Three dollar stablecoin positions, one date. On 31 October 2026, DefiLlama’s yield data showed 5.43% on USDC lent on Aave, 7.21% on USDC staked in Aave’s Umbrella safety module, and 28.58% on a Morpho vault holding USDG. All three are denominated in dollars. So why does one pay more than five times another? Yield farming, the habit of moving tokens between protocols to collect the best return, only makes sense once you can answer that question.
The answer is that every yield is a payment, and every payment has a payer. Traders pay it, or borrowers do. Sometimes the protocol pays with tokens it creates for the purpose, and sometimes it pays with a promise. Sort returns by payer and the line between real yield and subsidised yield draws itself. The case that made the question famous is Anchor, which offered about 20% on a stablecoin until the money behind the rate ran out. Nothing here ranks products or says what to buy.
The previous lesson on prediction markets showed a market where every winner is paid by a loser. Yield is usually less tidy, because the payer can be hidden several layers down. The glossary entry on yield farming gives the short definition; here we follow the money.
Yield farming starts with finding the payer
A bank pays interest on your savings because it lends them out at a higher rate. Nobody needs a white paper to see the payer. DeFi front ends show a single percentage instead, often in large green type, and that number can mix money earned today with tokens issued today and rewards expected some day. Two pools can display the same figure and share nothing else.
So take the percentage apart before judging it. The question has two halves. Who pays tells you whether the income has a reason to continue. What they pay in tells you whether it keeps its value by the time you collect it. Five percent paid in USDC by borrowers and five percent paid in a new token by its own issuer are different products.
The four places DeFi yield comes from
Usage fees: traders pay liquidity providers
When someone swaps on an automated market maker, a fee stays in the blockchain that runs exactly as written when it is called, holding and moving funds according to its code, without an intermediary able to change the outcome.">smart contract holding reserves of two or more tokens, supplied by users, that a decentralized exchange uses to execute trades automatically at a price set by a formula.">pool for the people who supplied it. On DefiLlama’s 31 October 2026 reading, the Uniswap v3 USDC-WETH pool on Ethereum with a 0.05% fee tier showed 13.88%, and the WETH-USDC pool on Base with a 0.3% tier showed 50.63%. Both figures are entirely fees, with no reward token, and the second one swings hard: DefiLlama’s 7-day version of it was 22.88%, and the spot number had lost about 82 points in a day. The mechanics are in our guide to liquidity pools and impermanent loss, and they carry a warning that matters here: a fee APY is a gross figure. It takes no account of wallet. It becomes permanent when the provider withdraws.">impermanent loss, which can consume all of it when the two assets drift apart.
This is the cleanest payer in DeFi, because a trader chose to pay for a service. It is also the most cyclical. Fees track volume, and when volume falls the yield falls with it, usually before the page has caught up.
Borrowing interest: borrowers pay lenders
In a lending pool, borrowers pay interest and suppliers collect it. Our guide to crypto lending and borrowing covers the pool model, so one reminder is enough. Aave’s documentation puts it plainly: “Supplier yields are funded by borrower interest net of the reserve factor.” The rate follows utilisation, with two slopes around an optimal point: below it, borrow rates rise gently, and above it they rise faster.
This explains why rates jump. When demand for borrowing a stablecoin surges, lenders are paid more within hours, and when borrowers leave the rate sinks. On 31 October 2026 Aave V3 on Ethereum showed 5.43% on USDC against a 30-day mean of 4.40%, and 6.01% on USDT against 3.81%. Those are real payments from real borrowers, taken on a day well above the past month’s average.
DefiLlama’s lending data shows what sat behind the USDC rate: about 2.29 billion dollars supplied and 2.19 billion borrowed, so roughly 96% utilisation, with borrowers paying 6.28%. Suppliers received less than that because the reserve factor comes off the top and the interest is spread over the whole pool, including the part nobody has borrowed. It also means only about 93 million dollars was free to withdraw at that moment.
Token emissions: the protocol pays with its own new tokens
Here the payer changes. A protocol issues its token and hands it to users as a reward for depositing. The modern version began with Compound, which started distributing COMP on 15 June 2020, according to The Block: about 2,880 COMP a day across eight markets, shared in proportion to interest accrued. Users suddenly earned a token for lending or for borrowing, and a borrower could end up net positive.
