# Introduction

The tradeoffs of protocol fees on Uniswap have been a topic of active debate for several years. Charging LPs a percentage fee for the yield they accrue on Uniswap could be a substantial source of revenue for the protocol, which could then be used to fund various strategic initiatives. On the other hand, there are also significant costs that have so far been difficult to quantify.

To make the best community decision, UNI holders must be equipped with robust research and predictive analysis to better understand the range of outcomes. This report builds on our prior work on DEX economics with the Uniswap Foundation to consider the tradeoffs of protocol fees. We provide estimates of the risks and benefits of implementing a fee to help inform discussions in the community on this important strategic tradeoff.

# Goals

We use economic models and simulations of Uniswap and competing DEXs to evaluate the impact of fees on Uniswap V3 in three different areas:

**1. Quantifying the potential benefits of a protocol fee**
- *How much revenue can a protocol fee generate?*

**2. Quantifying the potential downsides of a protocol fee**
- *How much liquidity and volume would be lost?*
- *Is there a threshold where fees create a major negative flywheel for users?*

**3. Assessing the impact of a protocol fee across different pools**
- *Are some pools disproportionately affected?*

In answering these questions, we take into account many critical details, such as the possibility of triggering negative feedback loops in liquidity and volume, the different types of pools and trading strategies active on Uniswap, and the dynamics of competing exchanges that may attract cost-sensitive traders. However, we do not look to propose a specific protocol fee solution in this report. Since the optimal tradeoff involves many qualitative and strategic factors, we will not make any concrete recommendations until we can further evaluate these in the context of the analysis presented here. Instead, we will focus on forecasts that can help clarify the pros and cons and inform further discussions.

# Methodology

The models used in this report are the latest iteration of tooling that has evolved throughout our work with the Uniswap Foundation. The [liquidity-volume model](https://gov.uniswap.org/t/uniswap-incentive-design-analysis/21662/2#simulation-11) initially developed for incentive analysis is combined with the [swap simulator](https://www.gauntlet.xyz/resources/uniswap-price-execution-analysis) from our more recent work on price execution. This combination allows us to explore detailed counterfactuals within a robust economic framework for how a DEX operates at a higher level.

## Liquidity Model

We start by modeling the effect of protocol fees on liquidity, assuming LPs adjust efficiently to the initial yield drop. As LPs withdraw, the APY for the remaining liquidity should gradually return toward the pre-fee level, provided this level represented an efficient market price. If the yield is instantaneously cut by 1% due to a protocol fee, this would require removing 1% of the initial TVL, conservatively assuming that LPs will withdraw enough liquidity to bring their APY fully back to pre-fee levels.


It is possible that some LPs would accept a lower APY or adjust their liquidity to a tighter range to return to a higher APY instead of withdrawing. Either of these cases would result in less TVL being removed to restore market equilibrium than we estimate. However, these would also imply that either the efficient level of APY or users' tolerance for impermanent loss has changed. To make our estimates more conservative and avoid excessive assumptions on LP behavior, we exclude changes in LP yield preferences and tick ranges from the impact forecasts. This means that, to first-order, the liquidity model estimates a linear impact of 1% TVL drop uniformly across the tick distribution for every 1% of protocol fee.

## Volume Model

The volume model uses the results from the liquidity model to create a counterfactual DEX pool, called the Mock DEX, within our swap simulator. The simulator can then backtest swaps that were originally routed to the real Uniswap pool on the Mock DEX and competing DEXs. By observing the performance of the Mock DEX in this realistic environment, we can then estimate how much volume would be lost due to its reduced competitiveness caused by the protocol fee.

The diagram shows the component parts of the volume model and how it fits together with the liquidity model. In the following sections, we will break down each of these parts, starting with the core volume simulation.


### Core Volume Simulation

The core volume methodology is extended from our earlier [Price Execution Analysis](https://www.gauntlet.xyz/resources/uniswap-price-execution-analysis), with the addition of the Mock DEX allowing us to explore counterfactual liquidity conditions as a result of the protocol fee. For each competing DEX, we replicate the precise liquidity conditions at the time of each swap and the specific DEX pool logic that generates the swap output. Based on the simulated swap outputs, we can then identify which swaps would be re-routed if the real Uniswap pool were replaced with the Mock DEX.

