A stablecoin does not "maintain" a peg the way a central bank maintains a currency band. There is no committee, no rate decision, no press conference. The peg is a contract — usually a redemption right, sometimes a smart-contract auction, occasionally a reflexive incentive loop — and it holds exactly as long as the counterparty on the other side of that contract can honor it. When Silicon Valley Bank failed on March 10, 2023, USDC de-pegged to $0.87 within 36 hours. Not because the code broke. Because the redemption counterparty broke. That is the entire subject of this piece.
Methodology: What "Maintaining the Peg" Actually Means
I want to be precise about the audit frame here, because most explainers collapse four completely different mechanisms into one word — "peg" — and then act surprised when the mechanisms fail differently.
The peg is not the price on Coinbase. The price on Coinbase is a consequence. The peg is the redemption contract that arbitrageurs enforce against, and the mechanism you are actually auditing is the enforceability of that contract under stress. Everything else is downstream.
So the frame I use has four questions and only four questions. First — who is on the other side of a redemption at par, and can they honor it in the size a real depeg would demand? Second — what is the collateral, where does it sit, and can it be liquidated on the same clock as the redemption window? Third — what happens if the mechanism goes reflexive against the peg instead of toward it? Fourth — if the issuer disappears tomorrow, does the contract still execute?
Everything downstream of those four questions — the transparency page, the attestation cadence, the marketing about "1:1 backing" — is signal about the answers, not the answers themselves. I will not treat an attestation as a fact. I will treat it as an assertion by a specific accounting firm about a specific dollar balance at a specific point in time, which is what it is.
I am also not covering every stablecoin. I am covering the four mechanism archetypes — fiat-backed reserve, crypto-overcollateralized, algorithmic/seigniorage, and hybrid — because every stablecoin you can name is a variation on one of those four. And I am grounding claims in the public record: transparency pages, on-chain traces, published post-mortems, and the two depeg events that revealed how each mechanism actually behaves when it is being tested rather than described.
Finding #1: Fiat-Backed Pegs Are an Arbitrage Contract, Not a Promise
Here is what a fiat-backed stablecoin actually is, mechanically, once you strip the marketing.
An entity — Circle, Tether, Paxos, whoever — accepts one U.S. dollar via wire, mints one token, and posts that token to your address. Later, an entity accepts one token back, burns it, and wires you one dollar minus fees. The peg is that redemption contract. Not the reserve. Not the attestation. The redemption contract, at par, in the size and speed a stressed market demands.
Now — and this is the part that gets skipped — you, the retail buyer on Kraken, do not have that contract. Only whitelisted institutional counterparties do. So when the secondary market price drops to $0.97, the peg is maintained not by the retail seller but by an arbitrage desk that can redeem 50 million tokens directly with the issuer for 50 million dollars, then buy back at 97 cents, and pocket the three-cent spread. That desk is the peg's enforcement mechanism. Retail is not.
Two things follow from that, and they are both unpleasant.
First — the peg is only as fast as the slowest link in the redemption chain, which is almost never the smart contract and almost always the banking rail. USDC's 2023 depeg is the canonical case. The token did not break. Circle held $3.3 billion of its $40 billion reserve at Silicon Valley Bank, SVB failed on a Friday, and the redemption counterparty — Circle's ability to wire out dollars over the weekend — became a real question mark. The peg cracked from $1.00 to roughly $0.87 by Saturday night. Reserve composition and redemption clock, not code, decided the price.
Second — the attestation you are reading is a monthly or quarterly photograph of the reserve. Between photographs, you are trusting the operator. That is not a criticism. It is a description. If you are underwriting a fiat-backed stablecoin, you are underwriting an issuer's operational risk and their banking counterparty risk. Nothing else.
Finding #2: Overcollateralized Pegs Trade Capital Efficiency for Survivability
DAI is the working example. Also the honest one, because MakerDAO has published enough on-chain telemetry over the years that you can actually audit the mechanism instead of taking the marketing on faith.
The contract, in one sentence — you lock up more than a dollar of crypto collateral, mint less than a dollar of DAI, and if the collateral value falls toward the debt value, an on-chain auction sells your collateral to keep the system solvent. The peg is enforced by (a) the redemption incentive of being able to close your position at par and (b) the liquidation mechanism that keeps the aggregate collateral ratio above 100%.
Let me show the math on a single vault, because "overcollateralized" is one of those words people say without ever writing down what it means.
