How to tier vault strategies by risk

Vault strategies sort into three tiers by how much borrowed money sits behind each dollar of your capital. That ratio is readable onchain, which makes it a better tier boundary than any adjective.

An onchain vault programme tiers cleanly by leverage, running from unlevered lending through levered carry to concentrated looping. You get the multiple by dividing a vault's gross position value by its net asset value, which means the tier comes out of the balance sheet rather than out of how the strategy describes itself. Higher multiples raise the yield and, in the same motion, the speed at which a bad day compounds.

What you're really pricing is the spread over a risk-free reference rate. That spread pays you for taking smart contract risk, and a couple of hundred basis points rarely covers it on a strategy you can't exit. Tiering by leverage tells you how much of the spread came from taking more of the same risk rather than a better trade.

The reason to use leverage rather than a label is that labels don't survive contact with a portfolio. Two vaults can both call themselves conservative stablecoin strategies. One supplies USDC; the other borrows against a tokenised treasury position five times over. Their yields differ by a factor of four, and so does the size of move that wipes out a year of return. Leverage is the number that separates them, and anyone can read it.

What are the three risk tiers of a vault programme?

Three tiers cover the range that institutional capital actually uses today, and each one has a live example with a measured track record. The figures below come from the vault history on 4 August 2026, with returns taken from the asset-to-share ratio rather than a quoted rate.

The tier-three row carries a range rather than a single rate because its trailing figure reads 9.5% over seven days and 15.6% over thirty, and 52 days of history can't settle that spread. Two of the three rows also sit at a denomination the other doesn't, which is the first reason to distrust a league table of headline yields.

A large vault can also sit in two tiers at once, and Sentora USD is the working example. It runs a carry against Maple's syrupUSDC and a tokenised treasury position, and at the same time supplies $29.2M of PYUSD while borrowing $39.4M of PYUSD, which is the same asset on both legs. Blended, it comes out at 1.97x. No label on that vault tells you it holds a loop inside a carry book, so read the positions.

How do you measure leverage in a vault?

Leverage is gross position value divided by net asset value, and both numbers are readable from the vault's onchain positions. Vault positions are readable from the venues themselves, such as the Morpho app. Sum what the vault has supplied across every venue to get gross assets. Subtract what it has borrowed to get net assets. The ratio between them tells you how much of the position is financed rather than owned.

Two mistakes are common enough to name, and the test for both is whether the same token appears on the supply and the borrow side. A high loan-to-value on a single position doesn't by itself mean the vault is looping. A levered carry trade borrows one asset and deploys it into something different, the way Sentora ETH posts weETH and borrows PYUSD. Looping is recursive on a single asset. It shows up as gross assets several times net assets, with the same token on both sides of the balance sheet. Reading only the LTV misses the distinction entirely.

Does realised return rank with risk tier?

Over short windows it doesn't, and an allocation sized off the tier alone will be mispriced. Tier one returned 3.45% on a trailing 30-day basis against tier three's 9.5% to 15.6%, which fits the pattern you'd expect. Tier two came in at 3.51% on the same trailing basis, below tier one despite carrying 1.73x, partly because it earns in ETH rather than dollars.

The ordering that does hold is the failure ordering. A tier-one position loses money when a borrower defaults or a lending market goes bad. The loss is bounded by the size of the position. A tier-three position loses money on the same events, multiplied by the leverage. A move against the collateral can force liquidation before anyone chooses to close. Return varies with market conditions; the shape of the downside is structural.

Why a near-zero drawdown isn't a safety claim

All three vaults show max drawdowns between 0.00% and -0.02%, and that says less than it appears to. Max drawdown measures the worst peak-to-trough move that has actually happened inside the measured window. None of these three has been through a stress event of the kind that tests a levered stablecoin position. The number reflects a quiet period as much as a sound design.

Two questions turn the metric into something useful. Ask how long the series is, because 52 days of calm is not evidence and 202 days is only slightly more. Then ask what a stress event would have to look like for the strategy to lose money. Ask how far the collateral would need to move before a liquidation triggers. A curator who can answer the second question precisely is telling you they've modelled it.

Share prices in these vaults rise by design, which flatters ratio-based metrics like Sharpe and Sortino to the point where they stop discriminating. Max drawdown, cumulative return since inception and the exposure itself carry more signal. They're the three columns in the tier cards above.

Always make sure to do your own research and be aware of the above and any other risks before depositing.

What should you ask for instead of an APY?

A quoted APY is a forecast dressed as a fact. It's also the number most likely to be stale by the time it reaches a committee. Five items replace it, and each one is either readable onchain or answerable in a sentence by a curator who knows their own book:

• Cumulative return since inception, with the start date. Annualising a short window inflates whatever happened recently.
• Max drawdown and the length of the series. The two travel together, and the second one tells you how much the first is worth.
• Gross over net leverage. One number that places the strategy in a tier.
• What the vault is holding right now. Positions by venue and token, not a strategy description.
• The failure scenario. What has to happen for this to lose money, and how far it would go.

