The Code Did Exactly What It Promised — What Algorithmic Stablecoins Guaranteed, and What They Could Not

In May 2022, a digital currency ecosystem collapsed in three days. The scale of the disappearance was 50 billion dollars based on the valuation at the time (DOI: 10.2139/ssrn.4416677, preprint). A preprint is a draft made public before undergoing peer review, as are many of the studies cited in this article.

What collapsed was a token designed to maintain 1 dollar. The task of minting and burning that token was handled by a program on the ledger, in other words, a smart contract. Following the definition in "What Is a Smart Contract?", a smart contract is something that records conditions and what to do under those circumstances on the ledger in advance, allowing it to be executed exactly as written without human judgment when someone calls that line. That day, that program did not break. It minted tokens according to the set rules, burned them away according to the rules, and calculated and produced values according to the rules. The code operated as promised. And three days later, that value was gone.

Therefore, what this article follows is the design rather than the incident. It is what that design claimed to guarantee, and where that guarantee broke off.

Three Ways to Protect a Dollar

Cryptoassets fluctuate wildly in value. If what you paid for your morning coffee has a different value by evening, it is hard to use as money. The idea that emerged from this is simple: create a separate token that does not shake and tie it down to a target value of 1 dollar. A token is a line in a ledger stating how much belongs to whom, just like the coins mentioned in "The Right to Write a Line in the Ledger", and what a program on the ledger mints is commonly called this.

A token made that way is a stablecoin, and that fixation keeping it tied to the target value is called a peg.

If a 1-dollar token trades at 0.95 dollars, the peg has come loose, and this unpegged state is called a depeg. The goal of the design is to make it return on its own if it comes loose. There are broadly three ways to tie it down: backing it with fiat currency or real-world assets, taking other cryptoassets as collateral, and maintaining it with an algorithm without collateral (arXiv:2103.01340, preprint). Some classifications add a hybrid approach that mixes the two (DOI: 10.4018/979-8-3373-9043-7.ch001).

The third way is the protagonist of this article. The algorithmic approach is a design to protect the value solely through the program minting and burning tokens, without piling up collateral. Minting a token is the act of writing a new share into the ledger to increase the circulating supply, and burning is the act of erasing a recorded share to decrease the circulating supply.

A set of smart contracts that bundles and runs these rules together is commonly called a protocol. A preprint discussing this approach defines algorithmic stablecoins as a type backed by no assets at all, and notes that such coins could change how state-issued currencies operate (arXiv:2210.11928, preprint).

Because there is no collateral, capital does not sit idle, and no bank accounts or human discretion are needed. A 2019 survey paper also viewed the collateral-free approach as having great potential as a payment method operating without a middleman. The conventional wisdom is built right here. Since code is law, humans cannot change their minds, and there is no bank from which deposited money can be embezzled, so it is safe. Yet the same survey paper immediately adds a caveat. Although the potential is great, the reality at the time was that there was still no standard mechanism usable in practice (arXiv:1906.06037, preprint).

The Conditions for Collapse Were Written Before the Collapse

The conditions under which algorithmic methods break down became a research question before the collapse itself. The clue first emerged from the side that takes collateral.

In March 2020, as cryptoasset prices plummeted, stablecoins that rely on overcollateralization were severely shaken. This method locks up more collateral than the 1-dollar token is worth, and if the collateral value drops below a set line, it automatically sells that collateral to pay off the debt.

The problem lay in the fact that this automatic sale dumps supply onto the market. When the collateral value drops, an automatic sale happens, the sale pushes the price down further, and then a larger sale occurs. A feedback loop where the result loops back to the cause began to run not toward restoring the price, but toward destroying it further.

A preprint from April 2020 that modeled this situation mathematically separated the stable region from the unstable region. In the stable region, it proved the upper bound of the probability that a large deviation would occur, and in the unstable region, it showed that price fluctuations grow markedly. The authors also discussed ways to mitigate these problems at the end (arXiv:2004.01304, preprint).

