Leveraged Yield Farming: Loops, Health Factor, Liquidation

Leverage on Stellar by the numbers (October 2026)

Measure	Value	Source
Blend V2 TVL / borrowed	$146.6M / $43.8M	DefiLlama, 9 Oct 2026
Blend Fixed Pool USDC supply APY	6.67%	DefiLlama yields
Blend Fixed V2 pool status	"On ice": no new borrowing	On-chain read, 8 Oct 2026
XOXNO Lending TVL / borrowed	$129K / $48K	DefiLlama
XOXNO Aquarius LP collateral	50% LTV, 60% liquidation threshold, 10% bonus	XOXNO configuration
XOXNO USDC in Stables & FX spoke	88% LTV, 92% threshold	XOXNO configuration

The headline is that Stellar's deepest lending market has paused new borrowing in its main pool, and the venue with the most leverage features is small. That shapes what is practical today more than any formula. Market overview: best lending protocols on Stellar.

How to choose

• You want to understand a position before opening it: work out the three numbers below with your own rates, then halve the leverage you first had in mind.
• You want leverage on an Aquarius LP: XOXNO accepts several Aquarius LP tokens at 50% LTV; see Aquarius LP tokens as collateral.
• You want a stablecoin loop: it only pays if the supply rate plus rewards beat the borrow rate; check both live.
• You do not want to monitor a position: do not use leverage. Unlevered auto-compounding, covered in auto-compounding explained, has no liquidation price.

What a loop is

A loop deposits an asset as collateral, borrows against it, converts the borrowed funds into more of the collateral, deposits that, and repeats. Each round adds less, because you can only borrow a fraction of each new deposit. A flash loan does all the rounds at once by borrowing the final amount upfront.

Take $1,000 of collateral at a 50% loan-to-value. Round one borrows $500 and deposits it; round two borrows $250; then $125. The total converges to $2,000 of collateral and $1,000 of debt, which is 1 ÷ (1 − 0.5) = 2x. Doing it by hand means several transactions, a fee and price impact on every swap, and a position that is briefly riskier between steps. Doing it with a flash loan is one transaction, checked once at the end; see what is a flash loan.

The maths of net APY

With equity E and leverage L (total collateral ÷ E), you earn the collateral yield y on L × E and pay the borrow rate r on (L − 1) × E. Net APY on equity is L × y − (L − 1) × r, before price moves, fees and rewards on the debt side.

Illustrative numbers: an LP earns 12% and borrowing costs 7%. At 2x, net is 24 − 7 = 17%. At 3x, 36 − 14 = 22%. Now let the borrow rate rise to 12%: 3x returns 36 − 24 = 12%, no better than not borrowing, with liquidation risk added. Above 12%, leverage loses money. Lending rates on Blend and XOXNO move with utilisation, so r is not fixed; the mechanics are in DeFi lending platforms.

Two corrections matter. Use APR, not compounded APY, for both sides, or convert consistently (APY vs APR). And if the collateral is an LP token, impermanent loss is part of y, and leverage multiplies it too.

The health factor

Blend's whitepaper defines borrowing capacity as the sum of each collateral position's value times its collateral factor, minus the sum of each liability's value divided by its liability factor. Expressed as a ratio, health factor = (collateral × c) ÷ (debt ÷ l). Below 1, a position can be liquidated.

Lenders that use loan-to-value and liquidation thresholds, as XOXNO does, compute it as collateral × liquidation threshold ÷ debt. Example on XOXNO: $2,000 of an Aquarius LP at a 60% threshold against $1,000 of USDC debt gives 2,000 × 0.6 ÷ 1,000 = 1.2. A health factor of 1.2 means the collateral can lose one-sixth (16.7%) of its value before liquidation.

Liquidation is not the end of the position, but it is expensive. Blend uses Dutch auctions in which the liquidator's terms improve block by block over about 200 blocks; XOXNO's configuration pays a 10% bonus on LP collateral. Either way, you lose more than the debt you shed.

