The best academic measurement found liquidity providers lost money: across 43 percent of Uniswap v3 TVL they earned $199.3m in fees and suffered $260.1m in impermanent loss. Providing liquidity is market making with stale quotes, not yield.
The best available academic measurement of this activity found that liquidity providers lost money doing it.
A 2021 paper on impermanent loss in Uniswap v3, cited more than a hundred times since, analysed 17 pools representing 43 percent of the protocol's total value locked. Over the period studied, those liquidity providers collected $199.3 million in fees and suffered $260.1 million in impermanent loss. Net, they would have been better off by roughly $60.8 million if they had simply held their tokens and done nothing.
That is the aggregate result across nearly half the protocol, not a cherry-picked bad pool. A separate analysis by Topaz Blue and the Bancor Protocol, widely reported at the time, found that around 49.5 percent of Uniswap v3 liquidity providers had negative returns once impermanent loss was accounted for.
Neither finding means the activity is impossible to profit from. Both mean the default outcome is a loss, which is the opposite of how liquidity provision is usually presented, and it is the reason this guide spends most of its length on the mechanism rather than on the returns.
It is worth saying why a losing default is worth documenting at all rather than skipping. Liquidity provision is genuinely useful infrastructure, the fees are real, and there are configurations where the arithmetic works. What makes it dangerous is that the interface reports a rising number continuously while the position may be losing, so the feedback a person receives is positive throughout. Almost no other activity on this site has that property. A failing shop shows empty shelves and a failing freelance business shows an empty calendar, but a failing liquidity position shows an accumulating fee counter. Understanding the mechanism is the only available substitute for feedback that tells you the truth.
What You Are Actually Selling
The word "yield" does a lot of damage here. Providing liquidity is not lending, and the fee is not interest.
When you deposit two assets into an automated market maker, you are posting a standing offer to trade either asset for the other at prices determined by a formula. You will buy when someone wants to sell and sell when someone wants to buy, automatically, at whatever the curve says, regardless of what the wider market thinks the price is.
a16z crypto's analysis puts the consequence plainly: liquidity providers "suffer losses from adverse selection, which is part of the price of doing business as an LP. By virtue of offering to take either side (buy or sell) of a trade at a given price, every LP in an AMM runs the risk" of trading against someone who knows more than the pool does.
That counterparty has a name. When the market price of an asset moves, your pool's price is stale until someone corrects it, and the person who corrects it is an arbitrageur who buys the underpriced side from you and sells it elsewhere. Their profit is your loss, and it happens on every meaningful price move, automatically, whether or not you are paying attention.
So the honest description is this: you are running a market-making operation with no ability to update your quotes, and collecting a fee for the privilege. The question is whether the fee covers what the arbitrageurs take. Everything else is detail.
Why Impermanent Loss Is the Wrong Number
Impermanent loss is the metric everyone learns first and it is a poor tool, for a reason worth understanding because it changes how you evaluate a pool.
Impermanent loss compares your position's value against simply holding the same tokens, and it depends only on the starting price and the ending price. a16z's write-up flags exactly why that should bother you: "The independence of impermanent loss on the price trajectory (other than its initial and final values) should strike you as fishy."
It should. Consider a price that rises sharply and then falls back to where it started. Impermanent loss says you lost nothing, because start and end prices match. But arbitrageurs traded against your pool on the way up and again on the way down, extracting value each time. The metric reports zero while your capital was being drained twice.
The better measure, introduced in that same analysis, is loss versus rebalancing, or LVR, pronounced "lever". It compares your AMM position against a reference strategy that makes the same trades the pool makes, but executes them at the true market price on an external venue instead of at the pool's stale price.
The worked example in the paper is compact. Take a constant-product pool holding 1 ETH and 1,000 USDC, and suppose the market price of ETH jumps from $1,000 to $4,000. An arbitrageur buys 0.5 ETH from the pool. The reference rebalancing sells that same 0.5 ETH, but at the real market price of $4,000. The difference between what the pool got and what the market would have paid is the LVR, and it is the arbitrageur's profit.
Two properties make LVR the right tool. It depends on the price path, so a round trip up and back down accumulates cost rather than reporting zero. And it accumulates trade by trade, which matches how the loss is actually inflicted.
The practical test, in a16z's words, is whether "the fees collected exceeded the LVR suffered". That is the entire question. A pool advertising a high APR is telling you about one side of that inequality.