Emissions are not free money. Each new token dilutes those who already hold it, so the cost falls on holders, and the income depends on someone buying the token when you sell it. That makes this source different in kind from the first two. Fees and interest arrive from outside the protocol. Emissions are created inside it and only turn into dollars when the market absorbs them.
Emissions can have a fair use. A young protocol can pay for liquidity it does not yet have, as a marketing budget, and some of that liquidity may stay once it is useful. The Morpho vault from our opening shows the pattern: 28.58% on 31 October 2026, made of 4.96% base and 23.63% reward, so about 83% of the headline came from rewards. Its 30-day mean was 7.33%, a quarter of the figure on display that day, and the total had risen about 23 points in 24 hours. DefiLlama classes that part as “reward APY”, which can mean the protocol’s token or a partner’s. The split here is DefiLlama’s; we did not check it against the vault’s own page.
Points and airdrops: a promise, not a payment
A points programme pays nothing today. You deposit, a dashboard counts points, and the team suggests, usually without committing, that points may matter when a token appears. The payer is a future token holder, the amount is unknown and the date is unknown. There may never be a payment at all.
Points still have a cost: your capital sits in the contract, exposed to every risk of that contract, while it could be earning something measurable. No percentage can honestly be attached to them, which is why a page that converts points into an APY is guessing. A later lesson in this path deals with farming airdrops as an activity. For the moment, treat points as a lottery ticket whose price is your forgone yield.
Two neighbours that are not quite DeFi: staking and reserve income
Staking looks similar, and the payer is different again. On Ethereum, validators are rewarded by the protocol for proposing blocks and attesting, and proposers also collect transaction fees and MEV. ethereum.org showed a current APR of 2.5% when we opened it on 31 October 2026. Part of that is issuance, so ETH holders who do not stake carry a form of dilution, and part is fees paid by users.
Stablecoin issuers are the other neighbour. Tether showed 521.7 million dollars in fees over the past 30 days and Circle’s USDC 208.6 million (DefiLlama, 31 October 2026), mostly because they hold interest-bearing reserves and by default keep what those reserves earn. Our guide to yield-bearing stablecoins explains the tokens that pass part of such income to holders. Ethena’s sUSDe, which showed 5.04% with a 30-day mean of 4.98%, has a payer of its own, derivatives traders, as set out in our Ethena USDe explainer.
Real yield and subsidised yield, side by side
Real yield means a return paid out of a protocol’s actual income. Subsidised yield means a return paid out of newly issued tokens or a reserve set aside for the purpose. The label has no official definition, and analysts disagree on the details, such as whether emissions should be deducted at market price before calling anything real. The grid below is ours, and it describes the logic rather than any single product.
| Source | Who pays | Paid in | Survives without new tokens? | How it usually fails |
|---|---|---|---|---|
| Usage fees | Traders | Pool assets | Yes, while volume lasts | Volume dries up; impermanent loss exceeds fees |
| Borrowing interest | Borrowers | The lent asset | Yes, while demand lasts | Utilisation falls; bad debt |
| Token emissions | Holders, through dilution | Protocol or partner token | No | Reward token falls; capital leaves |
| Points and airdrops | Future token holders | Nothing until a token exists | Not applicable | No airdrop, or a small one |
| Staking | Network issuance and users’ fees | Native asset | Partly | Slashing; price of the asset |
| Issuer reserves | Reserve income, such as government bills | Stablecoin | Yes, while rates last | Rates fall; depeg |
Now the same logic with figures. All rows come from DefiLlama’s yields API, read on 31 October 2026 at about midday UTC. These numbers move by the hour, and they are a teaching sample, not a selection. We left out pool sizes on purpose: for lending markets DefiLlama’s TVL column leaves out borrowed coins, so Aave’s USDC row showed 92.7 million dollars against 2.29 billion supplied. Our guide to measuring DeFi TVL goes further into that.
