Where crypto yield comes from

A return in crypto is a transfer. Someone pays it, for a reason, and the rate on the screen does not say who or why. Staking, lending, liquidity provision, airdrops and points are paid by different parties for different things, and each has its own way of failing. For the two kinds of stablecoin that the EU’s markets regulation defines, the issuer and regulated providers may not pay interest at all. The order that works is payer first, rate second.

Yield has a payer

The rate on a yield dashboard is an output. It is computed from something, whether a reward schedule, a pool’s loan book or a window of trading fees, and that something is what deserves reading. Four questions turn the output back into an input: who is paying, what for, in which asset, and what happens to the holder if the payer cannot or will not.

Staking: paid by issuance

On Ethereum, validators lock up ether and are paid in ether for checking and proposing blocks. The mechanics here are Ethereum’s, taken from ethereum.org’s rewards and penalties page. Other proof-of-stake networks set their own rules, so a rate quoted for one says nothing about another.

The protocol is the first payer. A validator’s base reward is proportional to its effective balance and inversely proportional to the square root of the total amount staked, so total issuance rises as more ether is staked while each validator’s share of it falls. Network users are the second payer: the priority fee goes to the validator that includes a transaction, and the base fee is burned. Issuance is a transfer from everyone who holds the asset to everyone who stakes it, so the supply arithmetic in the tokenomics piece applies: a holder who does not stake loses share to those who do.

The cost sits on the other side. A validator that signs conflicting blocks or votes is slashed: part of its stake is destroyed and it is forced out of the network. The penalty is small for an isolated incident and grows with the number of validators slashed around the same time, up to the entire stake, so the loss is correlated by design. Leaving is not instant either, because exits wait in a queue whose length depends on how many others are leaving.

Liquid staking turns the locked position into a token. In ethereum.org’s description the token is a claim on the staked ether and its rewards, and the protocol “doesn’t know your token exists”. The holder is a claimant on a pool’s contracts, governance and operators, not a staker. The token’s price can fall below the value backing it under stress, and slashing of the pool’s validators is typically shared across the holders. The reward is the protocol’s, less the pool’s fee, and the loss case is different.

Lending: paid by borrowers

In an over-collateralised lending pool, lenders deposit an asset, borrowers take it out against collateral worth more than the loan, and the interest borrowers pay is the lenders’ yield. The details below come from Aave v3’s contracts, release notes and technical paper, as a worked example rather than a guide to any other pool.

The rate follows utilisation, the share of the pool that is lent out. In the rate-strategy contract the borrow rate climbs with utilisation up to a configured optimal point, the contract requires the rise beyond that point to be at least as large as the rise up to it, and the supply rate is the borrow rate multiplied by utilisation and by what remains after the protocol’s reserve factor. Two things follow. The lender’s yield measures how much borrowers want the asset at that moment, so it moves with no one deciding, and it is highest when the pool is most drained; at full utilisation the pool holds none of the asset to hand back. And the curve is a parameter: the contract lets only the pool configurator change it, and the technical paper lists a risk-admin role that can change the reserve factor and the liquidation terms.

Lenders are protected by liquidation. When a position’s health factor falls below 1, a liquidator can repay part of the debt and receive a proportional amount of the collateral plus a bonus, which the code describes as covering market risk. The bonus is the wage: nobody is obliged to liquidate, so the mechanism depends on someone finding it worth the cost. It is the mechanism the stablecoin piece describes for crypto-collateralised designs, where a fast, correlated fall is the case that breaks it.

The release notes describe the outcome. Where the collateral liquidated is not enough to repay all the debt, the account is left with no collateral and some debt remaining, “bad debt” that the notes say is unlikely to be repaid. The shortfall is recorded as a deficit on the reserve, and a permissioned mechanism can burn aTokens against it. The notes do not state what an uncovered deficit means for a lender. Small positions fail differently: the notes give the lack of any incentive to liquidate dust-sized debt as the reason for allowing a 100% close factor on small positions, and acknowledge that liquidations remain strongly influenced by gas prices, the liquidation bonus and secondary-market liquidity.