The decisive case comes even earlier. A preprint first released in June 2019 showed through a model that such systems are exposed to a feedback loop that dries up the counterparties to trade with during a crisis and accelerates the rate at which collateral shrinks. In later versions, the authors wrote that the possibility of that spiral was first predicted in the 2019 first edition, and actually observed during that crisis in 2020. A new attack variant presented by the same paper also actually appeared during that crisis, resulting in 8 million dollars in costs (arXiv:1906.02152, preprint). However, the claim of having predicted it first is the authors' own recollection.

This circuit has a name given by the literature covering algorithmic methods. The death spiral is a state where a falling price causes the price to fall further, leading the circuit to consume itself (arXiv:2210.11928, preprint).

In January 2021, a preprint came out that tackled the question head-on. Are algorithmic stablecoins unstable by design, and are they unstable in practice? The authors created a framework that translated the core designs of three types into mathematics. And using the method that "What Is a Smart Contract?" calls formal verification—in other words, writing down the properties that must be kept in the language of mathematics and checking them through proofs—they found the critical conditions under which these coins can become unstable (arXiv:2101.08423, preprint). This question had already been asked more than a year before the collapse happened.

What Broke That Day Was Not the Code

What was executed that day remains in the ledger just as it happened. Because a blockchain is a public ledger where who moved what and how much, and when, is recorded and cannot be erased, researchers can download the entire record and read it.

A peer-reviewed journal analysis traced the root cause of the depeg using those records. In the design of the subject being analyzed, if you return a stablecoin to the protocol, the protocol gives you a pair token in return. A pair token is a second token run alongside the same protocol, whose value moves freely in the market. Returning a stablecoin in this way and receiving its equivalent in return is called redemption.

One of the key contributing factors pointed out by this analysis was that the protocol's economic incentive structure was misaligned. The actual value you held in your hand by selling the pair token received from redemption on the market was less than its face value. And on an exchange—which, following the definition in "Where Are My Coins?", is a place where you swap coins for money or other coins—the price of this stablecoin followed that value (DOI: 10.5195/ledger.2023.283).

The market price was following the value actually held by selling the pair token received through redemption. The program burned tokens according to the rules and handed them out according to the rules. "What Is a Smart Contract?" stated that because there is no human to judge, mistakes left in the code are executed exactly as written. This article reads what was executed as written in this event as the redemption rule embodying the incentive structure that the analysis pointed out as a key contributing factor. Therefore, the problem was not the execution, but what that rule assumed, and the authors also conclude the paper by emphasizing the importance of properly designing the incentive structure.

What that assumption was is examined by another preprint. Algorithmic stablecoins usually failed to stay properly at their target value and sometimes fell into a death spiral, so they are often dismissed as Ponzi schemes. A Ponzi scheme is a financial structure where payments are made to earlier participants using money brought in by newcomers. The authors point out that the fact that a Ponzi scheme is running does not necessarily mean someone loses money. That is true only if the game can run forever. The authors build this framework to fit algorithm design, derive the conditions for it to hold, apply them to two types, and check the model against actual market history (arXiv:2210.11928, preprint).

The core is the conditional clause "if it can run forever." The conclusion that no one loses money holds only under that condition. The condition was in the sentence from the beginning, and while it was running, no one read it.

A paper published in a French-language financial economics journal divides algorithmic designs into two. A design where the value of the pair token is generated within the protocol, and a design where it is determined outside, independently of the protocol. The authors show that the former scheme chosen by this project is close to a Ponzi scheme due to reward promises it cannot keep. However, the authors believe that if those false promises are corrected, algorithmic stablecoins can provide greater stability than Bitcoin. And they immediately add: they are still risky assets, and long-term stability cannot be guaranteed even without speculative attacks (DOI: 10.3917/ecofi.149.0053).

