Uniswap Liquidity Mining Analysis
By Shishi Feng and Vishesh Choudhry. All graphs mentioned in the article can be found here.
Introduction
Although the concept of liquidity mining was only introduced last summer, it soon became the de facto way for projects to create economic incentives for their stakeholders. Liquidity mining’s main value proposition is to help projects bootstrap liquidity through token incentivization. Users deposit liquidity in exchange for the protocol token with the expectation that it will accrue in value in the future. The more subtle value proposition of liquidity mining is that it is also a campaign to increase project awareness which in turn should attract more users and developers.
However, the juxtaposition of a limited period of incentives distribution and a liquid project token in most liquidity mining programs seems to leave limited room for user stickiness and long-term interest alignment. This article aims to assess the cost-effectiveness of Uniswap’s liquidity mining program in terms of project liquidity, user acquisition, and user behavior before, during, and after the program.
Token Distribution
In the initial genesis of 1 billion UNI, 150 million UNI was allocated as rewards for Uniswap liquidity providers. 400 Uni was set aside for each address that interacted with the Uniswap contract. The remaining UNI was set aside for liquidity providers. Of that 150 million, about 17 million is still unclaimed. Of the ~133 million that was claimed, only about 6 million is still held by the original recipients. Roughly 84% of the claimed UNI has been transferred out from the recipient addresses. This does not necessarily mean it was sold as it may have just been transferred to another user-owned wallet.
Here we can see how the % of rewards held by the original recipient addresses fluctuates over time. There was an asymptotic transfer of UNI tokens out of the recipient addresses immediately as rewards were released. Some of this activity may have been due to consolidation and wallet management. UNI prices in this same timeframe initially spiked and then began to fall steadily. It is therefore likely that a large portion of this movement was due to selling activity.
On the address level, around half of the addresses who claimed the airdrop held after 1 year. From the above graphs, we can infer that a large amount of UNI was airdropped to whale users of the protocol. Soon after the airdrop, the whale users sold their UNIs, resulting in a decrease in % of UNI held and price while those who got airdropped small amounts of UNI tend to hold.
Protocol Liquidity & Yield
The four pools that were incentivized at the time of the Uniswap v2 liquidity mining program were ETH-DAI, ETH-USDC, ETH-USDT, and ETH-WBTC. Here we can see the TVL of those four pools over time. It is evident to see that TVL in each of these four pools jumped while the program was active (September 18th to November 17th) and fell sharply as soon as the rewards ended.
The proportion of total TVL in the incentivized pools grew during the incentive time period of course. Note that after the incentives had ended, all pools (both incentivized and non-incentivized) sustained higher levels of liquidity than prior to the rewards program. This suggests that the incentive program increased LP awareness and led to liquidity growth in the long run. Further, after the incentives had ended, the incentivized pools sustained higher levels of liquidity than the non-incentivized pools (eventually this evened out). Though the incentive program did not necessarily create sticky near-term liquidity, this data implies that it may have a broader impact on how liquidity providers use the protocol.
Here we can see the cost in rewards per million dollars of liquidity provided. This cost decreased as TVL started to flow into the incentivized pools. The nominal amount of UNI rewards issued per day was constant and the UNI price slightly decreased. The daily cost of liquidity decreased from $1500/million to $500/million about three weeks into the program and then leveled off until the last day. This shows that it pays to be early as the first few LPs to mobilize captured an extra $1000/million per day. Some of the LPs may have initially held off due to lack of comfort with impermanent loss risk but eventually joined the pool upon seeing the yield opportunity.
The above graph shows the TVL in the incentivized pools versus cumulative rewards issued in USD terms. As we can see after the initial inflow, liquidity growth plateaued near the end of the program while the total cost of the program continued linearly. Though we saw that the daily cost of liquidity leveled off, the total cost of UNI rewards for the protocol outpaced liquidity growth in the incentivized pools during the reward period.
Here we can see rewards distributed daily and daily swap fees. By and large, UNI liquidity mining rewards eclipsed trading fees. By the end of the rewards period, this margin narrowed slightly as UNI prices fell.
This graph shows the same rewards ratio on a broader time scale. Though revenues during the reward period were obviously elevated, revenues from trading fees resumed their initial growth rate after the end of the program. Initially, TVL started to track LP revenues, but starting at the end of this program, the ratio of TVL to LP revenue also grew.
This can be interpreted as a lasting impact of the liquidity mining program. Dollars of liquidity post-program per dollar of LP revenue increased. In other words, as LP awareness improved there was reduced opportunity for individual LPs; this kind of discovery is a relatively expected impact of such programs as well as a general product and brand growth for Uniswap.
User Behavior (LPs)
The average size of LP positions in the four incentivized pools dramatically increased during the liquidity mining program, indicating that whales moved into these pools to farm the outsized yield. This trend (particularly with the ETH-WBTC pool) tracked with the growth of TVL in late October. There was a bias for these whales to target ETH-USDC and ETH-WBTC pools in particular, perhaps because they were already holding significant quantities of these assets to farm with.
Here we can see how the median amount of liquidity provided by each LP changed and how that compares to movements in ETH price. This shows that although the average position size increased during liquidity mining, the median position size fell. One potential explanation is that when whales moved into the pool to earn incentives, they pushed out some casual mid-sized LPs by reducing their relative yield. Another potential explanation is that a large quantity of smaller addresses moved into the pool as a result.
Here we can see the number of LPs in each bucket of position size over time. During the event, there was a significant uptick in LP positions of all sizes. This number peaked at the tail end of the program and then dipped back down after the rewards program ended. That said, the total number of LPs was significantly higher after the end of the rewards program than before it started. The number of LPs providing $100-$1000 in liquidity comprised the bulk of this increased activity. The number of LPs in larger buckets simply rose back to similar levels as during the program. This further indicates that liquidity mining was fairly successful at raising LP awareness.
