Stablecoins: Collateralized vs Algorithmic

The lending protocols page ended with a machine that lets untrusting strangers borrow against collateral without a loan officer: deposit an asset worth more than you borrow, and a liquidation engine watching an on-chain price will seize your collateral the instant your loan is underwater. Hold that mechanism in your head — over-collateralization plus liquidation. This page reuses it directly.

The problem this page solves is different from lending, but it leans on the same parts. Ether is money you can send to anyone on the shared world computer, but its price against the dollar swings wildly from hour to hour. That makes it a terrible unit of account: no one wants to price a salary, a loan, or a coffee in a token that might be worth 30% less by dinner. A stablecoin is a token engineered to trade at a fixed value — almost always $1 — while still living on-chain, transferable and composable like any other ERC-20. The entire design question is: how do you make a freely-traded token hold a price it has no natural reason to hold? There are three families of answer, and one of them is a cautionary tale.

What “the peg” actually means

A stablecoin is not a dollar. It is a token that anyone can buy and sell on the open market, so its price is whatever buyers and sellers agree on at any moment — exactly like ETH. Calling it “stable” is a claim about market price: the token should trade at ~$1 despite being a free-floating asset. That target price is the peg.

Nothing about being called “USD-something” forces the market price to $1. If a stablecoin trades at $0.97, it is de-pegged — and staying pegged is not a property the token has, it is a property the system around the token must continuously defend. So the real object of study is the peg-maintenance mechanism: the thing that pushes the market price back toward $1 whenever it drifts.

Every peg-maintenance mechanism works the same abstract way — it creates an arbitrage incentive that a profit-seeker will act on, and that action moves the price back:

   price DRIFTS below $1                price DRIFTS above $1
   ─────────────────────               ─────────────────────
   token is "cheap"                    token is "expensive"
        │                                   │
        ▼                                   ▼
   arb buys it cheap,                  arb creates a new token for
   redeems for $1 of value            $1 of value, sells it for more
        │                                   │
        ▼                                   ▼
   buying pressure ► price up          selling pressure ► price down
        │                                   │
        ▼                                   ▼
        └──────────►  back toward $1  ◄──────────┘

The families differ entirely in what “$1 of value” a token can be redeemed for, and therefore in who you have to trust for that redemption to be real. That “who do you trust, and where do they live” question is the throughline of this whole page.

Family 1: Fiat-collateralized (USDC, USDT)

The simplest answer: for every on-chain token, keep one real dollar (or a dollar-equivalent asset like a short-term Treasury bill) in a bank account off-chain. This is how USDC (Circle) and USDT (Tether) work, and together they are by far the largest stablecoins by supply.

The mechanism is a promise of redemption by a company:

   ON-CHAIN                        OFF-CHAIN (a company's bank / custody)
   ────────                        ──────────────────────────────────────
   1,000,000 USDC   ◄──backs──►    $1,000,000 in cash + T-bills

   mint:   give the issuer $1  ──►  issuer creates 1 new token
   redeem: give the issuer 1 token  ──►  issuer wires you $1, burns token

The peg holds because a large trader can always mint at $1 or redeem for $1 with the issuer. If USDC drifts to $0.99 on an exchange, an authorized party buys the cheap tokens, redeems them with Circle for $1 each, and pockets the spread — and their buying lifts the price back to $1. The arbitrage is only as good as the issuer’s willingness and ability to honor redemption.

Under the hood — the trust is centralized and off-chain

This design is technically the least interesting and politically the most fraught, because the peg does not rest on any on-chain guarantee. It rests on:

A custodian actually holding the reserves and not lending them out into something risky.
Audits / attestations that the reserves exist and match the supply. USDC publishes regular attestations; USDT’s reserve composition has been a running controversy for years.
Legal willingness to redeem — and the issuer can freeze addresses. USDC and USDT both maintain blocklists and have frozen funds in response to law enforcement and sanctions.

