Uniswap ALM Analysis

Note: This report was prepared prior to the exploit that occurred on Gamma on 3-Jan-2024, affecting LPs and leading to loss of funds. Through the manipulation of specific vault parameters that respond to excessive price movements, the attacker was able to “mint a disproportionately high number of LP tokens”. The effects of the exploit and the risks related to similar events are not covered here.

Introduction

The ability of any user to conveniently provide liquidity has been one of the defining characteristics of DeFi markets throughout their development. The first generation of Automated Market Makers (AMMs), which supported only passive liquidity, allowed many users to access convenient LP strategies for the first time. As AMMs have become more flexible and capital efficient, liquidity provision has evolved from strictly passive to more complex, active strategies. While these changes have allowed significant efficiencies in pricing and execution, they have also reduced the opportunity set for passive LPs. Automated Liquidity Management (ALM) protocols exist to bridge the gap between a passive LP experience and the active strategies employed in current markets.

ALM protocols implement active strategies for users who seek a better risk-adjusted return than passive LPs while maintaining the same level of convenience. Due to their popularity and significant market share in some areas, ALM protocols impact the overall functioning of many DeFi markets, including Uniswap. To support the Uniswap Foundation’s efforts to provide valuable insights to the Uniswap community and better understand ALM’s impact on Uniswap, Gauntlet has prepared this report on their roles in the market and relative positioning. The goals of this report are to:

1. Highlight the unique features of each ALM and how they relate to LP objectives
2. Outline the strengths and weaknesses of each ALM
3. Provide high-level insights on the asset types and typical returns in the ALM market

We explore three ALM protocols that represent different approaches and levels of scaling. In choosing these protocols, we aimed to include a wide range of strategy and asset types to provide a broad market perspective. The three protocols covered are Arrakis, Gamma and Mellow.

Executive Summary

This report delves into the diverse landscape of the ALM market, providing essential insights for LPs within the Uniswap ecosystem. It offers a comprehensive foundation for evaluating and classifying different ALM protocols, each tailored to meet specific investor needs and risk profiles. Below we describe each ALMs specialization, as well as estimated yield returns for various categories of assets.

• Arrakis stands out for its focus on low-risk strategies, particularly appealing to LPs interested in stablecoin pairs.
• Gamma offers a diverse array of options, characterized by strong external incentives, catering to a wide range of LP preferences.
• Mellow is noted for its specialization in ETH and BTC pools, employing more complex strategies, including the use of leverage and allocating capital to additional yield-generating protocols.

The ALMs examined showcase a range of specialized strategies, with a common thread being the predominance of range-based liquidity provisioning. These strategies dynamically rebalance positions in response to market movements, a feature seen in the majority of the ALMs analyzed. However, there are also more exotic approaches, as exemplified by Mellow, which involve leveraging positions and deploying capital to yield-generating protocols.

Methodology

Strategy Assessment

In preparing this report, we researched each ALM’s strategies and compiled overviews of their goals and mechanics. We reviewed all publicly available documentation and held several follow-up calls with strategy developers to discuss the finer details of their work. Some automated liquidity protocols typically classified as ALMs use proprietary offchain strategies to provide a similar user experience. However, since these “black box ALMs” are functionally more similar to offchain market makers than to onchain ALMs, we will mention them briefly in this report, while noting this as a potential area for future research.

At a glance, the strategies deployed by the ALMs examined in this report can be classified into two categories:

1. Range-based liquidity provisioning strategies: optimizing for providing liquidity across different price ranges and adjusting according to market movements or other logic.
2. Capital efficiency maximizing strategies: strategic allocation of funds across different fee tiers or narrow ranges, and reallocating inactive capital towards other yield-generating protocols (not necessarily through LPing). 

The vast majority of strategies examined here fall within the former category where the primary distinction is the logic that dictates rebalancing. 

Pool Types

To better evaluate LP performance across many asset types, we group ALM pools into categories depending on the tokens involved. We use four categories of tokens, which we identified as substantially different in terms of liquidity management behavior.