Those five turn a yield conversation into an underwriting conversation. They travel well internally too. A risk committee can act on a drawdown figure and a leverage multiple; a headline rate gives it nothing to act on.

Where do tokenised fixed income and equity index exposure fit?

Tier one is where tokenised fixed income belongs, and the vault structure already supports it. Vault shares themselves follow the tokenised vault standard, so a fixed-income underlying needs no special accounting. A vault can hold a tokenised treasury or money market position as its underlying. The tier-two and tier-three examples above already use one as collateral. Holding one unlevered as the whole strategy is the more conservative use of the same asset.

Equity index exposure is a structure Upshift can build rather than something running in a live vault today. That distinction is worth stating plainly. The vault contract is asset-agnostic. A mandate can be written around a tokenised equity index position, with the policy engine restricting which venues and functions the curator may touch. It needs a tokenised instrument with reliable pricing and a custodian willing to hold the receipt token. That makes it a build with a lead time rather than a deployment.

For a treasury team, tier one is available now across cash-like assets. Index exposure follows the same operational path described in the launch process. The mandate work is the same; the instrument availability is the variable.

How do you size a first allocation across tiers?

Most programmes sequence it the same way, and the reasons are operational rather than a matter of risk appetite:

• Start in tier one with a single vault and a conservative mandate. Your finance team needs to close a period against a vault's NAV, and your custodian needs to have held the receipt token through a redemption cycle.
• Add tier two once reporting has been reconciled for a quarter and the operational path is proven.
• Size tier three as a sleeve, small enough that a total loss on it costs less than a year of the tier-one return. That sets the size arithmetically instead of by argument, and it matches how the same teams treat levered strategies in their existing books.

The other decision to make early is whether you want a shared vault or one where you're the only depositor. A whitelisted single-depositor vault gives you the mandate and the reporting to yourself, at the cost of seeding the liquidity alone. A shared vault gives you a track record to underwrite before you commit, which is why most first allocations go there.

Frequently asked questions

What leverage is normal in a stablecoin yield vault?

Unlevered strategies run at 1.00x, and levered carry typically sits between 1.5x and 2.5x. As of 4 August 2026, Sentora ETH held $2.07M of assets against $0.87M of debt, which is 1.73x, and Sentora USD came out at 1.97x across a blended book. Looped strategies run considerably higher, with Sentora PRIME Looping at 5.52x on the same date.

How do I check a vault's leverage myself?

Read the vault's supplied and borrowed positions from its onchain allocations, then divide gross assets by net assets. Vault shares follow the ERC-4626 standard, so the asset-to-share ratio is readable from the contract too. Both figures are available through the Upshift API and through the vault's positions on the venues it uses. A vault that won't show its live positions is asking you to underwrite a description.

Is a tokenised treasury position a conservative holding?

Held unlevered, it behaves like the underlying instrument plus smart contract risk. Issuers such as Superstate publish the fund terms behind those tokens. Used as collateral for borrowing, the same asset sits inside a levered strategy. That risk comes from the leverage rather than the treasury exposure. The asset doesn't determine the tier; what the strategy does with it does.

What does a max drawdown of zero actually mean?

It means the vault's asset-to-share ratio never fell from a prior peak inside the measured window. On a 52-day series that reflects a calm period rather than a tested design. Always read drawdown alongside the length of the history, and ask what event would have to occur for the strategy to lose money.

Can we run a mandate that only permits tier-one strategies?

Yes. A policy engine checks every transaction against approved chains, protocols, tokens and functions. A mandate that excludes borrowing therefore prevents leverage mechanically rather than by instruction. Widening it later is a parameter change subject to the vault's timelock, which gives depositors notice.

Why not just compare Sharpe ratios across vaults?

Vault share prices are designed to rise, so volatility-adjusted ratios come out flattering across the board and stop separating good strategies from lucky ones. Max drawdown, cumulative return since inception and the current exposure give a clearer picture. Short series make fitted statistics unstable, which is another reason to prefer measured figures.

How often do these numbers change?

The asset-to-share ratio updates daily, and leverage changes whenever a curator adjusts positions, which can be intraday. Every figure on this page carries its measurement date for that reason. Pull current values before using any of them in a decision.

Keep reading

• DeFi yield strategies: mechanics, risk and returns covers the seven strategy types behind these tiers, with indicative ranges and how each one fails.
• What is an onchain vault: a treasury guide covers the wrapper itself and who holds the assets.
• Who does what in an onchain vault maps the four parties and which of them can widen a mandate.
• The risk management framework documents the policy engine that enforces a tier boundary in code.
• The vaults API reference lets you pull current TVL, share price and allocations for any figure on this page.