Since we have observed in [prior work on incentive design](https://gov.uniswap.org/t/uniswap-incentive-design-analysis/21662#measuring-swapper-price-elasticity-6) that onchain order routing is very efficient, we generally assume that swaps are routed to the DEX that provides the best swap output, with a few notable exceptions that we will discuss shortly. A typical swap simulation in the model would proceed as shown in the diagram below.


The example shows a real swap in the WETH / USDT 0.3% pool that was re-routed by the swap simulator. Though the swap was optimally executed in the actual WETH / USDT 0.3% pool, the lower liquidity version in the Mock DEX would no longer be the optimal venue. The simulation routing logic then removed this swap from the post-fee volume estimate.

### MEV Volume Model

One of the core assumptions in the volume simulation is that volume exists independent of any specific venue. If price execution was better at a competing DEX, some trades would be re-routed to that competitor and volume would stay unchanged in aggregate across all DEXs. While this is a reasonable assumption for most retail and institutional traders, it does not hold for MEV volume. Because MEV traders seek to arbitrage mispriced liquidity in DEX pools, a smaller opportunity for arbitrage would simply reduce MEV volume overall. This means that MEV volume can be more accurately modeled by estimating the change in the opportunity set, so we treat it separately from the core volume simulation.

In contrast to retail and institutional (ie. core) volume, which is typically profitable for LPs, MEV volume is not a significant contributor to LP yields. In our earlier [User Cohort Analysis](https://www.gauntlet.xyz/resources/uniswap-user-cohort-analysis), we found that MEV trades make up between 40% and 80% of volume on Uniswap and other major DEXs. However, since this volume is highly selective for liquidity that can immediately be resold at a better price elsewhere, it does not generate substantial returns for LPs despite its significant share of overall volume. Traders that are highly informed about minor inefficiencies are unprofitable for LPs due to an effect known as "Order Toxicity", which is discussed in much more detail in [our incentive design research](https://gov.uniswap.org/t/uniswap-incentive-design-analysis/21662#what-is-payment-for-order-flow-9).

For the purposes of this model, it is sufficient to estimate the impact of protocol fees on the quantity of mispriced LP liquidity. If we conservatively assume MEV traders will only trade against mispriced liquidity, we can then scale MEV volume proportionately to estimate the impact of the fee. This means that MEV traders are much more sensitive to liquidity than core users in our model, which corresponds to our observations in the [User Cohort Analysis](https://www.gauntlet.xyz/resources/uniswap-user-cohort-analysis).

We can then proceed to derive a formula for mispriced liquidity based on the pricing mechanics of Uniswap V3 pools. For a given discrepancy between the pool price and market price, the maximum arbitrage size is given by the formula below, where L is the liquidity available at the current tick bucket.


The mispriced liquidity available for MEV traders is thus directly proportional to the liquidity around the current tick. As our liquidity model reduces all ticks uniformly, we reduce MEV volume by the same percentage, which is substantially higher than the core volume loss in most cases. **Notably, this does not represent a re-routing of MEV volume to competing DEXs, but a drop in MEV activity overall due to a reduction in opportunities.** As with the liquidity model, we see this as a conservative approach that seeks to avoid excessive assumptions while providing a robust estimate based on a direct economic relationship.

### Non-Optimal Swaps

For completeness, another edge case to note are swaps that were not optimally routed to begin with. Though the amount of inefficient volume on DEXs is low, there will always be some cases due to noise and unaware traders. For example, in our [Price Execution Analysis](https://www.gauntlet.xyz/resources/uniswap-price-execution-analysis), we found that about 83% of volume on Uniswap V3 was optimally routed, which leaves 17% as non-optimal trades. In the simulation, we re-route this volume in proportion to the overall reduction of core volume, since most of it likely originates from retail traders who are relatively indifferent to price execution. The full routing logic, including MEV volume and non-optimal swaps, is shown in the diagram below.


