A stablecoin holds a peg through arbitrage, not through the number printed on it. Every design works by making it profitable for someone to buy the token when it trades below the target and sell it when it trades above. What differs is what backs that trade — and each backing has a specific way of failing.
Fiat-backed tokens are the simplest. An issuer takes dollars, issues tokens, and holds the dollars in cash and short-term government paper. Authorised participants can redeem at par, so a token trading below a dollar is a profitable purchase for anyone with redemption access. The peg is therefore as strong as the redemption promise, and the three things that weaken it are reserve composition, banking access and the breadth of redemption rights. When Silicon Valley Bank failed in March 2023, a stablecoin issuer disclosed that part of its reserve was held there; the token traded below a dollar over a weekend when redemption could not clear and recovered when the deposits were guaranteed. Nothing about the token had changed — the banking rail had.
Crypto-collateralised tokens hold volatile assets in excess of the debt they issue. A holder deposits collateral worth more than the stablecoin minted against it, and if the ratio falls below a threshold, the position is liquidated by third parties competing for a discount. The peg holds because minting and redeeming against the collateral is arbitrageable. The failure mode is the liquidation machinery under stress: on Ethereum's worst day in March 2020, congestion and thin auction participation meant some liquidations cleared at effectively zero, leaving the system undercollateralised and requiring a governance-run recapitalisation. The mechanism did not fail quietly; it failed in exactly the conditions it existed for.
Algorithmic designs hold no external collateral. The best-known depended on a mint-and-burn relationship with a second, freely floating token: the stablecoin could always be redeemed for a dollar's worth of that token, and vice versa. This is a peg that works while the market believes the second token is worth something, and it is reflexive — falling demand for the stablecoin means printing more of the volatile token, which depresses its price, which requires printing more. In May 2022 that spiral ran to completion in days and destroyed both assets. The design's collapse was not an implementation bug; it was the mechanism operating as specified in the direction nobody had priced.
Two structural risks cut across all three. Concentration: a token used as the quote asset in most trading pairs is a single point of failure for the venues that use it, and a depeg propagates into every position priced against it. And redemption asymmetry: retail holders in most designs cannot redeem with the issuer directly and must sell into the market, so the arbitrage that defends the peg is available to a small set of participants while the price risk is borne by everyone.
This is the context in which reserve disclosure and regulation became the sector's central argument. Attestations state what is held at a moment; audits examine controls over time; and the frameworks now in force in the European Union and the United States impose reserve composition and redemption requirements directly. What all of them are trying to fix is the same thing: the peg is a promise, and a promise is only as good as the ability to test it on demand.