Assume ETH is at $3,000. You lock 1 ETH — $3,000 of collateral. The liquidation ratio is 150%, meaning your debt cannot exceed collateral / 1.5. That gives you a debt ceiling of $2,000 in DAI. You mint a comfortable $1,500. Your effective collateralization ratio is 3000 / 1500 = 200%. Now ETH falls to $2,250. Your ratio is 2250 / 1500 = 150%. You are at the liquidation threshold. If the price ticks another dollar down, keepers trigger an auction, your collateral is sold, the auction proceeds burn your $1,500 of DAI, and the remaining collateral (minus a liquidation penalty of, in Maker's schedule, 13%) is returned to you. The system stays solvent. The peg holds — because for every DAI in circulation, more than a dollar of ETH (or WBTC, or LSTs, or whatever the collateral schedule allows) is sitting in a vault, redeemable through the same mechanism in reverse.
That is the design. The failure mode is chain-native — if collateral price falls faster than keepers can liquidate, or if gas prices spike enough that liquidation transactions can't confirm, the aggregate ratio can dip below 100% and the peg breaks from the inside. Black Thursday, March 12, 2020, is the receipt. ETH fell 43% in 24 hours, Ethereum gas prices spiked to a level where keeper bots' bids failed to land, and the resulting undercollateralization forced Maker to auction MKR to recapitalize — the "peg" survived, but only because MKR holders were diluted to plug the hole.
That is what "overcollateralized" actually buys you. It does not remove issuer risk. It transforms it into liquidation-mechanism risk, which is a different thing that fails on a different clock.
Finding #3: Algorithmic Pegs Are a Reflexivity Bet, and Reflexivity Cuts Both Ways
Terra's UST is the fossil everyone points to. Fine — but the mechanism deserves an honest description before we bury it, because the reason it failed is instructive for what the class actually is.
An algorithmic stablecoin has no external collateral. The peg is maintained by a two-token seigniorage loop. In Terra's case, one UST could always be swapped, on-protocol, for one dollar's worth of LUNA. If UST traded at $0.98, arbitrageurs bought UST for $0.98, redeemed it on-protocol for $1.00 of newly minted LUNA, sold the LUNA for $1.00. UST supply contracts, price rises back to peg. If UST traded at $1.02, arbitrageurs did the reverse — burned $1.00 of LUNA to mint one UST, sold at $1.02. Elegant on paper. Reflexive by design.
The problem is that the reflexivity is symmetric. On the way up, expanding UST supply is met by absorbing LUNA supply, and the flywheel spins. On the way down, contracting UST supply mints new LUNA — and if the market bid for that newly minted LUNA is not there, LUNA's price falls, which means it takes more LUNA to absorb each redeemed UST, which mints even more LUNA, which drives the price further down, which requires more LUNA per redemption. That loop, once it starts, does not stop until either the LUNA market cap goes to zero or the mechanism is halted.
Between May 9 and May 12, 2022, UST fell from $1.00 to under $0.10. LUNA's supply expanded from roughly 350 million tokens to over 6.5 trillion in the same window — I say "roughly" because the exact numbers depend on where in the collapse you snapshot, but the on-chain traces on Terra explorer show the mint volume in real time. The mechanism did what it was designed to do. The design was reflexivity, and reflexivity in the wrong direction is called a bank run.
The lesson is not "algorithmic stablecoins are bad." The lesson is narrower — a pure algorithmic peg with no exogenous collateral is a bet that the demand curve for the seigniorage token is deep enough to absorb any conceivable redemption. That bet is fine in a bull market. It is provably wrong in a bear one. Every algorithmic design since Terra has tried to add some form of exogenous backing precisely because the pure form is unbacked, and unbacked reflexivity is a dominant strategy for a bank run.
Finding #4: The Custody Layer Decides Whether the Peg Survives Its Issuer
This is the finding that gets skipped in every stablecoin explainer, and it is the one that matters most for anyone holding stablecoins as savings rather than as a trading rail.
The peg is a contract between the token and its redemption mechanism. Your ownership of the token is a separate contract — between you and whoever holds the private key. Those two contracts fail in different ways, and confusing them is expensive.
If you hold USDC on a centralized exchange and the exchange fails, the peg holding at $1.00 is irrelevant to you. Your claim is on the exchange, denominated in USDC, and it takes its place in a bankruptcy queue. If you hold USDC in a self-custody wallet where you control the seed, the peg holding at $1.00 is the entire question, because your token is under your control and the only remaining risk is the issuer's ability to honor redemption or the market's willingness to price the token at par.
So the custody stack matters as much as the peg mechanism. A qualified custodian — Coinbase Custody under NY DFS trust charter, Fidelity Digital Assets under NY DFS trust, Anchorage Digital under an OCC federal trust charter — is a bankruptcy-remote structure where client assets are legally segregated from the custodian's balance sheet. If the custodian fails, your USDC is still your USDC. If your exchange fails and your USDC was sitting in the exchange's operational wallet, that legal segregation does not exist.