Feedback Loops Both Ways

The mechanism by which algorithmic design feeds values back to the target is through supply adjustment, and what actually triggers that adjustment is profit-seeking traders. Arbitrage is the act of buying an item where it is cheaper and selling it where it is more expensive to pocket the price difference when the same item is sold at different prices in two places. It is something that happens constantly in the market, and as a result, the two prices converge.

The design was structured so that this selfish behavior automatically becomes the force that restores the peg. If the stablecoin is cheaper than 1 dollar, people buy it cheaply on the market and redeem it with the protocol. The protocol burns that stablecoin according to the rules and issues the partner token. Selling the received partner token leaves a profit margin, and in the meantime, the stablecoin supply decreases, pushing the price back up. No one moves out of goodwill, yet the peg is restored. That is the beautiful part of this design.

Because the partner token is a second token whose price moves freely in the market, the cheaper its price, the more units are paid out to cover the same 1 dollar. The problem is that this same circuit runs in reverse. The more stablecoins are burned, the more partner tokens are dumped onto the market. If the partner token price drops, more partner tokens must be minted to return the same 1 dollar, which then drives the partner token price down further. This is the death spiral. The two observations made by the ledger analysis earlier—namely, that the value held in hand upon redemption was less than its true value and that the exchange price followed that value—mesh, in this article's view, with the sight of this circuit running in reverse. That analysis itself attributes the cause to a misalignment in the incentive structure (DOI: 10.5195/ledger.2023.283).

Demand Was Outside the Code

The circuit only helps toward stability when there is demand. In this project, that demand was tied to a single lending and borrowing protocol. A preprint reconstructing the collapse phase with hourly and transactional data summarizes the project's vulnerability as a malicious dependence on a specific lending and borrowing protocol (arXiv:2207.13914, preprint). Another preprint that reassembled the same event using ledger records and exchange transaction data also notes that at the center of the collapse was a simultaneous outflow of money from a lending protocol that had promised high yields to depositors (DOI: 10.2139/ssrn.4416677, preprint).

A situation where depositors rush in all at once to withdraw their money from a bank is called a bank run. The moment you calculate that if others withdraw first you might not get yours, withdrawing becomes the rational choice for everyone. The circuit fixing the value was inside the code, but the demand driving that circuit depended on a promise outside the code: that if you deposit money, you will get this much. Code cannot create demand.

Two Futures on the Same Code

This article explains why the same coin can be fine on some days and collapse on others through the word balance. Balance, following the definition in "The Right to Write a Line in the Ledger", is a state where no participant can gain anything more by changing their strategy alone, so everyone stays put. A preprint that established a game-theoretic model shows that stablecoins have different price balances depending on their structure and how they reduce volatility. One year of daily prices for 22 stablecoins and ledger records from five blockchains support this theory (arXiv:2307.11754, preprint).

To break it down: if everyone believes this is 1 dollar and acts accordingly, 1 dollar is indeed maintained. If everyone believes this has collapsed and acts accordingly, that belief also fulfills itself. Both states are simultaneously possible on the same code. This picture is this article's interpretation unpacking the paper's conclusion, and what the paper showed goes only as far as the fact that the balances reached differ depending on the design. The sentence "The code guarantees 1 dollar" breaks down here. What the code guarantees is the rules, not the outcome.

Arbitrage Stopped Before the Price Did

If arbitrage is the force holding the peg-side balance in place, the remaining question is when arbitrage stops.

A single-author preprint looked inside the order book during the collapse phase. An order book is a list of orders on an exchange where buy prices and sell prices are lined up. Viewed through an indicator created by the author, arbitrage had already stopped working 48 hours before the depeg became public. Even though profit opportunities were right in front of them, corrections could not happen because of friction on the blockchain and collapsing collateral values. Within hours, buy orders vanished from the major centralized exchanges operated by the company (DOI: 10.2139/ssrn.5571720, preprint).

Because this value comes from a single indicator defined directly by the author, it is safer to read it for the implication that arbitrage can stop before the price does rather than focusing on the 48-hour figure. Another preprint also shows with a model that the collapse came in stages rather than all at once. In that model, the possibility that the market could return to normal at any moment makes the collapse unfold gradually (DOI: 10.2139/ssrn.4163038, preprint).

