Stablecoin Depegs: How Reserves Failed and the Signs That Came First

A stablecoin is worth $1 only if someone will reliably hand you $1 for it. When that promise weakens, the price drifts. When it breaks, the price drops fast, and the people left holding are usually the ones who skipped the warning signs.

Those signs were there almost every time. Below, I walk through the four ways stablecoins have actually failed, using real episodes. Then I give you a checklist for working out which risk the coin in your wallet carries.

How a peg is supposed to work

Most stablecoins hold their price through arbitrage. If the coin trades at $0.98, a trader buys it cheap and redeems it with the issuer for $1.00, pocketing two cents. That buying pushes the price back up. If it trades at $1.02, traders deposit $1 with the issuer, mint a new coin and sell it for $1.02. That selling pushes the price back down.

The mechanism rests on three things:

• Something real backs each coin. Cash, short-term government debt, crypto collateral, or, in the algorithmic case, a promise.
• Redemption actually works. Someone can turn the coin into $1 quickly, at scale, without a long queue.
• People believe the first two. Confidence is part of the backing, whether issuers admit it or not.

Remove any one and the peg wobbles. Each failure below is a story about one of them going missing.

Failure mode 1: Reserve losses and opacity

This is the classic fiat-backed risk. The issuer says each coin is backed by a dollar or something close to it. Then it turns out some of those dollars are missing, tied up, or were never there.

The historical case: Tether in 2018 and 2019. Tether's USDT spent stretches of late 2018 trading noticeably below $1 on some exchanges. Questions had been piling up for a long time: no full audit, a parade of changing banking partners, and statements about backing that shifted over time. A 2019 investigation by the New York Attorney General later alleged that a large sum belonging to the affiliated exchange Bitfinex was held by a payment processor, Crypto Capital, and couldn't be accessed, and that Tether reserves had been used to cover the gap. Tether and Bitfinex settled without admitting wrongdoing. USDT recovered and grew. But for a while, holders were trusting a balance sheet nobody outside the company could fully see.

Warning signs that showed up first:

• Attestations instead of audits. An attestation is a snapshot on one date. An audit tests controls and the full picture.
• Reserve descriptions that changed wording over time, such as going from "cash" to "cash and cash equivalents and other assets".
• Frequent bank changes and trouble keeping banking relationships.
• Redemption limited to large institutional accounts, with high minimums.

The lesson: surviving a scare doesn't prove the risk was imaginary. It means the run never got big enough to test it.

Failure mode 2: Algorithmic death spirals

An algorithmic stablecoin tries to hold $1 without holding $1 in reserves. Usually it pairs the stablecoin with a second, volatile token. You can always burn $1 of stablecoin to mint $1 worth of the volatile token, and the reverse. The idea is that arbitrage keeps the peg. The flaw is that the "backing" is a token whose value depends on people believing in the stablecoin.

The historical case: TerraUSD (UST) in May 2022. UST could be burned for $1 of LUNA. Demand for UST was propped up by Anchor, a lending protocol paying close to 20% a year on UST deposits. In early May 2022, large withdrawals from Anchor and big UST sales pushed the coin below $1. Arbitrageurs burned UST to mint LUNA and dumped the LUNA. That crashed LUNA's price, so every redemption needed more LUNA, which crashed it further. Within days UST was worth cents and LUNA was close to zero. Tens of billions of dollars in combined value vanished.

It wasn't the first time. Iron Finance, in June 2021, ran a partially algorithmic coin, IRON, backed partly by USDC and partly by its own TITAN token. A wave of selling triggered the same loop: redeem IRON, receive TITAN, sell TITAN, TITAN collapses. TITAN went close to zero in a single day. Earlier still, NuBits lost its peg in 2016 and again in 2018 once demand dried up.

Warning signs that showed up first:

• A yield far above anything the real economy pays, with no clear source. Close to 20% on a "stable" dollar was a subsidy, not a business.
• Backing made of the protocol's own token.
• Growth driven by the yield rather than by people using the coin.
• Emergency reserves bolted on late, in assets that could themselves fall in a panic.

The core problem is reflexivity. The backing loses value at the exact moment you need it.

Failure mode 3: Bank and custodian exposure

Here the reserves are real and well managed. The trouble is where they sit. A fiat-backed stablecoin holds cash at banks, and banks can fail.

The historical case: USDC in March 2023. On a Friday in March 2023, Silicon Valley Bank collapsed. Circle, the issuer of USDC, disclosed that around $3.3 billion of its reserves were held there. That was a minority of total reserves, but nobody knew whether the cash would be recovered. Over the weekend, with banks closed and redemptions effectively frozen, USDC traded as low as roughly 87 to 88 cents on some venues. The panic spread to stablecoins that held USDC as collateral, including DAI. When US authorities guaranteed SVB deposits on Sunday evening, USDC returned to $1 within days.

Warning signs that showed up first:

• Concentration in a small number of banks, disclosed in the issuer's own reports.
• Banking partners with heavy crypto or tech exposure, which tends to move together under stress.
• Redemptions that only work on business days, so weekend panics have no release valve.

This one is sneaky because the issuer did little wrong. The risk sits a layer down. It also shows how risks chain: a coin backed by a coin backed by a bank inherits the bank's problems.

A related variant is regulatory exposure. In February 2023, US regulators ordered Paxos to stop minting new BUSD. The coin didn't crash, because reserves were intact and redemptions continued, but it began an orderly wind-down. Your stablecoin can be fully backed and still stop being useful.

