All data used in this analysis is maintained in this realtime dashboard for future reference.
- MEV bots make up the majority of trading volume on Uniswap. Approximately 55% on Ethereum Mainnet and 75% on layer 2’s.
- On Ethereum aggregators and the Uniswap front end originate a similar share of trading volume, low 20% each.
- Aggregators play a smaller role in originating trades off Ethereum (3%-7%).
- On all chains bot trade sizes are the largest across the distribution of trade sizes.
- On Ethereum Mainnet router trade sizes are significantly smaller than aggregator trades, while on L2’s it is the other way around. This confirms the lesser developed aggregator ecosystem off mainnet.
- Stablecoin pairs are far more likely to be originated by an aggregator than a router.
- All other pairs see an approximately equal share of volume going through the router and aggregators.
- There are clear patterns of how Uniswap performs by trade size in aggregator trades when differentiating by pair type. Uniswap performs better the larger the trade in USDC-WETH, while less well the larger the trade on stablecoin pairs.
- Outdated routers typically continue to see volumes after they are removed from the front end. These appear to be used by trading bots to execute their strategies. Even some router volume appears to be non-MEV bot driven.
Uniswap needs no introduction. Over 95% of Ethereum DEX traders over the past 6 months have traded on Uniswap. This post is all about getting to know their behavior a little better.
In particular the focus is on:
- Who owns the user? Uniswap or the aggregator?
- Who is the user? A human or a bot?
- How are they interacting with Uniswap? What sizes are they trading? Which tokens?
- What are the differences between different token categories?
- And more…
Different Ways of Trading on Uniswap
At a technical level there are two ways to swap tokens on Uniswap. Transactions can send a swap instruction to one of the swap routers which will optimise the trade across available pools, or the swap instruction can be sent to a pool/s directly. This can be used to identify trade sources.
Aggregators operate similarly to routers except they route trades across all DEX pools, not just Uniswap’s pools. When a trade is initiated by an aggregator the aggregator contract determines the optimal route across all DEX pools and sends swap instructions directly to the pool contracts. This means aggregators do not interact with the Uniswap router contract and send trades directly to the pools.
MEV bots also take advantage of movements in individual pool prices. They monitor the memory pool for trades routed to specific pools and run a range of strategies to profit off these flows. This means they need to be very specific when routing their trades, interacting directly with liquidity pools.
Currently MEV bots typically deploy smart contracts representing a strategy onto the blockchain while running off-chain searcher bots watching for opportunities to profit. When an opportunity is detected the off-chain bot initiates a transaction against its deployed, on-chain contract which then trades directly against one or more liquidity pools.
Trades initiated against one of the router contracts are most often initiated through a front end or other UI. There are clearly also cases where non-MEV bots execute their strategies against a router contract. More on this later.
This is how we are going to classify the different categories of trade origination on Uniswap.
What Does the Origination Breakdown Look Like on Different Chains
For a clearer view of the medium term dynamics our dashboards measure activity over a rolling 6 months.
Bots make up approximately 75% of Uniswap volume on all chains except Ethereum where aggregators and router traders make up a larger share. It is reasonable to assume that the other category is mostly made up of front end executed human traders. This assumption is analysed in more detail below.
It is noticeable that on Ethereum router traders and aggregator traders produce a similar amount of trading volume while on other chains the aggregator share is far lower. The fragmented nature of the aggregator space, particularly off Ethereum with no dominant players, contributes to users migrating to the Uniswap brand on these chains directly by using the front end.
The greater share of bots off the Ethereum main chain is likely due to cheaper transactions offering more opportunities for arbitrage. Bots are able to take advantage of smaller profit opportunities as the fixed cost of gas is significantly lower. Excluding the much smaller BNB deployment, bot activity makes up an approximately 10%-15% larger share of activity on L2’s.
When looking at trade sizes we like to measure trade sizes across the trade size distribution. Using the 25th, median and 75th percentile allows us to compare the full spectrum of trading activity across chains.
Trade sizes are also predictably lower off Ethereum. This is particularly evident at the 25th percentile of the trade size distribution. Gas almost certainly influences this behavior as small trades are just not viable.