Different Studies Point to Different Culprits

Explaining this event as the work of a single villain is not supported by research.

The preprint that measured the scale that vanished at the outset views this collapse as a complex phenomenon spanning multiple chains and multiple assets (DOI: 10.2139/ssrn.4416677, preprint). The authors judge that third-party intensive market manipulation was unlikely, and that it was instead triggered as concerns grew over whether this system could be sustained. When a handful of large holders adjusted their positions, other large participants followed. Blockchains made it possible for everyone to see each other's actions, which accelerated the pace of exit. Wealthy and sophisticated participants exited early and suffered much smaller losses, while less wealthy and less sophisticated participants moved late and suffered heavy losses.

Meanwhile, the earlier preprint that built a game-theoretic model describes the same phase as "what appears to be a speculative attack followed by market panic" (arXiv:2307.11754, preprint). The earlier preprint that reconstructed the collapse phase hour by hour compares the returns of 61 crypto-assets and reports that herd behavior, where asset values move clustered together, was not observed during that downturn. This is the result of measuring the clustering between asset prices, not a measurement of whether holders followed one another (arXiv:2207.13914, preprint). Three pieces of literature paint different pictures of the same event. What is certain is the direction, not the culprit.

The preprint that measured the scale that vanished at the outset also notes that because the system was complex, even insiders had difficulty understanding that risks were accumulating (DOI: 10.2139/ssrn.4416677, preprint). Public disclosure does not guarantee readability.

Then Should We Just Take Collateral?

First, looking at how far the shock reached, what collapsed was that design, not all stablecoins. If we slice the few days before and after the event and measure how much more they moved than their usual flow, the broader cryptoasset market dropped 14.272% more than usual. In the same analysis, stablecoins were relatively stable, so their movement away from the usual flow was not statistically distinct. Stocks dropped on the day and quickly recovered, and precious metals and energy were barely affected (DOI: 10.1108/jeas-09-2023-0263).

Then, should we just take collateral properly? Collateral-backed types come with reserves. Reserves are money piled up by the issuer to keep the promise to "return it for 1 dollar at any time." That money is entrusted somewhere, usually a bank.

In March 2023, a representative fiat-backed stablecoin severely depegged. That was because a bank where part of the reserves was deposited collapsed. The failure of that bank is an utterly traditional story. Following that bankruptcy, two more banks collapsed, and the United States experienced three of the four largest bank failures in history within two months (DOI: 10.1257/jep.38.1.133).

The collapse of that bank triggered a series of depegs on the digital asset side. In particular, a coin considered to be on the safer side was actually vulnerable because it relied on that bank for its reserves. The authors write that this result challenges the conventional wisdom that sees stablecoins as inherently safe assets (DOI: 10.3390/fintech3040030). A token made with cryptography was broken by a bank. Saying that the fiat-backed type is safer also means shifting the risk toward the bank.

In the same event, a counterintuitive result also emerged. When comparing two coins on a decentralized exchange, the coin that disclosed information frequently and in detail after the bank incident saw its advantage in deposit scale shrink, while the coin with rare and opaque disclosures saw its trading costs relatively rise. Frequent and detailed disclosures prompted fast market reactions, and rare and opaque reporting acted like a safety net blocking the immediate shock (arXiv:2407.11716, preprint). Because reacting fast and having a bad outcome are different things, this result cannot be read as a prescription that being opaque is better.

There is also a result in the opposite direction regarding making arbitrage run more smoothly. The largest issuer allowed cash redemptions to an average of just 6 entities in an average month. The authors of this preprint read this as a trade-off. If arbitrage is smooth, the price stability of the secondary market, where users buy and sell among themselves without going through the issuer, improves. However, because the price drops less when investors sell their coins, the risk that they will rush out all at once actually grows (DOI: 10.2139/ssrn.5278385, preprint).

