is sfrxETH, which has rapidly accumulated $7.4M in collateral deposits to mint $5.1M of active loans. The pool has facilitated $3.3M of swap volume (~45% of TVL) in just one week, showing steady demand from arbitrageurs to maintain the pool’s health. The pool offers an 89% max LTV, which is significantly higher than other lending protocols offer for USD loans against LST collateral. For example, Aave V3 offers a max LTV of 68.5% against stETH collateral if the debt asset is anything other than ETH. Expanded risk tolerance is possible with LLAMMA because gradual liquidations minimize insolvency risk to protocol.

Future Collateral Listings

Although the core team selected sfrxETH as the first collateral type, the Curve DAO must approve all future collateral listings. Other LSTs like stETH, rETH, and cbETH seem to be the current priority based on community discourse. While top tier assets ETH and WBTC will of course be added, Curve has an opportunity to dominate the LST lending market. Much of that market is borrowing ETH against LSTs to generate leveraged ETH staking yield, and LLAMMA cannot offer this as the debt asset must be crvUSD. However, the amount of native ETH in lending pools is shrinking as users demand to generate at least the ETH staking yield. For this reason, we expect the ETH borrow rate in peer-to-pool models to converge with the ETH staking rate over the long term. As ETH borrowing rates move higher, Curve becomes an increasingly attractive solution for borrowers because other lending protocols cannot compete with LLAMMA’s improved risk tolerance.

In the future, Curve LP tokens like the tricrypto pool (ETH/WBTC/USDC) will be listed as collateral, improving the stickiness of Curve’s DEX liquidity. LP tokens are an interesting collateral to explore given the yield bearing nature of these assets, but using the protocol’s LP tokens as collateral for loans is clearly reflexive. While this would bolster Curve’s liquidity in a bull market, a large sell-off could cause a significant portion of Curve’s DEX liquidity to unwind. However, soft liquidations and proper risk parameters around max LTV help mitigate the risks of intertwining the DEX and the lending market, and no other protocol can compete with this feature set today. Given the cautious approach taken by the team when creating and launching the contracts, continued testing will likely be advised before new assets are listed and debt ceilings are raised.

Areas for Improvement

High gas costs are the biggest shortcoming of LLAMMA as complex computation is required to interact with the contracts. The Curve team is considering expansion to L2s to alleviate this, though they have expressed a desire to keep the main deployment on Ethereum mainnet. The crvUSD factory is controlled by the Curve DAO and mints crvUSD (up to the debt ceiling) to LLAMMA pools containing different collateral. Therefore, one potential option for L2 expansion is to mint a crvUSD allocation to an auxiliary LLAMMA deployment on an L2. This would allow users to avoid high gas costs by interacting with the contracts on an L2, yet the core contracts would live on Ethereum mainnet and receive its liveness and security guarantees. Additionally, Curve can manage its risk exposure to nascent L2 networks by throttling the debt minted to these auxiliary contracts.

crvUSD Monetary Policy

A stablecoin is only as useful as its ability to remain pegged to its underlying target. crvUSD is pegged to $1.00 through a basket of fiat stablecoins, and the protocol uses “stability pools” and “peg keepers” to maintain the price. Stability pools are stableswap pools that pair crvUSD alongside another stablecoin, and peg keepers are contracts with permission to mint a capped amount of crvUSD to improve price stability.

Stability Pools

Currently, there are four stability pools for each of the largest fiat-backed stablecoins USDC, USDT, USDP, and TUSD. The idea is for crvUSD to be pegged to this basket of stablecoins. The aggregate crvUSD price from these pools determines the crvUSD price used in the borrower interest rate calculation.

To seed the initial capital of the stability pools, Curve Founder Michael Egorov borrowed over $1M of crvUSD and deposited 250,000 crvUSD and 250,000 stablecoins into each pool. Gauges for the stability pools are currently undergoing a governance vote and are slated for activation on May 26. With gauges, the pools become eligible to earn CRV emissions that LPs can farm. Since the only way to mint crvUSD is to borrow against listed collateral, CRV emissions to stability pools spin the borrowing flywheel. There is currently over $11.6M of TVL across the four stability pools, with the USDC pool being the deepest at $3.9M.

Peg Keepers

Somewhat similar to the AMOs used by Frax Finance, peg keepers are smart contracts that work to anchor the price of crvUSD to its peg. Peg keepers are assigned to specific stability pools and can perform open market operations under certain conditions. When (and only when) the price of crvUSD is above 1.0 in a stability pool, the assigned peg keeper can mint and deposit crvUSD into the pool. Adding single-sided crvUSD liquidity into a pool when the crvUSD price is above the peg puts downward pressure on the price, moving it back towards its peg. If (and only if) the price of crvUSD later moves below the peg, the peg keeper can remove and burn single-sided crvUSD liquidity to put upward pressure on the crvUSD price, pushing it back to the equilibrium price. Peg keepers have a ceiling on the amount of crvUSD liquidity they can deploy into their assigned stability pool. Currently, each peg keeper has a debt ceiling of 2.5M crvUSD.