| Position (31 Oct 2026) | Total APY | Base | Reward | 30-day mean | Main payer |
|---|---|---|---|---|---|
| Aave V3 USDC, Ethereum | 5.43% | 5.43% | none | 4.40% | Borrowers |
| Aave V3 USDC, Umbrella staking | 7.21% | 5.43% | 1.78% | 5.98% | Borrowers, plus safety incentives for taking on slashing risk |
| Sky sUSDS, Ethereum | 3.80% | 3.80% | none | 3.76% | Sky Savings Rate |
| Ethena sUSDe, Ethereum | 5.04% | 5.04% | none | 4.98% | Mainly derivatives traders |
| Hyperliquid HLP vault, USDC | 4.34% | 4.34% | none | 3.81% | Vault trading results, which can be negative |
| Uniswap v3 USDC-WETH 0.05%, Ethereum | 13.88% | 13.88% | none | 10.56% | Traders, before impermanent loss |
| Uniswap v3 WETH-USDC 0.3%, Base | 50.63% | 50.63% | none | 56.50% | Traders, before impermanent loss |
| Morpho USDe, Base | 5.00% | 0% | 5.00% | 4.81% | Rewards only |
| Morpho STEAKUSDG, Robinhood Chain | 28.58% | 4.96% | 23.63% | 7.33% | Mostly rewards |
Read the second row carefully, because the extra 1.78% is the trap in this table. It looks like a free bonus on top of Aave’s lending rate. In fact, Aave’s Umbrella documentation says stakers accept slashing if USDC bad debt exceeds a set offset, that in extreme cases this can reach the full stake, and that leaving requires a 20-day cooldown, which governance can change, followed by a 2-day window. Aave adds that offsets make slashing much less likely in typical deficits. Even so, the reward pays for insurance you are writing.
What real yield does not promise
Real does not mean safe, and it does not mean profit. A liquidity provider can collect genuine fees and still end up behind someone who held the two assets. A lender can receive genuine interest and lose the principal to bad debt or a contract bug.
It also matters who receives the income. Fees are what users pay; revenue is the share a protocol keeps; holders’ revenue is what reaches token holders. Those are three separate numbers, and DefiLlama’s fee data for the last 30 days, read on 31 October 2026, shows how far apart the first two can sit.
| Protocol (30 days to 31 Oct 2026) | Fees paid by users | Revenue kept by the protocol |
|---|---|---|
| Hyperliquid Perps | $83.4M | $65.9M |
| Uniswap V4 | $53.3M | $0 |
| Lido | $47.6M | $2.9M |
| Aave V3 | $36.8M | $4.7M |
| Uniswap V3 | $35.3M | $8.3M |
| Sky Lending | $28.3M | $13.3M |
| Ethena USDe | $20.7M | $0.09M |
| Aerodrome Slipstream | $7.2M | $4.5M |
| Curve DEX | $4.1M | $1.7M |
A low revenue line is not a bad sign in itself. Lido and Aave pass most of what they collect to stakers and lenders, which is what depositors want. For a token holder the same line reads differently. Dashboards do not always agree either, so compare like with like. Uniswap shows how the split can change: its “UNIfication” vote passed on 25 December 2025 with 99.9% support, The Block reported, sending to the protocol part of the trading fees that had gone entirely to liquidity providers, with UNI to be burned on an ongoing basis. Liquidity providers give up that share, and token holders gain only indirectly, through a smaller supply.
Case study: Anchor and the 20% that borrowers never paid
What Anchor was, and who covered the gap
Anchor was a money market on Terra, built around the UST stablecoin. Rather than pay whatever the market would bear, the protocol, in Haseeb Qureshi’s words, “decided on a fixed target yield for depositors, rather than pay the market rate”. That target was close to 20%, and by May 2022 CoinDesk still described yields of up to 19.5%. Set against a bank savings account of the time, the gap was enormous, and UST supply grew from 2 billion to a high of 18.5 billion dollars within a year, by CoinDesk’s count.
Borrowers did not pay that rate on their own. Qureshi’s account, published on 16 May 2022, describes two extra supports: income from yield-generating collateral such as staking derivatives, and an on-chain reserve of UST capitalised by Terraform Labs that made up the difference. None of this was hidden. Anyone could read the reserve balance on-chain and watch it fall as deposits grew, and his account says the reserve was draining quickly and had to be refilled by Do Kwon in early 2022.