So a fiat-backed stablecoin is a bearer claim on a company. It is capital-efficient (one dollar backs one token — no waste) and the peg is rock-solid as long as you trust the custodian. But you have re-introduced exactly the trusted third party that the rest of this book spends its time removing. The chain agrees perfectly on who holds the token; it cannot verify that a dollar sits behind it. That verification lives off-chain, in an auditor’s report.

Family 2: Crypto-collateralized / over-collateralized (DAI, MakerDAO)

The second family refuses to trust an off-chain custodian and asks: can we back a dollar-token with on-chain collateral instead? The obstacle is that on-chain collateral (ETH, and other volatile crypto) can crash. You cannot back $1 with exactly $1 of ETH, because if ETH drops 20% your backing evaporates.

The answer is the mechanism you already met on the lending page: over-collateralization plus liquidation. This is DAI, issued by the MakerDAO protocol, and its peg is defended entirely on-chain — no bank, no custodian, no blocklist for the collateral logic itself.

   You want $100 of DAI.                Price of your ETH falls.
   ─────────────────────                ──────────────────────────
   Lock $150+ of ETH in a vault.        Vault's collateral ratio drops
   Protocol mints 100 DAI to you.       toward the minimum (e.g. 150%).
        │                                   │
        ▼                                   ▼
   You now hold a dollar-token,         LIQUIDATION: anyone can repay
   over-collateralized 1.5×.            your DAI debt and take your ETH
                                        at a discount — the debt is
                                        covered before it goes bad.

The DAI you minted is a debt you owe. To get your ETH back you repay the DAI (plus a stability fee, MakerDAO’s interest rate — the same lever the lending page’s rate curve pulls). Because every DAI in existence is backed by more than a dollar of locked crypto, and because liquidation forcibly closes any position before its backing falls below its debt, the system as a whole always holds more collateral than DAI outstanding.

Where the peg pressure comes from

Over-collateralization guarantees DAI is backed, but backing alone does not pin the market price to $1. Two forces do that:

The redemption floor. A holder can always repay $1 of DAI to unlock $1 of their collateral. If DAI trades below $1, borrowers buy cheap DAI to repay their debts at a discount — buying pressure that pushes it back up.
Monetary levers governed by the DAO. MakerDAO governance (holders of the MKR token — see DAOs & Governance) adjusts the stability fee and other parameters to tighten or loosen DAI supply, nudging the peg. Raising the fee makes minting expensive and contracts supply (price up); lowering it does the reverse.

The trade-off is capital efficiency: to put $100 of DAI into the world, someone must lock up $150+ of crypto that now sits idle. That is deeply inefficient compared to USDC’s 1-for-1. What you buy with that inefficiency is a peg whose collateral you can verify on-chain — you can read the vaults, the total debt, and the liquidation rules directly from the contracts. The trust did not vanish; it moved from a custodian’s balance sheet to the price oracle feeding the liquidation engine (the subject of the next page) and to the DAO’s governance.

The over-collateralization ratio is a bet on how far collateral can fall before liquidators can close the position. Work it through:

   Debt minted:            100 DAI  (target: $100 of value)
   Collateral required:    150%  → must lock $150 of ETH

   ETH price drops 30%:    $150 → $105 of collateral
   Still covers the $100 debt + gives liquidators a margin
   to act and take their liquidation bonus.

   ETH price drops 40%:    $150 → $90 of collateral  ◄── now UNDER the debt!
   The position is "underwater": backing < debt.

The buffer between the minting ratio (150%) and insolvency (100%) is the time-and-price room liquidators need. Set it too low and a fast crash leaves debt un-backed; set it too high and you waste even more capital. This is the same tension as a lending protocol’s liquidation threshold — a stablecoin is a lending protocol pointed at a peg.

Family 3: Algorithmic / seigniorage — and why it collapsed

The third family is the most seductive and the most dangerous. It asks: do we even need collateral? Can code alone hold the peg? An algorithmic (or seigniorage-style) stablecoin tries to defend $1 not with reserves but with a minting-and-burning rule against a second, volatile “sister” token.

The canonical example is Terra, whose stablecoin UST was paired with a governance/absorber token called LUNA. The rule was a fixed swap the protocol always honored:

   The protocol always lets you swap, at the PEG price:

       1 UST   ◄────────►   $1 worth of LUNA

   UST below $1  (say $0.98):
       buy 1 UST for $0.98, swap it for $1 of LUNA, sell → +$0.02
       (burns UST, mints LUNA)  ► UST supply shrinks ► price up

   UST above $1  (say $1.02):
       give $1 of LUNA, mint 1 UST, sell for $1.02   → +$0.02
       (burns LUNA, mints UST)  ► UST supply grows ► price down

On paper this is elegant: no idle collateral, perfect capital efficiency, a peg defended by pure arbitrage. The problem is what the peg is ultimately backed by. When you redeem UST, you receive LUNA — a token whose only value comes from the belief that UST will stay pegged. The two props hold each other up. That circularity is called reflexivity, and it is the fatal flaw.

The reflexive death spiral

Everything works while confidence is high. The mechanism inverts the moment confidence breaks:

   1. UST slips below $1 (large sell-off, or yield dries up).
   2. Arbers redeem UST → protocol MINTS fresh LUNA to pay them.
   3. New LUNA supply floods the market → LUNA price FALLS.
   4. Falling LUNA means each UST is "backed" by less value,
      so holders panic and redeem MORE UST.
   5. More redemptions → even MORE LUNA minted → LUNA falls further.
        │
        └──────────►  loop tightens with each turn  ◄──────────┐
                       (hyperinflation of LUNA;                 │
                        UST peg is never recovered) ────────────┘

There is no external anchor to stop the loop. In a crypto-collateralized system, a crash eventually hits a floor where the locked ETH is still worth something and liquidation clears the debt. In an algorithmic system, the “collateral” is a token being printed in unlimited quantity precisely as its price collapses — the defense mechanism is the thing destroying the value. The faster it tries to defend the peg, the faster it dilutes the asset backing it.

In May 2022, UST lost its peg and the death spiral ran to completion in days. LUNA’s supply hyperinflated from roughly 350 million tokens to over 6 trillion, its price falling from around $80 to fractions of a cent. UST never recovered. An estimated ~$40 billion+ of value evaporated across UST and LUNA, wiping out retail holders worldwide and triggering a cascade of failures (Three Arrows Capital, Celsius, and others) across the industry that summer.

The lesson is structural, not incidental: a peg backed by a sister token is backed by nothing external. No amount of clever minting logic escapes reflexivity. Terra had scaled to a top-10 crypto asset while carrying a design flaw that only shows up under stress — which is exactly when a stablecoin must not fail. Contrast this with the USDC de-peg, where real reserves existed the whole time and the peg recovered; here there was no floor to catch the fall.

The trade-off summary: where each family hides its trust

Every stablecoin is a bet on a different scarce resource. Read this table as “what did I give up, and where did the trust go?”

   Family                Peg backed by         Trust lives in         Weakness
   ──────────────────────────────────────────────────────────────────────────────
   Fiat-collateralized   off-chain $ / T-bills  a custodian + auditor  bank/issuer
   (USDC, USDT)          1:1, capital-efficient (centralized, off-chain) can freeze/fail

   Crypto-collateralized on-chain crypto,       the price oracle +     capital-
   (DAI)                 over-collateralized     the DAO's governance   inefficient;
                         1.5×+                   (verifiable on-chain)  oracle risk

   Algorithmic           a reflexive sister      pure confidence        DEATH SPIRAL
   (UST — collapsed)     token (nothing real)    (nothing external)     under stress

The three families trade off along the same axis the whole book keeps returning to — decentralization vs. capital efficiency vs. peg robustness, where you cannot maximize all three:

Fiat-backed buys robustness and efficiency by sacrificing decentralization: it works beautifully until an off-chain institution fails, and it can censor.
Crypto-backed buys decentralization and robustness by sacrificing capital efficiency: verifiable on-chain, censorship-resistant, but locks up far more value than it issues, and it lives or dies by its oracle.
Algorithmic promised all three at once by sacrificing the collateral itself — and paid for that promise with a total collapse. The efficiency was real; the robustness was an illusion that only held while no one tested it.

