Leverage loops, visualized
A leverage loop multiplies the yield by N — and the liquidation risk by N. Here is exactly what that means, and what it actually cost in our advisory lab: a real, dated liquidation tail. The desk refuses this for live capital.
What it is
Deposit → Borrow → Redeposit, ×N
Each turn of the loop stacks another layer of the same position on borrowed funds. N turns ≈ N× exposure: N× the yield, but also N× the sensitivity to a drop in the collateral price.
Deposit
Supply collateral (e.g. wstETH) to a lending market.
Borrow
Borrow a stablecoin against it, up to the LTV limit.
Redeposit
Swap back to collateral and supply again — repeat ×N.
Grows with N — the attractive side
N× yield
A 3rd turn can turn ~5% into a double-digit headline — that is the lure.
Grows with N — the dangerous side
N× liquidation risk
The same multiplier hits the tail. A smaller drop in the collateral price is enough to wipe the whole position.
The risk mechanic
Health factor & liquidation price
While the collateral value covers the debt with margin, the health factor is > 1. More turns → thinner margin → higher liquidation price → closer to the cliff. Example S73: a wstETH 2× loop on Aave v3, liquidation tail at 0.825 LTV.
More turns slide the marker left, toward the cliff. At 1.0 the position is liquidated — the collateral is sold at a penalty and the loss is locked in. No kill switch can "undo" a liquidation that already fired.
What looping actually cost
The real, dated liquidation tail
This is not an illustration. It is the leverage_loop row from the honest multi-metric scorecard of our advisory lab — the headline yield next to the tail it paid for.
Headline APY (the lure)
15.0%
What the loop advertises.
Backtest realized (the truth)
-8.95%
2024-07 → 2026-05: the return was negative. Sharpe ≈ -0.67.
Max drawdown
-27.94%
Realized peak-to-trough in the backtest.
Worst liquidation tail (stress)
-35.22%
During the USDe leverage-unwind window (October 2025) — the dated, worst test for the loop — the in-sample drawdown reached -13.02%, and the modeled shock crashed the position by ~35% and did not recover. That is the price the loop paid for its 15% headline.
Why this is class C
The headline yield is risk-compensation: the market pays more precisely because it carries a liquidation tail. Class C labels this honestly — the yield is not "free edge," it is payment for a tail that sometimes arrives — as it did here.
The desk's stance
We refuse this for live capital
| Property | Value | What it means |
|---|---|---|
| Lives in | ADVISORY AGGRESSIVE LAB | Paper-only. Shown so the owner can SEE the risk — not for allocation. |
| Live allocation | NEVER | outside_riskpolicy=true, live_eligible=false. Never touches the $100k go-live track. |
| Risk class | C | Risk-compensation — yield = payment for the liquidation tail. |
| Promotion to live | OWNER + CUSTODY GATED | Even a "select" under SPA_AGGRESSIVE_LAB_SELECT allocates nothing — requires an explicit owner + custody decision. |
What the desk holds instead
The desk's live capital is ~4.5% stablecoin lending with no open liquidation risk. Leverage loops are something we assess in the advisory lab, eyes open on the tail — and do not hold.