How to screen an onchain carry trade before you join it

Which onchain carry trades are taking in borrowed money, and how do you size one before your desk joins? A DeFi trading analyst post Flowdesk published on 28 October 2026 asks for exactly this: "Track new listings, venues, yield sources and pricing gaps across centralised and on-chain markets. Size the best ideas, outline risks and requirements, and present clear go/no-go proposals to the desk."

A carry trade earns a yield above what it pays to borrow. The most visible one onchain is the fixed-yield loop: a desk buys a Pendle principal token (PT), which pays a fixed yield if held to maturity, posts it as collateral, borrows stablecoins against it and buys more. We screened all 42 active PTs on Ethereum on 6 November 2026. Four checks decide what the trade is:

1. Most PTs are not levered, and the borrowed money sits in a few. 28 of the 42 have none on any lender. $237.9M is borrowed against the other 14, and three of them carry $208.4M of it. All of it is on Morpho. Aave holds no PT that has not matured.
2. The yield is fixed. The borrow rate is not. One PT pays 15.3% to maturity. At the same moment, its three borrowing markets charged 13.0%, 9.5% and 3.5%. On 30 October all three were fully borrowed and the largest charged 38.5%.
3. Count the borrowers, not only the dollars. In one trade, five wallets hold 87% of the debt. In another, 228 borrowers hold more than $10,000 each and the top five hold 28%.
4. Price the exit at the size of the largest position. Selling the largest USD3 borrower's PT on Pendle would get 17% less per token than a small sale and leave about a fifth of that borrower's equity.

Figures are read from the chain at block 26,356,300 (6 November 2026, 14:07 UTC) unless marked. Prices and exit quotes come from Pendle's API on the same afternoon.

The three trades that pass

The screen keeps a PT when at least 30% of its supply is posted on a lender and at least $10M is borrowed against it. PT-sUSDS, with $13.8M borrowed, misses at 29.8%. The screen also drops any market whose debt is larger than the collateral behind it, because what grows there is interest on a loan nobody can repay. Every PT with borrowing against it had 1.1 to 1.5 times that debt in collateral. Outside PTs, Morpho's wstUSR/USDC market reports $519.5M borrowed, up from $6.2M after Resolv's March exploit, and Morpho holds no wstUSR at all. A screen that sorts by "borrowed" puts it first.

Both large trades are built on a senior tranche: the layer of a pool that takes losses last. reUSD is Re Protocol's senior tranche, whose offchain capital, in Re's words, is "funding the reinsurer". USD3 is 3Jane's, which lends USDC to borrowers it underwrites for credit. A desk joining either trade is taking reinsurance or credit risk at leverage, not stablecoin risk.

1. Check the borrow rate, not only the fixed yield

A PT trades below its face value and, in Pendle's words, "Upon maturity, PT can be redeemed at 1:1 for the accounting asset": the asset the yield is counted in, USDC for PT-USD3. That discount is the fixed yield. PT-USD3 pays 15.3% a year to 17 December.

The borrowing side floats. Morpho's rate model raises the rate as a market fills and keeps raising it while the market stays above 90% utilisation, the share of supplied money that is lent out. At 14:23 UTC on 6 November the three PT-USD3 markets charged 13.0%, 9.5% and 3.5%, all at about 91% utilisation. Each market's rate depends on how full that market has been.

At five times leverage, a position earns roughly five times the fixed yield minus four times the borrow rate, so it breaks even at a borrow rate of 19.1%. Borrowing in the 3.5% market gives about 62% a year on the desk's own capital. Borrowing in the 13.0% market gives about 24%. On 30 October at 11:26 UTC all three markets were fully borrowed. The 13.0% market charged 38.5%, where the same position loses about 78% a year for as long as it lasts, and the 9.5% market charged 22.7%, a loss of about 14%. In Morpho's daily data the largest market's rate was above the fixed yield on three days since August, and above the five-times break-even on one of them.

Pendle's docs call the threshold the "break-even borrow": "the borrow rate at which looping APY hits 0%". The screen needs it per market and per leverage, checked against how often the market has been full, not one rate per token.

