What actually causes one
Three mechanisms, and they differ enough that treating them as one phenomenon is a mistake.
Doubt about the reserve
If holders stop believing the issuer can redeem at par, they sell, and the price falls before anyone has proved anything either way. This is a confidence problem and it resolves the moment the doubt does.
A liquidity crunch
Redemption may work fine while the market for the token temporarily has more sellers than buyers. The peg holds in theory and slips in practice until arbitrage catches up.
A design that never had reserves
Algorithmic stablecoins maintained their peg through mechanisms rather than assets, and when confidence went, there was nothing underneath to arrest the fall.
The two cases worth knowing
USDC, March 2023
Circle held part of its reserve at Silicon Valley Bank, which failed. USDC fell to roughly 0,87 dollars as holders questioned whether that portion was recoverable. Once US authorities guaranteed the bank’s deposits, the price returned to par within days. A reserve problem with a reserve solution.
Terra UST, May 2022
An algorithmic stablecoin backed by a mechanism rather than by assets. When the mechanism came under pressure it accelerated the collapse instead of arresting it, and roughly forty billion dollars in value disappeared over about a week. The token never recovered.
The difference between those two outcomes is the whole reason the distinction between backed and algorithmic matters.
How exposed a merchant actually is
Less than the topic’s prominence suggests, and the reason is timing.
Exposure lasts exactly as long as you hold the asset. A business that converts on receipt and settles out is holding stablecoins for minutes, and no depeg in history has moved meaningfully within that window.
A business that keeps its treasury in stablecoins is a different case. There the exposure is real and continuous, and the sensible responses are ordinary: understand whose reserves you are relying on, read the attestations, and avoid concentrating everything in one issuer.
What no merchant should do is hold reserves in an algorithmic stablecoin. That category has an unblemished record of ending badly.
How the three types of stablecoin differ on this
Worth separating, because the depeg risk is structurally different for each.
Fiat-backed
stablecoins hold cash and short-dated government debt against every token. The risk is that the reserve turns out to be less liquid or less real than stated, or that the bank holding it fails. USDC and USDT are here.
Crypto-collateralised
stablecoins hold other cryptocurrency as backing, over-collateralised to absorb volatility. The risk is a sharp market drop outrunning the liquidation mechanism.
Algorithmic
stablecoins hold nothing and maintain the peg through supply mechanics. The risk is that the mechanism relies on confidence, and confidence is precisely what disappears in a crisis.
The first two have survived stress events. The third category has an unbroken record of failing under them.