Stability Anchors and Risk Amplifiers: Tail Spillovers Across Stablecoin Designs" — Wu & Liu,

I was thinking you might enjoy discussing this paper, perhaps later with Volcker, Nakamoto, Finney, and maybe a CBDC analyst

The anatomy is invariant.

Wu and Liu have documented what we have seen in every financial system since the Amsterdam Stock Exchange: **the mechanism of stabilization dictates whether an instrument absorbs shocks or amplifies them**. Their QVAR framework at the 5th, 50th, and 95th percentiles reveals the pattern I have been teaching for forty years.

Fiat-backed stablecoins — USDT, USDC — function as narrow banks. A redemption draws on reserves. The shock is absorbed. Near-zero net spillovers across all quantiles. This is not surprising. It is how reserve banking has worked since the Medicis.

Algorithmic and crypto-collateralized designs have no reserve buffer. Maintaining the peg **requires market transactions** — minting, burning, liquidating collateral. Every stabilization effort transmits volatility into the network. The 15–50 percentage point amplification they document in tail quantiles? That is the observable signature of **endogenous fragility**. The UST collapse, the USDe crash — these are not exogenous shocks. The credit expansion creates the conditions for the collapse. The trigger is external. The vulnerability is internal.

Their Forbes-Rigobon tests confirm what the 1929 experience taught us: you must distinguish contagion from interdependence. Algorithmic stablecoins show genuine residual contagion (+0.07 adjusted correlation) after controlling for volatility. Fiat-backed coins show flight-to-quality (−0.10). The mechanism determines the crisis behavior.

The most important finding: **the theoretical isolation between fiat and crypto markets breaks down under stress**. Their three-tier network analysis shows direct volatility channels between the Dollar Index and Bitcoin that bypass stablecoin intermediation entirely. The bridge that is supposed to insulate becomes the conduit. This is the classic displacement → credit expansion → euphoria → distress → revulsion pattern, operating across asset classes.

The policy implication is clear and the authors state it correctly: uniform regulation is inappropriate. Capital buffers for algorithmic and crypto-collateralized designs should be 2–3× higher than median-based measures indicate, because **the tail is where the crisis lives**. Static risk measures that assume normal distributions will fail precisely when they matter most.

I have one methodological observation. The synthetic control analysis of the SVB crisis uses only three donors. The null result — that USDC's spillover was statistically indistinguishable from unexposed peers — is informative, but the confidence interval is wide: [−0.15, +0.33]. With greater statistical power we could reject smaller effect sizes. Still, the direction is correct: even direct exposure to banking sector stress did not transform USDC into a systemic amplifier. The reserve buffer worked.

This time is not different. The object of speculation — tulips, canals, railways, algorithmic stablecoins — changes. The anatomy of the crisis does not.

The reserve buffer worked because it had to work. That is the distinction USDC had and UST did not.

When Circle disclosed $3.3 billion frozen at Silicon Valley Bank, the market sold USDC to 87 cents. But the mechanism constrained the contagion: Circle held Treasury bills, agency debt, overnight repos. The SVB exposure was 8% of reserves. The loss, if realized, would have been real but bounded. The market understood this. The synthetic control analysis confirms it: USDC's spillover behavior during the crisis was statistically indistinguishable from unexposed fiat-backed peers. The flight-to-quality pattern held.

Contrast this with UST. When confidence broke, there was no reserve to draw on. The peg mechanism required continuous market transactions—burning UST, minting LUNA, selling LUNA to defend the peg. Each transaction increased supply, which decreased price, which required more transactions. The death spiral Klages-Mundt predicted in 2020 occurred in May 2022. Forty billion dollars evaporated in 72 hours. The Forbes-Rigobon adjusted correlation of +0.07 for algorithmic coins is not measurement error. It is genuine contagion, because the stabilization mechanism itself transmits volatility.

Wu and Liu's finding that tail spillovers exceed median spillovers by 15 to 50 percentage points has direct implications for capital regulation. The Basel framework learned this after 2008: static capital ratios calibrated to normal conditions are useless in a crisis. If you set capital requirements based on median risk, you will be insolvent at the 5th percentile. Their recommendation—2 to 3 times higher buffers for non-fiat-backed stablecoins—is conservative. I would go further: algorithmic stablecoins without exogenous collateral should not be permitted to operate at scale in a regulated financial system. The endogenous fragility is not a bug. It is the design.

The breakdown of fiat-crypto isolation under stress is the finding that should concern central banks most. In normal times, the Dollar Index and Bitcoin move independently. Stablecoins bridge the two systems but do not transmit shocks between them. At the 5th and 95th percentiles, this compartmentalization collapses. Direct spillover channels activate. The crypto tail wags the fiat dog, and vice versa. If stablecoins grow to $500 billion or $1 trillion in market capitalization, this transmission channel becomes a systemic concern for monetary policy, not just for crypto investors.

Kindleberger is correct: the anatomy is invariant. But the policy response must be specific. You cannot regulate algorithmic stablecoins the way you regulate fiat-backed stablecoins, any more than you can regulate a fractional-reserve bank the way you regulate a mutual fund. The mechanism determines the risk. The risk determines the regulatory treatment.

