11 March 2023. Circle disclosed that $3.3B of USDC's reserves, about 8 percent of roughly $40B, was sitting as cash deposits at Silicon Valley Bank. Regulators had closed SVB the previous day. Wires sent on the Thursday to move the balance had not processed before the bank was shut.
USDC traded down to about $0.87, at times as low as $0.86, and stayed under a dollar through the weekend. It regained the peg after US regulators said on the Sunday that all SVB depositors would be made whole. About three days, start to finish.
This is the autopsy for the lesson rather than Terra, and the reason is that USDC was fully reserved and the reserves were real.
It depegged anyway. Because reserves have to be held somewhere, and somewhere failed.
Which is a more useful lesson than a fraud would be. Full backing does not remove risk. It relocates it, onto whoever is holding the backing, and the market prices your ability to redeem today rather than your assets on a balance sheet.
If the phrase "fully backed" was doing reassurance work in your head, this is the incident that should replace it with a question: backed by what, held where, redeemable by whom, and how fast.
A stablecoin is a token that is supposed to be worth one unit of something, usually a dollar. They are the most used thing in crypto by a wide margin, they are how most people actually move value, and almost nobody who uses them can say what makes them stable.
The answer differs completely between them, and the difference is the whole risk.
Two separate questions, which everyone conflates
A peg is a price. It is what the token trades at, maintained by somebody being willing and able to buy at that level. Nothing in the token enforces it. There is no mechanism inside a stablecoin that makes it worth a dollar, any more than there is inside a share certificate.
A redemption right is a claim. It is the question of who owes you an actual dollar, under what terms, on what timeline, and whether you personally qualify.
These come apart constantly and the gap is where losses live. Most retail holders have no redemption right at all: the issuer redeems for institutions above a minimum size, and you are relying on someone else's arbitrage to keep your price at a dollar. Which usually works, and is a different arrangement from the one the word "stablecoin" implies.
So the question is never "is it stable". It is who is holding the thing that makes it stable, and can I get to it.
Four models, four failure modes
Fiat-backed. A company holds cash and short-term government debt and issues tokens against it. Fails at: the custodians of the reserves, which is the autopsy, and at the issuer, through freezing, insolvency or regulatory action. Note that most fiat-backed issuers can freeze your specific tokens, which is a property you may or may not want.
Crypto-backed. You lock volatile collateral, over-collateralised, and mint against it. Fails at: collateral price and liquidation mechanics. When the collateral falls fast, liquidations fire into a falling market and can fail to clear, which is how these break.
Algorithmic. No meaningful collateral. Stability is a mechanism, usually minting and burning a second token. Fails at: reflexivity. The mechanism requires the second asset to hold value precisely when everybody is trying to leave, which is the one condition under which it cannot.
Yield-bearing. Any of the above, paying a return. Fails at: wherever the yield comes from. This is the one that has to be asked about explicitly, because the yield is the marketing and the source is the risk.
The course opens on Terra for a reason, and it belongs to the last two categories at once.
This is the calculation to run on any depeg, and it takes a minute.
Step one: the worst-case backing floor. Assume the impaired portion is a total loss.
Total reserves: $40B At risk: $3.3B Surviving: 40 - 3.3 = $36.7B
Against roughly $40B of tokens outstanding, backing per token in the worst case:
36.7 / 40 = $0.9175
So even assuming SVB paid out nothing, each USDC was backed by about 92 cents.
Step two: compare to the market. USDC traded at $0.87, and touched $0.86.
0.87 is below the 0.9175 floor. By:
(0.9175 - 0.87) / 0.9175 = about 5.2 percent
Step three: read the result. The market was pricing something worse than a total loss of the disclosed exposure. That is not a solvency calculation. It is some combination of doubt about whether the disclosure was complete, doubt about the timeline to redeem, and people who needed out at any price over a weekend when redemptions were closed.
This is the distinction to carry. A solvency discount says the assets are impaired. A liquidity discount says the assets are fine and you cannot reach them today. They look identical on a price chart and they call for opposite behaviour.
