DeFi yield strategies: mechanics, risk and returns
Vaults for treasury and fund teams · reference
Seven strategies account for most of what onchain vaults do. Each earns its return by taking a different risk, and the headline APY won't tell you which.
Onchain vaults run a small number of recognisable strategies: over-collateralised lending, looping, levered carry, institutional lending through CeFi venues, tokenised real-world credit, basis and funding-rate trades, and liquidity provision. Each produces a return for a different reason. So comparing headline rates across them tells you very little until you know which risk sits behind each one.
That's why yield numbers travel badly. Two vaults can both pay 7% while doing unrelated things. One lends stablecoins into an over-collateralised market; the other borrows against tokenised credit at two turns of leverage. They fail in unrelated ways too. If you're allocating corporate balances or client money, you need the strategy description before the rate.
What yield strategies do onchain vaults actually run?
The table below maps each strategy to its mechanism, main risk, indicative range and a live example. The examples come from vaults running on Upshift, which is non-custodial vault infrastructure for onchain yield. Measured figures are trailing 30-day annualised as of 3 August 2026 and they move. Indicative ranges are directional and depend on utilisation, leverage and incentive programmes.
Strategy | Where the return comes from | Main risk | Indicative range | Live example, 3 Aug 2026 |
Over-collateralised lending | The borrow rate paid by borrowers who post more collateral than they draw | Oracle failure, bad debt at market level, rate compression as supply arrives | Low to high single digits on stablecoins | Upshift USDC's Morpho allocation on Monad, $2.13M supplied |
Looping | Supplying an asset, borrowing the same asset against it, and re-supplying, which multiplies exposure to one rate | The borrow rate crossing above the supply rate, and liquidation as the gap thins | Two to three times the unlevered spread, and it can invert | Sentora USD's PYUSD leg: $29.2M supplied against $39.4M borrowed |
Levered carry | The spread between what the collateral yields and what a different borrowed asset costs, multiplied by leverage | Borrow rate rising above collateral yield, liquidation, credit risk inside the collateral | Mid to high single digits on stablecoins | Sentora USD overall: $79.9M gross against $40.5M net, roughly 1.97x. Measured: 7.19% |
Levered carry keeping base-asset exposure | Yield on borrowed stablecoins while your deposited asset stays posted as collateral | A drawdown in the collateral asset forcing liquidation at the worst moment | Low single digits, in the base asset | Sentora BTC: WBTC collateral, PYUSD and RLUSD borrowed. Measured: 1.96% in BTC |
Institutional lending through CeFi | Term loan interest from institutional borrowers, with restrictions enforced onchain | Borrower credit, and lumpy interest that makes short windows misleading | High single digits | Upshift USDC: three active loans, $2.43M principal. Measured: 7.63% |
Tokenised real-world credit | The coupon on a tokenised treasury or credit instrument, passed to the share price | Issuer and settlement risk, and redemption on the issuer's calendar | Roughly 3% to 5%, tracking short rates | Upshift Clear RWA, $706k in vault. Measured: 3.45% |
Basis and funding-rate trades | Funding paid by leveraged longs to the short side of a perpetual, held against spot | Funding turning negative for a sustained run, which converts carry into cost | ~7.8% to 9% average on BTC and ETH over three years | Not run in the vaults above |
Liquidity provision | Trading fees for quoting both sides of a pair | Adverse selection: informed flow trades against your quote | Widely dispersed, and often negative against simply holding | Not run in the vaults above |
The last two rows have no live example here, but their risk signatures are unusually well documented and worth knowing:
- Basis trades. Across a three-year sample, Ethena's own funding data shows average annualised funding of roughly 7.8% to 9% on BTC and ETH. Funding was also negative on 8.84% of days for ETH and 15.9% for BTC, with a longest negative run of 13 consecutive days.
- Liquidity provision. A 2021 study of 17 major Uniswap v3 pools holding 43% of protocol value found $199.3M in fees earned against $260.1M of impermanent loss. Those providers ended $60.8M worse off than if they'd simply held the two assets.
Is looping the same thing as levered carry?
They're close relatives, and the difference changes what you monitor.
- Looping is recursive on one asset. Supply USDC, borrow USDC against it, supply the borrowed USDC, repeat. Each turn multiplies exposure to the same rate, so the position lives on the gap between one supply rate and one borrow rate.
- Levered carry borrows a different asset and deploys it elsewhere. You carry the spread between two positions, plus whatever the second position is exposed to.
Sentora USD runs both at once, which is common and is why one label misleads. As of 3 August 2026 it supplied $29.2M of PYUSD while borrowing $39.4M of PYUSD. That's the loop. It also held Hastra PRIME and Maple's syrupUSDC as collateral against those PYUSD borrowings, which is the carry leg. The credit exposure of that collateral is what you ultimately own. Gross assets of $79.9M against $40.5M net puts the whole position near 1.97 turns.
