Reserve Mix and Run Thresholds: Calibrating Stablecoin Liquidity

The 2023 de-pegging of USDC, even briefly, exposed a fundamental tension in stablecoin design: the tradeoff between yield and liquidity.

Circle holds its reserves primarily in short-term U.S. Treasury bills and cash. This is conservative by design, prioritizing safety and redeemability. But it also means USDC holders earn zero yield. Meanwhile, DeFi protocols offer 5-15% APY on stablecoin deposits by taking on additional risk.

This raises a question: what's the optimal reserve composition for a stablecoin issuer? How much liquidity is "enough" to withstand redemption pressure, and how much risk can you take to generate yield?

The Liquidity Ladder Framework

Think of stablecoin reserves as a liquidity ladder:

• Tier 1: Cash and overnight deposits (instantly liquid)
• Tier 2: Short-term Treasuries, <3 months (highly liquid, minimal price risk)
• Tier 3: Longer-term Treasuries, 3-12 months (liquid but subject to interest rate risk)
• Tier 4: Investment-grade corporate bonds (less liquid, credit risk)
• Tier 5: Yielding DeFi positions (illiquid, protocol risk)

The further down the ladder you go, the higher the yield, but also the higher the risk that you can't access funds when needed.

A purely Tier 1+2 reserve structure (like USDC) maximizes safety but earns minimal yield. A structure that includes Tier 4-5 assets (like early algorithmic stablecoins) offers higher returns but creates redemption risk.

Modeling Redemption Dynamics

To optimize the reserve mix, we need to model redemption behavior under stress. The key question: at what point do redemptions accelerate into a "run"?

I built a simple agent-based model with three types of holders:

• Sticky holders (40%): Long-term believers who only redeem under extreme stress
• Opportunistic holders (40%): Monitor the peg; redeem if it breaks 0.98 or if redemption queues form
• Algorithmic holders (20%): Programmatic traders who exit immediately on any deviation

Under normal conditions, daily redemptions are 0.1-0.5% of supply. But if the peg breaks, algorithmic holders exit immediately (20% redemption spike), which triggers opportunistic holders to follow (another 10-20% within 24 hours).

So the critical threshold is: can you handle 30-40% redemptions within 24 hours?

Calibrating on USDC Data

During the Silicon Valley Bank collapse in March 2023, USDC temporarily de-pegged because Circle had $3.3B (8% of reserves) stuck in SVB. Redemptions spiked to $10B over 48 hours, about 13% of total supply.

This was contained because:

• The majority of reserves (92%) were in liquid Treasuries
• Circle maintained access to banking partners for redemptions
• The crisis resolved quickly (SVB depositors were made whole)

But what if reserves were less liquid? Say 20% in longer-dated Treasuries or corporate bonds? The math suggests a full-blown run would have been likely, because Circle couldn't meet redemptions fast enough.

The Optimal Reserve Mix

Based on this analysis, here's my proposed framework for a stablecoin reserve targeting stability + moderate yield:

• 40% Tier 1 (cash, overnight): Handles daily operational redemptions
• 40% Tier 2 (T-bills <3 months): Can be liquidated within 24 hours with minimal loss
• 15% Tier 3 (Treasuries 3-12 months): Adds yield, liquidate-able within 48 hours
• 5% Tier 4 (corporate bonds, repo): Additional yield, backstop only

This structure ensures that 80% of reserves are accessible within 24 hours, which should handle even a severe run. The Tier 3-4 allocation generates 50-100 bps of additional yield without materially increasing risk.

The Yield-Safety Tradeoff

Of course, this is more conservative than what some issuers might prefer. Tether, for instance, holds a portion of its reserves in corporate loans and Bitcoin, higher-yielding but much less liquid.

This works for Tether because:

1. It has a long track record, so redemption pressure is lower
2. It primarily serves exchanges and institutions, not DeFi contracts
3. It can be selective about redemptions (slow-walk them if needed)

But for a new stablecoin trying to build trust, or one heavily used in DeFi (where algorithmic redemptions are common), a more conservative structure is necessary.

Implications for DeFi

This analysis also has implications for DeFi protocols that hold stablecoins. If you're building a lending market or DEX, you need to understand the liquidity profile of the stablecoins you support.

USDC is highly redeemable, if there's a crisis, users can exit to fiat quickly. DAI is backed by a mix of USDC, ETH, and other assets, redemption is slower and subject to price volatility. Algorithmic stablecoins have no reserves at all, "redemption" is just selling at whatever price the market offers.

Each of these has different systemic risk properties. A lending protocol that accepts all three as equal collateral is taking on hidden tail risk.

Conclusion

Stablecoin design is fundamentally about liquidity management under stress. The goal isn't to maximize yield, it's to ensure you can meet redemptions when it matters most.

A well-designed reserve structure holds enough liquid assets to handle predictable stress scenarios (30-40% redemptions over 24-48 hours), while optimizing the remaining allocation for yield.

Getting this balance right is the difference between a stablecoin that survives crises and one that doesn't. And in DeFi, where leverage compounds quickly, the stakes are existential.