## Flywheel Effects

Though we have treated them as a one-way linkage so far, volume and liquidity are interdependent and may exhibit feedback channels that amplify the effects of falling activity. If the volume drops due to lower liquidity lead to a further reduction in LP yields, liquidity may also drop further in an adverse feedback loop. This would be the opposite of the beneficial flywheel that comes up in our [work on liquidity mining](https://www.gauntlet.xyz/resources/intro-to-amm-incentives). To account for the potential downside scenario, we include a feedback mechanism that adjusts the APY in the liquidity model based on the change in core volume.


The nature of the volume loss influences the effect on APY and thus the likelihood of a significant flywheel effect. A loss of core volume is problematic to the overall health of the DEX ecosystem since it reduces the primary source of LP profits and is likely to have further negative effects. MEV volume, on the other hand, does not provide substantial LP profits and its drop is unlikely to have an impact on LP yields. This means that lower MEV volume is not economically equivalent to a loss of core volume and is not a significant driver of negative flywheel effects. This point is crucial to the overall economic tradeoff, since it tells us that **the volume that is most sensitive to protocol fees is also the least harmful type of volume to lose.** In the flywheel model, we calculate the APY impact on LPs based on only core volume and iterate until the incremental losses are minimal.

Though core volume losses in our simulations are fairly small, they directly affect LPs and may be amplified by flywheel effects due to their role in generating LP returns. The MEV volume losses are larger, and may amount to a material share of overall volume in some cases. However, because MEV volume is mostly unprofitable for LPs, these losses should be evaluated with the context of their low economic significance to the Uniswap ecosystem.

# Direct Impact Results

To start, we present and discuss our projections for only the direct impact of the protocol fee, excluding any flywheel effects. The table shows the predicted effect of different protocol fees on all the [whitelisted pools](https://support.uniswap.org/hc/en-us/articles/20131678274957-What-are-Uniswap-Labs-fees) at the time of writing.


We note that liquidity and MEV volume both fall in line with the protocol fee, as we conservatively estimated that they are linearly related. Core volume is relatively insensitive to fees, with less than 1% lost at a 15% protocol fee. This is likely due to Uniswap's existing market share and substantial price improvement for swaps that are optimally routed to the target pools, which we described in the [price execution analysis](https://https://www.gauntlet.xyz/resources/uniswap-price-execution-analysis). Even a substantial decline in liquidity does not affect execution enough to re-route much core volume to competitors. However, due to flywheel effects being more likely for core volume, it is possible that this version of the analysis somewhat underestimates the true impact.

The table also shows estimates for annual revenue generated by each version of the protocol fee. These revenues are based on Uniswap's usage from Aug-Oct 2023 and should be seen as ballpark numbers. Actual revenues could be substantially higher or lower depending on macro factors such as overall DEX trading volumes and crypto market prices.

To better understand where the loss of core volume is occuring, we can look at these projections on a pool-by-pool basis. We would expect pools in the most competitive markets to losing more core volume, and pools that are dominant in their markets to lose very little.


The table above shows the annualized volumes for each pool over the Aug-Oct 2023 period. We split each pool into MEV and core volume and show the impact figures for both categories. Since the MEV impact scales linearly, it is largest in the pool with the most MEV volume (USDC / WETH 0.05%). More importantly, we can see that the core volume losses are quite small, and mostly concentrated in one highly competitive pool (USDC / USDT 0.01%). Since the USDC / USDT 0.01% pool has a well-known competitor (the Curve 3pool), we will now go through this case in detail to sanity check our results.

## Model Validation

It may seem surprising that the impact of protocol fees on core volumes is fairly small. To illustrate why this is the case, we go through a sample of swaps routed to the USDT / USDC 0.01% pool and the competing Curve 3pool. The chart below shows the size of each swap and the price improvement that was obtained by routing to the optimal venue vs the competitor.