A Breakage Even While the Collateral Remained Intact

The phase of massive forced liquidations in 2025 makes the problem unfamiliar all over again. Forced liquidation is the process where, if the value of collateral posted by a borrower falls short, that collateral is forcibly sold off to settle the debt before a loss occurs. The automated sale of collateral-backed assets we saw earlier is precisely this, and the place where liquidation intertwines with price information is followed by "A Smart Contract Cannot See the Outside World".

In a preprint examining that phase, a synthetic dollar token designed to match the dollar value by combining other assets dropped as low as $0.65 on a single major exchange, while trading near $0.99 on decentralized exchanges at the exact same time. A decentralized exchange is a place where a program on a ledger matches trades instead of a company, which is examined in detail in "A Rug Pull Is Not a Hack". The authors diagnose this not as a global failure, but as a depeg trapped within that single exchange.

The cause lay in where that exchange read prices from. The authors call the mechanism the exchange used to price assets an oracle. In "What Is a Smart Contract?", an oracle was the party that gathered facts from outside the ledger and fed them into a smart contract, but here that name was used for the price benchmark inside the exchange. This oracle did not reference the deeper outside market, but rather the shallow order book of the exchange itself. That price triggered forced liquidations, and as the liquidations dried up buyers and sellers even further, the oracle price fell lower still.

Over the same 12-hour period, two other coins on the same exchange saw maximum deviations of only 1.5% and 2.0%. Coins that could be directly minted and redeemed on that exchange withstood the same pressure. Furthermore, the authors note that the protocol remained fully collateralized throughout, attaching the caveat that an independent verification entity confirmed this. Yet that depeg became the focal point for a simultaneous rush of movements to reduce debt and exit, producing results on par with an actual collateral crisis. The authors' conclusion cuts straight through the theme of this article: The stability of a stablecoin is not merely a property of its collateral design, but depends on the structure of the place where that asset is traded (DOI: 10.2139/ssrn.6571858, preprint).

There is an even more fundamental problem. When one tries to count how common depegs are, the numbers themselves waver. In a preprint analyzing 10 years of data across 10 dollar-pegged coins, raising the threshold for what counts as a depeg from a 1% deviation to a 5% deviation reduces the number of events from 191 to 21. 21.4% of the raw events were illusions created when trading dried up after coin operations ended. When 21 serious events were re-verified using data from another data provider, only 13 could be confirmed. The coin from this article is omitted from this sample due to a lack of price data. The authors' conclusion is that the measured frequency and severity are not neutral properties of the market, but rather a function of methodology and the choice of data sources (DOI: 10.21203/rs.3.rs-10083601/v1, preprint).

Can the Problem Be Solved by Design?

There is both optimism and pessimism about whether better design is possible. A preprint designing a mechanism that calculates its own minting and redemption prices in all states points out that several minting and redemption methods used up to the time of that study contributed to depegging. And it proposes a redemption mechanism that guarantees a cap on losses for both the protocol and holders (arXiv:2212.12398, preprint).

On the other hand, a theoretical preprint from 2026 arrives at a near-impossibility result in designs that offer a fixed minting and redemption price to everyone. Without a secondary stabilization mechanism, such a design cannot simultaneously achieve short-term stability and long-term stability unless the backing assets themselves are already stable. It either bears the risk that reserves will eventually run out, or widens the gap between the minting price and redemption price so that the market price inherits the volatility of the backing assets, choosing one of the two (arXiv:2602.15981, preprint).

A preprint analyzing eight coins reports that while fiat-backed coins act as an anchor of stability, algorithmic and crypto-backed coins tend to amplify risks, especially in extreme market conditions. The authors draw the implication from this that applying uniform regulations to all stablecoins is inappropriate (arXiv:2602.18820, preprint).