The Round Trip, Worked Through
The gap between the two metrics is easiest to see with actual numbers. Take a constant-product pool holding 1 ETH and 1,000 USDC at a 0.30 percent fee tier, and send the price up to $4,000 and back to $1,000.
The pool always rebalances so that the ratio of its holdings matches the new price, and arbitrageurs are the ones who force it there.
| Price move | What the pool does | Effective price | Market price | LVR | Fees at 0.30% |
|---|
| $1,000 to $4,000 | sells 0.50 ETH | $2,000 | $4,000 | $1,000 | $3.00 |
| $4,000 to $1,000 | buys 0.50 ETH | $2,000 | $1,000 | $500 | $3.00 |
On the way up, the pool sells half an ETH at an average of $2,000 while the market is paying $4,000. On the way down, it buys that same half back at an average of $2,000 while the market is charging $1,000. It sold low and bought high, in that order, automatically.
The position ends holding exactly what it started with: 1 ETH and 1,000 USDC. So:
- Impermanent loss reports $0. Start and end prices are identical, and IL sees nothing.
- LVR totals $1,500, transferred to arbitrageurs across the two trades.
- Fees earned: about $6, which is 0.4 percent of what was extracted.
Nothing unusual happened here. One round trip in a volatile pair, no exotic event, no exploit. The metric everyone is taught reported a perfect outcome while the position leaked value on both legs.
This is why an APR that counts fees and ignores LVR is not an incomplete picture. It is a picture of the wrong quantity.
The Structural Insight That Follows
Once you see LP as selling to arbitrageurs, the profitable and unprofitable configurations become predictable rather than mysterious.
Volatility is the cost driver. LVR scales with how much the price moves, because every move creates an arbitrage opportunity against your stale quote. A volatile pair generates more fee revenue and more LVR, and the second usually grows faster.
Correlated pairs are structurally safer. Two assets that track each other closely leave little room for the price ratio to move, so there is little for an arbitrageur to extract. This is why stablecoin pairs and pairs of assets pegged to the same underlying behave completely differently from a volatile pair, and it is the single most important selection criterion.
Fee tier must match volatility. A pool's fee is your compensation for adverse selection. A volatile pair in a low fee tier is a structurally losing position, because the compensation was set for a risk that is not the one you are taking.
Uninformed flow is what pays you. Fees from someone swapping because they want the other token are pure revenue with no adverse selection attached. Fees from an arbitrageur are a partial refund of what they just took from you. Analysis comparing Uniswap versions has noted that v2 positions performed better than v3 largely because v2 sees more of this ordinary flow. A pool where most volume is arbitrage is a pool where the reported volume is a warning rather than a selling point.
Concentration multiplies both sides. Concentrated liquidity lets you supply within a chosen price band, earning far more fees per dollar while the price stays inside it. It also multiplies your adverse selection exposure within that band, and pays nothing at all when the price leaves. It is leverage on the same trade, not a way around it.
Where This Genuinely Works
Three configurations survive the analysis, and they have far less exciting yields than the ones being advertised.
Correlated and pegged pairs. Stablecoin pairs, and pairs of assets that track the same underlying, keep LVR structurally low because the price ratio barely moves. Fees are modest, and modest fees against near-zero adverse selection is a real return. This is the closest thing in this space to the "yield" framing, and it is the least promoted because the numbers are small.
The risks here are not price risk, they are peg risk and contract risk. The failure mode is not a slow bleed; it is one asset losing its peg and your position converting almost entirely into the broken one. Correlated pairs trade continuous small losses for occasional large ones, which is a different shape of risk rather than less of it.
Active management with a real edge. Professional market makers profit by rebalancing, hedging the directional exposure elsewhere, and choosing when not to quote. This works, and it requires infrastructure, capital and continuous attention. It is a trading operation. Anyone doing it as a side activity is competing against firms doing it as a business.
Incentivised pools, with the incentive counted honestly. Protocols pay token rewards to attract liquidity, and those rewards can genuinely exceed LVR. The discipline is to price the reward token at what you could actually sell it for, not at the quoted price, and to assume the emission rate falls. A position that is profitable only while rewards last is a position with an expiry date you do not control.
What does not survive is the common case: supplying a volatile pair in a standard fee tier, passively, because the displayed APR looked good.
Reading an Advertised APR
Pool APRs are the most misleading number in this space, and the reasons are mechanical rather than dishonest.