Bitcoin - Pairs involving WBTC or a similar token that tracks the price of native Bitcoin, usually paired with ETH or a stablecoin.

Ethereum - Pairs involving ETH or similar token paired against a stablecoin. For instance, we do not include ETH/UNI pairs in this category but we would WETH/USDC. 

Stablecoins - Pairs involving purely stablecoins. These pools can include any wrapped variants of stablecoins, different stability mechanisms, and different stablecoin providers. Examples of tokens in these pools would be FRAX, DAI, AGEUR, USDT, and USDC.

Other - All other token pairs that do not fall into the preceding categories, ranging from obscure (AXLLQDR/WBNB) to more known (LINK/WETH).

LP Revenue Data

We also gathered data on the yields generated by ALM pools, both from public sources like DefiLlama and through direct conversations with ALM developers. As a metric to assess the returns generated for LPs, we focus on the pool APY, which normalizes for pool sizes and their evolution over time. As a caveat, we note that there are different ways to calculate the APY in use by different protocols. Due to the fluctuating prices and quantities of the assets in the LP pool, there are many different ways to define a benchmark to serve as the “neutral” or 0% APY level. We will present APY data in 3 different ways, which are not necessarily directly comparable:‍

Price Neutral APY - The annualized return that a user would generate if they were fully hedged on price changes of the pool assets at all times. This is not a value typical ALM users directly experience since perfect hedging is not realistic in most cases. However, it reflects a theoretical view of the ALMs performance without external market price moves.

US Dollar APY - The annualized return a user would generate from participating in an ALM pool over a certain period, assuming USD stablecoins as a base currency for the start and endpoints. This measure includes the impact of any appreciation or depreciation of the pool assets against the US dollar and is not a reflection of the ALM's performance in isolation. However, it more accurately reflects the returns that a passive retail user would directly experience due to participating in an ALM pool.

Quoted APY - In some cases, protocols provide APY statistics that do not exactly reflect either the USD or Price Neutral APY frameworks. If the available data was not sufficient to calculate either of our above metrics, we used the values as provided by the protocol’s public UI or strategy developers. We note that these values are closely related to our metrics but not necessarily directly comparable.

A possible area for future work is building a backtesting simulator capable of calculating standardized yields for ALMs based on historical onchain data. This would allow us to more accurately compare ALM protocols that use different yield calculation methodologies, standardize measures for impermanent loss, and evaluate protocols for which data is currently insufficient.

Arrakis

Arrakis V1

Arrakis is designed to primarily serve Uniswap V3 on Ethereum, Polygon, and Optimism. Arrakis V1 vaults use a permissionless single-position strategy, which allows a whitelisted vault manager to execute rebalancings within a defined rule set.

Arrakis V1 Strategy - A vault manager is appointed to rebalance the vault’s single LP position as market conditions evolve. The manager can specify a new range and the token amounts they want to swap in order to adjust the position to the new range. Anyone can be a manager on Arrakis V1, so the precise triggers and amounts for rebalancing are at the manager’s discretion.

Grandfathered Leveraged Positions - Originally on Arrakis, LPs could leverage up to 50x in stablecoin vaults like USDC/DAI through Maker, significantly amplifying yields on these positions. However, Maker ceased new leveraged entries by setting the debt ceiling to zero. This left existing users with legacy leveraged positions that are still profitable but cannot be increased – they are in a "close only" mode. As some leveraged users withdrew, the remaining ones now hold a larger pool share, making their positions more attractive. This has led to the USDC/DAI vaults accounting for over 90% of Arrakis V1’s current TVL despite support for V1 migrating to V2 innovations.

Arrakis V2

The majority of our analysis involving Arrakis focused on V1, given that it has more historical data and vaults to analyze. However, it is worth talking about V2 and the innovations it brings. 