2. Count the borrowers, not only the dollars

Dollar size says how much capital agrees with a trade. The number of borrowers says how many separate decisions are behind it.

reUSD has 228 borrowers above $10,000 and the five largest hold 27.7% of the debt. USD3 has 136, but the largest holds 24.7% alone, $10.78M. USDat has 17 and five hold 87.2%; the top three each hold about 21% at similar loan-to-value ratios, which can mean one entity running several wallets.

For a desk, a concentrated book means one decision can move the rate and the exit. For a lender, it means most of the loan book depends on a handful of borrowers' risk limits.

3. Price the exit at the size of the largest position

A looper can leave in two ways: sell the PT on Pendle, or wait for maturity and redeem.

Selling. For each trade we asked Pendle's API for a quote to sell the largest borrower's whole PT position. reUSD (15.86M PT) and USDat (3.92M PT) came in within 0.2% of a small sale, mostly filled by limit orders rather than the pool. USD3 (13.84M PT) came in at $0.818 per PT against $0.984 for a small sale. That borrower owes $10.78M; Pendle quotes $11.32M in USDC for the PT, leaving about $0.54M of the roughly $2.8M of equity in the position. To get to USDC, the quote redeems USD3, which draws on the same payout limit as every other USD3 holder.

Redeeming. At maturity each PT redeems for USD3 or reUSD worth one dollar, and the desk still needs USDC to repay. USD3 has $115.0M of assets and can pay out $11.1M at once today. 3Jane's docs say USD3 funds "credit assets that are longer-dated than instant redemption". Yearn's review of USD3 adds: "Redemption is atomic only while idle reserves suffice; when the buffer is depleted (as during the April 2026 shutdown, when isShutdown() blocked the path entirely) redemptions are delayed." Morpho holds 56.5M PT-USD3. reUSD caps instant redemptions at 20% of its redemption reserves per day, 10% per wallet and 1,000,000 reUSD per transaction, and moves to a quarterly queue if its buffer falls below 1%.

4. Read what the lender's oracle watches

An oracle is the price feed a lending market uses to value collateral. The three PT-USD3 markets price the same collateral three ways. The 13.0% market uses a fixed discount of 30% a year that never reads a market price. The 9.5% market takes the lower of a 15-minute Pendle price average and a 20% fixed discount; today the fixed discount is the lower one, so two of the three markets price PT-USD3 by formula. The 3.5% market uses a 25-minute Pendle price average.

From the desk's seat, a fixed-discount oracle never liquidates a position because the PT's market price fell. From the lender's seat, that is the risk: if the PT's value fell, or USD3 itself were written down, the oracle would keep valuing the PT near face value and losses would stay with lenders. The wstUSR market shows what an oracle that stops reading the market hides once the asset breaks: its USR feed has not updated since 21 March and still reads $1.00.

Running the screen from either seat

If your desk would borrow:

• Which PTs have at least 30% of supply on a lender and $10M borrowed, after removing markets whose debt is larger than their collateral?
• What is the break-even borrow rate at your leverage, and how many days has each market been full in the last 90?
• How many borrowers hold more than $10,000, and what share do the top five hold?
• What does a quote to sell the largest position cost in the currency you owe, and how much can the underlying pay out at once?

If you lend stablecoins into these markets:

• What does each market's oracle read, and does it ever read a market price?
• What is the underlying exposed to, and who takes losses before it?
• What share of the loan book sits with the top five borrowers?

Limits

• This covers Ethereum and Pendle PTs. The same trade runs on other chains, and on Euler and others where they list PTs.
• Rates move by the hour. The rates and quotes here are one afternoon's readings. The daily rate history comes from Morpho's API; we checked 30 October onchain.
• Wallets are not entities. One firm can run several borrowers, and one address can pool many.
• The five-times figures are arithmetic. They leave out fees, slippage and the gap between the oracle price and the market price.
• Capital in a trade is not a forecast. It shows where borrowed money is today, not what it will return.

Screening every trade, every chain

One chain and one token type took reads across 42 PTs, 22 lending markets, their oracles and their largest borrowers. A desk's pipeline covers every chain and every kind of carry: lending loops, basis trades, staking. Datablocks indexes every EVM chain from genesis and shows new strategies and yield sources, and how much capital is already in them. If your desk sizes onchain carry, book a call below and tell us which trades you need to screen.