A stablecoin is not a currency. It is a promise with a mechanism attached, and the mechanism is the only part worth analysing. Two coins can both trade at $1.00 for years and fail in completely different ways — one when its issuer's reserves wobble, another when its collateral ratio hits a reflexivity loop it cannot escape.
The trilemma, stated honestly
You can have price stability, decentralisation, and capital efficiency. You can have two.
Fiat-backed coins pick stability and efficiency and pay in trust: you are betting on a regulated issuer's reserves and its willingness to redeem. Crypto-collateralised coins pick stability and decentralisation and pay in capital: you need more collateral than the coin is worth, which makes the system expensive to run. Algorithmic designs tried to pick efficiency and decentralisation, and the ones that abandoned collateral entirely are mostly dead. The honest reading is that the trilemma is real and every current design has picked its corner.
The three archetypes
Fiat-backed. USDT (2014) remains the deepest, backed by cash, Treasuries, reverse repos, and some corporate paper — a broad mandate with historically opaque disclosure. USDC holds cash and short-term Treasuries only and publishes monthly attestations. PYUSD, issued by Paxos for PayPal, runs the same narrow-reserve model and exists mainly to move PayPal's user base on-chain. Reserve quality is the whole question, and a third-party attestation is weaker than a GAAP audit but considerably stronger than a blog post.
The reserve yield is the issuer's actual business. Interest earned on Treasuries belongs to the issuer unless the coin passes it through, which is why newer entrants compete by sharing it — Ethena's USDe and Ondo's USD0 are built around that arbitrage. That competitive pressure is also the source of the odd regulatory question: is a yield-bearing dollar a deposit?
Crypto-collateralised. MakerDAO's DAI (and its successor USDS) is the reference implementation: overcollateralised debt positions, a liquidation engine, and a governance-controlled stability fee that steers the peg. Liquity's LUSD is the austere version — immutable, no governance, a fixed collateral type. These coins are transparent and expensive, and their peg depends on arbitrage plus the willingness of holders to participate when the mechanism demands it.
Algorithmic. UST was the large-scale experiment and it failed in May 2022, not because of an oracle bug but because the design needed continuous demand growth to hold. FRAX began partially algorithmic and steadily raised its collateral ratio, which is the market's verdict stated as a product decision: the hybrid won on the way to being fully collateralised.
How a peg is actually maintained
Three forces, in order of reliability.
Redemption. If you can always exchange the coin for the underlying at par, arbitrage does the work. Fiat-backed coins have this; it is why they hold. Note the qualifier: "can always" means the issuer honours it, which is a legal and operational claim rather than a cryptographic one.
Arbitrage. If the coin trades below par, buying it and redeeming (or borrowing against it and closing) is profitable, which creates the bid. This works only if redemption exists, or if the mechanism pays someone to absorb the imbalance.
Monetary policy. Raising the stability fee to make holding the coin expensive, or opening a debt auction, steers the peg without a direct redemption path. Slower, and it depends on governance acting, which in a fast depeg is a real latency problem.
How each one breaks
The archetypes fail differently, and knowing which failure you own is most of the risk work.
Fiat-backed coins fail on issuer risk. The March 2023 USDC depeg to roughly $0.87 was a solvency scare at the reserve custodian, not a flaw in the token. Beyond that: freeze functions mean an issuer can blacklist addresses, and OFAC compliance makes that routine rather than theoretical; multichain deployments add the bridge as a new attack surface and fragment liquidity; and because USDC and USDT are the base layer of lending markets and DEX pairs, a depeg is never just a stablecoin problem — it triggers protocol-wide liquidations.
Crypto-collateralised coins fail on reflexivity. The collateral is often the very asset the coin is used to lever into, so a price drop forces liquidations that push the price down further. Oracle failure can make a solvent system insolvent on paper for a block, which is long enough to liquidate everyone.
Algorithmic coins fail on demand. The mechanism works while new money is entering, and the mechanism is the thing that breaks when it stops.
What to check before you hold one
Concentration of the collateral or reserves, and who verifies it. Whether redemption is available to you specifically, or only to authorised participants. The freeze and blacklist surface, because it determines whether your balance is truly yours. Oracle design and update cadence if the coin is collateralised. And chain distribution, because a coin that is 60% on one chain has a bridge risk you did not sign up for.
References
Next, read Crypto Insurance 101. More in DeFi Markets.