Trade sizes on bot trades on Arbitrum, and to a lesser extent Polygon are interestingly higher than other L2’s to the point where they compete with Ethereum up the size distribution. Perhaps there is something about trading activity in these chain’s native tokens that lends them to seeing a reasonable amount of larger trade sizes. Our initial thoughts is that their more diverse DeFi ecosystems lead to larger trade sizes. The pool type breakdown between Arbitrum and Polygon vs Optimism does not indicate conclusively that this is driving the difference in trade sizes.
On Ethereum, aggregator traders trade larger sizes on average than router traders using the Uniswap routers. Interestingly this is not the case on other chains. Both router and aggregator originated trades attract a similar breakdown of trading pair types so differences in trade composition is not driving this behavior. We put this down to the less developed and more siloed aggregator landscape off Ethereum.
The breakdown of trade origination by pool type also shows some interesting nuance.
The USDC-WETH pair, which sees enough volume for its own category, has a far higher share of bot volume than the other pool types. This is due to a combination of most trading going through the relatively low fee 5bp pool and relatively high volatility creating many opportunities.
Stablecoin pools see a much higher (34.4% vs 23.1%) percentage of volume from aggregators than through the router. Stablecoin traders clearly shop around.
This is in contrast to the other pool types where aggregator and router trades see a similar share.
Predictably stablecoin pools also have a little less relative bot activity. There are just not as many opportunities for arbitrage or sandwiching despite the typically lower fees.
Volatile pair traders lean a bit more towards trading through a router. This is a combination of retail speculators and what looks like automated strategies (see below).
Activity by Router Contract
Uniswap has updated its router contracts in each new version of the protocol. There were also additional upgrades. First in late 2021 with the release of the auto-router, capable of routing across v2 and v3 pools as well as splitting orders. Then in late 2022 the Universal Router was released, routing across NFT and ERC20 tokens and improving the approval process. The Universal router has seen two upgrades since.
Despite each upgrade being implemented as default in the Uniswap front end, older router versions continue to see volumes. Closer examination shows these appear to be trading bots executing non-MEV trading strategies. These strategies are mostly focused on meme-coin pairs with a little WETH-USDx activity as well.
This is confirmed by the resurgence of activity on the v2 router in 2023 corresponding to the resurgence in meme-coin activity.
The share of trading activity by router provides another datapoint to measure Uniswap front end activity. The speed
Data from the past three months shows a concentration of meme-coins at the top of the old router volumes when sorted by total volume over the period. The Uniswap liquidity in these tokens is clearly a drawcard for automated trading strategies. The gas savings of using the v2 router could be the reason for sending transactions there rather than the Universal router or the auto-router.
Looking at the top accounts calling older routers the distribution is fairly diverse. The largest traders contribute 1.5% of total volume, trading the USDT-WETH pair over medium time frames, holding positions for an average of around 3 hours.
Most of the other large traders focus on individual or multiple meme-coins. It is possible that many of these are controlled by a single entity executing the same strategy across different pairs.
Looking at the distribution of trade sizes against non-default front end routers it is interesting to notice the smaller sizes across the distribution vs the router average above.
The most likely conclusion to draw is the obvious one that aggregator human traders are on average more sophisticated and trading with more capital. The smaller sizes of bot trades through older routers muddies this analysis a little, but the concentration of meme-coin strategies amongst this category could be to blame.
Bot activity on Uniswap ranges from around 55% on Ethereum mainnet to 75% on Polygon. This clear majority of trading volume originating from inorganic sources begs the question why are other DEX’s not seeing the same gross volume of bot flow as Uniswap, they typically see a similar share of bot volume.
Our conclusion is that the power of the much larger volume of Uniswap router flows through the front end differentiates Uniswap. It seems that bot activity is inevitable in any competitive DEX for now. Uniswap sees more organic volume in all pool types and therefore generates more bot opportunities.
Uniswap still manages to win a lot of aggregator volume. The combination of low fees, gas efficiency and the v3 concentrated bonding curve creates a marketplace that has bootstrapped itself.
All data used in this analysis is maintained in this realtime dashboard for future reference.