PhilidorPhilidor Research
Thematic Report

Three Ways a Stablecoin Breaks: Reflexivity, Reserves, and Contagion

  • Philidor ResearchRisk Desk, Philidor Labs
7 min read

The assumption underneath on-chain credit

Lending markets quote a stablecoin as the risk-free leg. Vaults denominate returns in it. Collateral is valued against it. When the peg holds, the assumption is invisible; when it breaks, it is the only thing that matters. The useful question is therefore not "is this stablecoin safe" but "how would it break, and is that failure mode correlated with the others I hold." Three real events answer the first half precisely.

Failure mode 1 — Reflexivity (TerraUSD, May 2022)

TerraUSD (UST) was an algorithmic stablecoin: its peg was defended not by reserves but by an arbitrage with its sister token, LUNA. One UST could always be burned for $1 of newly minted LUNA, and vice versa. In calm markets the arbitrage works. Under stress it inverts into a death spiral — the peg is defended by a token whose value depends on confidence in the peg.

When UST slipped below $1 around 7–9 May 2022, redemptions minted LUNA at an exponential rate. LUNA's supply hyperinflated from roughly 350 million to over 6.5 trillion tokens in days; UST fell to about $0.10 by 13 May; and an ecosystem worth tens of billions — sustained partly by Anchor Protocol's ~20% yield — evaporated, with estimates of ~$40–60B in value destroyed. The Luna Foundation Guard deployed roughly $3B in Bitcoin to defend the peg and failed (Harvard Law CGR, ecos.am).

The signature: an endogenously-backed peg — defended by a token the protocol itself can mint — has no exogenous floor. Once confidence breaks, the defence accelerates the collapse. This is a pure asset-vector failure, and it is non-recoverable: there is no reserve to redeem against.

Failure mode 2 — Reserves (USDC, March 2023)

USDC is the opposite design: fully reserved by off-chain cash and short-dated treasuries. Its March 2023 depeg had nothing to do with its mechanism and everything to do with where the reserves sat. Circle disclosed that $3.3 billion of USDC's backing — roughly 8% of reserves — was held at Silicon Valley Bank, which had just failed. By ~2am on 11 March, with the cash apparently trapped, USDC traded down to $0.87 (CNBC, Circle / CoinDesk).

The recovery was equally telling about the failure mode. USDC re-pegged not through any on-chain mechanism but when US regulators backstopped SVB depositors over the weekend and Circle resumed redemptions on Monday 13 March (CoinDesk). A later Federal Reserve note studied exactly this transmission from a bank run to stablecoins (Federal Reserve).

The signature: a fully-backed coin can still depeg if its reserves carry counterparty and banking risk. The peg is only as good as the weakest custodian of the backing. The failure is recoverable — the assets existed — but the holder bears mark-to-market and gating risk in the interim.

Failure mode 3 — Contagion (DAI, March 2023)

DAI is the case that ties the room together. It is overcollateralized and nominally decentralized — yet it dropped to about $0.88 the same weekend USDC did. The reason: a large share of DAI's backing was USDC, held through MakerDAO's peg-stability module. A "decentralized" stablecoin had inherited a centralized stablecoin's banking risk wholesale (CoinDesk). MakerDAO subsequently debated emergency measures and ultimately voted to retain USDC as its primary reserve (CoinDesk).

The signature: a stablecoin that is collateralized by another stablecoin does not diversify the peg — it transmits it. DAI's depeg was not an independent event; it was USDC's event, repriced one hop downstream.

The common thread: concentration meets correlation

Read together, the three failures expose the property that makes stablecoin risk systemic rather than idiosyncratic:

  • Concentration — a small number of coins back the majority of on-chain credit, so a single failure is large.
  • Correlation — the failure modes are linked. DAI was correlated to USDC by collateral; the entire fiat-backed class is correlated by shared banking rails; algorithmic designs are correlated to each other by the same reflexive logic.

A portfolio that diversifies across vaults but holds USDC, DAI, and a third USDC-collateralized coin is not diversified against the risk that matters. It is three expressions of one banking-reserve exposure. The unit of diversification that actually reduces risk is the peg-defence mechanism, not the issuer brand.

Implications for risk scoring

The taxonomy maps cleanly onto how a peg should enter a risk model:

  1. Classify by failure mode, not by name. Reflexive/algorithmic, reserved/ counterparty, and collateral-contagion are different risks with different tails. Philidor's methodology distinguishes synthetic and fiat-backed stablecoin classes for exactly this reason — the class determines the failure shape.
  2. Score the peg as a live input, not a static label. A coin is not "a stablecoin" forever; it is at $1.00 today and could be at $0.87 tomorrow. Philidor runs live peg-stability monitoring for stablecoin-class assets, with a cooldown and automatic recovery on re-peg, feeding the asset vector continuously rather than assuming par.
  3. Look through to collateral. A stablecoin backed by other stablecoins inherits their failure modes. The asset vector has to score what backs the backing — DAI's risk in March 2023 was legible only to anyone who looked through to its USDC exposure.

Peg events are continuous, not occasional. Recent critical incidents across the tracked universe — overwhelmingly depeg events — illustrate the point:

Loading events…

Conclusion

"Is this stablecoin safe?" is the wrong question because it has no static answer. TerraUSD failed because its peg had no exogenous floor; USDC wobbled because its floor sat in a failing bank; DAI dropped because its floor was USDC. Three designs, three failure modes, one lesson: price the mechanism and the collateral path, monitor the peg as a live signal, and treat issuer brand as the least informative variable in the set.

References

Print / Save as PDF