And a caution against reading this as a trading lesson, because it is not one. The floor calculation is only as good as the disclosure it uses. Everyone running this arithmetic in March 2023 was trusting Circle's statement of what and where the reserves were, and if that statement had been wrong the floor would have been fiction. The calculation tells you what the numbers imply, not whether the numbers are true. P10 applies here as everywhere: nothing in this section is a suggestion to buy a depegged asset.
Why Terra is the course autopsy
Run the same test on Terra and it fails at step one, which is the point.
UST was the third largest stablecoin at about $17.5B, and roughly 75 percent of it sat in Anchor, a protocol paying about 19.5 percent. That yield was not earned from anywhere. It was subsidised from a reserve, topped up by $450M in February 2022 to a total of $507M, which the foundation judged sufficient for a year. By April the subsidy was burning around $6M a day.
$507M at $6M a day is about 85 days. They projected twelve months and the arithmetic said under three.
When the peg broke on 7 May, the defence mechanism minted LUNA to absorb redemptions. LUNA's supply went from 343 million to 6.53 trillion in a week, dilution by a factor of roughly nineteen thousand.
There was no step one. Under USDC there were assets, in the wrong place. Under UST there was a mechanism that required a volatile asset to hold its value at the exact moment everyone was selling it.
If a stablecoin pays a return, that return comes from somewhere, and the somewhere is the risk you are holding. There are only a few honest answers: short-term government debt, over-collateralised lending, or a subsidy.
The first two are real and their rates are visible and boring. If the yield materially exceeds what a government bill pays, the difference is being generated by taking a risk, and you should be able to say which one.
The third has an end date you can compute, and Terra's was public the whole time. Every input was published: the rate, the deposit base, the reserve size, the burn rate. Nobody needed inside information. They needed two minutes and a willingness to have the answer come out badly.
Reading a disclosure
Three questions, in order of how often they find something.
What is in it, exactly? Cash and short-dated treasuries are the boring answer and boring is correct. Commercial paper, loans to affiliates, or the issuer's own token are progressively worse answers, and the last one means the reserve falls when the token does.
Where is it held, and across how many counterparties? This is the question SVB made compulsory. One bank holding a large share is a single point of failure regardless of how good the assets are.
Who verifies it, and how often? An attestation, where an accountant confirms a balance at a moment, is weaker than an audit, and both are weaker than the same information published on-chain where anyone can check it continuously. Note the irony that F102-02 already flagged: verifiable settlement is on the list of things this technology actually offers, and most fiat-backed issuers hold their reserves in exactly the place where nobody can verify anything.
Stablecoins are the safe part of crypto, so a stablecoin balance is like cash.
It is the low-volatility part, which is not the same as the safe part, and the difference has a specific shape.
Volatile assets fail continuously and visibly. You watch the number move and you can act at any point along the way.
Stablecoins fail discontinuously. They sit at exactly $1.00 for years, which trains you to stop thinking about them, and then they move to $0.87 over a weekend when redemptions are shut. All of the risk arrives at once, at the moment you are least able to respond.
They also carry a risk that volatile assets do not: most major issuers can freeze specific addresses, and have. A stablecoin balance is a claim on a company that can decline to honour it for you specifically, which puts you back in F102-01's position with fewer protections than a bank depositor.
None of that means avoid them. They are genuinely useful and the alternative for most purposes is worse. It means treating a stablecoin balance as a short-term position in a specific issuer's credit, sized accordingly, rather than as cash that happens to live on a chain.
A peg is a price somebody maintains, and a redemption right is a claim most retail holders do not have, so the question is never whether a stablecoin is stable but who holds the thing making it stable and whether you can reach them. Fiat-backed fails at the custodian, crypto-backed at liquidation, algorithmic at reflexivity, and yield-bearing wherever the yield comes from. USDC was fully reserved and still traded to $0.87, below its own 92 cent worst-case floor, which is what a liquidity discount looks like next to a solvency one. And Terra's end date was computable from public figures: $507M of reserve burning $6M a day is 85 days, against a projection of a year.