Sentora BTC is the cleaner carry example, because the borrowed asset leaves the collateral behind. WBTC stays posted, PYUSD and RLUSD are borrowed against it, and the borrowed RLUSD goes into a separate Sentora RLUSD vault while Merkl incentives accrue. You keep bitcoin exposure and earn roughly 1.96% annualised on top of it in BTC terms, sized so a bitcoin drawdown leaves the position well clear of liquidation.
Both structures carry an exposure that position-level numbers won't show you. When collateral is itself a claim on another pool, leverage builds a chain. A New York Fed staff report puts it in central-bank terms. Rehypothecation "creates a collateral chain", where one liquidation "could cause a cascading pattern of selling as users unwind positions across different platforms". A vault's own loan-to-value says nothing about how many links sit beneath its collateral. Ask the curator directly.
How does over-collateralised lending produce yield?
This is the simplest strategy in the set. The vault supplies an asset into a lending market where borrowers post collateral worth more than they draw, and their interest accrues to suppliers. Liquidations are automated, so a single borrower walking away isn't your concern.
The exposures that matter sit one level up:
- Oracle failure. If the price feed a market depends on misprices collateral, liquidations fire late and the market can take on bad debt that suppliers absorb.
- Rate compression. A market paying an attractive supply rate attracts supply, and the rate falls toward the cost of capital.
- Fire-sale dynamics. The Financial Stability Board flagged the liquidation machinery itself in 2023, noting that automatic liquidation of collateral is "a primary reason why deleveraging dynamics in DeFi can be especially vehement".
As of August 2026 a large share of onchain lending sits on Morpho. Its markets are isolated, so an incident in one stays in one.
Where does institutional lending yield come from?
Institutional lending swaps the automated liquidation engine for underwriting. The vault extends term loans to institutional borrowers through CeFi venues, and Upshift Lend enforces the restrictions onchain through the policy engine rather than in a contract clause. Upshift's prime stack, which serves over $7B in monthly transaction volume and more than $800M in loans originated, underwrites and services the book.
Upshift USDC held three active loans totalling roughly $2.43M as of 3 August 2026, alongside a DeFi supply allocation. Its trailing 30-day return annualises to 7.63%. Watch the measurement window here: the same vault's trailing seven-day figure annualises to 21.51%, which is one interest payment landing inside the window rather than a run rate. Read a lending vault on 30 days at minimum, and on cumulative return since inception where you can get it.
How do tokenised real-world assets fit in?
Tokenised credit brings a coupon onchain and passes it to the share price. Returns are lower and steadier than the DeFi strategies, which is usually what you want from the conservative sleeve. Upshift Clear RWA returned 3.45% on a trailing 30-day basis as of 3 August 2026 on $706k in vault.
The structural problem with tokenised assets shows up at exit rather than in the yield. Most tokenised treasury and credit instruments redeem on the issuer's calendar, and many settle on a five-day week while the balances funding them move daily. Upshift RWA Clear holds a buffer in front of the position so you can exit on demand while the instrument settles on its own schedule.
How do you compare returns across strategies that take different risks?
Risk-adjusted return is the only comparison that survives strategies this different. The standard measures need care on a vault, though, and the reason is structural. A vault's return series is its share price, and a share price is built to grind upward: yield accrues continuously and the strategy is run to avoid marking down. That gives you a series with very little measured volatility, which flatters any ratio built on volatility whether or not the underlying risk is small.
Sample length compounds it. Sharpe's own write-up notes that the t-statistic for whether a Sharpe ratio is meaningful equals the ratio multiplied by the square root of the number of observations behind it. A ratio from 30 daily readings carries a fraction of the weight of one built from three years. Both print as a single number.
So lean on the measures that describe what actually happened.
Measure | What it tells you | Where it misleads on a vault | Ask for |
Sharpe ratio | Return per unit of total volatility | Share prices are built to rise smoothly, so low volatility can mean a hidden tail rather than low risk | The number of observations behind it, and whether the window contains a loss |
Sortino ratio | Return per unit of downside deviation only | Same defect, sharper: with few down days the denominator approaches zero and the ratio inflates | The count of negative days in the sample |
Maximum drawdown | The worst peak-to-trough the share price actually took, and when | Needs a window long enough to contain a real event. A young vault's zero drawdown says nothing | Size and date of the largest drawdown since inception |
Cumulative return since inception | What a depositor holding from the start actually earned | Nothing, which is why it belongs beside every trailing APY | Cumulative return with the inception date, next to the trailing figure |
Exposure by strategy, venue and collateral | Where the money sits right now | A "diversified" or "multi-strategy" label can hide concentration in one venue or collateral type | The position list with venue, collateral, size, and gross assets over net value |
Maximum drawdown and cumulative return are the pair that catch what the ratios miss. A strategy can post an excellent Sharpe ratio for eighteen months, then lose more in a week than it earned across all of them. Exposure levels catch a third thing. A vault concentrated in one venue can look stable until that venue has an incident.
