DeFi yield: fully auditable, entirely unforgiving

What DeFi yield actually is

Almost every on-chain yield is one of two things dressed in different clothing: someone is paying interest to borrow from you, or a blockchain is minting rewards for helping secure it. Those are the same two engines that drive centralised earn products. The difference is not the economics — it is that here the machinery is open.

When you supply USDC to Aave, you can see how much of the pool is currently lent out, what collateral backs those loans, what the liquidation thresholds are, and the exact formula converting utilisation into your supply rate. If utilisation climbs past 70–80%, the model accelerates rates sharply to attract more deposits and discourage further borrowing. You are not taking anyone's word for any of it.

That is a genuinely significant advantage, and it is not hypothetical. Through 2022, while several centralised lenders froze withdrawals and then filed for bankruptcy, Aave and Compound liquidated positions on schedule and let depositors leave whenever they wanted. Nothing about that outcome required trusting a management team.

The cost is that the system assumes competence. There is no support desk, no chargeback, no account recovery. Send funds to the wrong address and they are gone. Sign a malicious approval and your tokens can be drained. This is the honest trade, and it is why we do not think DeFi is simply better than CeFi — it is better for a specific kind of user.

What the protocols actually pay

Rates as of 16 September 2026. Every one of these floats, and the lending rates in particular can move several points within a week when borrowing demand shifts.

Five kinds of on-chain yield, ranked by how well we understand them

Lending supply. Deposit into Aave, Compound, Morpho or Spark and earn what borrowers pay. Over-collateralised, transparent, and the most boring thing in DeFi — which is exactly why it is the right starting point. Expect 3.5% to 5.5% on stablecoins in normal conditions.

Liquid staking. Deposit ETH, receive stETH or rETH, keep earning consensus and MEV rewards while the token stays tradable and usable as collateral. Lido charges a flat 10% and pays 3.8% to 4.1%; Rocket Pool's more distributed operator set pays 3.5% to 3.9%. Both track the same underlying Ethereum issuance, so net APRs across the major liquid staking tokens have converged tightly. See our liquid staking guide.

Yield-bearing dollars. sUSDS pays a governance-set rate of 3.60%; sUSDe harvests perpetual funding and paid 4.75% in September 2026 after a long compression from 10–15%. One token, no claim step, no active management. Covered in depth on our stablecoin yield page.

Liquidity provision. Supply a pair to an automated market maker and collect trading fees. This is where impermanent loss lives, and where a great many people have discovered that a 20% advertised APY produced a negative return. Stable-to-stable pairs are the defensible version.

Incentive farming. A protocol emits its own token to attract deposits. The headline APY can be enormous and is usually measured in a token that falls faster than it pays. This is a trade, not a yield, and it should be sized like one.

Impermanent loss, explained properly

This concept costs people more money than any other misunderstanding in DeFi, largely because the name is terrible. Nothing about it is impermanent once you withdraw.

Suppose you supply equal values of ETH and USDC to a liquidity pool. ETH then doubles. The pool does not simply hold your original ETH — it has been selling ETH to arbitrageurs throughout the rise, keeping the two sides balanced. You withdraw with less ETH and more USDC than you put in, and the total is worth less than if you had simply held the original two assets in your wallet. That gap is impermanent loss.

Trading fees and incentive emissions push the other way, and for a heavily traded pair they can more than compensate. For a volatile pair in a trending market, they frequently do not. The version that genuinely works for most people is a pool of two assets that track each other — two dollar stablecoins, or ETH against a liquid staking token — where divergence is structurally small.

Our honest view: if a strategy requires you to model impermanent loss to know whether it is profitable, and you are reading an introductory guide to it, that strategy is not yet for you. The lending and liquid staking routes get you most of the available yield with none of this.

Smart-contract risk, oracle risk, and the risk that actually gets people

Smart-contract risk is the possibility that the code contains an exploitable flaw. It is real, and it has cost the sector billions over the years. It is also concentrated in new and unaudited protocols. Aave and Lido have operated for years, through multiple market crashes, with extensive audit histories and — in Aave's case — a funded safety module. That is not a guarantee, but the difference between a six-year-old audited protocol and a six-week-old fork is enormous.

Oracle risk is the possibility that a price feed reports something wrong and triggers liquidations that should not have happened. Major protocols use redundant feeds with deviation thresholds specifically to contain this.

User error is the one that should worry you most, because it is by far the most common. Phishing front-ends, malicious token approvals, wrong-network transfers and mistyped addresses account for more retail losses than protocol exploits. All of them are avoidable with discipline: bookmark real URLs, verify before signing, revoke stale approvals, and test with small amounts.

How to start without learning the expensive way