As expected for stablecoin pools, the price improvement is usually fairly small, and rarely exceeds 0.01%. The pools are highly competitive and receive comparable shares of swaps across a wide range of trade sizes.

The effect of a protocol fee in this picture would be to reduce the price improvement of Uniswap. If any Uniswap trades have their price improvement fall below zero, they would switch over to Curve due to re-routing to the optimal venue. The chart below shows this impact as the white line. Any purple dots below the line would be re-routed in the 30% protocol fee scenario.


We note that for trades sizes below about $100k, the protocol fee impact is much smaller than the typical price differences between the venues. Between $100k and $1M, we note a few swaps that would be rerouted, which represent the core volume loss in the simulation. At very large trade sizes the fee impact starts increasing more quickly, as liquidity in the Uniswap pool starts to become a limiting factor.


To set a benchmark for reasonable price differences, we add two blue lines for typical arbitrage thresholds in the chart above. Any differences above the solid blue line can likely be arbitraged against stablecoin liquidity on centralized exchanges, while the area above the dashed blue line can likely be arbitraged directly onchain. As expected, we see very few swaps executing above the blue lines, since such a large price improvement would be unsustainable in an efficient market. Below the blue lines there is not enough price difference to cover the operational and transaction costs of arbitrage, so this area is where most core volume ends up occuring.

The chart also includes the green lines which show the swap size needed to reach the tick boundary of the Uniswap pool. The solid green line is the actual average value over the sample period, and the dashed green line is adjusted for the protocol fee. This marks the point where slippage starts to increase notably in the Uniswap pool due to liquidity being a limiting factor. We note that the fee impact rises quickly after this point, as we would expect from the primary mechanism through which the fee affects the ecosystem being decreased liquidity. To the left of green lines, the fee impact is significantly less than one tick (0.01%), which is also reasonable given the swaps here are occuring within a single tick bucket.

So we conclude that the low core volume losses are due to Uniswap's strong existing market position and liquidity rarely being a limiting factor to core users, even at a fairly high 30% protocol fee. Having confirmed that the first-order model is reasonable, we can now move on to the flywheel case, which is our best forecast for what would actually occur.

# Including Flywheel Effects

The flywheel version of the analysis begins where the first-order model left off. Starting from those earlier results, we iterate the model to find the eventual compounded effect of the initial loss in core volume. Since the initial effect on core volume (and thus LP profitability) is relatively small, these results do not differ radically from the earlier version. However, we do observe a larger drop in volume and liquidity and a slight erosion in expected revenue, as shown in the table below.


When interpreting these results, we again emphasize the importance of core retail and institutional volume to the health of the Uniswap ecosystem. To maintain strong liquidity and high-quality price execution on Uniswap, it is essential to retain core volume that generates competitive returns for LPs. In this context, the modest declines in core volume are encouraging and suggest that the risk of significant economic disruption from a 5-15% protocol fee in the Uniswap ecosystem would be minimal.

While we consider MEV volume to be much less economically meaningful than core volume, we recognize that headline DEX statistics do not differentiate between the two types. Thus, a significant drop in MEV volume may still be perceived as a serious economic issue by public opinion, despite causing minimal tangible harm. Though we do not attempt to quantify brand or PR impact in this report, we note that it may be an important qualitative factor for further discussion.

To get a closer look at the tangible impact, we again break down the impact by pools. We would expect the pools that saw the most core volume loss in the previous results to also see the most pronounced flywheel effects. For pools that had minimal core volume loss initially, adding the flywheel should not change anything significantly.


We see that this is indeed the case at the pool level. The USDC / USDT 0.01% pool, which showed the highest core volume loss earlier, also shows the biggest flywheel effect, though it still retains almost 96% of its original core volume. The MEV volume losses in this version are amplified by the liquidity loss due to the flywheel effect, and are thus largest for the pools with the most core volume impact.

# Tradeoff Charts

To complete the review of results, we can visualize the pros and cons of a protocol fee discussed so far in a series of charts. On the cons side, we can plot the reductions of core and MEV volume for a given level of protocol fee, as shown for the core volume below.