The actions taken by peg keepers generate revenue for the protocol. Depositing single-sided liquidity for the asset with a smaller token balance (when crvUSD price is greater than 1.0) in the AMM returns more LP tokens relative to a balanced deposit. It will then take fewer LP tokens to single-sided withdraw the same amount of crvUSD when the price falls below 1.0. Therefore, the difference in stability pool LP tokens is the revenue generated by the peg keeper. 20% of the revenue is shared with the user that calls the contract to update the debt balance, while the remaining 80% is paid to veCRV holders.

A major depeg event for one of the stablecoins used in a stability pool poses a risk to peg keeper operations. In a depeg scenario, the peg keeper buys the depegging fiat stablecoin until it hits its debt ceiling. Using the recent USDC depeg as an example, the price of crvUSD in the USDC/crvUSD stability pool would have lurched higher as USDC holders panicked to sell their fiat stablecoin. Given the length and severity of the depeg, it is likely that the peg keeper would have acted as USDC exit liquidity, deploying its entire 2.5M crvUSD debt. The incident highlights the danger of pegging to other stablecoins, but this scenario would have actually resulted in profits as USDC successfully repegged. Curve must maintain a strict preference for reliable stablecoins as stability assets.

Variable Interest Rate Calculation

Like other peer-to-protocol designs, borrowers pay variable rate interest to the protocol based on their total outstanding debt. The current interest rate is dependent on the current crvUSD price, while the shape of the interest rate curve is a function of the risk parameters (sigma and rate0) and the activity of the peg keepers (debt ratio and target debt). Interest accrues to borrowers in the pool by ratcheting the bands of the liquidation range higher.

The interest rate model operates under the assumption that borrowers generally sell their crvUSD debt after opening a loan. For example, a LLAMMA borrower is expected to take out a loan in crvUSD and sell it on the open market to perform some action with the borrowed funds. Therefore, when the crvUSD price falls below its 1.0 peg, the borrow rate rapidly increases to incentivize borrowers to repurchase crvUSD from the market and close their loans. The design aims to create buy pressure on crvUSD to help tighten the peg. Conversely, when the crvUSD price rises above its 1.0 peg, the borrow rate rapidly decreases to encourage more loan creation.

The Curve DAO sets the initial risk parameters (sigma and rate0) based on the riskiness of the asset. The parameters change the shape of the interest rate curve, which influences properties like the interest rate at equilibrium (rate0) and how rapidly the rate increases when the crvUSD price falls below 1.0 (sigma). Tighter parameters for riskier assets lead to higher interest rates, influencing borrowers to repay risky loans when the price of crvUSD declines. Intuitively this makes sense, as the system works to reduce risk when the market is dumping crvUSD.

The aggregate activity of the peg keepers plays an important role in the interest rate calculation through two variables, the debt ratio and the target debt. The debt ratio is the amount of crvUSD actively deployed by the peg keepers relative to their combined debt ceiling. The target debt is the intended percentage of debt deployed. With four peg keepers that have a 2.5M debt ceiling, the current max debt in the system is 10M crvUSD. The target debt is currently set to 10% or 1M crvUSD. Considering the peg keepers mint unbacked tokens into stability pools, this is an important metric to measure the risk in the system.

The interest rate calculation is in equilibrium when the debt ratio equals the target debt. All else equal, peg keepers deploying crvUSD into stability pools increases the debt ratio, which decreases the interest rate and makes borrowing cheaper. Borrowers sell crvUSD on the market to put their additional capital to work, lowering the current crvUSD price. If the crvUSD price falls below peg, peg keepers withdraw their debt from the stability pools, decreasing the debt ratio and shifting the interest rate curve back to equilibrium. Therefore, the interest rate formula adjusts the borrow rates to rebalance peg keeper involvement towards the target debt ratio.

While the images above are meant to illustrate that more debt deployed leads to lower rates, the interest rate curve is built using the current settings of the sfrxETH LLAMMA pool, only modifying the debt ratio. Further, a higher target debt steepens the slope of the curve when crvUSD price is below peg, maintaining higher interest rates. Higher interest rates push the peg keepers to continue deploying debt under a wider range of market conditions.

Overall, the interest rate formula balances risk preferences with current market dynamics to incentivize desired borrowing behavior. The Curve DAO should continue measuring the impact of different parameter settings during different market conditions to ensure the durability of the system.

