The impact of protocol fees in Reclamm pools is much larger than for other pool types such as a regular weighted pool. From simulations we observe that at the current level of 50%, the impact on a Reclamm can be ~5% annually deducted from the LPs position. From recent real-world data we even observed larger impacts corresponding to periods of very high volatility.
In this report we present quantitative results relevant to decide the value of the protocol fee parameter for Reclamms, from now on denoted as reclamm_protocol_fee.
(a) Simulations
Through arb-only simulations we obtain a good approximation of the effect of protocol fees in weighted and Reclamm pools. By comparing the two cases we can make an informed decision for reclamm_protocol_fee. Notice that the presence of organic-flow, expected in some cases like BTC/ETH, ETH/USD and hopefully other, would lead to a better performance for both the pool and the collected protocol fees.
By preliminary inspection of some cases we started by observing that reclamm_protocol_fee = 10% is a possible candidate. Therefore we use this value for the simulations, as a reference. The behavior for a different value, like 5%, 15% or 25%, is proportional (geometrically, though approximately additive, since the impact is ~1%). So different values for reclamm_protocol_fee are immediately intelligible given the data presented here.
For different pairs we simulate the positions:
Find the results of the simulations in this file:
From the numbers obtained we can easily compute the effect of the protocol fees. For some pairs it's close to 1% annually while for other pairs (like COW/ETH or ETH/USDC) it's close to 2%.
From the charts, we can easily visualize that the effect of a 10% protocol fee in a Reclamm is comparable to the effect of a 50% protocol fee in a weighted pool. For some pairs it's very close, like GNO/ETH or ETH/USDC, while for other pairs the effect on ReClamm is more like 65% of the effect on Weighted pools.
Note: even though for most pairs we see ReClamms performing better than weighted, we should be careful, since the parameters are optimized given the past price evolution. The future performance for parameters based on the past could be worse than this.
(b) Simulations alongside live pools data
We have compared the evolution of the value of most of the live Reclamm pools with the corresponding simulated arb-only dynamics. We also compared these with the value for standard positions such as “hold” and “feeless weighted”.
In some cases (namely WXPL/USDT0 and AAVE/WETH) the reality closely matched the simulation results. For other cases (cbBTC/WETH and EURC/USDC) the value from the actual pool surpassed the simulated one, indicating the presence of organic flow. Finally for GNO/USDT and GNO/wstETH, the opposite holds, indicating inefficient arbitraging, possibly caused by insufficient liquidity outside the pool. However the case GNO/USDT is notable because the live pool outperformed “hold” by a significant (annualized) 3.5% approximately.
Given the present results, the experimental nature of the Reclamm pool type, and the DAO's potential long-term gains from having good results for the LPs, it seems like a good idea not to use a value larger than 25% for reclamm_protocol_fee. Moreover, a value lower than 25% could be an excessive deviation from Balancer’s standard value of 50%. With the new value, we will see better LP performance and, presumably, profitable positions for more pools. We will continue monitoring the evolution of the pools in order to reassess this decision.