Arrakis V2 is a new core framework that was initially focused on PALMs (Private ALMs). PALMs are a type of ALM that manages liquidity primarily to bootstrap new asset pairs. Protocols and DAOs use this approach as an alternative to efforts like Liquidity Mining. Instead of bootstrapping liquidity through rewards, some protocols may allocate a portion of their treasury (called protocol-owned liquidity) to a PALM designed to support market depth in the new token. Since the goals of PALMs are different from the goals of ALMs that manage liquidity for yield generation, they are not directly comparable. PALM strategies are typically highly customized and not publicly disclosed, as they usually source liquidity from a single whitelisted user (the protocol or DAO).  

More recently, Arrakis V2 has expanded outside of PALMs by launching their Liquid Staking Token (LST) pools, which are public ALM pools open to all LPs. These pools focus on LST tokens issued by Lido and receive an external incentive allocation from Lido with the goal of reducing slippage in the staking tokens. Arrakis has historically relied on range-base strategies in V1 and that continues to be the case with V2, with the notable exception of the Cross-Fee tier strategy. Details of the new strategies in V2 are described below: 

Multi-Positions Strategy - This relatively simple strategy provides liquidity to a single Uniswap V3 pool at different price ranges. The primary benefit is more fine-tuned price exposure than providing liquidity in a single range.

Cross-Fee Tier Strategy - As the name suggests, this strategy allows LPs to provide liquidity across multiple fee tiers under one position. For example, half of the funds can be allocated towards the 0.05% pool for more concentrated liquidity, and the rest can be allocated to the 0.3% fee tier. In the case of stable asset pairs, this approach can allow for large-sized trades close to the peg, while compensating for liquidity providers’ risks further away from the peg by providing liquidity in different fee tiers.

Arrakis V1 Performance

At the time of writing, Arrakis has a median APY varying between 1% and 3%, which is not entirely surprising given the majority of Arrakis’ TVL is composed of stablecoins (52% DAI and 42% USDC). However, it places Arrakis V1 firmly In the category of relatively low-risk and low-return ALMs. In the chart below, we look at the quoted APY of Arrakis pools by pool type and the overall median across all pools. 

While the majority of TVL in Arrakis is made up of stablecoin pairs, they appear to perform at the lower end of Arrakis V1’s APY range. Next we can examine the performance of popular network tokens, which are primarily dominated by ETH.

We can see that historically yields were higher, peaking around March 2023. This corresponds to the spike in DEX volume during the USDC volatility event earlier this year that temporarily boosted the yields earned across most of the DEX ecosystem.

Arrakis V2 Performance

Arrakis V2 vaults are currently all ETH-related vaults (e.g., rETH, wstETH, etc), showing a strong median APY of 14%. Unlike Arrakis V1, these pools receive an external incentive yield roughly on par with the base yield generated from rewards. Currently, 60% of the yield generated from Arrakis V2 comes from external incentives (particularly from Lido). While this means that Arrakis V2 may be dependent on the continuation of these incentives to maintain competitive rewards, the pools currently generate fairly attractive returns relative to other opportunities in ETH.

Arrakis Key Takeaways

Arrakis V1 primarily serves stablecoin-related pools. Given its general low volatility, Arrakis V1 may be most suitable for LPs with a low tolerance for price fluctuations and impermanent loss. However, higher yields could be available from competing ALMs within the same asset types. While their exact returns are unclear, the remaining LPs who have leveraged positions in stablecoin vaults are likely earning very favorable APYs from their grandfathered positions in USDC / DAI. These leveraged positions are no longer available for new users but make up a large portion of Arrakis V1’s current TVL.