A pool’s rules can be read in its code. An exchange’s “earn” product runs on the venue’s terms. ethereum.org’s guidance on pooled staking says of centralised-exchange programmes that the provider holds the keys and the ETH, that rates and lock-ups can be revised at any time, that the yield may come from lending or trading rather than staking, and that if the provider fails there is nothing on-chain to redeem. In a January 2023 ruling on the failed lender Celsius Network, a US bankruptcy court held, on the evidence before it, that the terms of use of its Earn programme transferred title to deposited assets to the company, leaving account holders as unsecured creditors (page 30). At the petition date the programme had about 600,000 accounts, and they held crypto worth about $4.2 billion as of 10 July 2022 (page 5). The court added that even if the deposits were loans the holders would still be unsecured creditors (page 40). The ruling concerns one company’s terms. What carries over is the mechanism: if title passes to the venue, the holder is its creditor, as what happens when an exchange fails sets out, and the exchange liquidity piece calls such yields a credit product.

Liquidity provision: paid by traders, taxed by price movement

An automated market maker holds two assets in a pool and prices trades with a formula instead of an order book. People who deposit both assets become liquidity providers and share the trading fees; in Uniswap v2 the fee was a flat 0.3% of trade volume, according to a post by the writer Pintail, which the protocol’s documentation quotes. The payer is the trader and the thing bought is immediacy, the price of which the exchange fees piece describes; a pool’s depositors collect the fee for it.

The cost comes from the formula. When one asset’s price rises against the other elsewhere, traders buy it from the pool until the pool’s price catches up, so the pool ends up holding less of the asset that rose and more of the one that fell, and the depositor’s share is worth less than the same two assets held. The post, as the documentation quotes it, works it through for a constant-product pool: ignoring fees, a 2x price change in either direction leaves the depositor 5.7% below simply holding, and the shortfall grows with the size of the move. It is called impermanent because it vanishes if the price returns to its starting ratio.

The more exact description is that the depositor has sold volatility. Milionis, Moallemi, Roughgarden and Zhang summarise earlier work, which they credit to Clark, Fukasawa et al. and Deng et al.: ignoring fees, a constant-product pool breaks even if the price ends where it began and loses otherwise, which makes the position equivalent to giving away a bundle of European options (section 9.1). In their model the loss against a portfolio that made the same trades at market prices, which they call loss versus rebalancing, is one-eighth of the price’s instantaneous variance per unit of time, as a fraction of the pool’s value (Example 3). That benchmark is not holding, which is how the two results fit (section 5.1): a pool whose price comes back to where it started is level with holding but behind that portfolio. Fees are the premium for that exposure. Because the cost depends on variance, doubling the volatility quadruples it; that follows from the formula and is not a figure the paper reports. The position earns from volume and pays for movement.

A quoted rate can therefore exceed what depositors keep. DefiLlama’s guidelines for the adaptors behind its yield table state an aim of showing the minimum attainable yield, unboosted, yet ask that fee-based APY be calculated over a 24-hour window, so one unusual day can set an annualised figure, and the pool schema they describe has a field for fee yield and one for reward yield but none for what the position lost against holding. The offset can be large. A 2021 study of 17 Uniswap v3 pools, non-correlated pairs only and each pool above $10m in value locked, together about 43% of v3’s locked value at the time, counted $199.3m in fees against $260.1m in impermanent loss between 5 May and 20 September 2021. Fees exceeded the loss in only three of the pools. That is one study of one window, and its loss is measured against simply holding, which the loss-versus-rebalancing authors argue mixes in the price risk of the assets themselves. It shows that the offset can exceed the fees, not how often it does. A protocol’s fees in aggregate, which say nothing about any one depositor’s result, are listed beside its total value locked in the DeFi protocols table, as of the latest snapshot.

Airdrops and points: paid by future token holders

An airdrop is a distribution of tokens to addresses that meet a rule. It is not income from outside the system: the tokens come out of supply the project controls, so the payers are the holders of the rest of that supply and, once recipients sell, whoever buys. How much of a supply is committed to recipients, and when, is the subject of the tokenomics piece.

An empirical study of nine major airdrops by Messias, Yaish and Livshits gives the aims as bootstrapping communities, rewarding early adopters and decentralising distribution. The nine are 1inch, Arbitrum, Arkham, dYdX, ENS, Lido, Optimism, Tornado Cash and Uniswap, and the study measures the share of tokens that recipient accounts sold at exchanges as their first transfer after claiming, as of October 2024. The highest was 65.75% and another reached 58.67%.

The simplest design is a retroactive snapshot. When Uniswap launched its token in September 2020, 400 UNI were claimable by each of 251,534 addresses that had ever called the v1 or v2 contracts, based on a snapshot ending 1 September 2020, or 10.06% of the 1 billion UNI minted at genesis. Recipients qualified on use that predated the announcement. The same study notes that farmers can prepare in advance by using protocols before any airdrop is announced, as they did before dYdX’s, so a snapshot measures ordinary use only until people expect one. Among the responses it lists to sybil farming, in which one person operates many wallets to meet the criteria, are task requirements, distributions in rounds so the criteria can be refined, and detection and exclusion of wallet clusters.

