An algorithmic stablecoin holds its peg by adjusting supply through smart contracts rather than by holding reserves, expanding supply when the price runs above the peg and contracting it when the price falls below. Few or no assets sit behind it. The Federal Reserve's own taxonomy splits the category into two mechanism families, rebase and coupon, and states that both hold the peg only while enough market participants expect them to.
Expanding supply is trivial and contracting it is not, so every design in this category answers one question: what will a holder accept in exchange for handing back a stablecoin below par? Answer with a second token you also issue, and the peg is collateralised by demand for that token.
The definition, from a central bank rather than a whitepaper
The Federal Reserve Board's FEDS Note of 16 December 2022 puts it plainly. Uncollateralised stablecoins, also known as algorithmic stablecoins, aim to maintain exchange rate stability by employing a set of rules and strategies, usually implemented by smart contracts, that dynamically match the supply of the stablecoin with user demand, and in contrast with collateralised stablecoins few or no assets are held in reserve to support the exchange rate.1
The stabilising loop is a supply rule. Above $1, new coins are issued to devalue each existing token; below $1, coins are removed from circulation to raise the value of each one.1 The asymmetry is the whole story, and the Fed states it directly: increasing supply can be implemented by simply distributing new coins, while decreasing supply can be more challenging.1
Issuing is a mint call. Removing requires persuading somebody to hand a token back, which means offering them something in return, and the entire design space lives inside that exchange.
Rebase and coupon, and where seigniorage sits
The Fed splits algorithmic designs into two subgroups. The rebase model adjusts the total supply at regular intervals across every wallet holding the coin, proportional to the price deviation from the peg, and continues until the peg is reached. Ampleforth is the named example.1 Nobody is asked to do anything; balances simply change.
The coupon model instead gives holders an incentive to change their holdings deliberately. Supply contracts by inducing users to exchange stablecoins for interest-bearing coupons or bond tokens, which pay out during later periods of expansion when the price is back at or above the peg. The Fed notes that this is also referred to as the seigniorage model, because the stabilisation relies on issuing new coins and selling bond tokens in a manner similar to a central bank's open market operations, and names Terra Classic USD as its example.1
The dual-token form of this is what most founders mean when they say seigniorage. The stablecoin's supply is paired with a secondary share token, and holders may mint one by burning the other, which creates the arbitrage that is supposed to hold the peg. Terra's UST and LUNA operated on this model before their collapse in May 2022.2 The seigniorage entry covers where that value actually accrues; what matters here is that the share token is the balance sheet.
Scroll to see the full diagram
Both families fail on belief, and the mechanism documents say so
This is the part that surprises people, because it is not a critique from outside. The Fed's own analysis states that supply adjustment returns the price to the peg only if enough market participants expect it to happen, and that if enough of them believe the value will drift away, each holder has an incentive to sell before the others do, generating a self-fulfilling run.1 It describes this for the rebase model and then repeats it for the coupon model.
So the peg is not maintained by the algorithm. It is maintained by a shared expectation that the algorithm will work. Once that goes, the same code that stabilised the peg accelerates the exit, because contracting supply in a falling market means issuing more claims into a market that no longer wants them.
That is a design constraint, not a moral judgment. An algorithmic design cannot be assessed on whether the arithmetic balances at the median, which it does. It has to be assessed on what the mechanism prints at the tail.
What the UST postmortems actually establish
One evidence note first. Terraform Labs' own pre-collapse description of the UST and LUNA mint-and-burn mechanism could not be located at any verifiable live or archived URL in the research behind this page, and neither could the original Basis whitepaper that named the seigniorage-share pattern. What follows comes from independent academic and institutional postmortems instead.
Uhlig's NBER working paper records the sequence. UST held close to one US dollar from its inception in November 2020, then crashed in the two weeks of 9 to 15 May 2022, losing more than 75% of its value while LUNA fell more than 99.9%, LUNA supply increased by a factor of 19,000, and more than $50 billion, around 90% of market value, was erased.3 His model of why is the useful part: holders sold their UST once the probability of an eventual suspension of convertibility crossed an individual threshold, and he finds that the majority waited until that probability was already rather high before burning their holdings.3
Liu, Makarov and Schoar reach the collapse from the run side. Terra's combined market capitalisation was $50 billion before the crash with average daily trading volume of $1 billion, making it the third-largest ecosystem, and within three days its value fell to zero.4 Their finding that should change how founders think about redemption design: although all investors had access to the same blockchain, wealthier and more sophisticated investors processed information more efficiently, ran more decisively and realised much smaller losses, and system complexity made it difficult even for insiders to assess the buildup of risk.4 A redemption mechanism open to everyone was not usable by everyone at the same speed.