Arrakis V2 remains focused on PALMs and has a majority of its TVL (about $20M) in private strategies tailored for specific DAOs or protocols. Though public ALM pools are not the main goal of Arrakis V2, it does maintain a few Ethereum-related vaults that have demonstrated highly competitive performance, in no small part due to Lido incentives. Nevertheless, LPs who are interested in maintaining a position in Ethereum (or its staking derivatives) could consider Arrakis V2 as a very attractive option. ‍

Gamma

Gamma serves a large variety of DEXs (including Uniswap) across many networks (including Ethereum, Polygon, Optimism, Arbitrum, Moonbeam, and Base). The Dynamic Range and Stable strategies are specifically tailored to reduce impermanent loss while maximizing fee earnings for LPs. Gamma’s strategies are considered range-based for the most part, but it can be argued that overly narrow ranges requiring frequent rebalancing would involve some elements of capital efficiency maximizing strategies: 

Dynamic Range Strategy - These strategies involve automated rebalancing of liquidity ranges when certain triggers (e.g., price movements) are hit. They also come in two flavors: narrow and wide. As the names suggest, they indicate the spread of liquidity provided. Narrow ranges cater to risk-tolerant LPs, providing more fees but more subject to impermanent loss (IL) during volatility. Conversely, wide ranges are aimed at less risk-tolerant LPs, minimizing IL during periods of volatility at the cost of earning somewhat lower fees. Accrued fees are compounded back into positions regularly, enabling a passive LP experience.

Stables Strategy - Similar to Dynamic Range but tailored for stablecoin pairs. Unlike the standard Dynamic Range strategy, liquidity ranges are based on historical data rather than dynamically adjusting to live price movements.

Gamma Performance

Gamma’s current median APY sits around 15%. Far more than other ALMs, Gamma supports a wide diversity of DEXs, chains, and asset types. Its largest category of pools is “Other” (119 pools), which includes pairs like WMATIC/GNS, USDC/GIDDY, and many ANKRBNB pairs. The next two largest pool categories are ETH-related (49 pools) and stablecoins (29 pools).

Looking at Gamma’s TVL breakdown by token, we see that WETH makes up 27%, followed by “Others” at 24%. Indeed, not only is Gamma’s preference towards more niche pairs present in its pool selection, but also evident in its TVL breakdown. The skewed preference towards such pools suggests that Gamma’s dynamic range strategies may be well-suited to such tokens with volatile price movements. Stablecoin pools currently offer an APY of around 4%, which is about average for such opportunities. ETH-related pools also show decent but unsurprising performance of around 12% APY, while the “Other” category shows a slightly higher APY of about 15%.

Having looked at many vaults on Gamma, we noticed a pattern where vaults without external incentives typically perform significantly worse than those with incentives. This suggests that incentives contribute substantially to Gamma’s performance. When we broke down Gamma’s returns to isolate the external components of APY, we saw that external incentives made up about half of the pool APY on average. 

However, this share has declined in recent months without a notable drop in usage, suggesting that Gamma is not necessarily dependent on external incentives to provide a competitive user experience. The role of external incentives in ALM competitiveness is also a potential area for further research.

Gamma Key Takeaways

Gamma has by far the most diversified selection of DEXs, networks, and asset types, and offers relatively high APYs as well. However, the high APY is significantly dependent on incentives sourced from external protocols and not from Gamma itself, as this is Gamma’s explicit business model. Without these incentives, Gamma may not be as competitive, though recent declines in the share of external APY appear to have minimal negative effects so far. As with Arrakis, higher volatility pairs on Gamma typically offer higher yields. However, Gamma has a far larger selection of high-volatility pairs, possibly allowing for greater diversification.

Mellow

Mellow deploys a variant of different strategies, most of which are range-based and rely on sophisticated rebalancing logic. A couple notable exceptions are the Uni V3 Boosted and Fearless Gearbox strategies which involve integrations with other protocols to bolster yield through non-liquidity providing means. Both are capital efficiency maximizing strategies and make use of leverage and allocate funds to other yield opportunities, such as staking.

Fearless Gearbox Strategy - This strategy mainly targets the wstETH-USDC pool on Curve. It maintains a position in the pool using leverage via the Gearbox V2 protocol, and stakes the Curve LP tokens via Convex to obtain a boosted return. This strategy is effectively a leveraged version of typical staked LP positions on Curve and Convex. 

