Rebalancing vs Passive strategies for Uniswap V3 liquidity pools.
Is auto rebalancing the Holy Grail of yield farming? Does it really add value versus passive strategies?
A flurry of investment strategies have recently emerged on Uniswap V3 to improve yield farming returns. The most notable include the auto rebalancing of liquidity positions. Many investors are attracted to the idea of using an auto rebalancing strategy on their LP in order to keep in their positions inside the liquidity range and maximize the fees they receive.
However, such strategies increase complexity, by introducing path dependence as well as additional gas fees and operational risk. Hence they have been met by incomprehension and caution by some members of the DeFi community.
In this paper, we present an evaluation of rebalancing strategies for Uniswap V3, a decentralized token exchange protocol built on Ethereum. We first go through the mechanics of rebalancing and highlight the main risk exposures added versus an equivalent passive strategy. In the second part of this paper, we run Monte Carlo simulations to model returns and understand which environment is the most conducive to rebalancing. We conclude that:
- Auto rebalancing was attractive when Uniswap v3 launched in May when fees APR were high. Since then, volumes on most pools have plummeted and no longer compensate for the extra volatility risk.
- Auto rebalancing strategies suffer a volatility drag versus passive strategies as they force investors to crystallize their impermanent losses. They increase investors gamma exposure versus passive. These strategies need a high level of fees to be profitable. Therefore one should turn on rebalancing only during periods of high volumes.
- As long as the price remain within our initial liquidity range, we are still collecting the same amount of fees. So there is little incentive in rebalancing our positions. We will significantly increase our gamma risk without a meaningful increase in fee collection.
- Liquidity providers should choose their LP managers with extra care as the selection of the rebalancing range and frequency are significant drivers of returns. Investors should only choose pairs with an attractive fee to token volatility ratio.
Preliminary Remarks
In this paper we model a rebalancing mechanism close to the Alpha Vault proposed by Charm Finance. Please note that our model will defer versus what is actually implemented but the purpose of this article is not to judge the Alpha Vault Strategy but use a concrete example for our analysis. We will benchmark all our results versus 100% USDC and not 50/50 as it is often used by many DeFi providers. our aim being to make an absolute return vs USDC or fiat.
Principles of auto rebalancing
In a passive LP strategy, the investor will usually set up a liquidity range [Pmin,Pmax] and do nothing until liquidity is withdrawn. The profit of a standard LP is a function of the terminal price and the amount of fees collected. The amount of fees you will receive will depend on the % of time the token price has spent in the liquidity range (and also the width of the range).
Through time, it is not uncommon to have a price drift i.e. the token price has moved out of your initial liquidity range, meaning that you have stopped earning fees. So at first sight it would make sense to rebalance your position to avoid that drift ad continue collecting more fees.
The Alpha Vault Concept
The concept of the Alpha Vault has been documented here by Max https://medium.com/charmfinance/introducing-alpha-vaults-an-lp-strategy-for-uniswap-v3-ebf500b67796
The idea is to choose a base liquidity range symmetric around the current price [Current -B, Current+B] and to rebalance regularly (let’s assume twice a day). As the token price moves, our initial ratio (50/50 each asset) will drift. As our token price decreases, we buy more of the token and inversely as it increases.
Now let’s assume that the price moves to X and that our portfolio is now composed of 70% T and 30% USDC. The strategy will post a new LP position at [X-B, X+B] for 30%USDC/30% T. The remaining extra 40% T is placed as a passive rebalancing order at [X,X+R] with R being the rebalancing range. At every rebalancing, we count the number of token we hold and redo the same process.
Increased fees for increased Volatility exposure
Let’s put that into practice with a concrete example.
Let’s assume we start T/USDC=100, we put our liquidity inside the range [90..110] and the price at the rebalancing date is 105. Our portfolio is now composed of 163.3 USDC and 0.4821 T. Expressed in USDC, our portfolio is now 76.3% USDC and 23.7% T. So we have an extra 52.6% of USDC which will translate into a buy order of Token T at a range [104,105]( range is just an illustration). The rest of the position will be placed in a 23.7%/23.7% range[95,115].
Below is the comparison between active and passive strategies depending on what is the next price of Token T.