The run of May 2022 and the court decisions that followed
At its height, by Qureshi’s count, Anchor held more than 14 billion dollars of UST. CoinDesk reported on 9 May 2022 that deposits had dropped to 8.7 billion dollars that Monday from 14 billion the previous Friday, and that the reserve had gone to 176 million dollars from 323 million over 30 days. UST had already slipped to 0.98 dollars over the weekend and had lost its peg twice in three days. What followed wiped out about 40 billion dollars, a figure used in coverage of the sentencing.
The civil case came first. In June 2024 the SEC announced that Terraform Labs and Do Kwon had agreed to pay more than 4.5 billion dollars after a unanimous jury found them liable: for Terraform, 3,586,875,883 dollars in disgorgement, 466,952,423 dollars in prejudgment interest and a 420 million dollar civil penalty; for Kwon, 110 million dollars in disgorgement and 14,320,196 dollars in interest, joint and several with the company, plus an 80 million dollar penalty. Kwon then pleaded guilty in August 2025 to conspiracy and wire fraud, and on 11 December 2025 Judge Paul Engelmayer of the Southern District of New York sentenced him to 15 years in prison, more than the 12 prosecutors had sought. “Your fraud was unusually serious,” the judge told him, according to Cointelegraph.
So is 20% on stablecoins a scam?
Not automatically, and the honest answer is more useful than a yes. A 20% rate can be a short promotion openly funded by a token budget, or a borrowing spike that lasts a week, or a risky strategy described as such. It can also be a scheme that pays old depositors with new deposits, the pattern covered in our guide to crypto Ponzi schemes. We do not put Anchor in that last box: its subsidy was public, and a high rate is not itself a crime. What Anchor shows is narrower. A rate that nobody earns has an end date, and if the rate is the main reason people hold a stablecoin, the end of the rate threatens the stablecoin.
Four checks separate those cases, and all four can be done before depositing anything.
- Name the payer in one sentence. If you cannot, or if the answer is vague (“our trading strategies”, “arbitrage”), stop there.
- Size the payer against the payment. Compare a year of income with what the rate costs at current deposits, and look for a reserve that is shrinking.
- Find out what happens when deposits double. A real market rate falls as supply arrives, whereas a fixed rate gets more expensive to keep.
- Look at what the rate is holding up. A rate that exists to create demand for the same token that pays it has a circular payer.
The pattern did not end with Terra. In early November 2025 Stream Finance said an external fund manager had lost about 93 million dollars of its assets, and suspended withdrawals and deposits. Cointelegraph reported that its xUSD token fell as low as 0.10 dollars, that debts to lenders were estimated at 285 million dollars, and that Elixir, owed about 68 million dollars, was shutting down its deUSD stablecoin. The APY on xUSD depended on an outside manager, and holders learned of the loss only when it was disclosed.
What DeFi summer taught about farms
The Compound launch set off the farming wave of mid-2020, told in full in our history of DeFi. One lesson of that period belongs here: capital that arrives for emissions leaves with them. Farmers learned to deposit, claim and sell the reward at once, which is rational for each of them and hard on the token. A protocol could show a large total value locked figure bought with its own treasury, then lose most of it when the schedule tapered off.
That is why a TVL number on its own proves little in a farm. The useful question is how much remains after rewards are cut, and the only way to learn that is to look at a protocol that has already cut them.
A ten-minute method to test any yield
This works on any pool page and needs nothing more than DefiLlama and the protocol’s documentation. The next lesson in this path, on APR and APY, covers the compounding arithmetic.
- Split the headline. On DefiLlama Yields, read base APY and reward APY separately, and write down what share of the total is reward.
- Compare spot with the 30-day mean. A wide gap means today’s number is an event, not a rate.
- Identify the reward token and the date its schedule ends or falls.
- Open the protocol’s fees and revenue on DefiLlama and put a year of income next to a year of emissions, valued at the current token price.
- Name the counterparty for your principal: a borrower, a pool, a strategy or a fund manager.
- Read the exit terms, including cooldowns, withdrawal queues and any slashing of staked funds.
- Write down the risks the rate is paying for, such as contract failure, depeg, oracle error or bad debt.
- Estimate gas both ways and the tax on what you receive, then redo the sum on the net figure.