Failure mode 4: Collateral gaps in crypto-backed coins

Crypto-backed stablecoins, like DAI in its original design, lock up more than $1 of volatile crypto for each $1 of stablecoin, say $150 of ETH. If the collateral price falls too far, the position gets liquidated: the collateral is auctioned to cover the debt. The danger is a crash so fast that liquidations can't keep up.

The historical case: MakerDAO's "Black Thursday", March 2020. ETH fell sharply in a matter of hours. The Ethereum network clogged and gas fees spiked. Many liquidation bots failed to bid, and in some auctions a few participants won ETH collateral for bids of zero. The system was left with a few million dollars of bad debt, which Maker covered by minting and auctioning its governance token, MKR. Oddly, DAI traded above $1 during the chaos, because borrowers scrambled to buy DAI to repay loans and there wasn't enough supply.

Warning signs that showed up first:

• Collateral concentrated in one volatile asset.
• Liquidation systems that depend on network conditions staying normal.
• Thin overcollateralization buffers for the asset's actual volatility.
• Heavy reliance on another stablecoin as collateral, which imports that coin's risks.

The four failure modes side by side

A checklist you can actually run

Before you hold a meaningful balance in any stablecoin, whether it's a trading float, a casino bankroll or savings parked between positions, work through this.

1. Name the backing. Find the issuer's reserve page. Write down in one line what backs the coin: cash and short-term treasuries, crypto collateral, its own token, or a mix. If you can't write that line, stop here.
2. Check the reporting. Is there a full audit or only monthly attestations? Who signs them? Read the most recent one and note the date. Reports that are months stale are a flag.
3. Look for concentration. How much sits at any one bank, custodian or in any one collateral asset? If one failure could knock out a large share, note the exposure.
4. Test redemption in your head. Who can redeem directly? What's the minimum? Does it work on weekends? If you, personally, can't redeem, you're relying on exchange liquidity and arbitrageurs.
5. Find the yield source. If anyone pays you a high rate to hold it, trace exactly where that money comes from. "Protocol revenue" or "reserves" isn't an answer.
6. Check for reflexive backing. Does the backing include a token whose value depends on this stablecoin's success? If yes, treat it as algorithmic risk, whatever the marketing says.
7. Map the dependencies. Is the collateral another stablecoin? Then repeat this checklist for that coin too.
8. Watch the price on-chain and on exchanges. Persistent trading at $0.995 or below on major venues, especially with large redemptions, is the market telling you something.
9. Decide your exit in advance. Set the price at which you'll swap out, such as two cents below peg for a fiat-backed coin, and the coin you'll swap into. Panic decisions are bad decisions.

Worked example: two stablecoins, one bankroll

Hypothetical numbers for illustration only.

Say you keep a $10,000 trading float and you're choosing between two made-up coins.

Coin A is backed by cash and short-term treasuries, with monthly attestations from a named accounting firm. Its last report shows 30% of cash at a single bank. Redemption is business days only, with a $1,000 minimum for verified users. It pays no yield.

Coin B is backed 60% by USDC-style fiat stablecoins and 40% by the protocol's own governance token. A partner platform pays 18% on deposits.

Run the checklist. Coin A's main risk is bank exposure. Suppose that 30% at the one bank were frozen indefinitely and fully lost. That's the worst case, and backing per coin would fall to $0.70. In a more realistic scenario, half of it is eventually recovered, so the coin is worth about $0.85. Your $10,000 becomes roughly $8,500 on paper during the panic, and likely recovers if the deposits are made whole.

Coin B stacks two failure modes. It inherits the fiat coin's bank risk on 60%, and the 40% in its own token is reflexive. In a run, that token could go to near zero, leaving at most $0.60 of backing, and that's before any bank trouble. Your $10,000 could become $6,000 or less, and history says algorithmic portions rarely recover. Meanwhile the 18% yield earns you about $150 a month. You'd need more than two years of yield to cover a single bad week.

A sensible split might be keeping most in Coin A, adding a second fiat-backed coin that uses different banks, and skipping Coin B. Spreading money across coins only helps if they fail for different reasons.

Common mistakes

• Treating a recovery as proof of safety. USDC and USDT came back. That tells you the backing held that time, not that it always will.
• Diversifying into correlated coins. Holding three stablecoins that all sit at the same banks, or all use the same collateral, isn't diversification.
• Chasing yield on a "stable" asset. A dollar doesn't naturally earn double digits. Someone is taking a risk to pay you, and it may be you.
• Ignoring weekends. Several panics hit when banks were closed and redemptions couldn't happen.
• Assuming an exchange balance is the same as holding the coin. If the exchange fails, the coin's peg is irrelevant to you.
• Selling at the bottom of a temporary depeg. For fiat-backed coins with intact reserves, the worst prices have often been brief. Know which failure mode you're facing before you dump at 88 cents.

FAQ

Is a small depeg, like $0.997, something to worry about?

Usually not on its own. Tiny deviations come from normal trading and fees. What matters is direction and persistence. A coin sliding lower over days while redemptions spike is very different from a brief dip during a volatile hour.

Are fully reserved fiat-backed stablecoins risk-free?

No. They still carry bank risk, custodian risk, regulatory risk and the risk that the issuer can freeze addresses. They've failed less catastrophically than algorithmic designs, but "fully backed" describes the assets, not where those assets sit or who controls them.

Can an algorithmic stablecoin ever be safe?

Designs keep evolving, and some now add real collateral or hedges. The test is the same one from the checklist: if the backing loses value at the same moment people want out, a run can feed itself. Ask what happens to the backing when everyone redeems on the same day.

Every depeg in this guide was explained afterwards by facts that were public beforehand. The yield was advertised. The bank was named. The reserve wording was on the website. So the question isn't whether you could have known. It's whether you'll check next time, before the price moves.