How much of the yield comes from incentives?
Incentive rewards are a real component of several strategies, not a rounding error. Sentora BTC accrues Merkl incentives in RLUSD on top of its carry, and the Upshift USDC allocation on Monad accrues WMON alongside its supply rate. You receive them, so an honest APY includes them.
Duration is what separates an incentive from a strategy return. A supply rate persists as long as borrowers want the asset. An incentive programme runs until its budget is spent or its sponsor changes plan. If a vault's return leans heavily on incentives, its economics reset when the programme ends, and you want to know that in advance.
What should you diligence before allocating?
Ask | Why it separates one vault from another | A good answer looks like |
Is the position levered, and by how much? | A single loan-to-value reading won't show it, and leverage changes every other number | Gross assets over net value as a multiple, with the target range |
Which venues and which collateral? | "Diversified" can mean one venue and one collateral type | Named venues and named collateral with sizes, not categories |
Is it looping, carry, or both? | Looping concentrates on one rate; carry adds a second position's risk | Both legs traced, with the supplied and borrowed asset named on each |
What does the policy engine restrict? | A curator confined to pre-approved protocols, tokens and addresses is a different counterparty | Restrictions at chain, protocol, token and function level, and what widening them takes |
What's the largest drawdown since inception? | It's the only number describing a loss that actually happened | Size, date, cause, and what changed afterwards |
What does redemption look like on a Saturday? | Balances that move daily can't wait for a five-day settlement calendar | Whether a buffer covers it, how it's sized, and the fee |
How much of the return is incentives? | Incentive programmes end on a schedule someone else controls | Strategy yield and incentive yield split out, with programme end dates |
Upshift's risk management framework documents the controls behind several of those answers. The contracts have been through 10 audits by 6 independent firms as of August 2026.
Always make sure to do your own research and be aware of the above and any other risks before depositing.
Frequently asked questions
What is the safest DeFi yield strategy?
Over-collateralised lending and tokenised real-world credit have the fewest moving parts, and both returned less than the levered strategies on a trailing 30-day basis as of 3 August 2026. Neither is risk-free: lending markets depend on price feeds and can accumulate bad debt, and tokenised instruments carry issuer and settlement risk.
How much yield can a stablecoin vault realistically earn?
There's no fixed number, and the range is wider than the examples here. A conservative tokenised-credit vault tracks short rates in the 3% to 5% area. A lending or carry vault sat near 7% on a trailing 30-day basis as of 3 August 2026. Strategies running more leverage, more concentrated collateral or an active incentive programme can pay materially more. Each step up buys the extra return with a different risk rather than a better version of the same trade, and no vault yield is guaranteed.
Is a looping strategy the same as borrowing?
Borrowing is one leg. A loop needs both: the vault borrows against supplied collateral and re-supplies the same asset, which pushes gross assets above net value. A position can borrow at a high loan-to-value without looping, so gross assets over net value is the figure that identifies leverage.
Why isn't the Sharpe ratio enough to compare vaults?
A vault's share price is built to rise smoothly, so its measured volatility stays low whether or not its tail risk does. The Sharpe ratio divides by that volatility, so a concentrated or levered strategy can score well right up to the event that hurts it. Maximum drawdown, cumulative return since inception and the current exposure list describe the risk more directly.
What happens to a vault strategy if the borrow rate rises?
For looping and levered carry the spread narrows and can invert, at which point leverage works against the position and the curator reduces it. For unlevered lending a higher borrow rate lifts the supply rate, so the same move helps you.
Can a vault strategy change after launch?
A curator allocates within whitelisted protocols, tokens and addresses, and changes inside that perimeter don't require depositor action. Widening the perimeter is a parameter change subject to the vault's timelock, and neither Upshift nor the curator can move depositor funds to an external wallet.
How do I see what a vault currently holds?
Allocations are visible onchain and in the Upshift app, including venue, collateral and any borrowings. Read the supplied and borrowed sides together, because the net figure is what you're exposed to.
Keep reading
This series: Part 1 covers what a vault is. Part 2 covers using one inside a treasury product. Part 3 covers launching a vault for your own clients. This piece is the strategy reference the series points back to.
- What is an onchain vault: the wrapper these strategies ship inside, written for a treasury or fund team.
- What are DeFi yield vaults: how to evaluate a vault, and what the share token represents.
- How to earn yield on stablecoins: the methods available for stablecoin balances and the rate ranges each produces.
- How traditional asset managers are controlling risk in DeFi: what the policy engine restricts and why allocators ask about it first.
- Multi-asset vault architecture: how a single vault holds several strategies at once.
- The Axis Origin vault: a live partner vault launch and how it was structured.
- earnUSDnr on Fluent: an ecosystem stablecoin vault as a chain's anchor yield product.
Create a vault with Upshift
Share your use case and we’ll get back to you shortly