Step two can be startling. DefiLlama showed Lido‘s stETH at 15.07% on 31 October 2026, with a 30-day mean of 2.55%. Read that as an anomaly, not a rate. Lido’s own API gave 15.08% for its latest daily reading, taken on 30 October, after six days between 2.24% and 2.78%, and ethereum.org still showed about 2.5% for staking. The API gives the number and not the reason, and we could not establish the cause, but anyone comparing spot figures that day would have drawn the wrong conclusion. We also dropped one row from our own table during reporting: a large Curve stablecoin pool showed 0% on both DefiLlama and Curve’s own API, even though that API reported about 17 million dollars of volume. A zero next to real volume is more likely a data gap than a fact.
Fees against emissions, with a rule of thumb and its limits
Step four gives a simple ratio: annual revenue divided by annual emissions. Take an example with invented round figures. A protocol earns 6 million dollars a year and emits tokens worth 30 million at today’s price, so its ratio is 0.2. It is paying five dollars for every dollar it earns, and the difference is coming from token holders. Above 1, it could in principle stop issuing and still pay the same amount.
This is a common rule of thumb, not a proven law. Emissions are valued at a price that their own selling pushes down, revenue may not reach the people receiving the yield, and a young protocol can run below 1 for a good reason.
Worked example: a 12% pool over three months
These numbers are invented for the example, and describe no real pool. You deposit 10,000 dollars in a stablecoin pool showing 12%, made of 4% base and 8% reward. You stay three months, and the reward token loses half its value by the time you sell.
| Line (example) | Calculation | Result |
|---|---|---|
| Base yield | $10,000 × 4% × 3/12 | +$100 |
| Rewards at the starting price | $10,000 × 8% × 3/12 | $200 |
| Rewards after a 50% token fall | $200 × 0.5 | +$100 |
| Gas in and out, assumed | -$30 | |
| Net before impermanent loss and tax | +$170, or 6.8% annualised | |
| Impermanent loss of 1.5%, assumed | $10,000 × 1.5% | -$150 |
| Net before tax | +$20, or 0.8% annualised |
The displayed 12% became 0.8%, and nothing went wrong. No hack, no depeg, only a falling reward and a small price gap in the pool. With a stablecoin pair that gap would normally be smaller, and with a volatile pair it could be far larger.
Yield farming profits depend on what is left after costs
There is no general answer, and this lesson does not offer one. For large sums, gas is a small line and a small rate difference can matter. For small sums, gas on a busy chain can consume a quarter’s income, while lower-fee networks change that sum. The same position can be sensible for one reader and pointless for another, and past yields say little about future ones. Compare the net figure with what the same stablecoin earns in a lending market, and ask whether the difference pays for the extra risk. Searches for the best DeFi yield tend to end at a table sorted by APY, and that sort says nothing about who pays.
Rules and taxes for readers in the US, UK, France and Italy
United States: Rev. Rul. 2023-14 says a cash-method taxpayer includes staking rewards in gross income at fair market value in the year they gain dominion and control over them, meaning they can sell or otherwise dispose of them. The ruling addresses staking, not farming rewards, so the US crypto tax guide is the place to check how that applies. On the regulatory side, centralised yield has already drawn action: on 14 February 2022 the SEC announced that BlockFi had agreed to pay a 50 million dollar penalty, plus 50 million dollars in fines to 32 states, and to stop offering or selling its interest accounts in the United States, without admitting or denying the findings.
United Kingdom: HMRC weighs several factors. Its cryptoassets manual says a return to a lender or liquidity provider points to income when it is agreed in advance and paid periodically by the borrower or platform, and to capital when it is speculative and realised by disposing of an asset. No single factor decides. Our UK crypto tax guide goes further. The FCA tells consumers that crypto is high risk, that they should be prepared to lose all their money and that protection from the Financial Services Compensation Scheme is highly unlikely.
France and Italy: individuals in France pay the 31.4% flat tax on capital gains from digital assets, per impots.gouv.fr, unless they opt for the progressive scale, and total sales of 305 euros or less in a year are exempt. In Italy the Agenzia delle Entrate states a 33% rate on gains and other income from crypto-assets realised from 1 January 2026, with no exempt threshold. How individual DeFi rewards are classified is less settled in both countries; our guides to crypto taxes in France and crypto taxes in Italy set out the general rules. Where rewards are taxed on receipt, the bill rests on their value that day and can exceed what the tokens fetch when sold.