Uni V3 Boosted Strategy - It operates on the wstETH-USDC pool on Uniswap V3, aiming to maximize capital efficiency and mimic the bonding curves of Uniswap V2. This strategy maintains a single position where only a small portion of liquidity is provided within a narrow range, which adjusts to the current price.  This strategy benefits from a reduction in the amount of "inactive" capital (capital in the pool but outside the current price range) and therefore potentially increases the yield. The remaining capital is allocated towards a yield-generating protocol, with some additional capital reserved for transaction costs.

Pulse Strategy - Designed for volatile pools, this strategy maintains a position in the wstETH/USDC Uniswap V3 pool and rebalances liquidity to a new interval when the price approaches the current interval margin. If rebalancing is not needed, it collects and compounds fees. It also utilizes an integration with 1inch to reduce swap fees during rebalances.

Pulse Strategy V2 - Similar to the original Pulse Strategy but with a more efficient rebalancing process that expands the existing position uniformly in both directions instead of minting a new position. This reduces the amounts to be swapped and decreases the impact of impermanent loss (IL) on profitability. However, once a position's width limit is reached, a new position is minted around the current tick just as with the original Pulse strategy

Tamper Strategy - This strategy is for the wstETH/WETH pair on Uniswap V3. It operates with three vaults: one ERC20 vault and two UniV3 vaults. The strategy holds two positions in the Uniswap V3 pool and rebalances liquidity between them depending on the pool's price. The ERC20 vault contains a small portion of unused capital that is used to facilitate rebalancing. Three types of rebalances are possible, depending on different scenarios of price positioning relative to the intervals set for the lower and upper vaults. This strategy additionally hedges against oracle risk by obtaining estimations from three separate oracles

Mellow Performance

Mellow provides extensive data on their vault performance to the public, with price-neutral and USD APYs displayed directly on their website. Mellow has the most named strategies among the ALMs we observed, though the strategies operate over a relatively limited selection of pools. Some of the pools examined here have only been very recently deployed, resulting in limited historical data.

There are a number of vaults that show negative APYs over certain periods. However, this is likely a function of the strategies that are employed. Many volatile token pairs on Mellow use the Pulse strategy, which rebalances positions frequently. Impermanent losses in these pools are thus locked in during rebalancing which otherwise may have never been realized. The V2 iteration of the Pulse strategy, which improves the efficiency of rebalancing, appears to be much more performant and has been profitable overall so far.

Mellow Key Takeaways

Mellow’s strategies vary in performance, with some more susceptible to realized losses from rebalancing than others. The performance of Mellow’s vaults is very transparently available, which makes them easier to evaluate than some competitors. It is worth noting that the performance self-reported by Mellow specifically includes the effect of impermanent losses, while other ALMs do not necessarily provide this information upfront. In terms of actual performance, Mellow’s BTC vaults are quite compelling. Even after controlling for price effects, the average yield recently hovers around 5% APY, which is fairly competitive for BTC opportunities. Also, their ETH-related vaults have done well recently despite having periods of negative returns historically. Mellow does not use external incentives so these yields are generated solely from Mellow’s liquidity management activity, which may make them more sustainable if external incentives decrease overall in the longer term.

Conclusions

In conclusion, the ALM market available to prospective LPs is fairly diverse. Each ALM has its own area of focus and tailors its approach to a particular type of user experience. Arrakis focuses on low-risk stablecoin pairs, Gamma offers a wide range of options with strong external incentives, and Mellow specializes in ETH and BTC pools with more intricate strategies. The choice of ALM depends on the LP's risk tolerance, performance objectives, and the specific asset types they wish to participate in. Due to the significant variety in ALM design, reporting metrics, and performance, we look to provide a foundation for the Uniswap community to evaluate and classify different protocols. Though ALMs may become more standardized over time, the current market is much too varied for a single broad metric to capture all the relevant details. As a result, we took a more qualitative approach in this report than we do in most other research. Further work in simulating ALM returns could also help develop more standardized comparisons, as it would allow all protocols to be measured on the same APY scale. The role of external incentives and less public strategies in the ALM market are also areas that could be examined in follow-up research.