Of the LPs who provided liquidity during the incentive period (09/18–11/17), only 34% never provided liquidity again (on those addresses) after 11/17/2020. This implies that LP growth was fairly sticky and the majority of LPs either remained in the incentivized pools or returned at some point after the rewards ended. Although the 34% removed a significant amount of liquidity as they exited, a great number of LPs joined after the rewards program. Broadly, a number of whales exited when the rewards ended and the number of smaller LPs increased commensurately.
This graph shows the amount of liquidity in each bucket by LP size. During the reward period, there was a significant uptick in the amount of liquidity in positions over $10M. After these larger LPs (whales) pulled out a chunk of their capital in December 2020, the mid-sized LPs ($1–10M) picked up some of the revenue opportunity from the decreased liquidity. It is also reasonable to assume that this was in part due to price appreciation of previously smaller positions in the pools.
When we combine all of the above data, a narrative emerges. With the announcement and branding of the liquidity mining incentives, a large number of small LPs entered the pools. Hence, we see a decrease in median LP position across all four incentivized pools. Meanwhile, a few LPs ($10M+) deployed large positions into the pools, increasing the average position size. A large share of incentives provided during the reward period went to these larger LPs. Once these incentives ended, these whales pulled out and more mid-sized LPs began providing liquidity, both due to increased relative opportunity and probably due to increased awareness of the revenue opportunity from providing liquidity.
The number of smaller LPs broadly remained elevated even though total liquidity fell. In other words, though whales exited after rewards ended, there was not much churn of smaller LPs. This strengthens the case that incentives successfully drive persistent, long-term behavior changes.
The above chart shows the 7-day average number of distinct addresses that made LP action calls as well as the 7-day average number of LP action calls. Notably, the commencement of the rewards period catalyzed a spike in LP activity, after which activity sustained overall higher but decreasing levels for 9 months as it tapered off, ultimately converging on activity levels slightly higher than the period right before rewards.
Combined with the growth in the absolute number of LPs throughout the rewards period and afterward, this suggests that LM had a relatively sustained positive impact on LP engagement. The gradual slowdown in activity after the rewards period suggests that the LP base trended towards more passive LP strategies. This may represent a period of experimentation and eventual discovery of more optimal LP strategies. Also, there may have been more churn of old addresses and cycling in of new addresses in late 2020.
User Behavior (Trading)
Daily trade volume in incentivized pools was muted and stayed relatively the same before, during, and after the rewards period. This indicates that the rewards program didn’t have any immediate impact on volume. The trade volume then started to increase approaching the end of 2020 in both pools and remained high until the end of May 2021.
This uptick in volume is likely a combination of increased recognition of Defi and DEXes in general, and increased trading of extremely hyped tokens at the time(i.e. SHIB). Since there are many factors at play during this time period, it is difficult to conclude that the rewards program had any discernable impact on volume long-term.
The number of traders in incentivized pools was elevated upon the beginning of the reward period, and notably, remained elevated after. This further supports the case for the LM program having had a lasting impact on trading behavior that is not specific to the targeted pairs.
We can speculate that the increase in liquidity in the incentivized pools caused new traders to begin using Uniswap over other venues. When the incentives were removed, those traders remained, as they had now already crossed the biggest barrier to use (onboarding). Regardless of the reason, we can at minimum conclude that liquidity mining resulted in an increase in the number of traders using Uniswap’s incentivized pools.
Similar to the number of traders, the number of trades in all pools saw an increase at the start of the reward period and remained elevated even after the liquidity mining program ended. This trend holds for incentivized and non-incentivized pools alike. We can infer that the liquidity mining program, to a certain degree, is successful at attracting and retaining broad users of the protocol. Presumably, the improved liquidity initially helped provide an attractive trading experience (e.g. lower slippage).
Conclusions
- For airdrops, less is more. Power users that received large airdrops were far more likely to sell, whereas smaller users were more likely to hold. This is fairly intuitive as a lot of the power users were likely businesses who might see it as irresponsible to hold an outsized amount of tokens on their balance sheets longer-term, whereas smaller users tended to be individuals.
- Liquidity isn’t sticky, users are. The majority of the uptick in liquidity during the rewards period was whale-driven, evidenced by the steep uptick and drop-off in TVL around the rewards period. Whales tend to be trading firms who are generally short-term oriented due to the nature of their businesses. The TVL did not continue to grow as more rewards were paid out, so you could argue that much of the value was leaked to actors that structurally need to sell. However, as with the retroactive airdrop, smaller users were much more likely to hold.
- In both of the above cases, potential remedies to value leakage to mercenaries might include introducing vesting to align power users on a longer horizon or more granular targeting of individual users.
- Increased project awareness post-incentive. The rewards program seemed to improve trader and LP awareness of the protocol significantly. Even though TVL from the incentive period was not sticky, general usage and the number of addresses providing liquidity remained elevated afterward. After the end of the rewards program, the ROI for LPs generally speaking was lower than before, likely due to increased awareness of yield potential on Uniswap and a higher saturation of the opportunity.
- Change in type of usage. The ratio of TVL in incentivized pools to TVL in non-incentivized pools has remained elevated since the incentive program ended. This supports the notion that behaviors learned as a result of incentives are sticky — even if the majority of liquidity that is attracted isn’t — and suggests that using token incentives, even if applied temporarily, are effective at curating user behavior on a protocol.
Huge thank you to Aleks Larsen for extensive feedback on this article.
