Yield farming is supplying capital to a decentralized application in return for rewards, usually paid partly in that application's own token. It sits at the application layer, not the consensus layer, and that is the entire distinction worth holding on to. Staking is paid out of a protocol's issuance and the fees users already pay for securing the network. Farming is paid out of an incentive budget a treasury has decided to spend. Issuance and fees continue for as long as the chain does. An incentive budget runs out.
Split any farm's advertised rate into two lines: the part paid by people using the protocol, and the part paid out of the treasury's token budget. Only the first line is still there after the budget is gone, and it is usually the smaller of the two.
Scroll to see the full diagram
The boundary: who is paying you
Ethereum pays validators from a formula. The base reward per validator is its effective balance multiplied by a base reward factor of 64, divided by four times the square root of the total active balance on the network, with the total split across five weighted components that sum to 64.6 That is newly issued ETH, topped up by priority fees that users pay to transact. Both continue as long as the chain produces blocks.
Yield farming, also called liquidity mining, rewards users for provisioning liquidity or providing other value-adding services to an application's ecosystem, with farmers rewarded in proportion to their deposit and paid in the application's native token.1 The money comes from a treasury allocation that somebody sized and somebody can stop.
The mechanics of locking capital look similar from the outside, which is why the two terms get used interchangeably and why the confusion is expensive. The question that separates them is not what you do with the tokens. It is who is on the other side of the payment, and whether that party has a schedule.
Fee yield and emission yield are different instruments
Fee yield is revenue. Aave's interest rate strategy computes rates from the reserve's utilization against two slopes, one below an optimal usage ratio and a steeper one above it, so the supply rate is funded directly by what borrowers are paying.4 Nobody prints anything. If borrowing stops, the rate goes to zero honestly and immediately, which is the correct behavior.
Emission yield is a budget. Curve routes its inflation to users through gauges, with the rate each gauge receives depending on the current inflation rate and on gauge type weights that are voted on, and rewards boosted by up to a factor of 2.5 for participants who vote-lock tokens.3 Every part of that is a governance decision about how to spend a fixed pool, and none of it is revenue.
So a farm quoting one blended number is adding a revenue line to a marketing line and presenting the sum as a rate. In our view the fix is not a disclaimer, it is two lines. We have not yet met a protocol that split its rate honestly and lost deposits it wanted to keep.
The budget is finite, and sometimes it is finite in code
Curve wrote its taper into the token itself. The CRV contract sets INITIAL_SUPPLY at 1,303,030,303 tokens, an INITIAL_RATE of 274,815,283 tokens per year, RATE_REDUCTION_TIME at 365 days, and RATE_REDUCTION_COEFFICIENT at 1189207115002721024, which is the fourth root of two scaled by 1e18.2 Every year the mining rate is divided by that coefficient. Curve's own documentation states the same thing in one sentence: the inflation is dropping by the fourth root of two every year.3
Work out what that means and the whole argument for this page falls out of it. Dividing by the fourth root of two is a 15.9 percent cut each year and an exact halving every four years, so year six pays about 42 percent of year one. More usefully, the sum of every future emission is computable in closed form on day one: 274,815,283 divided by 0.159, which is roughly 1.73 billion CRV. Add the premine and the eventual supply is about 3.03 billion tokens, which is exactly what Curve's documentation says it converges to.3
That is the entire incentive budget, for all time, knowable before the first deposit arrived. Most programs do not offer that. They put the taper in a governance post, which is a promise rather than a constraint, and a promise can be extended by the same vote that made it. Our preference is straightforward: if a schedule matters, put it where governance has to work to change it, and if you are unwilling to do that, say so plainly rather than publishing a chart.
Scroll to see the full diagram
Impermanent loss is part of the return, not a footnote
Impermanent loss is the difference in value over time between depositing tokens into an automated market maker pool and simply holding those tokens, and it occurs whenever the price ratio between them diverges in either direction. It is called impermanent because the gap closes if the ratio returns to where it started, which Chainlink notes can be rare, making the loss permanent in practice.1
The size is calculable rather than vague. Uniswap's constant product invariant holds the product of the two reserves fixed as trades move the price.5 Working the standard derivation from that invariant, a position's value relative to holding is twice the square root of the price ratio, divided by one plus that ratio. Run the numbers: a 25 percent relative move costs about 0.6 percent, a doubling costs about 5.7 percent, and a fourfold move costs 20 percent.
Now put that against a headline. A pool advertising 30 percent APR, held for a year, across a pair where one asset doubles against the other, has already given up 5.7 percent of position value before anyone asks what the reward token did over the same year. Realized return is fees plus emissions minus impermanent loss minus gas, and only the first two appear on the page you deposited from.
Mercenary capital is a design outcome, not a character flaw
Capital that leaves when the rewards stop is capital doing precisely what the incentive told it to do. Nobody was misled and nobody defected. The program specified a payment for a deposit, the payment fell, and the deposit went to where payments were higher. Calling that behavior mercenary is a way of blaming users for a mechanism.
The market for emissions is genuinely competitive, and it has its own layer of infrastructure. Chainlink documents how protocols compete to capture Curve governance power in order to direct CRV emissions toward their own pools, and how Convex accumulated a significant portion of the vote-locked supply, letting protocols rent that direction rather than acquire it.1 When the ability to point emissions is itself an asset class, depositor capital is being priced against every other program continuously.
The design consequence is uncomfortable and worth stating anyway. TVL bought with emissions reprices daily against every alternative, and the depositor's switching cost is a gas fee. Across the projects we have worked on, the question that predicts what happens at the end of a program is never how much TVL it bought. It is what would still be here at zero emissions, and teams that cannot answer it usually have not asked.