Protocol Revenue

The lending market and stablecoin generate revenue for the Curve DAO in two meaningful capacities, interest paid by borrowers and stability revenue from peg keeper operations. Nearly 100% of this revenue flows back to the Curve DAO consisting of veCRV lockers. In the first week of deployment, 85% of revenue was generated from interest fees, while the remainder came from peg keeper operations.

Variable rate interest is collected on outstanding crvUSD loans, with all interest paid to veCRV lockers. The sfrxETH pool has an equilibrium interest rate of roughly 3%, so every $100M of outstanding loans would generate an additional 0.5% APR for veCRV lockers based on the current amount of locked CRV. While 0.5% is a relatively small number, it would provide a ~40% increase to the current veCRV yield of 1.25%. Scalability will determine the revenue potential of Curve’s new lending market. Aave and Compound currently have a combined $3.4B of assets borrowed, so Curve only needs to compete for a portion of this market to have a staggering impact on veCRV yield. Given the user experience enhancements and focus on adding more LST collateral, Curve is well-positioned to compete for the future demand for leverage.

Stability revenue is collected from the operations of the peg keepers, and 80% of this revenue flows to veCRV lockers. Deeper stability pool liquidity and crvUSD price volatility increase the size and frequency of peg keeper activity. While stability revenue comprised a majority of the total revenue in the test deployments, the target debt ratio was set to 50%, 5x higher than the current value. Setting a higher target debt ratio allows peg keepers to deploy more crvUSD liquidity into the pools, which increases the revenue potential of the peg keepers. This is a risk parameter the Curve DAO can, and will, expand as the LLAMMA design becomes more battle-tested.

Further, the Curve DAO has additional potential streams of revenue generation in the future. Since arbitrage transactions through LLAMMA pools are technically swaps, the traders pay fees just like regular Curve swaps. Currently, the swap fee on the sfrxETH pool is set to 0.6%, and all swap fees flow to the borrowers (LPs) in the active band. The swap fees are significantly higher than a regular Curve pool in an attempt to throttle the amount of arbitrage that occurs. Borrower collateral is eroded as arbitrage transactions occur, so the high transaction fees work to preserve borrower capital. As more LLAMMA pools are added and liquidity deepens, the protocol can explore a revenue share on these fees, but it must be cognizant that these fees mitigate the impact of collateral erosion. Additionally, the Curve DAO can consider auctioning the rights to arb LLAMMA pools and share profits with borrowers in the pool or the veCRV lockers.

Impact on CRV

Curve’s lending mechanism and stablecoin ultimately stand to increase the total revenue generated by the protocol, improving its sustainability by increasing the amount of veCRV locked. Curve uses CRV emissions to incentivize LPs to provide liquidity to its pools and attract traders. A further explanation of the tokenomics can be read in the “CRV veTokenomics” section of our recent Curve report.

The daily lock rate (CRV locked/CRV emitted) serves as a health check on the Curve protocol. When veCRV is in demand, more tokens are locked, greatly reducing the impact of daily emissions. As trading activity diminished during the bear market, the revenue stream for veCRV lockers dried up and the yield fell. The lock rate has continued to slide since making an all-time high in July 2022. Before this upgrade, veCRV yield came only from the 50% share of protocol swap fees, but now it will now also consist of 100% of interest revenue and 80% of peg keeper stability revenue. If the lending protocol scales, veCRV yield stands to increase dramatically, attracting more CRV holders to lock their tokens and decreasing the circulating supply. Tracking the change in the lock rate is an indicator that protocol health is improving, making the token more attractive.

Final Thoughts

Scalability is everything, and the demand for leverage will likely remain muted for the remainder of the bear market. As the leading DEX by TVL, second most active DEX by volume on Ethereum, and a revolutionary lending mechanism, Curve is well positioned to be the battleground of the next bull market. Curve’s ability to drive incentives at crvUSD stability pools is an advantage it has over all competitors in the lending space. As market sentiment transitions to risk-on, tracking the outstanding debt on Curve’s lending market will indicate if it is seeing meaningful adoption that will drive fees to veCRV lockers. More importantly, tracking how the outstanding debt grows as a percentage of total DeFi debt will prove if the market favors the LLAMMA design over traditional mechanisms.

If the LLAMMA design scales, protocol revenue and veCRV yield will increase, attracting more demand to buy and lock CRV. Token holders would likely continue to lock CRV until veCRV yield returns to its historical average between 1% and 3%, potentially pushing CRV price higher. The deepest source of liquidity for CRV, both onchain and offchain, is the CRV/ETH Curve pool that has a +/- 2% depth of about $1.2M. Watching for abnormal activity in this pool and tracking the lock rate is a reasonable way to watch for signs of CRV accumulation.

Curve's Stablecoin and Lending Market

Active Subscription Needed