What a liquidity pool actually holds
An automated market maker (AMM) pool is a shared pot of two assets — say ETH and a stablecoin — that traders swap against directly, with no order book and no counterparty on the other side of the trade. When you "provide liquidity," you deposit both assets at whatever ratio the pool currently holds, and in return you get a share of the pool represented by LP tokens. Your reward is a cut of the trading fees every swap generates.
The mechanism that keeps the pool priced correctly is simple math, usually some version of x × y = k: the product of the two asset quantities stays constant, so as one asset is bought out of the pool, its price rises along a curve. That curve is also the source of the problem.
Why the pool's mix changes without you doing anything
Suppose you deposit into a 50/50 pool of ETH and USDC when ETH is $2,000. You put in 1 ETH and 2,000 USDC — $4,000 total. Now ETH rallies to $3,000 on outside exchanges. The pool's price hasn't moved yet, so it's temporarily cheap to buy ETH out of it. Arbitrageurs notice instantly and buy ETH from the pool until its internal price matches the market price of $3,000.
That arbitrage trade is what rebalances the pool — and it does so at your expense as a depositor. Every ETH the arbitrageur buys comes out of your share of the pool, replaced by USDC at the old, lower price. By the time the pool's price catches up to $3,000, your share of the pool no longer contains 1 ETH and 2,000 USDC. It contains less ETH and more USDC than before, because you sold ETH on the way up without deciding to.
This is impermanent loss: the difference in value between what you'd have if you'd simply held the two assets separately, and what you actually end up with by having deposited them into the pool.
The math, without calculus
The size of the loss depends only on how much the price ratio between the two assets has moved, not on the direction. Whether ETH doubles against USDC or ETH halves against USDC, the percentage loss versus holding is identical — the pool punishes divergence in either direction.
| Price ratio change |
Value if held |
Value in pool |
Impermanent loss |
| 1.25x |
$4,500.00 |
$4,472.14 |
0.6% |
| 1.50x |
$5,000.00 |
$4,898.98 |
2.0% |
| 2.00x |
$6,000.00 |
$5,656.85 |
5.7% |
| 4.00x |
$10,000.00 |
$8,944.27 |
10.6% |
The pattern to notice: loss is small for modest divergence and grows faster than the divergence itself. A 25% price move costs you well under 1%. A 4x move costs over 10%. This is why IL is often dismissed as trivial by people who've only watched sideways pairs, and then discovered the hard way during a real trend.
Why "impermanent" is doing a lot of work in that name
The loss is called impermanent because it's only a paper loss while your funds remain in the pool. If ETH's price relative to USDC drifts back down to exactly where it was when you deposited, the pool's holdings rebalance back to your original ratio and the loss disappears — you'd have earned fee income for free, with no lasting cost.
The catch is in the word "exactly." Asset prices are not mean-reverting on a schedule that cares about your entry point. A price ratio wandering in a range might genuinely cycle back through your entry level multiple times, in which case IL really does behave as impermanent for anyone who stays in and out at the right moments. But a price ratio in a sustained trend — an altcoin grinding higher over a year, or a token declining toward zero — has no mechanical reason to revisit your entry ratio ever again. At that point the loss is impermanent only in a technical sense. Practically, once you withdraw — or once enough time passes that reversion becomes unlikely — it's just a loss.
Why the pair you choose matters more than anything else
IL is a function of correlation between the two assets, and nothing else about the protocol matters as much as this one variable.
- Two stablecoins, or a stablecoin pool of pegged assets designed to trade near 1:1: the price ratio barely moves, so IL is close to zero almost all the time. The main risk here isn't IL, it's one side of the pair losing its peg entirely.
- Two wrapped or staked versions of the same underlying asset (a liquid staking token and its base asset, for example): these trade in a tight band because arbitrage between them is easy, so IL stays small under normal conditions.
- An altcoin paired with a stablecoin: the altcoin can move 30% against the dollar in a week with nothing unusual happening. This is the pairing where IL becomes a serious, recurring cost rather than a rounding error.
- Two uncorrelated altcoins: the worst case, since either leg can move independently and the ratio between them can swing hard in both directions within the same month.
The general rule: the more correlated the two assets, the smaller the ratio moves, and the smaller IL gets, in a nonlinear way — cutting expected divergence in half cuts expected IL by much more than half.
Fees are supposed to pay for this — and often don't
Trading fees exist specifically to compensate liquidity providers for taking on IL risk. Every swap that hits the pool pays a small fee, typically a fraction of a percent, split among LPs proportional to their share. In a pool that trades sideways with high volume, fee income can comfortably outrun IL, and providing liquidity is a genuinely good trade — you're earning yield for absorbing volatility that ultimately nets out to nothing.
The problem is the two forces operate on different clocks. Fee income accrues slowly and steadily, proportional to trading volume over time. IL can happen almost instantly, in a single sharp move, proportional to how far price traveled regardless of how much volume went with it. A pool can go months collecting modest, satisfying fee income and then give back the equivalent of a year of fees in a single volatile week when one asset breaks out. LPs who eyeball their historical fee APR without stress-testing it against a plausible price move are pricing the yield correctly and the risk not at all.
There's also a selection problem: pools offering the highest fee yields are frequently the ones pairing volatile, low-liquidity tokens — exactly the pairs where IL is largest and fastest. The advertised APR and the underlying risk tend to move together, not in opposition.
A separate risk category from hacks and rug pulls
It's worth being precise about what impermanent loss is not. It isn't a smart contract exploit, it isn't a rug pull where a team drains the pool, and it isn't an exchange getting hacked. Those are custody and code risks — the protocol behaving in a way it wasn't supposed to. IL happens even when the protocol works exactly as designed, with an audited contract and no malicious actor anywhere in the chain. It's a structural, quantifiable cost of the AMM mechanism itself, present in every price divergence, every time, on every pool. Treating it as a tail risk alongside hacks understates how routine it actually is; treating it as equivalent to a hack overstates how dangerous any single instance tends to be.
When providing liquidity is actually a reasonable trade
The decision comes down to comparing two numbers you can estimate in advance: expected fee yield over your holding period, and expected IL given how correlated (or not) the pair is likely to be over that same period.
Liquidity provision tends to make sense for stable or tightly correlated pairs with real trading volume, where fee income is close to free money because the IL side of the ledger rarely moves. It also makes sense for volatile pairs when you have a genuine, specific view that the price ratio will stay range-bound — not just a hope that it will, but a reason to expect it, over the exact window you'd be providing liquidity.
It stops making sense the moment the advertised yield is the only reason you're looking at the pool. A 40% APR on a new altcoin/stablecoin pair is a number quoted for volume conditions on the day it was measured, not a forecast, and the pair that generates that kind of fee income is almost always volatile enough to erase it in a single directional week. Before depositing, run the pair through a version of the table above using a price move you'd consider plausible over your intended holding period, not just possible in general — if the fee yield you'd realistically collect over that same window doesn't clear the IL at that move size, the pool is a directional bet on the pair staying calm, wearing a yield-farming label.