wstETH/WETH loop — borrow-rate spikes and the hold-vs-unwind decision

TL;DR — the decision rule

Hold through. Every observed spike in the 2-year window mean-reverted in 1–3 days, and the wstETH/WETH on-chain discount widens during these stress windows (precisely because panicking loopers are dumping wstETH). At the worst observed spike (Aave WETH borrow 17.54% on 2025-07-16), holding one day cost ~4 bps per dollar of debt while the DEX exit cost was 8–28 bps. The decision rule:

Unwind only if discount_bps + fee_bps < (peak_rate − stETH_yield) × expected_spike_days / 365 × 10_000.

With historically observed spike durations of ≤3 days and DEX discounts that widen during spikes (not shrink), the right-hand side is almost never larger than the left. Hold.

The single scenario where unwinding pays: discount stays narrow (< 5 bps) AND the rate sits >5pp above stETH yield for >5 days. We have not observed that combination in this dataset.

Question

Looping wstETH against WETH is structurally exposed to the WETH borrow rate. When utilization on Aave/Spark/Morpho spikes (typical cause: a wave of new loop entries pushing supply utilization toward 100%), the IRM kinks borrow rates upward — sometimes dramatically (peak observed: 17.54% on Aave).

A position holder has two options:

• Hold through the spike, paying the elevated rate while it lasts.
• Unwind — sell wstETH on a DEX to repay WETH and exit the loop. This crystallises the wstETH/WETH discount (DEX prices wstETH below the on-chain stEthPerToken rate during stress).

We wanted a defensible decision rule for the operator on call, parameterised by observable inputs (peak rate, current DEX discount, stETH yield).

Methodology

Findings

Five worst spike weeks in the 2-year window

“Break-even hold-through duration” = the unwind cost (discount + ~3 bps fee) divided by the daily hold cost ((peak_rate − stETH_yield) / 365 expressed in bps). If the actual spike duration is shorter than this, holding through is the right call.

The worst spike densified — Aave WETH borrow + wstETH DEX discount, daily

The peak was a single-day event. By July 18 (2 days later) the rate had dropped from 17.54% to 4.70%; by July 20 (4 days later) it was 2.66%, fully normal. The wstETH discount peaked on day 2, after the rate had already partially mean-reverted — confirming the pattern that loopers panicking is what drives the discount, and that the panic itself is poorly timed.

Two-year averages (n = 53 bi-weekly samples)

All three venues’ borrow rates spike together (2025-07-22 saw Aave 7.62%, Morpho 6.52%, Spark held lower at 3.24%; 2025-03-17 saw Spark 5.97% with the others near 2.7%). Cross-venue spike correlation is high but not perfect — a spike on one venue does not guarantee a spike on the others, which is itself useful information (relocating across venues is a partial hedge).

Decision framework

Per dollar of WETH debt:

• Hold cost per day = (borrow_rate − stETH_yield) / 365, expressed in bps. At the worst observed spike (17.54% rate, 2.5% stETH): 15 pp negative carry → 4.1 bps/day.
• Unwind cost = wstETH/WETH DEX discount + Curve fee (~3 bps). Observe discount in real time; do not assume the historical median.

The rule:

unwind if:  discount_bps + 3 < (peak_rate − stETH_yield) × expected_spike_days / 365 × 10_000
hold otherwise

For the operator on call: if the DEX discount is wider than the gross negative-carry over 3 days, hold. It always has been.

Failure modes (re-run the study when any of these change):

• Spike durations have lengthened. The empirical pattern (≤3 days, fast mean-reversion) is driven by the symmetric IRM curves on Aave/Spark/Morpho and by competing loopers all unwinding simultaneously. A regime change — e.g., a venue switches to a more aggressive utilization-2 kink, or a structural shift in looping demand — would invalidate the duration assumption.
• DEX discount no longer widens during stress. Today the discount is concurrent with the spike (panicked loopers sell). If wstETH gets new natural buyers during stress (e.g., a market-neutral arbitrage product), the discount could compress and unwinding could become cheaper than holding.
• Cross-venue rates decorrelate enough that relocation pays before unwinding does. Today, peak rates often coincide across two venues; if one stays low while the other spikes, moving the loop is cheaper than either unwinding or holding.

Caveats

• We measured the marginal DEX price (1-stETH probe on Curve), not the size-impact cost of unwinding $5M–$50M of position. For larger positions, the actual unwind cost is higher than the discount-bps shown. Spark and Morpho main markets currently support $9M–$50M of instant exit before per-trade impact rises materially; sizing the unwind cost properly for $30M+ positions requires a separate liquidity study.
• We did not measure the impact of unwinding across venues — e.g., partially unwinding the Aave leg while leaving the Spark leg open. The decision rule above treats each loop as a single block.
• The bi-weekly sampling can miss intra-fortnight peaks. The densification around 2025-07 showed the peak was on a non-sampled day (2025-07-16) and that the sampled day captured a partial recovery. Worst observed in the dataset is therefore a lower bound on the actual 2-year worst rate.
• We did not estimate the probability that any future spike will mean-revert as fast as observed ones. The conclusion is empirical, not model-based.