Evidence

Why a forward-looking measure

Impermanent loss is easy to compute after the fact and hard to anticipate. The question is whether the option market knows something the past does not. Over three and a half years of hourly data, it does.

The problem

Yesterday's loss is a poor guide to tomorrow's

The obvious way to judge a liquidity position is to measure what impermanent loss has been recently and assume the next period looks similar. That is what a realized measure does, and it is what most dashboards show.

The trouble is that impermanent loss is driven by volatility, and volatility is not persistent enough for this to work well. It arrives in bursts, decays quickly, and the periods that matter most to a liquidity provider are precisely the ones the recent past failed to signal.

Option markets face the same problem and solve it by quoting forward. IILX reads that quote and converts it into the quantity a liquidity provider actually cares about. The rest of this page tests whether that conversion carries information.

Result one

Implied explains about twice as much as realized

For each day we compare two forecasts of the impermanent loss that materialized over the following seven days: the option-implied value IILX, and the realized value RILX. Higher R² means more of the subsequent variation is explained.

Seven-day horizon Implied (IILX) Realized (RILX) Both together
BTC, full range0.2740.1410.276
BTC, ±50% range0.2720.1420.274
ETH, full range0.2200.1070.220
ETH, ±50% range0.2140.1050.214

The third column is the more telling one. Adding the realized measure alongside the implied one moves R² from 0.274 to 0.276, and its coefficient is statistically indistinguishable from zero. Once IILX is known, the backward-looking estimate adds nothing. The reverse is not true: adding IILX to a realized measure nearly doubles what it explains.

Hourly data, January 2023 to July 2026, roughly 1,300 overlapping windows per series. Newey–West standard errors.

Result two

Fee income follows implied impermanent loss

Sorting every day in the sample into five buckets by that day's IILX, and then measuring the fee income Uniswap V3 pools earned over the following week, gives a monotone relationship. Median annualized fee APR, ±50% range:

BTC pools
Lowest IILX
6.2%
2
8.4%
3
12.3%
4
12.9%
Highest IILX
18.5%
ETH pools
Lowest IILX
7.3%
2
13.9%
3
17.6%
4
23.1%
Highest IILX
25.1%

The gap between the highest and lowest bucket is +15.6 percentage points for BTC and +23.3 for ETH, both comfortably significant. Sorting on the realized measure instead produces the same shape but a smaller spread — +10.3 and +20.4 points. Here too the forward-looking measure is the sharper instrument.

Result three

And the two cancel out

High IILX brings more fee income. It also brings more impermanent loss. Sort the same days by IILX and look at the net of the two, and the pattern disappears entirely.

That is what a competitive market should produce, and it is the reason the index exists. A pool advertising a high APR is not offering free money; it is quoting the price of a risk that arrives at the same time. Whether the fee actually compensates for that risk is not visible in the APR, and it is not visible in what impermanent loss happened to be last week.

Across the sample, implied impermanent loss exceeded what subsequently materialized in roughly three quarters of seven-day windows. Liquidity provision is compensated for bearing this risk — IILX is the measure of what that compensation is worth.

See the live index How it is constructed

Caveats

What this does not show