What the incentive was supposed to buy
The programs with the clearest documentation had a job. Synthetix launched liquidity mining in 2019 rewarding sETH and ETH liquidity providers with SNX, and the reward did two specific things: it offset impermanent loss for those providers, and it lowered the barrier for traders entering the ecosystem by making it possible to convert ETH into sETH with much lower slippage.1 That is a bootstrap with a named target, a named pair and a mechanism by which it stops being necessary.
The test we apply is three questions long. Name the thing the budget is buying. Name the level at which it is bought, so you know when it worked. Name the date you stop. If the answer to the first is total value locked, to the second is as much as possible, and to the third is when we decide, that is not a program, it is a spend.
Underneath sits the anchor for all of this. Emissions can make a market deep enough for a real business to operate in. They cannot substitute for the business. If the fee line beneath a farm is near zero and stays there, the protocol is paying people to hold an empty pool, and the token is financing that arrangement out of every other holder's position.
The disclosure standard we hold clients to
Four lines, published together. Split the fee-funded rate from the emission-funded rate. State the emission rate, the schedule it follows and the date it ends, with a pointer to wherever that schedule is enforced. State whether the number is APR or APY and the compounding assumption behind it, which is the whole argument of the APR versus APY entry. And state impermanent loss exposure by pair composition, since a stable pair and a volatile pair carry entirely different risk under the same headline.
None of that requires a protocol to be conservative. It requires it to be legible. A team confident that its fee line is growing has every reason to show the split, because the split is the evidence. A team unwilling to show it is telling the market something anyway.
This page describes mechanisms and the design questions they raise. It is not investment advice, not a recommendation to supply liquidity to any protocol or hold any asset, and nothing here should be read as a projection of returns. Rates quoted by any farm are estimates that depend on conditions nobody controls.
Common questions
What is the difference between staking and yield farming?
The source of the money. Staking rewards come from protocol issuance and from fees users already pay to transact, computed by a consensus formula that runs for as long as the chain does.6 Yield farming rewards come from an application's incentive budget, paid in that application's own token, and a budget is bounded by construction.1 Locking capital looks similar in both cases. What funds the payment, and whether it has an end date, does not.
Where does yield farming yield actually come from?
Two places that should be quoted separately. Fee yield is revenue: on Aave, supply rates are set from utilization against a two-slope model, so what suppliers earn is what borrowers pay.4 Emission yield is a budget: Curve routes inflation to liquidity providers through gauges whose weights are voted on, with boosts of up to 2.5 times for vote-locked participants.3 The first survives the incentive program ending. The second is the incentive program.
What is impermanent loss and how big can it get?
It is the gap between depositing into an automated market maker pool and just holding the same tokens, and it appears whenever the two prices diverge in either direction.1 Working the standard derivation from the constant product invariant, a 25 percent relative move costs about 0.6 percent of position value, a doubling costs about 5.7 percent, and a fourfold move costs 20 percent.5 It is only impermanent if the price ratio comes back.
Why does TVL fall when incentives end?
Because the deposit was priced against the incentive, not the product. When the emission rate steps down, the same capital reprices against every competing program, and the switching cost is a gas fee. The competition is real enough to have its own infrastructure: protocols compete to capture Curve governance power specifically to direct emissions to their own pools.1 The useful design question is what would remain at zero emissions.
Is yield farming the same as liquidity mining?
They are used interchangeably, and Chainlink's definition treats them as the same thing: rewarding users for provisioning liquidity or other value-adding services to an application's ecosystem, paid proportionally in the application's native token.1 Where a distinction is drawn, liquidity mining tends to mean the narrower case of rewarding liquidity provision specifically, while yield farming covers the broader practice of moving capital between protocols to capture whichever rewards are highest.
See Tokenomics Design Services for how this applies in practice.
Sources
- What Is Yield Farming? Meaning and Definition
Chainlink, 2023
Core definition of yield farming and liquidity mining, the impermanent loss definition, the Synthetix 2019 liquidity mining program and its stated purpose, and the competition to direct CRV emissions. - ERC20CRV.vy, Curve DAO token contract source
Curve Finance, curve-dao-contracts repository, 2020
Emission constants as deployed: INITIAL_SUPPLY 1,303,030,303, INITIAL_RATE 274,815,283 per year, RATE_REDUCTION_TIME of 365 days, and RATE_REDUCTION_COEFFICIENT of 1189207115002721024, the fourth root of two scaled by 1e18. - Curve DAO documentation
Curve Finance, curve-dao-contracts repository, 2020
The inflation schedule dropping by the fourth root of two every year, the asymptotic supply of approximately 3.03 billion tokens, gauge-directed distribution with voted type weights, and the boost of up to 2.5 times for vote-locked holders. - Interest Rate Strategy
Aave Docs, 2026
The two-slope utilization model that sets borrow and supply rates, the primary-source basis for treating fee yield as revenue funded by borrowers rather than by emissions. - Uniswap v2 Core whitepaper
Uniswap Labs, 2020
The constant product invariant underlying the standard impermanent loss derivation. The closed-form loss expression is a derivation from that invariant, not a quotation from the paper. - Proof-of-stake rewards and penalties
ethereum.org, Ethereum Foundation, 2026
The consensus base reward formula and its five weighted components, cited here to fix the contrast between issuance-and-fee funded staking rewards and budget-funded farming rewards.
Last reviewed 2026-08
Know the terms but not sure how they apply to your project? That is what an engagement is for. We design, document, and stress-test the whole token economy inside the Tokenomics Data Room.
80+ projects advised. Complete tokenomics in 4 to 6 weeks.