Stablecoins: four designs, four different promises
Every stablecoin says the same thing on the label and means something different underneath. The four designs fail in four distinct ways, and knowing which one you hold tells you which failure you are exposed to.
Not financial advice. This article is for informational purposes only.
The short version
“One dollar” is a claim, and the four stablecoin designs back that claim in incompatible ways: dollars at a bank, over-collateralised crypto, an algorithm with no collateral, or a commodity. Each has a characteristic failure. Fiat-backed depends on a custodian and its bank. Crypto-backed depends on liquidations clearing during a crash. Algorithmic depends on continued confidence, which is the thing that disappears first. Knowing your design tells you your risk.
A stablecoin’s job is to be boring. It is the unit people use to hold value between trades, price things, and move money without touching the banking system twice. Because it is boring, it gets treated as a single homogeneous thing — “a dollar on-chain” — and the differences between designs get flattened away.
They should not be. The word “stable” describes the intent, not the mechanism, and the mechanism is where the risk lives. There are four broad approaches.
1. Fiat-collateralised: dollars somewhere else
The most common design. An issuer takes dollars, holds them in reserve, and issues one token per dollar received. Redemption is the anchor: if a token can reliably be exchanged for a dollar, arbitrage keeps the market price near a dollar without any clever machinery.
What it depends on. Everything rests on the reserve genuinely existing, being liquid enough to meet redemptions in a rush, and being reachable. Two questions matter more than any others: what are the reserves actually invested in, and who is allowed to redeem at par? If redemption is only available to large institutional clients, retail holders depend on those institutions arbitraging on their behalf — which works until it does not.
How it fails. Not usually through fraud. The realistic failure is the reserve being held somewhere that becomes unavailable — a bank that fails, a jurisdiction that freezes, an asset that cannot be sold at par under stress. The token then trades below a dollar not because the dollars are gone but because nobody can get at them today. This design converts crypto risk into banking and custody risk. It does not remove risk.
2. Crypto-collateralised: over-collateralised and liquidated
Here the backing is other crypto assets, held in a smart contract rather than at a bank. Because the collateral is volatile, the system demands more than one dollar of it per dollar issued — often substantially more. If the collateral falls toward the debt it supports, the position is liquidated and the debt repaid.
What it depends on. Three things, all of which are stressed at the same moment: an accurate price feed, enough liquidity for liquidations to actually clear, and participants willing to buy the collateral being sold. It is transparent in a way the fiat-backed design is not — you can verify the collateral on-chain, right now, yourself.
How it fails. Through correlation. In a sharp decline the collateral falls, liquidations trigger, the liquidations sell collateral into a falling market, and that pushes collateral lower still. If prices fall faster than positions can be closed, the system ends up under-collateralised, holding bad debt. Every mechanism designed to protect it — the price feed, the liquidation queue, the block space needed to execute — is under maximum load exactly when it is needed. See liquidation for the mechanics.
3. Algorithmic: confidence as collateral
The most ambitious design and the one with the worst record. Here there is little or no collateral. Instead the protocol manages supply against a companion asset: when the stablecoin trades above target, supply expands; when it trades below, holders are given an incentive to swap into the companion asset, contracting supply.
What it depends on. That the companion asset has value. And the companion asset’s value derives largely from the expectation that the system will keep working — which makes the argument circular.
How it fails. Quickly, and in public. If the stablecoin trades below target and the swap mechanism requires issuing more of the companion asset, then a falling companion asset means issuing more of it to redeem the same value, which pushes it down further. Confidence and collateral are the same variable, so they fail together and there is no floor. This is not a hypothetical — the collapse of a major algorithmic stablecoin in May 2022 followed exactly this path and erased tens of billions of dollars of nominal value in days.
Some designs are partially collateralised, sitting between this and the previous category. They are better, and the honest way to assess them is to ask what happens at the collateral portion alone, because that is the part that does not depend on belief.
4. Commodity-collateralised: a claim on a physical thing
The token represents a quantity of a physical commodity, most often gold, held in a vault. It is not a dollar stablecoin at all — it is stable against the commodity, and therefore moves against the dollar.
What it depends on. Whether the metal exists, is audited by someone independent, is unencumbered, and can be redeemed at a size and cost that makes redemption real rather than theoretical. Storage costs money, so there is usually a fee that quietly erodes the holding.
How it fails. Through the custody chain, and through redemption terms that turn out to be impractical for anyone holding a normal amount.
The one comparison worth keeping
| Design | Backed by | Verify by | Characteristic failure |
|---|---|---|---|
| Fiat-collateralised | Cash and short-term instruments | Attestations, reserve reports | Custodian or bank becomes unreachable |
| Crypto-collateralised | Over-collateralised crypto | On-chain, directly | Correlated crash outruns liquidations |
| Algorithmic | A companion asset and confidence | Nothing external | Reflexive collapse with no floor |
| Commodity-collateralised | Physical metal in a vault | Independent vault audit | Custody chain, unusable redemption |
What to actually do with this
Find out which design you are holding — it is usually stated plainly in the issuer’s own documentation, and if it is not, that is informative. Then ask the one question that fits that design: for fiat-backed, where are the reserves and who can redeem; for crypto-backed, what is the collateral and how has the liquidation system behaved under stress; for algorithmic, what is the non-reflexive floor; for commodity, who audits the vault and what does redemption really cost.
And keep the general point in view. A stablecoin is not cash. It is a claim on someone or something, and “stable” is a description of how it usually behaves rather than a guarantee about how it must. None of this is advice about which to hold — see our disclaimer.
Key takeaways
- “Stable” describes intent, not mechanism. Four designs back the same claim in incompatible ways.
- Fiat-collateralised converts crypto risk into banking and custody risk. It does not remove risk.
- Ask who is permitted to redeem at par — if only institutions can, retail depends on their arbitrage.
- Crypto-collateralised is verifiable on-chain but fails through correlation, when liquidations cannot clear fast enough.
- Algorithmic designs use confidence as collateral, so the two fail together and there is no floor beneath them.
- A commodity-backed token is stable against the commodity, not the dollar, and storage fees erode it over time.
- A stablecoin is a claim on someone, not cash. Identify your design, then ask the one question that fits it.