Self-custody moves the problem one layer down. The peg mechanism is untouched — but now you are the custody counterparty, and the failure modes are seed compromise, firmware exploit, and operational error. Hardware wallets — Ledger, Trezor, GridPlus Lattice1 — reduce the attack surface but do not eliminate it, and each has a firmware audit history you can read before deciding which failure mode you prefer.
The point is this. When someone asks "is USDC safe" or "is DAI safe," the answer is always in two parts — the peg mechanism and the custody stack — and 90% of the risk sits in whichever part they didn't think about.
What This Does NOT Prove
This piece does not prove that any specific stablecoin is safe or unsafe. It proves that the four mechanism classes fail in four distinct ways, on four distinct clocks, and that "the peg held" and "the peg will hold next time" are different claims requiring different evidence.
It also does not settle the reserve-composition question for any specific fiat-backed issuer. I referenced Circle's SVB exposure in 2023 because it is on the public record. I did not run the analogous exercise for Tether's current reserve breakdown, because doing that honestly requires primary-source attestation review that would double the length of this article, and the point I am making about redemption-clock risk generalizes across issuers whether or not any particular reserve pie chart is accurate today. If you are underwriting a specific stablecoin as a large position, the reserve breakdown is not optional homework — but it is not this piece's job.
The Takeaway
The peg is a contract, not a property. Audit the contract's enforceability under stress, and the marketing takes care of itself.
FAQ
What is the difference between a stablecoin's "peg" and its "backing"?
The peg is the price target — one token equals one dollar. The backing is the collateral that makes redemption at par credible. They are related but not the same. A stablecoin can have full backing and still lose the peg temporarily if the redemption mechanism is bottlenecked (USDC in March 2023 is the canonical case — reserves existed, but the banking rail that connected them to arbitrageurs was frozen for 72 hours). Backing is a stock. The peg is a flow.
Why did USDC de-peg in March 2023 if it was fully backed?
Because "fully backed" and "immediately redeemable" are not the same claim. Circle disclosed that $3.3 billion of its reserve was held at Silicon Valley Bank, which failed on a Friday afternoon. Over the weekend, arbitrageurs could not confirm whether that portion of the reserve was recoverable, and the secondary market priced the uncertainty by discounting USDC to roughly $0.87. Once the FDIC guaranteed all SVB deposits on Sunday night, the peg recovered within hours. The mechanism was fine. The banking counterparty was the failure surface.
Is DAI fully decentralized, or does it depend on USDC too?
DAI is a mixed collateral system, and a meaningful share of its backing has historically been USDC held inside the Peg Stability Module — a swap contract that exchanges USDC for DAI at 1:1 to keep DAI tightly pegged. That composition changes over time based on Maker governance decisions, so any specific ratio is a snapshot. The honest description is that DAI's peg quality has, in practice, been correlated with USDC's peg quality during the periods when the PSM held significant reserves.
Can any algorithmic stablecoin work, or is the design fundamentally broken?
The pure form — no exogenous collateral, peg maintained only by seigniorage against a volatile governance token — is a reflexive bet that fails in a bank run. Every serious design proposed after Terra has added some form of exogenous backing precisely because the pure form is provably fragile under stress. Whether hybrid designs (partial collateral plus algorithmic elements) are durable is an open question that will be answered by their behavior in the next liquidity crisis, not by the current whitepaper.
Does holding stablecoins with a qualified custodian eliminate depeg risk?
No. It eliminates custody risk — your tokens are legally segregated from the custodian's balance sheet, so a custodian failure does not put your position in a bankruptcy queue. But if the stablecoin issuer's reserves are impaired, or the redemption mechanism is bottlenecked, the token itself trades below par regardless of who holds it. The two risks compose. Custody protection is necessary for large positions and not sufficient — the peg mechanism still has to work on its own merits.
What on-chain signals should I watch to spot a depeg early?
Three concrete ones. First — secondary-market price on the deepest liquidity venues (Curve pools for on-chain stablecoins, Coinbase/Kraken order books for centralized). A sustained deviation past 20-30 basis points is the market pricing counterparty doubt. Second — redemption queue depth at the issuer, when disclosed. If institutional redemptions are backing up, the peg is under strain even if the screen price still shows par. Third — for collateralized designs, the aggregate collateral ratio and the volume of liquidation activity on-chain, both of which are queryable in real time on Dune or the protocol's own dashboards.