Maximum leverage and liquidation price

For a loop where the borrowed asset buys more collateral, maximum leverage is 1 ÷ (1 − c × l) on Blend and 1 ÷ (1 − LTV) on LTV-based venues. Liquidation price depends on the collateral: a single volatile asset can fall by 1 − 1/h, while a 50/50 LP's volatile side can fall to 1/h² of its price.

Maximum leverage. On Blend with a collateral factor of 0.60 and a liability factor of 0.75, the limit is 1 ÷ (1 − 0.45) ≈ 1.82x. On XOXNO, an Aquarius LP at 50% LTV allows 2x, and USDC in its Stables & FX spoke at 88% LTV allows about 8.3x in theory. At the limit the health factor is already near 1, so the usable leverage is lower.

Liquidation price. If XLM is your collateral and USDC your debt, a health factor of 1.2 means XLM can fall 16.7%. If your collateral is a constant-product XLM/USDC LP, the LP's value moves with the square root of the XLM price, so XLM can fall to 1 ÷ 1.2² ≈ 69% of its starting price, a 30.6% drop, before the same health factor reaches 1. The LP cushions the move but adds impermanent loss. A worked Stellar example is in Aquarius LP tokens as collateral.

Managed vs self-managed leverage

Self-managed means you open the loop on a lending protocol and watch the health factor yourself. Managed means a vault opens and maintains it for depositors. Managed vaults save effort but add a contract, an operator and, depending on design, shared losses.

On XOXNO, positions are NFT accounts fixed to a mode and spoke, and its multiply and flash_position functions build Multiply, Long or Short accounts in one transaction. Each account carries its own health, so a liquidation touches only that account. On Blend you can attach supply, borrow and collateral requests to a flash loan for the same effect, where the pool allows borrowing.

Vaults differ in a way that matters. WhaleHub, the publisher of this article, is building a leveraged Aquarius-LP vault on Blend V2 that runs on testnet only. In its current design the vault holds one Blend position and gives each depositor shares of it, so interest and any liquidation are shared pro-rata across all depositors, including those who did not over-lever. Per-user isolation is planned, not built.

Stellar venues

Blend V2 is Stellar's largest lending protocol, with health-checked flash loans and isolated pools, but its main Fixed V2 pool was closed to new borrowing on 8 October. XOXNO Lending has more leverage tools (multiply, flash positions, LP collateral at 50% LTV) but held only about $129,000 on DefiLlama.

Blend: per-pool collateral and liability factors, Dutch-auction liquidations, a backstop that takes first loss, and a flash loan that opens debt only if the position ends healthy. Pools are isolated, so check the specific pool's status and oracle; see what is Blend. XOXNO: shared liquidity hubs with risk set per spoke, collateral-only RWA and LP markets, and flash loans at 0.09%; see XOXNO Lending on Stellar and Blend vs XOXNO.

How these fit together

The three numbers belong together. Net APY says whether the trade is worth doing, the health factor how close you are to the edge, and the liquidation price what market move pushes you over. Venue choice decides the parameters and the liquidation penalty; managed or self-managed decides who watches the position and who shares the losses.

The takeaway

Leverage turns a yield spread into a bigger yield spread and a price move into a bigger loss. Before looping, write down L, y and r, compute your health factor and liquidation price, and decide what you will do when rates or prices move against you. On Stellar today the deepest venue has paused new borrowing in its main pool and the most flexible one is small, so size positions accordingly.

Sources: Blend whitepaper; github.com/blend-capital/blend-contracts-v2 (pool flash_loan and submit code); github.com/XOXNO/rs-lending-xlm @ 598aa1f (configs/mainnet spokes.json and markets.json, skills/xoxno-lending-contracts positions.md and flash-loans.md); DefiLlama protocols and yields APIs, 9 October 2026; Blend Fixed V2 pool status read on 8 October 2026; WhaleHub leverage-vault source and design notes. Examples with round numbers are illustrative. WhaleHub is the publisher of this article.