It is backward-looking and annualised from a short window. A figure derived from a day of unusually high volume, multiplied out to a year, describes a year that will not happen.
It counts fees and ignores LVR. This is the fundamental problem. The advertised number is one side of the inequality that decides whether you make money.
It assumes your capital stays in range. For a concentrated position, the quoted return applies only while the price is inside your band. Out of range, you earn nothing and hold entirely the wrong asset.
It is diluted by everyone else who sees it. Fees are shared across all liquidity in the pool. A high advertised return attracts capital, and the return falls as it arrives. You are entering a number that your entry helps destroy.
Reward tokens are valued optimistically. An APR including emissions assumes you can sell the reward token at its quoted price, in your size, continuously. Thin liquidity in the reward token makes that assumption false exactly when you need it to be true.
A more useful evaluation is to look at realised fees over a meaningful period, against realised price movement in the pair over the same period, and ask whether the first plausibly exceeded the cost implied by the second.
The Costs That Do Not Appear in the APR
Beyond adverse selection, several real costs are routinely left out of the arithmetic.
Gas. Entering, exiting and any rebalancing all cost transaction fees. For concentrated positions, active management means repeated transactions, and on an expensive chain this alone can consume the return on a small position.
Rebalancing is a realisation event. Moving a concentrated range means closing at current prices and reopening, which converts paper loss into realised loss and usually creates a taxable event.
Smart contract risk. Your capital sits in code. Established protocols have long track records, and new pools frequently do not. This risk does not show up in any yield figure and it is the one that takes the whole position rather than part of it.
Bridge and chain risk. Chasing higher yields onto newer chains adds bridge risk, which has historically been where the largest losses in this industry occurred.
Tax complexity. Depending on jurisdiction, depositing into a pool may itself be a disposal, fees may be income as they accrue, and every rebalance may be another event. This is among the most burdensome record-keeping in crypto and it is almost never mentioned alongside an APR.
How AMMs Came to Exist
The design that produces all of this was a deliberate trade, and knowing what it bought explains why the costs are structural rather than fixable by picking a better pool.
Traditional markets run on order books. Market makers post bids and offers, update them continuously as conditions change, and pull them when they see danger. That works because a market maker can react. Their whole edge is the speed at which they revise a quote when the world moves.
An order book cannot run on a slow, expensive, public blockchain. Every quote update is a transaction with a fee and a delay, and everyone can see your order before it executes. Early attempts at on-chain order books were unusable for exactly that reason.
The automated market maker solved it by removing the quoting entirely. Instead of a human deciding prices, a formula decides them, and the canonical one is the constant product curve, \`x * y = k\
Primary sources for the figures on this page. Where a number is not covered below, it is an estimate rather than a measurement and is labelled as such in the text.
A realistic month-by-month plan for reaching $5K/mo with DeFi Liquidity Provision:
DeFi Liquidity Provision costs $100-$10000 to start. Many people start at the lower end.
Reported income: Negative for most providers, according to https://arxiv.org/abs/2111.09192, https://a16zcrypto.com/posts/article/lvr-quantifying-the-cost-of-providing-liquidity-to-automated-market-makers/. Results vary by effort and market.
Most people see first profit within unpredictable.
Yes, DeFi Liquidity Provision is a legitimate side hustle. Reported income is Negative for most providers (source: https://arxiv.org/abs/2111.09192, https://a16zcrypto.com/posts/article/lvr-quantifying-the-cost-of-providing-liquidity-to-automated-market-makers/). Like any business, success depends on your effort, skills, and market conditions. Start with $100-$10000 and expect first results within unpredictable.
Yes. Most successful DeFi Liquidity Provision practitioners started with no prior experience. The key is following a structured learning path, starting small, and iterating. Free resources on YouTube and blogs can teach you the fundamentals within 1-2 weeks.
DeFi Liquidity Provision offers higher income potential (reported Negative for most providers) and location freedom compared to most jobs, but requires self-motivation and involves more uncertainty. Many people start DeFi Liquidity Provision as a side hustle while keeping their job, then transition to full-time once income is consistent.
Startup tools for DeFi Liquidity Provision cost $100-$10000. At minimum, you need a computer and internet connection. As you scale, invest in specialized software and tools to automate workflows and increase efficiency.
If you're interested in DeFi Liquidity Provision, you might also like these similar opportunities: