[ARFC] Improve Liquidity Buffer for USDC on Ethereum Core – Raise Slope 2, lower optimal utilization
[This post is personal views only, not representing those of Circle]
Summary
USDC on Aave v3 Ethereum Core has been essentially pinned at full utilization for four days. Variable borrow rate has been flat at the post-kink ceiling (~14%) the entire time, and pool supply contracted by ~$60M in the last 24 hours as repayments were matched dollar-for-dollar by queued withdrawals. The rate is not clearing the market. I think a parameter recalibration is warranted. Specifically:
- Slope 2: ~10% → 50% target, with an interim Risk Steward step at 40%.
- Optimal utilization (U^*): 92% → 85% target, with an interim step at 87%.
- Slope 1: hold at 3.5%. Base rate: hold at 0%. Reserve factor: hold at 10%.
- Slope 2 Risk Oracle: pause or floor for USDC during this incident; re-engage when conditions normalize.
The intent is narrow: restore the post-kink region as a functional price discovery range so the pool can clear through price instead of through the withdrawal queue. The added benefit is also to shift the effective maturity of the pool from unknown to near-instantaneous again with under 100% utilization rate. The latter has important implications in attracting capital inflows to restore a healthy onchain market.
The practical path I’d suggest is a Risk Steward action by @LlamaRisk + @Aave Labs on the interim parameters, followed by full-governance ratification on the target within a week. LlamaRisk has the data, the authority (via the 2/2 Risk Steward multisig), and — per their continuity statement — explicit manual-AGRS coverage of Slope 1, Slope 2, Base, and U^*. I’d defer to their judgment on exact values; the numbers below are a starting point.
1. Current state
Aavescan snapshot, 2026-04-22:
| Metric | Value | 24h Δ |
|---|---|---|
| Utilization | 99.87% | essentially unchanged |
| Total supplied | $1.89B | ≈ −$60M |
| Total borrowed | $1.89B | ≈ −$60M |
| Available liquidity | < $3M | thin |
| Variable borrow rate | 13.82% | near ceiling |
| Supply rate | 12.42% | near ceiling |
Pool supply and debt have shrunk roughly dollar-for-dollar over the past 24 hours. Repayments are matched by queued withdrawals; net new deposits and net new borrows are close to zero.
Source: USDC on Ethereum V3 | Aavescan
2. Why the rate isn’t clearing
The marginal-borrower mix appears to have shifted since the April 18 rsETH event. External reporting (CoinDesk; Aave forum) describes ~$300M of incremental borrow flow in the 72h following the exploit, dominated by trapped-liquidity extraction:
- Stuck USDC/USDT suppliers borrowing other stables against their own deposits and exiting via DEX.
- Users whose WETH is on Core (unfrozen but with LTV=0) or on still-frozen Prime / L2 deployments, reaching for dollar liquidity through whatever non-WETH collateral they have available.
- Users holding non-WETH collateral (WBTC, other stables) drawing USDC to extract dollar liquidity from positions whose normal exit paths are temporarily blocked.
These borrowers are structurally rate-insensitive over short horizons. A user accepting a potential dobule digit headline loss to exit via swap is treating the borrow rate as a fee for bypassing the queue; at 14%, one week of carry costs ~27 bps and one month ~117 bps — small compared to the loss they’re already absorbing. They likely don’t unwind until rates are multiples of current levels.
On the supply side, the current rate ceiling is competitive but not a dominant strategy given liquidity issues. An allocator pricing in the observable illiquidity require much more of a premium that the current ceiling doesn’t clearly provide.
Even though supplying USDC on Aave is safe, the illiquidity of the pool requires more compensation, as the instrument have shifted from one with instantenous liquidity to one with uncertain maturity.
The active lever is supply attraction. If a meaningful share of borrowers are rate-insensitive, steepening Slope 2 works primarily by making the pool an irresistible destination for new LP capital. Lifting the rate ceiling to 35–50% range should incentivize inflows within hours, pushing utilization back below the kink and allowing rates to re-anchor automatically. This is how Treasury repo, fed funds, and commercial paper clear: price spikes, capital arrives, price normalizes. The kinked IRM was designed to reproduce this behavior; at the current S_2 = 10\%, it’s under-amplified for this specific shock.
Deadweight loss As with any price ceiling, restricting free market clearing leads to suboptimal consumer welfare and deadweight loss. This inefficiency leads to deterioration of markets and exit of participants that seek more competitive venues over time.
3. Parameters
3.1 Current parameters (inferred)
| Parameter | Value |
|---|---|
| Base variable borrow rate (R_0) | 0% |
| Slope 1 (S_1) | ~3.5% |
| Slope 2 (S_2) | ~10% |
| Optimal utilization (U^*) | 92% |
| Reserve factor | 10% |
3.2 Max-rate math at various Slope 2 values
With R_0 = 0 and S_1 = 3.5\% held fixed, variable borrow APR at U=100% equals S_1 + S_2. Supply rate at U=100% = borrow rate × (1 − reserve factor) = borrow rate × 0.9.
| Slope 2 | Max borrow rate at U=100% | Max supply rate | Assessment |
|---|---|---|---|
| 10% (current) | 14.0% | 12.6% | Current state — does not clear |
| 25% | 28.5% | 25.7% | Plausible but likely insufficient given rate-insensitive borrower share |
| 40% (proposed interim) | 43.5% | 39.2% | Likely clears via supply attraction within hours |
| 50% (proposed target) | 53.5% | 48.2% | Comfortable margin for additional stress |
| 75% | 78.5% | 70.7% | Incremental deterrence minimal; larger discontinuity risk at kink |
3.3 Proposed two-step path
Interim (Risk Steward action, same day):
| Parameter | From | To | Δ |
|---|---|---|---|
| Slope 2 | ~10% | 40% | +29.5 pp |
| Optimal utilization | 92% | 87% | −5 pp |
| Slope 1 | 3.5% | 3.5% | — |
| Slope 2 Risk Oracle (USDC) | active | pause or floor at static S_2 | — |
Target (full-governance ratification, 5–7 days):
| Parameter | Interim | Target | Δ |
|---|---|---|---|
| Slope 2 | 40% | 50% | +10 pp |
| Optimal utilization | 87% | 85% | −2 pp |
The interim lands closer to the target than to the status quo intentionally. If borrower-side demand is rate-insensitive and the active lever is supply attraction, a modest step (say S_2 = 25\%) may produce only partial rotation and leave the pool still pinned — burning a steward action window without producing clearing. A single decisive move, with the option to soften afterwards, looks more efficient. That said, this calculus depends on the marginal-borrower-elasticity assumption; LlamaRisk may reach a different conclusion on the evidence they hold.
3.4 Resulting rate curve
| Utilization | Current | Interim (S2=40%, U*=87%) | Target (S2=50%, U*=85%) |
|---|---|---|---|
| 50% | 1.9% | 2.0% | 2.1% |
| 80% | 3.0% | 3.2% | 3.3% |
| 85% | 3.2% | 3.4% | 3.5% (kink) |
| 87% | 3.3% | 3.5% (kink) | 10.2% |
| 90% | 3.4% | 12.7% | 20.2% |
| 92% | 3.5% (kink) | 19.1% | 26.8% |
| 95% | 5.4% | 28.7% | 36.8% |
| 97% | 7.1% | 35.1% | 43.5% |
| 100% | 14.0% (observed) | 43.5% | 53.5% |
Below 85% utilization — the pool’s normal operating regime — the proposed curves essentially coincide with the current curve. The change is structural only above the kink.
3.5 Why these specific values
S_2 = 50\% at target. At U=100% this produces a supply rate around 48%, competitive with every onchain and offchain allocation by a wide enough margin to pull capital in within hours. 75% adds little additional deterrent while worsening rate discontinuities near the kink. 30% risks leaving rate-insensitive borrowers in place.
U^* = 85\% at target. The current 8 pp of headroom above the kink was consumed in hours on April 18. 15 pp is closer to what a pool of this scale probably wants; cost is ~30–40 bps of incremental borrow rate at typical healthy-regime utilization. Reasonable cases exist for 80% (more defensive) or 88% (less disruption).
Slope 2 Risk Oracle pause for USDC. Two reasons. First, empirical: LlamaRisk’s own February 2026 retrospective documented that during the WETH utilization spike, the oracle’s first response was a reduction of Slope 2 from 8% to 6.5% (below its published floor), that Slope 2 reached only 9.75% some 28 hours after the breach, and that during a follow-on spike at 99.85% utilization the oracle “did not respond for 43 hours.” The retrospective’s language — the oracle “diverged from its specification” — is LlamaRisk’s. Second, operational: the oracle was built by Chaos Labs, whose orderly exit from Aave on 2026-04-06 means ongoing support is no longer contractually provided. During an active stress event, a static parameter under direct steward control is operationally simpler. The oracle can be re-engaged in calmer conditions once its maintenance story is settled. None of this is a criticism of Chaos Labs — they contributed three years of high-quality work to Aave and their offboarding was professional.
4. Risk considerations
4.1 Supply rotation
A supply rate approaching 40–50% under stress should pull USDC from sophisticated allocators across venues. This is the primary intended clearing mechanism. Rates remain elevated only as long as utilization stays above the kink; decay is automatic as supply arrives. My sense is the rotation takes hours, not days, but LlamaRisk’s elasticity views are more authoritative.
4.2 Repayment interaction
Borrowers repaying at elevated rate incur one-time interest at the rate they repay at; for repayment windows of hours to a day, the cost is basis points. The IRM applies continuously — no cliff.