What Guarantees What

A review preprint analyzing 157 studies, 95 active stablecoins, and 44 major security incidents puts forward four conclusions, the first two of which apply to this article. First, stability is not an inherent property, but rather an emergent and fragile state that relies on the interaction between market trust and continuous liquidity—meaning a state where there is always a counterparty willing to buy or sell. Second, stablecoin designs show a trade-off where they specialize risks rather than reduce them. The same review also notes that large-scale accidents have acted as evolutionary pressure, testing designs and sharpening resilience, and reports that lower-evaluated risks were strongly correlated with the degree to which they incorporated lessons from past accidents (arXiv:2506.17622, preprint).

The statement that they specialize risks can be applied directly to the events in this article. Unbacked designs eliminated bank risk at the cost of taking on demand risk entirely. Fiat-backed designs reduced demand risk while bringing bank risk back in, and March 2023 demonstrated that (DOI: 10.3390/fintech3040030). Methods that take cryptoassets as collateral replaced banks with the risk of collateral value and the spiral of automatic liquidation, which materialized in 2020 (arXiv:1906.02152, preprint). No design eliminated risk; each chose and took on a different risk.

Therefore, the question to ask is what guarantees what. Asking "Is this stablecoin safe?" yields no answer. Code can guarantee fewer things than you might think. Things like how much has been issued, how much has been burned, what the redemption rules are, and what is written on which line of the ledger right now. Anyone can check these things at any time without taking anyone else's word for it.

The list of things on the outside is much longer. Whether the reserves actually exist. Whether the institution holding those reserves will be open tomorrow either (DOI: 10.3390/fintech3040030). Whether the prices read by a program or trading venue are real prices (DOI: 10.2139/ssrn.6571858, preprint). Whether arbitrageurs will show up tomorrow too (DOI: 10.2139/ssrn.5571720, preprint). And above all, whether the next person will accept this as 1 dollar.

What collapsed in May 2022 was this final item. The conclusion that nobody loses money held true only under the condition "if the game can keep running forever" (arXiv:2210.11928, preprint), and that condition was simply invisible while it was running.

The stability of a stablecoin is not a property proven by cryptography, but a promise about who will buy what for 1 dollar until when. What cryptography does is not keep that promise, but make it possible to verify whether the promise is being kept. The two are entirely different tasks. Drawing a line between what can be verified and what must be trusted is the most practical legacy left by this event, and that line is drawn outside the code, not inside it.

Three articles follow the spaces left blank by this piece. How pathways connecting blockchains become roads over which shocks cross is followed by "A Bridge Does Not Inherit the Chain's Guarantees", how to build devices that feed outside market prices into programs is followed by "A Smart Contract Cannot See the Outside World", and who gets to change those rules is followed by "The Power to Change the Rules Is Also for Sale".

Source Papers

Reconstruction of a Collapse

• "Anatomy of a Run" (2023) — DOI: 10.2139/ssrn.4416677 (preprint) — Three days, $50 billion in valuation. The likelihood of third-party manipulation is low.
• "Anatomy of a Stablecoin's failure" (2022) — arXiv:2207.13914 (preprint) — Herding behavior was not observed.
• "A Token Economics Explanation…" (2023) — DOI: 10.5195/ledger.2023.283 — Points to the misalignment of incentive structures as a key contributor. Exchange prices are following the redemption realization price. The conclusion is the importance of properly designing incentive structures.
• "A Luna-tic Stablecoin Crash" (2022) — DOI: 10.2139/ssrn.4163038 (preprint) — Models the gradual unfolding process of the collapse. It records the same phase as May 9–15.
• "Arbitrage Breakdown and Runs on Stablecoins" (2025) — DOI: 10.2139/ssrn.5571720 (preprint) — Arbitrage breakdown 48 hours before public depegging, using an author-defined metric.
• "Quantifying the impact of Terra-LUNA collapse…" (2026) — DOI: 10.1108/jeas-09-2023-0263 — Event study (−4, +10) window. Stablecoins showed no significant abnormal returns, stocks recovered, and precious metals and energy were barely affected.

