Impermanent loss, in plain arithmetic, and when fees cover it

A constant-product market maker holds two assets and prices trades so that the product of its reserves stays fixed. When the market price of one asset rises, arbitrageurs buy it from the pool until the pool’s implied price matches the market, leaving the pool with less of the asset that rose and more of the one that fell. The provider who supplied the reserves has, in effect, sold the winner into its rally. The difference between the value of their position and the value of simply holding the original two assets is impermanent loss.

The word impermanent is misleading. The loss is unrealized while the position is open and reverses only if price returns to where it started. On withdrawal at any other price it is permanent.

The shape of the loss

For a full-range constant-product pool the loss depends only on how far the price ratio has moved from its starting point, not on which direction. A widely cited derivation shows that the position’s value relative to holding equals twice the square root of the price ratio divided by one plus the ratio. At a ratio of one there is no loss. Move the ratio a modest distance and the loss is small; move it several-fold in either direction and the loss becomes a meaningful share of the position. The curve is symmetric in the log of price and it is always at or below zero: a provider is never better off than a holder before fees.

Concentrated positions steepen the curve. Because the same capital is deployed over a narrower band, the inventory converts faster, and a position that has left its range is entirely in one asset. The researchers who studied Uniswap v3 pools in the protocol’s first months, Loesch, Hindman, Richardson and Welch, found that across the pools they analyzed, impermanent loss exceeded the fees earned, so that providers in aggregate would have been better off holding. Fees were substantial. The divergence was larger.

Why it is really a short volatility position

A liquidity position pays out most when price stays where it was and loses most when it moves far. That is the payoff of being short an option, and the fee income is the premium collected for writing it. The loss-versus-rebalancing framework of Milionis, Moallemi, Roughgarden and Zhang makes the point formally: the provider’s shortfall against a rebalancing portfolio is what arbitrageurs extract by trading against the pool’s stale prices, and it grows with the volatility of the pair. Fees have to be earned faster than volatility takes them away.

When fees cover it

Whether a position pays comes down to three quantities over the holding period: the fee income the position’s share of volume generates, the divergence loss the pair’s realized volatility produces, and the costs of entering, exiting and rebalancing. The first is high in pools with heavy, persistent volume and little competing liquidity. The second is low in pairs that move together. The best positions in the historical record are therefore the unglamorous ones: an asset against its staked version, two stablecoins, two wrapped forms of the same asset. Volatile pairs can pay in the right regime and can lose the position’s value in the wrong one.

• Model the position with fees against expected divergence at the pair’s realized volatility before entering
• Exclude incentive emissions from the base case; they are paid in an asset whose price the emission schedule pushes down
• Prefer correlated pairs for the core of a farming book; hold volatile pairs small and only where the fee tier pays for the divergence
• Track fees earned against divergence accrued continuously, and exit positions that are not covering it rather than waiting for price to return

What the newer designs change

Volatility-linked fees, available through Uniswap v4 hooks and native to designs such as Meteora’s dynamic liquidity market maker, raise the fee rate in exactly the conditions that produce divergence loss. That improves the odds without changing the structure. In a large one-directional move no fee schedule keeps pace, and a position that leaves its range stops earning entirely. Impermanent loss is not a flaw to be engineered away. It is the price of the fee income, and the discipline is in paying it knowingly.