Notably, the core volume loss is very non-linear, with the majority occuring at very high values of the protocol fee. At the extreme, a 100% protocol fee should result in a 100% loss of all volume, since such a DEX would be unable to retain any LPs. However, even for an extremely aggressive 80% protocol fee, the simulated DEX still retains substantial core volume. This once again highlights that liquidity is rarely a limiting factor for most of Uniswap's core users. For a more realistic protocol fee setting of 5-15%, the core volume loss curves are quite flat, in both the flywheel and non-flywheel cases. This corresponds to the encouraging core volume impact we observed throughout the earlier analysis.


Switching to MEV volume, we see a very different picture. Due to the liquidity-sensitive nature of this volume and the linear relationship between the protocol fees, liquidity, and MEV volume discussed earlier, the impact is mostly linear. While the dropoff in MEV volume represents a minimal economic loss, it is still material in headline terms and needs to be considered for a full analysis of the tradeoffs.

Finally, we can look at the revenue generated by the protocol fee as the primary benefit of the tradeoff. The chart below shows the annualized revenue adjusted for the volume impact of the fee. At some level of the protocol fee, the volume loss outweigs incremental revenue and leads to diminishing returns, similar to a [Laffer curve](https://www.investopedia.com/terms/l/laffercurve.asp) seen in some tax revenue models.


The revenue curve at higher values of the protocol fee depends significantly on the strength of the flywheel effect. Once core volume losses move into the steeper part of their impact curve, the downside risk of volume and revenue erosion starts to become more problematic. However, we note that for a more realistic 5-15% protocol fee, the revenue curves are fairly linear and far from the theoretical maximum. We conclude that a protocol fee in this range would be effective at generating revenue and is unlikely to suffer from diminishing returns due to core volume loss.

# Conclusions

The goal of this analysis was to evaluate the risks and benefits of protocol fees on Uniswap and evaluate their impact across different pools. We produced quantitative forecasts for liquidity, volume, and revenue, and now address the questions posed at the start.

**1. Quantifying the potential benefits of a protocol fee**
- *How much revenue can a protocol fee generate?*

The amount of revenue generated is a function of both the protocol fee set and the amount of trading volume. While we cannot predict macro factors that may affect the volume that is subject to the protocol fee, we can provide an estimate based on existing volume, adjusted for the volume impact of the fee.

Using our predictions of trading volumes at various protocol fees we are able to predict the amount of revenue generated at various fee tiers. In these estimates we use the 3 month period from Aug-Oct 2023 as a reference for volume on Uniswap. We can consider 3 fee scenarios to illustrate the mechanics of the tradeoff, noting that these are chosen to highlight significant inflection points and are not necessarily practical options for Uniswap.

- **Conservative (10% Protocol Fee) -** A 10% protocol fee on the currently whitelisted pools would generate an expected \$7.2M of annualized revenue, result in 10.7% in TVL loss, a reduction of 0.75% ($499M annualized) in core trading volume, and a reduction of 10.7% ($5B annualized) in MEV-related trading volume.
- **Flywheel Avoidant (20% Protocol Fee) -** We find that the flywheel effect of reduced volume driving reduced liquidity only kicks in significantly at an ~20% protocol fee. Here we would expect to see about $13M in annualized revenue.
- **Profit Maximizing (60% Protocol Fee) -** Taking the effect of negative volume-liquidity flywheels into account we would expect to maximize protocol revenue at a 60% protocol fee driving an expected ~$30M in revenue.

