Loss-Versus-Rebalancing: The LP Cost Nobody Prices
A preprint posted on October 5, 2026 measured realized loss-versus-rebalancing across 62 Uniswap v3 pools on Ethereum, Arbitrum and Base over the 257 days from January 1 to September 14, and found that sorting pools by headline fee APR produced worse forward returns than not sorting at all. It has not been peer reviewed. The direction matters more than the precision, because fee APR is the number on nearly every liquidity dashboard and fund fact sheet in this market.
The Benchmark Is the Whole Argument
Loss-versus-rebalancing is defined against a specific alternative. Take the pool's holdings, then run a self-financing strategy that continuously trades at the centralized exchange price to hold the same risky-asset position the pool holds. LVR is that strategy's value minus the pool's.
The benchmark choice is not cosmetic. The rebalancing strategy makes the same trades as the pool at better prices, so the gap between them carries no market exposure. It is the only candidate benchmark whose loss process has zero quadratic variation, which is the formal way of saying it isolates the cost of quoting stale prices from the cost of owning the asset.
Impermanent loss, benchmarked against holdings fixed at a start date, fails that test: it is not additive across intervals, its sign can depend on the start point chosen, and it mixes market risk into a figure presented as a cost of market making. Re-benchmark at each short interval and sum, and it becomes LVR.
One property does most of the practical work: LVR is non-decreasing. It accumulates and never reverses. A pool can show zero impermanent loss because the price returned to where it started, while LVR accrued every minute of the round trip.
Fees Scale With Volume, the Loss Scales With Variance
For a constant product market maker the instantaneous LVR rate has a closed form, and as a fraction of mark-to-market pool value it is variance over eight, per unit of time. At 5 percent daily volatility that is roughly 3.1 basis points of pool value per day.
Note what is absent. Volume does not appear. Fee income scales with what trades through the pool; the loss scales with how much the price moves. A passive provider is long volume and short variance, holding two exposures only loosely related and frequently anticorrelated in the painful direction.
That is also why no protocol upgrade fixes this. Volatility is exogenous to the venue, so a mechanism can change who absorbs the cost or what is charged for it, but nothing in the design space reduces the variance of the underlying.
What the 2026 Data Shows
Across those 62 pools the preprint reports $119.8 million of LP fees against $153.9 million of realized LVR on $154 billion of volume, netting to negative $35.6 million after gas. Negative weeks ran 75.9 percent on Ethereum, 83.3 percent on Arbitrum and 84.4 percent on Base.
The ranking result is the one worth reading twice. Sorting weekly on fee APR and holding a week gave a top tercile at negative 9.39 percent against a bottom tercile at negative 2.92. Sorting on realized LVR-net APR flipped the spread from negative 6.46 percentage points to positive 9.47. The rank correlation between fee APR and next-week net return was negative 0.173, correctly signed in 4 of 18 test weeks, while the LVR-net measure ran positive 0.580 and was correct in all 18.
Honesty requires the counterexample. The original study measured ETH-USDC on Uniswap v2 from August 2021 to July 2022 and found fees minus LVR averaging about 8.16 percent annualized once hedged. The sign is regime-dependent. What is not regime-dependent is that the cost exists, accrues continuously, and is invisible in the metric most people use.
Defenses That Work and Defenses That Do Not
Waiting is not a defense. The loss is non-decreasing, so there is no horizon over which it averages out, and a position held longer accrues more of it.
Raising a static fee tier is not reliably one either, since it reduces the volume generating the offsetting income without changing the variance driving the loss. Concentrating a range raises fee capture and adverse selection together. And hedging the pool's delta, which every serious provider should do, removes market risk rather than adverse selection: in the original data it cut annualized return volatility from about 42 percent to under half a percent while leaving the LVR term intact.
What works falls into two families. Charge for the staleness, through fees that rise with realized volatility so the price of the option the pool writes tracks its value. Or sell the staleness, by auctioning the right to trade first against the pool so the arbitrage profit returns to providers rather than leaving with the arbitrageur.
Three things are worth monitoring weekly, all computable: realized LVR in basis points of volume, the ratio of fees to realized LVR, and the share of weeks that ratio fell below one.
What Scaling Requires, and What Has to Be Disclosed
Because the cost sits on a parameter the venue cannot touch, growth cannot come from reducing it. It has to come from reallocating it, and every available reallocation either moves the quoting decision to a party that can reprice or auctions it to a party that profits from it.
That changes what the product is. A provider in a dynamic-fee or auction-managed pool is not earning a share of trading fees for passive inventory. The economics converge on financing a market maker for a share of what that operator earns, which is credit and performance exposure rather than a fee stream.
For anyone drafting or reviewing disclosure, that distinction is the work. A fee APR quoted without a stated benchmark is not a return figure, and the 2026 evidence suggests it is negatively informative about the next period. Three items belong in any material putting a yield number in front of an investor: the benchmark return is measured against, whether delta is hedged and how often, and an LVR estimate with its volatility assumption stated.
The residual risk is permanent for genuinely passive provision, because variance over eight is a floor rather than a bug to be patched. The constructive note is that this is measurable rather than theoretical. The closed-form rate has existed since 2022, realized LVR is now computed pool by pool against reference prices, and mechanisms that price or auction the cost are deployed rather than hypothetical. A cost that can be quantified weekly can be disclosed, and one that can be disclosed can be priced.
For informational purposes only. Not an offer to buy or sell any security. Available only to accredited investors who meet regulatory requirements.
