Thursday, March 07, 2024

LP Profitability on Long-Tail Pools

The crypto bull market is back, best exemplified by worthless meme coins PEPE and Shiba rising to market caps of $3B and $21B, respectively. These types of coins have lottery-like payoffs, making them enticing buys when crypto bros are flush with 'house money.' They could also highlight a worst-case scenario for liquidity providers (LPs), who are short volatility.

This is another example of the positive volatility/return correlation unique to crypto. In standard asset classes, higher volatility is correlated with negative returns, which is why a long volatility position has a negative beta in equity markets, and loses money on average (e.g., see VXX etf). An LP to a standard automated market maker (AMM) is short volatility, in that their best case scenario, the price stays put. LPs generally lose money when a token rises 50% in a single day. 

Here are the LP profits for 2 prominent and 4 lower-tier Uniswap v3 pools (mainchain) and the latest Total Locked Value. Pepe is the most profitable pool. However, it has only $5MM in TVL. The Chainlink-ETH pool has been unprofitable for the past year, and Matic and Shiba pools might have become unprofitable, given their performance over the past five months. As per the big price spike, Shiba lost a bundle, while Pepe has had a good month so far. On average, higher volatility does seem to hurt LPs, but not necessarily.

Average Daily USD LP Profitability 

The profitability applies to the LP collective, assuming they hedged their LP positions once a day. To see how this was calculated, take as an example the pool inflows for the PEPE pool on March 4, when Shiba rose 53% (Pepe and Shiba prices are hard to read because they have so many zeros between the decimal and the first natural number, but it's actually a smart move because I've seen several people post about how much money they would make if the price moved up to merely one penny):

 LP PnL Calculation Example

This profitability analysis looks at the total amount of Shiba and ETH going into and out of the pool. When examined daily, it is equivalent to assuming LPs hedge once a day. The nice thing about this is that you do not need the 'liquidity' number or look at LP mints and burns to calculate LP profitability. The net inflows include fees, creating an 'all-in' profitability metric for a hedged LP. On this day, the LPs took in $1.04MM worth of ETH and sent out $1.25MM worth of Pepe tokens for a net loss of $214k.   

The smaller pairs do seem profitable, unlike the major pair pools. If we try to identify the characteristics of profitable LP pools, one might think volume is key. Yet, looking at the volume data below, we see Pepe has a relatively high volume, and the LPs in the Pepe-ETH pool have done well, unlike the other popular pools. Volume rose significantly in the Link pool, but this coincides with its move to unprofitability. 

Average USD (thousands) Daily Volume

The data on trades per day is also ambiguous. Below, we see that the high-fee USDC-ETH pool trades less than the Pepe pool, and the money-losing BTC-ETH pool trades less than the Matic-ETH pool.

Average Trades per Day by Month

Profitable capital-efficient AMM pools are not exactly unicorns, but they aren't rats either (mooses?). A lot of development is aimed at better user interfaces and connections to other AMMs (composability), but these will not address the LP unprofitability problem. In the long run, we cannot assume people will want to provide liquidity if they lose money. Airdrop rewards are not a sustainable strategy. 

Development proceeds as if current AMMs are profitable, as there is no reason to restate and amplify an asset that loses money. The key to this thinking is that these LP convexity costs do not really exist. Some think it compares to an unrealizable ideal, the perfectly hedged position (see here). The standard way to present the abstruse convexity cost is to compare the LP's position with his initial position. Applying this to daily data is like assuming the LP hedges his position once a day. It is not an impossible benchmark like a frictionless medium, as obviously, the LP owned that position at inception.

Others note that most people they know don't seem to notice (see Guillaume Lambert here). For example, as in the recent case of Shiba LPs, the cost is forgone profit instead of actual losses (for LPs that don't hedge). Shiba rose 53% on March 4, generating a loss of $214k, assuming the LP hedged. Is missing an upside really a loss? 

Consider if you owned Apple stock and sold covered calls to generate revenue (option premium paid to you). If Apple's price doubles, you will not lose money; you will just miss out on capital gain. If you think that loss is merely hypothetical, remember why someone paid you for that option, so the buyer thinks it's real money; in cash-settled markets, on expiration, you have to send the option buyer real money if it expires in the money. The fact that many people don't consider this expense highlights the ignorance of most people selling gamma (Ribbon finance presents its options assuming the earnings premium is pure profit, presented as an APY like a bond yield).

Hopefully, more people will start addressing this fundamental problem. Interestingly, just this morning, several Uniswap researchers released an academic paper proposing an auction mechanism to 'reduce losses to informed order flow.' They do not state LPs are unprofitable, but it's a step in the right direction (see my take here). 

One point mentioned in that paper noted prior work on dynamic fees to reduce LP losses. Intuitively, this makes sense, generating the best of both worlds: low fees on average and high fees when prices are spiking. However, this fix is unlikely, given the losses for the 5 and 30-basis point fee Uniswap pools targeting the same pairs (see below). Indeed, the average loss on the higher fee pools is greater, highlighting the problem with assuming volume is exogenous. 

LP PnL in Basis Points per ETH Traded

No comments: