Impermanent loss is explained badly more often than any other concept in DeFi, usually because the explanation stops before the arithmetic. So here is one position, followed all the way through.
The setup
You provide liquidity to an ETH/USDC pool on a standard automated market maker. ETH is $2,000. You deposit 1 ETH and 2,000 USDC — a $4,000 position, half in each asset, as these pools require.
The pool keeps the product of its two reserves constant. Every trade against it moves the balances along that curve, which means the pool is permanently selling whichever asset is rising and buying whichever is falling. You are not choosing those trades. You are the counterparty to all of them.
ETH doubles to $4,000
Traders buy ETH from the pool until its price matches the wider market. When the dust settles your share of the pool holds:
- 0.707 ETH — worth $2,828
- 2,828 USDC
- Total: $5,657
Had you simply held the original 1 ETH and 2,000 USDC, you would have $6,000. The gap is $343, or 5.7%.
Notice what happened: you still made money. You made $1,657 instead of $2,000. Impermanent loss is not a loss of capital — it is underperformance against holding, and it is the fee the pool charges you for automatically selling into strength.
The same position if ETH halves
ETH falls to $1,000. Now the pool holds 1.414 ETH ($1,414) and 1,414 USDC, totalling $2,828. Holding would have left you with $3,000. The gap is again 5.7%.
That symmetry is the point: the loss depends on the ratio of the price change, not its direction. Halving and doubling cost exactly the same.
The full curve
Divergence between the two assets, and what it costs against holding:
| Price change | Loss vs holding |
|---|---|
| 1.25× (or −20%) | 0.6% |
| 1.5× | 2.0% |
| 2× (or −50%) | 5.7% |
| 3× | 13.4% |
| 4× (or −75%) | 20.0% |
| 5× | 25.5% |
| 10× | 42.5% |
The curve is gentle for small moves and brutal for large ones. This is why stablecoin pairs, which barely diverge, carry almost none of it, and why pairing a volatile token against a stablecoin carries the most.
Do the fees cover it?
This is the only question that decides whether providing liquidity was worth doing, and it is answerable.
Take the 2× example, where you need to earn more than $343 to break even. Suppose the pool holds $400,000, so your $4,000 is a 1% share, and it trades $200,000 a day at a 0.3% fee.
- Daily fees to the pool: $200,000 × 0.3% = $600
- Your 1% share: $6 a day
- To cover $343: about 57 days
So on those assumptions, if ETH doubles over two months you roughly break even against having held; if it doubles in a fortnight you are meaningfully behind. Both outcomes get advertised as “12% APY” and both are true statements about the fee income alone.
Two things this illustration simplifies: your share dilutes as others deposit, and volume itself changes with volatility — usually rising exactly when divergence is largest, which partially offsets the loss. Run the same arithmetic with the actual pool’s numbers rather than these.
What actually reduces it
- Correlated pairs. Two stablecoins, or an asset and its liquid-staking version, barely diverge. Lower fees, but far lower loss.
- Fee tier against volume. A higher fee tier only helps if the volume still comes.
- Time in position. Fees accrue continuously; the divergence cost is set by the ratio at the moment you withdraw. Longer horizons favour the fees — unless the ratio keeps moving.
- Concentrated liquidity, carefully. Narrowing your range multiplies fee income per dollar and multiplies the divergence effect inside that range, and pushes you fully into the losing asset once price leaves it. It raises both sides of the trade, not just the good one.
The honest framing
Providing liquidity is a real business: you are a market maker earning a spread, and impermanent loss is your inventory risk. It is profitable when fee income exceeds the cost of always selling the winner. Quoted yields describe only the first half of that sentence.
Nothing here is financial advice. See our risk disclaimer.