No stablecoin removes trust. It only relocates it — into a bank, into an oracle and a governance vote, or (fatally) into collective belief with nothing underneath.

The architect’s lens

Why does it exist? Because ETH is money you can send anywhere but a terrible unit of account — too volatile to price a wage, a loan, or a coffee in. A stablecoin gives the world computer a stable-valued asset while keeping it on-chain and composable.
What problem does it solve? It pins a freely-traded token’s market price to ~$1 by building a peg-maintenance mechanism — an arbitrage incentive that a profit-seeker restores whenever the price drifts.
What are the trade-offs? You cannot maximize decentralization, capital efficiency, and peg robustness together. Fiat-backed gives up decentralization; crypto-backed gives up capital efficiency; algorithmic gave up the collateral and paid with collapse.
When should I avoid it? Avoid algorithmic (uncollateralized reflexive) designs entirely — Terra proved they fail exactly under the stress a stablecoin must survive. Avoid fiat-backed when censorship-resistance is the whole point, since the issuer can freeze you. Avoid crypto-backed when capital efficiency is critical, since it locks up far more than it issues.
What breaks if I remove it? DeFi loses its unit of account. Lending, AMMs, and payments all quote and settle in dollar-terms; without a stable denominator, every position becomes a directional bet on ETH, and the ecosystem’s usefulness for ordinary finance collapses.

Check your understanding

A token is named “USD-Coin,” so does its market price have to be $1? Explain what “the peg” actually is and what defends it.
USDC is fiat-collateralized and DAI is crypto-collateralized. For each, name specifically who or what you are trusting for the peg to hold, and where that trust lives (on-chain or off-chain).
DAI reuses the lending page’s core mechanism. Name the two parts of that mechanism and explain how they guarantee DAI is always backed by more than a dollar.
Walk through the Terra/UST death spiral step by step. Why does minting LUNA to defend the peg make the collapse worse rather than better?
State the three-way trade-off every stablecoin faces, and say which corner each of the three families sacrifices.

Show answers

No. A stablecoin is a freely-traded token, so its market price is whatever buyers and sellers agree on — the name forces nothing. “The peg” is the target price (~$1), and it is defended by a peg-maintenance mechanism: an arbitrage incentive that pays a profit-seeker to buy when the token is below $1 (pushing price up) or create-and-sell when it is above $1 (pushing price down), restoring $1.
USDC: you trust a custodian (Circle) to actually hold $1 of reserves per token and honor redemption, plus auditors to verify it — trust that lives off-chain and can freeze addresses. DAI: you trust the price oracle feeding liquidations and the MakerDAO governance (MKR holders) setting parameters — trust that lives on-chain and is verifiable by reading the contracts.
The two parts are over-collateralization (lock more than $1 of crypto per $1 of DAI, e.g. 150%) and liquidation (anyone can forcibly close a position before its collateral falls below its debt). Together they guarantee that every DAI is backed by more than a dollar and that positions never stay underwater.
UST slips below $1 → arbers redeem UST for $1 of LUNA, so the protocol mints new LUNA → that new supply drives LUNA’s price down → less value now backs each UST, so holders panic and redeem more UST → more LUNA is minted → LUNA falls further. It worsens the collapse because the “collateral” (LUNA) is being printed in unlimited quantity exactly as its price crashes — the defense mechanism itself dilutes the asset backing the peg, and there is no external floor to stop the loop.
The trade-off is decentralization vs. capital efficiency vs. peg robustness — you cannot have all three. Fiat-backed (USDC/USDT) sacrifices decentralization (off-chain custodian, can censor). Crypto-backed (DAI) sacrifices capital efficiency (locks up 150%+). Algorithmic (UST) sacrificed the collateral itself, trading real robustness for an illusion that collapsed under stress.