DeFi yield farming and the difference between fees and subsidy

DeFi yield farming is the practice of depositing crypto assets into a permissionless protocol and collecting whatever that protocol pays out. The mechanics vary — supplying a lending pool, providing liquidity to an automated market maker, locking a governance token, depositing into a vault that does all three on your behalf — but the payout always decomposes into the same two parts. One part is organic: swap fees from traders, interest from borrowers, revenue from something a real user wanted. The other part is incentive: newly minted governance tokens, vote-bribes, or points redeemable for a future airdrop. Dashboards add them together and print a single percentage.

That single percentage is where almost all of the misunderstanding lives. A pool advertising 12% where 2% is fees and 10% is emissions is not the same instrument as a pool advertising 5% that is entirely fees, and it will not behave the same way when sentiment turns. This page is about telling them apart, and about the one cost that is specific to liquidity provision and never appears in the advertised rate at all.

Where the organic part of a DeFi yield comes from

There are only a few honest sources. In a pooled lending market such as Aave or Compound, the money comes from borrowers, and the rate you receive is mechanically derived from what they pay. Supply APY equals borrow APY multiplied by utilisation multiplied by one minus the reserve factor. Utilisation is the share of the pool currently lent out. The consequence is arithmetic rather than opinion: your rate is always materially below the borrow rate, and if borrowing demand disappears your rate goes to approximately nothing regardless of how much capital is sitting there. This is why wrapped Bitcoin supplied to Aave v3 on Ethereum was paying 0.00% on $2.58bn of deposits at a utilisation rate of about 2.3% when we checked. We take that apart properly in crypto lending.

In an automated market maker the money comes from traders. Curve's stable pools charge0.01% to 0.04% per swap, which is deliberately tiny because the whole product is low-slippage stablecoin routing at volume. Uniswap v3 processed$33.829bn of volume over 30 days, producing$52.36m in fees on $1.478bn of TVL, with protocol revenue of $14.58m of that. Those are large fee-to-capital ratios, and the reason is concentrated liquidity — which, as the next-but-one section explains, is not free.

In a yield aggregator or a wrapper such as Convex, Yearn or a Morpho curated vault, there is no new source of money at all. These protocols re-route someone else's cash flow and take a cut for doing it. Convex exists so that people who have not locked CRV for four years can still capture the veCRV boost; Curve directs 90% of protocol earnings to veCRV holders, and the vote-bribe market that sits on top is a genuine, growing revenue line rather than a curiosity. None of this creates yield. It redistributes it, and each layer charges.

The organic-versus-emissions test

You can do this test on any protocol in a couple of minutes, and it is the single most useful habit on this page. Take the fees the protocol generated over the last 30 days. Annualise them. Divide by total value locked. That gives you a crude but honest ceiling on how much of the advertised yield can possibly be organic. Whatever the advertised number exceeds it by is being paid in tokens, bribes or points.

Read the Pendle row carefully, because it is the one most likely to be misinterpreted. Pendle is not paying a 9.5% yield out of $628,765 of monthly fees. Pendle splits a yield-bearing token into a principal token, redeemable one-for-one at maturity, and a yield token that carries the income stream. The advertised pool APYs reflect the yield of the underlying assets plus incentives, not Pendle's own revenue. The point of the row is narrower and still useful: the protocol layer itself is thin, so anything you are paid is coming from somewhere below it and you should know where.

The cleanest demonstration of subsidy in the dataset is EigenLayer. Over 30 days it generated$236,015 in fees while paying out $225,318 in incentives, and DefiLlama's annualised protocol earnings figure for it was roughly−$2.75m. A protocol whose incentive spend consumes essentially all of its revenue is not producing yield. It is buying deposits, and the price it is willing to pay for them will change.

Impermanent loss, with the arithmetic shown

If you supply a two-sided liquidity pool, there is a cost that appears in no rate table anywhere, and it can comfortably exceed everything you earn. It is usually called impermanent loss, though divergence loss is the more accurate name. Here is the worked example in full.

Setup. A 50/50 ETH/USDC pool. ETH is trading at $2,000. You deposit1 ETH and 2,000 USDC, a position worth $4,000. The pool's constant-product invariant is k = x × y = 2,000.

ETH doubles to $4,000. Arbitrageurs rebalance the pool until the ratio of reserves matches the new price, so y ÷ x = 4,000 while x × y stays at 2,000. Solving gives x = √(2,000 ÷ 4,000) = 0.70711 ETH and y = 2,828.43 USDC.

• Your position is now worth 0.70711 × $4,000 + $2,828.43 = $5,656.85.
• Holding the original 1 ETH and 2,000 USDC would be worth $6,000.
• The difference is $343.15, or −5.72%.

Now look at where the money went, because this is the part that explains the mechanism rather than just naming it. The arbitrageur took 0.29289 ETH out of your position and left $828.43 behind for it. That is an average sale price of about $2,828 per ETH against a market price of $4,000. Impermanent loss is not a fee anybody charges you. It is the cumulative cost of your capital being sold at stale prices, over and over, by people whose entire business is spotting the gap between your pool and the market.