**2. Quantifying the potential downsides of a protocol fee**
- *How much liquidity and volume would be lost?*

The reduction in LP yields due to the protocol fee leads to an outflow of liquidity, which is the primary mechanism through which a protocol fee can negatively impact the rest of the ecosystem. We expect this is a roughly linear effect with increasing fees, so our projected outcomes for the three cases are:

- **Conservative (10% Protocol Fee) -** Reduces TVL by about 10.7%, roughly uniformly across the affected pools and tick distributions
- **Flywheel Avoidant (20% Protocol Fee) -** Reduces TVL by about 22%, roughly uniformly across the affected pools and tick distributions
- **Profit Maximizing (60% Protocol Fee) -** Reduces TVL by about 65%, roughly uniformly across the affected pools and tick distributions

The impact of liquidity on trading volume depends significantly on the origin of the affected volume. Core retail and institutional users, which make up roughly half of the total volume on Uniswap, are not directly sensitive to liquidity, but rather indirectly through price execution quality. Our estimates for the impact of a protocol fee on core volume across all whitelisted pools are:

- **Conservative (10% Protocol Fee) -** Reduces core volume by about 0.75%, including minor negative flywheel effects
- **Flywheel Avoidant (20% Protocol Fee) -** Reduces core volume by about 4%, including moderate negative flywheel effects.
- **Profit Maximizing (60% Protocol Fee) -** Reduces core volume by about 10%, including substantial negative flywheel effects. The likelihood of major negative flywheel effects that prevent further incremental profits increases quickly after this point.

On the other hand, MEV traders, who roughly make up the other half of total volume, are directly sensitive to liquidity. We project that reduced liquidity on Uniswap would reduce MEV activity by a proportionate amount, due to the smaller opportunities to profit from minor price dislocations. We then expect that a protocol fee would reduce MEV volume by the same percentages as TVL.

- *Is there a threshold where fees create a major negative flywheel for users?*

The economic costs of core and MEV volume losses are also not equivalent, and relate closely to the flywheel effect. MEV volume is highly adversarial and mostly unprofitable for LPs, so a reduction is unlikely to cause a flywheel effect or any further economic issues for the ecosystem. In contrast, core volume is the main driver of LP returns, and losses in core volume will likely affect APYs and liquidity negatively. The negative flywheel effect is thus tied to losses in core volume, while MEV volume losses can be large nominal amounts but are mostly harmless in comparison.

We estimate that a 10% protocol fee across all whitelisted pools would create a relatively minor flywheel effect, increasing core volume losses from 0.5% without the flywheel effect to 0.75%. We also project that the point of diminishing returns, where the negative effects erode any incremental benefit of the protocol fee, occurs at protocol fees of about 60%.

**3. Assessing the impact of a protocol fee across different pools**
- *Are some pools disproportionately affected?*

The core volume losses are highly concentrated in a few pools with strong external competitition. We especially note the USDT / USDC 0.01% pool, which competes closely with other stablecoin pools, such as the Curve 3pool. We estimate the USDT / USDC 0.01% pool would lose 4% of its core retail and institutional volume at a 10% protocol fee, compared to a 0.75% volume loss for the average pool.

Losses in liquidity and MEV volume are more widespread, as we expect them to reflect a uniform reduction in activity proportional to the protocol fee. A caveat to this would be for pools that see a strong flywheel effect due to core volume losses, which would see their liquidity and MEV volume losses amplified somewhat relative to others.

## Overall Recommendation

The ultimate choice for whether to institute a protocol fee depends substantially on Uniswap's valuation of MEV (non-core) volume. Under most fee scenarios the volume drop on its face may be substantial relative to the revenue generated. However, the volume lost will almost entirely consist of MEV volume rather than core user volume. MEV volume only exists to profit off of market inefficiencies and does not represent a long-term scalable market for Uniswap. However, it does contribute substantially to Uniswap's headline volume numbers.

From a long-term growth perspective for both non-MEV traders and LPs on Uniswap, a high (up to 80% for some pools) protocol fee is unlikely to substantially impact non-MEV traders and is capable of generating significant revenue. However, from a short-term growth perspective, the loss of MEV volume may cause messaging challenges as it may affect headline volume numbers substantially.

It is also important to note that these results are ultimately theoretical and represent our best estimate based on available data, while making highly conservative assumptions. The way to definitively uncover the relationhip between fees, TVL, volume, and revenue will necessarily require a long rollout period, experimentation, and further analysis. As next steps, we will work on exploring different rollout plans and optimization strategies that trade off the loss of volume against revenue potential.