4.3 Asymmetry with the WETH IRM flattening
The DAO flattened WETH’s IRM on April 20 (S1→2%, S2→3%, U*→94% on Core) to avoid rate-driven liquidations of the stuck rsETH-backed WETH positions during socialization. That action was pool-specific and, in context, reasonable. USDC is different: the asset is healthy; the only stress is withdrawal demand. Flattening WETH and steepening USDC are complementary responses to different problems, not contradictory signals.
4.6 What I’d welcome LlamaRisk’s view on
Short list where LlamaRisk’s data or judgment should override mine:
- The actual marginal-borrower composition post-April 18 (above is a read from public flows, not address-level data).
- Whether the interim should in fact be closer to 25% than 40% — a gentler step may be appropriate if my rate-insensitivity read is overstated.
- Cross-instance implications for USDC on Arbitrum, Base, Polygon, Avalanche, Optimism, Linea.
- Disposition of the Slope 2 Risk Oracle during the Chaos Labs transition window.
5. Execution
- Risk Steward action, same day. LlamaRisk and Aave Labs execute the interim parameters (S_2 = 40\%, U^* = 87\%) and the USDC-specific oracle pause/floor via the 2/2 multisig and manual AGRS. The scope is within the envelope of prior steward IRM actions (including the April 20 WETH flattening and the February 7 steward bypass of the oracle during the WETH incident).
- Full-governance ratification within 5–7 days. Standard ARFC → Snapshot → onchain vote targeting S_2 = 50\%, U^* = 85\%.
- Protocol Emergency Guardian fallback. The 5-of-9
EMERGENCY_ADMINmultisig is available if stewards decline to act and conditions continue to deteriorate. This is a contingency.
Follow-on proposals for USDC on Arbitrum, Base, Polygon, Avalanche, Optimism, and Linea should follow this one if those pools show similar pinning; I’d leave that call to LlamaRisk as the venue-specific data comes in.
6. Open to input
Pushback welcome — especially from LlamaRisk and Aave Labs on the inferences above, from Chaos Labs on the Slope 2 Risk Oracle’s intended operating range (their perspective would sharpen §3.5), and from other risk researchers on the clearing-rate estimate. The narrow question is whether current parameters produce a clearing market; if not, what parameter adjustment best restores that property. I’ve offered one answer above and would find a different one useful, not less so.
7. Specification (machine-readable)
market: aave-v3-ethereum-core
asset: USDC
reserve_address: "0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48"
# Interim — Risk Steward action (same day)
irm_interim:
baseVariableBorrowRate: 0.00 # RAY
variableRateSlope1: 0.035
variableRateSlope2: 0.40
optimalUsageRatio: 0.87
reserveFactor: 0.10
# Target — full-governance ratification
irm_target:
baseVariableBorrowRate: 0.00
variableRateSlope1: 0.035
variableRateSlope2: 0.50
optimalUsageRatio: 0.85
reserveFactor: 0.10
slope2_risk_oracle:
mode: "paused_for_USDC" # alternative: "floored" at interim=0.40 / target=0.50
8. References
- USDC on Ethereum v3 (Aavescan): USDC on Ethereum V3 | Aavescan
- rsETH Incident Report (Aave governance, 2026-04-20): rsETH Incident Report (April 20, 2026)
- rsETH incident — 2026-04-18 (Aave governance): rsETH incident — 2026-04-18
- LlamaRisk — Ensuring Continuity of Aave’s Risk Management: LlamaRisk: Ensuring Continuity of Aave's Risk Management
- [ARFC] Renew LlamaRisk as Risk Service Provider — Epoch 4: [ARFC] Renew LlamaRisk as Risk Service Provider - epoch 4
- Orderly Transition and Offboarding Plan for Chaos Labs: Orderly Transition and Offboarding Plan for Chaos Labs
- Retro: WETH utilization spike and Slope 2 Risk Oracle performance (Feb 2026): Retro: WETH utilization spike and Slope2 Risk Oracle performance
- Chaos Labs — Introducing the Aave Slope 2 Risk Oracle: Introducing the Aave Slope2 Risk Oracle | Chaos Labs - Industry Leading Risk Management, Economic Security, and Incentive Optimization
- Chaos Labs exits Aave (CoinDesk, 2026-04-06): Aave loses key risk manager Chaos Labs amid contributor exodus and disputes
- CoinDesk — “A $300M borrowing spike on Aave signals liquidity crunch after KelpDAO exploit” (2026-04-20): A $300 million borrowing spike on Aave signals liquidity crunch after KelpDAO exploit
- CoinDesk — “Aave faces ‘serious trouble’ as all its core markets hit 100% utilization” (2026-04-21): Aave faces 'serious trouble' as all its core markets hit 100% utilization. What this means.
- Aave v3 Interest Rate Strategy documentation: Interest Rate Strategy | Aave Protocol Documentation
- Aave Governance documentation: Governance | Aave Protocol Documentation
- Companion data analysis:
circle-research-usdc-aave-irm-analysis.md
