What exactly is impermanent loss? It's the hidden cost of the act of 'providing liquidity.' When you supply liquidity to an automated market maker (AMM) like Uniswap, you must deposit a pair of tokens (say ETH and USDC) so others can swap through your pool. The AMM uses a formula to keep a relationship between the two token amounts. When the market price moves, arbitrageurs buy low and sell high against your pool, automatically swapping away the token whose price went up and leaving you holding more of the one that went down. The result: the token mix you withdraw is often worth less than if you'd simply kept those two tokens in your wallet. That shortfall is impermanent loss. It's 'impermanent' because if the price returns to the ratio at which you deposited, the loss disappears — but if the price doesn't return and you withdraw, it becomes a very real loss.
Why does price divergence automatically cause a loss — what's the mechanism? The key is the AMM's constant-product formula x × y = k. Suppose you deposit 1 ETH (then $2,000) and 2,000 USDC, so x=1, y=2000, k=2000. Now ETH rises to $4,000 on the outside market, but your pool's ratio still reflects the old price, so arbitrageurs find they can buy ETH cheaply from your pool. They keep buying until the pool's implied ETH price also reaches $4,000 — at which point the pool holds about 0.707 ETH and 2,828 USDC (keeping k constant). Notice your ETH shrank, because part of it was 'sold cheaply.' The mix you withdraw is worth about $5,657; but if you'd done nothing and simply held 1 ETH + 2,000 USDC, you'd now have $4,000 + $2,000 = $6,000. The $343 difference (about 5.7%) is the impermanent loss. In essence: the AMM forces you to keep selling an asset as it rises and keep buying as it falls — the opposite of trend investing.
When is providing liquidity worth it, and when should you absolutely avoid it? Ask one core question: will the two assets you deposit diverge significantly in price? Three cases. First, two stablecoins (e.g. USDC/USDT): both hover near $1, divergence is tiny, impermanent loss approaches zero, and fee income is nearly pure profit — the safest LP scenario. Second, highly correlated assets (e.g. ETH/stETH): prices move almost in sync, IL is small, good for steady rent. Third, a Stablecoin paired with a volatile token (e.g. USDC/some new coin): this is where IL bites hardest, especially if you're bullish on that token — ironically, the more it rises, the bigger your IL, so you called the direction right but didn't capture the full upside. Rule of thumb: the more you believe a token will rally one-directionally, the less you should put it in an LP; just holding spot earns you more. LPs suit range-bound markets and assets where you want fees without betting on direction.
How do advanced users assess LPs more precisely? Don't chase the headline APY — run a simple 'net yield' estimate: expected fees + farming rewards − (impermanent loss + gas and opportunity cost), and only enter if it's positive. A few advanced checks. First, look at the pool's fee APR excluding token rewards: if real trading fees alone cover your estimated IL, that's 'real yield'; if rewards depend entirely on platform-issued governance tokens, a token dump can leave you net negative. Second, estimate IL from expected volatility: use the standard table — 2x ≈ 5.7%, 3x ≈ 13.4%, 5x ≈ 25.5% — and think through your expected range first. Third, consider concentrated liquidity (Uniswap V3): it boosts capital efficiency, but if price exits your chosen range you become 100% one asset and stop earning fees, amplifying IL. In one line: an LP is a 'short volatility' strategy — the calmer the market, the more you earn; the wilder it gets, the more you lose.
Here's a vivid example. Mei sees an ETH/USDC liquidity pool advertising 18% APY and thinks it beats a savings account by a mile, so at ETH = $2,000 she deposits 1 ETH and 2,000 USDC, total $4,000. Three months later ETH rises to $4,000 (doubles). Mei is thrilled — she called it right — but is stunned at withdrawal: her share is now about 0.707 ETH (worth $2,828) plus 2,828 USDC, totaling about $5,657. She does the math: if she'd done nothing and just held 1 ETH + 2,000 USDC, she'd now have $4,000 + $2,000 = $6,000. So 'providing liquidity' earned her $343 LESS than sitting still — that's the impermanent loss (about 5.7%). Of course she also collected fees over three months; say $250, leaving her net down $93. The lesson is clear: when you're strongly bullish on a coin, putting it in an LP means it's automatically sold off as it rises — right direction, but you didn't capture the gain. Mei would have done better simply holding ETH.
The trade-off of providing liquidity is accepting 'impermanent loss and the opportunity cost of one-directional upside' in exchange for 'steady passive fee income.' The upside fits when your assets are long-term holds you won't trade directionally, or when you supply a stablecoin pair / strongly correlated pair (small divergence) — there IL is minimal and fees are nearly pure profit, putting idle assets to work. The downside fits when you're strongly bullish on a volatile token expecting a one-way rally — in an LP it's auto-sold during the climb, and you watch spot holders earn more. In short: an LP is a tool for earning 'market-boredom fees,' not for betting on 'market melt-ups'; confusing the two means under-earning in exactly the rallies you most wanted to capture.