We can see that the impermanent loss of the Rebalance Strategy will underperform a passive strategy most of the time except in the case where Token T continues to appreciate (in our case finishes above 105). Versus a passive Strategy, the rebalancing strategy is short realized volatility and is expected to produce a higher impermanent loss. However, this strategy will have more concentrated liquidity giving us the opportunity to earn more fees.
Simulating Returns
As we can see the returns of a rebalancing strategy depend on the level of realized return volatility. The strategy is expected to outperform in a trending market but underperform in a choppy market.
In our previous article https://medium.com/@DeFiScientist/uniswap-v3-a-quant-framework-to-model-yield-farming-returns-941a1600425e, we concluded that the level of fees had to be high enough to compensate for the volatility risk taken by liquidity providers. We found out that on most pools the fees were too low for yield farming to be a viable strategy versus selling puts.
In the case of an auto rebalancing strategy, it seems that the level of fees relative to volatility is even more important.
Assumptions
We are going to explore the sensitivity of the relative returns versus passive with Monte Carlo simulations. These are more robust than the historical backtesting most people do provided you have the correct initial assumptions. Here are our base assumptions:
- Log returns follow a t student distribution wish degree of freedom 5. This looks to fit past returns for Bitcoin and Ethereum. A t student distribution is more realistic than a normal distribution as it allows for fat tails. Remember the BTC flash crash of May 2021 when BTC collapsed 25% in 1 hour this would have never happened in a normal distribution.
- Our simulations do not assume any prior knowledge of future returns on the pair. If we knew this, we would not use LP!
- We simulate returns every 6 hours this enables us to get a better modelling of the amount of fees received during the period.
- Rebalancing every 12 hours and simulate returns over 5 days (20 time steps)
- For simplicity, we assume flat liquidity all over the price range. We know that this is not the case but we do not want to incorporate additional complexity at this stage.
- Pair used : WETH/USDC (fees 0.3%)
- Implied volatility assumed 110% in line with option markets.
- Pool Fees: 400 000$ a day. Many of the simulations you see from providers are based on past data where fees and volumes were much higher. For example volumes were 7x higher in May than currently with a TVL only 50% higher on WETH/USDC. Their results will show very good results but we believe those results will not be replicated in the future due to increased competition.
- Base range +-9.6% for both strategies. Rebalancing range:+-4.8%
Sensitivity to Base range
In this simulation, we look at various base ranges but fix the rebalancing range at 4.8%. Both passive and rebalancing strategies have the same base range for like for like comparisons. The auto rebalancing strategy will underperform on average with the underperformance being particularly acute for small ranges. The extra fees we collect are not enough to compensate for our short volatility position.
In auto rebalancing strategies, the range should not be too narrow as every time we rebalance we crystallize our losses and risk getting consistently whipsawed.
Sensitivity to Rebalancing range
We fixed the base range at 9.6% and used various rebalancing ranges. From our simulations, there seems to be an inverse linear relationship between performance and the size of the range.=> the smallest the range the better.
Sensitivity to Rebalancing Frequency
In those simulations we have kept the rebalancing and base range constant and we use various ranges of rebalancing frequency (expressed as multiples of 6 hours). We also use two types of volatility regimes (high vol: 110% and low vol : 50%).
If we rebalance too frequently we are going to get whipsawed too often and our performance will materially suffer. The rebalancing strategy can add value provided 1) the level of volatility is low enough relative to the fees we collect 2) the rebalancing frequency is appropriate.
Conclusion
Fixed rebalancing parameters do not seem ideal. As rebalancing is a costly process, it should only be done when necessary. We also believe that the ideal solution lies in using dynamic parameters. In the upcoming weeks, we plan to carry out further research in that direction. I hope that you enjoyed our second analysis of Uniswap V3. We have more innovative quant research in the pipeline so do not forget to follow us on Medium and Twitter https://twitter.com/DefiScientist
Disclaimer: The content of this post is provided for informational purposes only. Nothing herein constitutes investment, legal, or tax advice or recommendations. Nothing on this site should not be relied upon as a basis for making an investment decision. It should not be assumed that any investment in the asset class described herein will be profitable and there can be no assurance that future events and market factors would lead to results similar to any historical results described.