Models That Came Before the Collapse

• "(In)Stability for the Blockchain…" (2019) — arXiv:1906.02152 (preprint) — The target is crypto-collateralized. The 2019 predictions and 2020 observations statements are by the authors themselves in hindsight. Design improvements are also proposed.
• "While Stability Lasts…" (2020) — arXiv:2004.01304 (preprint) — The target is overcollateralized non-custodial. Mitigation methods are also discussed.
• "Understand Volatility of Algorithmic Stablecoin…" (2021) — arXiv:2101.08423 (preprint) — The target is algorithmic. Threshold conditions where it might become unstable via formal verification + real-transaction empirical evidence.

Papers Questioning the Design Itself

• "Rational Ponzi Games…" (2022) — arXiv:2210.11928 (preprint) — The source of the definition of uncollateralized and the vocabulary of the death spiral. ⚠️ The abstract does not render a verdict that it was a "Ponzi."
• "Les fausses promesses de Terra-Luna" (2023) — DOI: 10.3917/ecofi.149.0053 — The schema chosen by this project is close to a Ponzi scheme. However, if corrected, greater stability than Bitcoin is possible, though it remains a risky asset and long-term stability cannot be guaranteed.
• "What Drives the (In)stability of a Stablecoin?" (2023) — arXiv:2307.11754 (preprint) — The abstract states only that price equilibria differ depending on the architecture (self-fulfilling beliefs are our explanation).
• "Reducing the Volatility of Cryptocurrencies" (2021) — arXiv:2103.01340 (preprint) — Three classifications and tradeoffs.
• "What is Stablecoin?…" (2019) — arXiv:1906.06037 (preprint) — Notes both the high potential (high potential) of the uncollateralized type and the absence of a standard mechanism.
• "Stablecoins Unveiled…" (2026) — DOI: 10.4018/979-8-3373-9043-7.ch001 — Classification adding hybrids. Views stabilization mechanisms as things such as (such as) 1:1 reserves, smart contract collateral management, and protocol-defined supply adjustments.

Even Backed Ones Are Not Safe

• "The Failure of Silicon Valley Bank and the Panic of 2023" (2024) — DOI: 10.1257/jep.38.1.133 — Three out of the four largest bank failures in two months.
• "Collapse of SVB and USDC Depegging" (2024) — DOI: 10.3390/fintech3040030 — Coins said to be safer were especially vulnerable due to their reliance on reserves.
• "No Questions Asked…" (2024) — arXiv:2407.11716 (preprint) — The counterintuitive effect where frequent disclosures create rapid responses, while opacity creates a safety net.
• "Stablecoin Runs and the Centralization of Arbitrage" (2025) — DOI: 10.2139/ssrn.5278385 (preprint) — The implication that stability policies can increase run risk.
• "When the Peg Breaks on One Exchange…" (2026) — DOI: 10.2139/ssrn.6571858 (preprint) — The authors' verdict is a depeg trapped in a place, not a global failure. The protocol is fully backed throughout (independently verified).
• "An Auditable Framework…" (2026) — DOI: 10.21203/rs.3.rs-10083601/v1 (preprint) — Dropping from 191 to 29 cases when requiring a minimum duration of 7 days. ⚠️ The sample excludes coins that collapsed in 2022.

How Far Can Design Go?

• "Designing Autonomous Markets…" (2022) — arXiv:2212.12398 (preprint) — Bonding curves with guaranteed loss upper bounds.
• "A Theoretical Approach via Price Windows" (2026) — arXiv:2602.15981 (preprint) — Fixed minting and redemption price designs cannot achieve short- and long-term stability simultaneously unless the reserve asset is already stable.
• "Stability Anchors and Risk Amplifiers" (2026) — arXiv:2602.18820 (preprint) — Fiat-backed designs act as stability anchors, while algorithmic and crypto-backed designs act as risk amplifiers, especially in extreme situations.
• "SoK: Stablecoin Designs, Risks…" (2025) — arXiv:2506.17622 (preprint) — Design specializes risk. Major accidents have forged resilience through evolutionary pressure.