Points are the same promise with fewer commitments. Galaxy Research’s July 2024 report on five points programmes, Tensor, Blur, Friend.tech, Blast and EigenLayer, describes a ledger kept off-chain by the issuer, with no historical guarantee that points convert into tokens or at what rate. It finds that programmes often offer points before saying how, for what or when they can be redeemed. Tensor’s second season ended with points not redeemable for tokens, and in the third Tensor removed participants’ real-time view of their points to stop them inflating their totals. The Eigen Foundation revised its airdrop on 3 May 2024, after criticism of its criteria, adding 100 tokens to a base of 10 for eligible users.

A points balance is therefore an IOU whose terms may be written after the points are earned. If it converts, its payer is the same as an airdrop’s, the holders of the token it converts into. DefiLlama’s guidelines exclude pre-launch points and non-transferable tokens from the reward yield they display.

Interest on e-money tokens: prohibited in the EU

The EU’s Markets in Crypto-Assets regulation, Regulation (EU) 2023/1114 (MiCA), takes the issuer and regulated providers out of the list of possible payers for the stablecoins it defines. Article 50(1) provides that issuers of e-money tokens shall not grant interest in relation to them, and Article 50(2) that crypto-asset service providers shall not grant interest when providing crypto-asset services related to e-money tokens. Paragraph 3 closes the obvious workaround: any remuneration or other benefit related to the length of time a holder holds the token is treated as interest, including net compensation or discounts with an equivalent effect, whether it comes from the issuer or from third parties. An e-money token is a crypto-asset that purports to maintain a stable value by referencing one official currency (Article 3(1)(7)), and what holds a stablecoin’s peg covers what else that regime requires of issuers, including redemption at par.

Article 40 is the counterpart for asset-referenced tokens, with the same three paragraphs, so the ban spans both kinds of stablecoin the regulation defines. Titles III and IV, which contain the two provisions, have applied since 30 June 2024 (Article 149(3)).

The ban attaches to particular payers: an issuer, or a service provider the regulation reaches. It does not speak to yield on crypto-assets generally. Staking rewards on ether and interest on a token that is neither an e-money token nor an asset-referenced token are outside the wording of these two articles, and so are points that do not depend on how long a holder keeps either kind of stablecoin. It also leaves open how the rule applies where nobody operates the service. Recital 22 says services provided in a fully decentralised manner without any intermediary fall outside the regulation, but a recital explains the text rather than operating as it, and whether a given protocol is such a service depends on who runs it.

For a holder of a token that references a single currency, the consequence is that a return advertised on the balance cannot be the token’s own interest, because an issuer the regulation covers may not grant any. Where one is offered anyway, the question is who is paying, under what arrangement, and whether the provider is one the rule binds. Which providers MiCA authorises, and for what, is covered in what a regulated exchange is regulated for.

How to read any yield figure

Six questions, in order of how much they change the answer.

1. Is it base yield or reward yield? Base yield is paid out of activity: fees, loan interest, issuance. Reward yield is an incentive paid on top, in whatever token its funder chooses. A figure that counts locked rewards is counting something that cannot yet be sold.
2. Paid in what? A reward in the staked asset, or in a project’s own token, carries that asset’s price risk. The rate is stated in tokens and the outcome is realised in currency.
3. Measured over what window, and is it fixed? Fee yields come from a recent window. Lending rates move with utilisation and staking rates fall as more is staked. A rate is a reading of a variable, not a term.
4. Who can change it? In staking, a formula in the protocol. In lending, parameters held by a role. At a venue, the provider’s terms. The further down that list, the less the rate is a property of the product.
5. Can the deposit be withdrawn, and how fast? Staking exits wait in a queue, a lending pool at full utilisation holds none of the asset to hand back, and a venue can revise its lock-ups at any time. A rate says nothing about the exit.
6. What is the loss case, and is it in the number? Each source above has its own, from slashing to an unsecured claim on a venue. A contract holding the deposit can also be exploited, a loss case that how crypto gets hacked traces through the recorded incidents. A figure that leaves it out describes only the upside.

Whether an idle balance should be earning anything at the venue that holds it is a question about custody before it is a question about rates, and exchange or self-custody sets out that trade.