What we tell founders who want this design
Start with the identity nobody writes down. If defending the peg mints share tokens, the cost of defence is denominated in an asset whose price falls as you spend it. Model that at the tail: what does the mint schedule print if the share token halves while redemptions run, and does the dilution close the loop or widen it? If the loop only closes while the share token holds a bid, the peg is collateralised by sentiment, and that belongs in the documentation rather than in a holder's discovery.
Second, separate the redemption promise from the redemption capacity. Terra's data shows that a formally open mechanism delivered materially different outcomes by sophistication.4 If your design has a queue, a cap, an oracle window or a gas-cost floor, those are rationing devices, and they decide who gets out. Publish them.
Our position is that the interesting work in stablecoin design is not the algorithm. It is what backs the promise and who is obliged to honour it, which is where most of a stablecoin engagement goes. This page is reference material for design work, not investment advice, and the regulatory treatment of any specific design is jurisdiction-specific and a question for your counsel.
Common questions
What is an algorithmic stablecoin?
It is a stablecoin that holds its peg by using smart contracts to adjust supply rather than by holding reserve assets. The Federal Reserve defines the category as stablecoins that dynamically match supply with user demand through rules implemented in code, with few or no assets held in reserve to support the exchange rate, issuing coins above the peg and removing them below it.1
What is the difference between a rebase and a seigniorage stablecoin?
A rebase model changes every holder's balance proportionally at set intervals until the peg is reached, with Ampleforth as the named example. A coupon or seigniorage model instead induces holders to swap stablecoins for bond or share tokens that pay out in later expansion periods, which is the family Terra Classic USD belonged to.1 The seigniorage version is usually implemented as a paired two-token mint and burn.2
Why did UST fail?
Because the mechanism depended on holders expecting to be able to monetise what they received on redemption, and that expectation broke. UST fell more than 75% between 9 and 15 May 2022 while LUNA fell more than 99.9% and its supply rose by a factor of 19,000, erasing over $50 billion.3 Holders sold once the probability of a suspension of convertibility crossed a personal threshold, and most waited until it was already high.3
Are algorithmic stablecoins inherently unstable?
They carry a run risk that reserve-backed designs do not, and the Federal Reserve's own analysis says so directly: supply adjustment holds the peg only if enough market participants expect it to, and otherwise each holder has an incentive to sell before the rest, producing a self-fulfilling run.1 Whether a particular design manages that risk acceptably is a question for its specific parameters and your own analysis.
See Tokenomics Design for how this applies in practice.
Sources
- The Stable in Stablecoins (FEDS Notes)
Board of Governors of the Federal Reserve System, 2022
Dated 16 December 2022. Defines uncollateralised or algorithmic stablecoins, sets out the rebase and coupon subgroups with Ampleforth and Terra Classic USD as the named examples, states that the coupon model is also called the seigniorage model, and records that both families hold the peg only while enough participants expect reversion. Read 3 August 2026. - Algorithmic Stablecoins
BitGo, 2026
Institutional custodian's description of the seigniorage or dual-token pattern: the stablecoin's supply is paired with a secondary token and holders may mint one by burning the other, with Terra's UST and LUNA named as the model's operators before their May 2022 collapse. Read 3 August 2026. - A Luna-tic Stablecoin Crash (NBER Working Paper 30256)
Harald Uhlig, National Bureau of Economic Research, 2022
July 2022. Records UST close to one dollar from November 2020 until the crash of 9 to 15 May 2022, a loss of more than 75% with LUNA down more than 99.9%, LUNA supply up by a factor of 19,000 and over $50 billion or 90% of market value erased, and models holders selling once the probability of a suspension of convertibility crosses an individual threshold. - Anatomy of a Run: The Terra Luna Crash
Jiageng Liu (MIT Sloan), Igor Makarov (LSE) and Antoinette Schoar (MIT Sloan and NBER), hosted by Sveriges Riksbank, 2023
Independent run analysis: $50 billion combined market capitalisation and $1 billion average daily volume before the crash, value to zero within three days, and the finding that wealthier and more sophisticated investors ran more decisively and realised much smaller losses despite equal access to the blockchain.
Last reviewed 2026-08
More in Token Types and Classification
Know the terms but not sure how they apply to your project? That is what an engagement is for. We design, document, and stress-test the whole token economy inside the Tokenomics Data Room.
80+ projects advised. Complete tokenomics in 4 to 6 weeks.