Two things follow from that table. First, the loss is called impermanent because it unwinds completely if price returns to where you deposited — and it becomes entirely permanent the instant you withdraw. Second, providing liquidity is profitable only when accumulated fees plus incentives exceed the divergence. For a stablecoin pair that is a low bar, because r barely moves. For a volatile pair against a stablecoin it is a demanding one, and a 5× move in the volatile leg costs you a quarter of the position against simply holding it.

Concentrated liquidity multiplies both sides of the trade

Uniswap v3 changed the shape of this. Instead of spreading your capital across every possible price from zero to infinity, you confine it to a band. Inside the band your capital is doing far more work, so you earn a far larger share of the fees. The capital-efficiency multiplier against a full-range position is 1 ÷ (1 − (Pa ÷ Pb) to the power of one quarter), where Pa and Pb are the bounds of your band. A band spanning a 2× price range gives roughly6.3×. A very tight band around a stablecoin peg gives something on the order of 4,000×.

The part that marketing material tends to skip is that the identical multiplier applies to divergence loss. Narrow ranges earn dramatically more fees and suffer dramatically amplified losses when price moves, because the same capital is being rebalanced across a much shorter distance. And there is a discrete failure mode at the edges: once price leaves your band, your position is 100% of the depreciating asset and earns zero fees until price comes back or you rebalance. Rebalancing crystallises the loss and costs gas. A concentrated position is an active trading strategy that happens to be settled by a smart contract, and treating it as a savings product is the most expensive mistake available in this category. Uniswap v4, which adds hooks and a singleton architecture that cuts pool-creation and multi-hop gas costs, held $1.053bn of TVL against v3's $1.478bn when we checked — migration is real but incomplete.

Expected value, honestly

Points are a lottery ticket attached to a deposit

Points programmes replaced one-off snapshot airdrops because they demand sustained engagement rather than a single qualifying transaction. Three routes dominate: layer-one and layer-two re-attribution campaigns, staking verified assets such as ETH, SOL or ATOM, and pre-mainnet testnet farming.

The published outcomes are not encouraging. Data cited in July 2026 reporting found that88% of airdropped tokens lost value within three months and64% of recipients sold immediately at the token generation event. Hacken's 2025 report put $3.1bn of losses to crypto scams in the first half of 2025, with airdrop phishing a primary vector.

The structural problem is simpler than the statistics. A point has no contractual claim, no published conversion ratio and no cap on how many more can be issued. You cannot calculate an expected value at the moment you deposit, which means you cannot size the position. That is a different activity from earning a rate.

Smart-contract risk, with dates attached

Every figure on this page assumes the contract holding your money does what its documentation says. Sometimes it does not, and the record is long enough now to be useful rather than anecdotal.

The aggregate picture is genuinely improving and genuinely fragile at the same time. Immunefi recorded $972m across 207 incidents in the first half of 2026 — the highest incident count ever measured and the lowest loss total since 2021, with DeFi-specific exploits at $680.3m, down 74% from the 2022 peak of $2.62bn. CertiK's Hack3D report for the same period gives $1.32bn. Those two numbers describe the same six months; the methodologies differ on what counts as an incident and what counts as a loss, and we quote both rather than picking the one that suits the argument. DefiLlama's all-time totals stood at $20.647bn hacked overall, $9.253bn of it DeFi-specific and $3.681bn bridge-related.

The uncomfortable detail is the timing. Both of 2026's largest incidents landed in the three weeks before we checked. A downward trend in annual totals is not the same as a downward trend in the risk you personally carry in any given month, andour risk guide turns that into a checklist you can apply before depositing rather than after.

How to read a farming opportunity without a spreadsheet

Pulling the threads together, there are four questions that between them explain almost every disappointing outcome in this category, and none of them require a model.

1. What fraction is organic? Annualised fees divided by TVL. Anything the advertised rate exceeds that by is a subsidy with an expiry date you do not control.
2. Am I taking divergence risk? A single-asset lending deposit has none. A two-sided pool has it, and a concentrated position has a multiple of it. Price the table above against the fee rate before you deposit, not after.
3. How many contracts deep is this? Each wrapper — aliquid staking token inside a vault inside an aggregator — adds a failure point and a fee layer while adding no new source of revenue.
4. What is the benchmark? The 3-month US Treasury bill was at3.97% on 15 September 2026, and USDC supplied to Aave v3 on Ethereum was paying 3.57%. Most mainstream DeFi stablecoin yield currently sits at or below the risk-free rate; the full ladder is on ourcrypto interest rates page. If a farm is paying materially more than about 6%, the excess is coming from leverage, credit risk, basis risk or emissions, and it is worth being able to name which.

None of this means yield farming is unusable. It means the honest version of the question is not "which farm has the highest APY" but "what am I being paid to absorb, and is the payment proportionate". Readers weighing this against simpler alternatives will find the comparison instaking vs yield farming and the custodial equivalents in crypto savings accounts. If the attraction is stablecoin yield specifically, stablecoin yield explains the four engines behind it, and tokenised treasuriescovers the version that holds actual government paper. For sizing rather than mechanism, seepassive income with crypto.