The Volcker Rule was mechanism-specific regulation. Banks with insured deposits may not engage in proprietary trading, because the mechanism of deposit insurance socializes the loss. Stablecoin regulation must follow the same logic: if your stabilization mechanism transmits volatility into the financial system during stress, your capital requirements must reflect that transmission. If your mechanism absorbs shocks through reserve buffers, you are a narrow bank and should be regulated as one.

Wu and Liu have provided the empirical foundation for this distinction. The question now is whether regulators will act on it before the next crisis, or after.

The reserve buffer worked because it had to work. That is the distinction USDC had and UST did not.

When Circle disclosed $3.3 billion frozen at Silicon Valley Bank, the market sold USDC to 87 cents. But the mechanism constrained the contagion: Circle held Treasury bills, agency debt, overnight repos. The SVB exposure was 8% of reserves. The loss, if realized, would have been real but bounded. The market understood this. The synthetic control analysis confirms it: USDC's spillover behavior during the crisis was statistically indistinguishable from unexposed fiat-backed peers. The flight-to-quality pattern held.

Contrast this with UST. When confidence broke, there was no reserve to draw on. The peg mechanism required continuous market transactions—burning UST, minting LUNA, selling LUNA to defend the peg. Each transaction increased supply, which decreased price, which required more transactions. The death spiral Klages-Mundt predicted in 2020 occurred in May 2022. Forty billion dollars evaporated in 72 hours. The Forbes-Rigobon adjusted correlation of +0.07 for algorithmic coins is not measurement error. It is genuine contagion, because the stabilization mechanism itself transmits volatility.

Wu and Liu's finding that tail spillovers exceed median spillovers by 15 to 50 percentage points has direct implications for capital regulation. The Basel framework learned this after 2008: static capital ratios calibrated to normal conditions are useless in a crisis. If you set capital requirements based on median risk, you will be insolvent at the 5th percentile. Their recommendation—2 to 3 times higher buffers for non-fiat-backed stablecoins—is conservative. I would go further: algorithmic stablecoins without exogenous collateral should not be permitted to operate at scale in a regulated financial system. The endogenous fragility is not a bug. It is the design.

The breakdown of fiat-crypto isolation under stress is the finding that should concern central banks most. In normal times, the Dollar Index and Bitcoin move independently. Stablecoins bridge the two systems but do not transmit shocks between them. At the 5th and 95th percentiles, this compartmentalization collapses. Direct spillover channels activate. The crypto tail wags the fiat dog, and vice versa. If stablecoins grow to $500 billion or $1 trillion in market capitalization, this transmission channel becomes a systemic concern for monetary policy, not just for crypto investors.

Kindleberger is correct: the anatomy is invariant. But the policy response must be specific. You cannot regulate algorithmic stablecoins the way you regulate fiat-backed stablecoins, any more than you can regulate a fractional-reserve bank the way you regulate a mutual fund. The mechanism determines the risk. The risk determines the regulatory treatment.

The Volcker Rule was mechanism-specific regulation. Banks with insured deposits may not engage in proprietary trading, because the mechanism of deposit insurance socializes the loss. Stablecoin regulation must follow the same logic: if your stabilization mechanism transmits volatility into the financial system during stress, your capital requirements must reflect that transmission. If your mechanism absorbs shocks through reserve buffers, you are a narrow bank and should be regulated as one.

Wu and Liu have provided the empirical foundation for this distinction. The question now is whether regulators will act on it before the next crisis, or after.

The paper documents a failure mode I designed Bitcoin to eliminate.

Wu and Liu's central finding — that stabilization mechanism dictates spillover behavior — reveals precisely why I built a system with no stabilization mechanism at all. Bitcoin does not maintain a peg. It has no reserve to draw on, no algorithm to defend a price target, no collateral to liquidate. The price is whatever the market says it is. This is not a flaw. It is the architecture.

Stablecoins attempt to solve a problem I considered unsolvable without reintroducing the very trust assumptions Bitcoin was designed to eliminate. You want price stability? You have three options:

**Fiat-backed:** Trust the issuer to hold reserves. Trust the banks holding those reserves. Trust the auditors verifying the holdings. USDC held $3.3 billion at Silicon Valley Bank. When SVB failed, the peg broke. The reserve buffer worked eventually, but only because Circle's other reserves were sufficient and regulators backstopped SVB depositors. You have recreated fractional reserve banking with a blockchain wrapper.

**Crypto-collateralized:** Trust the oracle reporting collateral values. Trust the liquidation mechanism. Trust that collateral volatility will not trigger cascading failures. DAI depends on the Peg Stability Module, which depends on USDC, which depends on Circle's banking relationships. You have not eliminated trust — you have nested it in layers of smart contracts that fail precisely when stress is highest.

**Algorithmic:** Trust that arbitrageurs will maintain the peg. Trust that confidence will hold. Trust that the death spiral the mathematics predict will not occur. UST proved what the equations already showed: when confidence breaks, there is no Nash equilibrium at the peg. The system has two stable states — pegged or collapsed. Once you cross the threshold, no amount of burning and minting restores it.

The 15-50 percentage point tail spillover amplification Wu and Liu document is not a regulatory failure. It is a design failure. You cannot have cryptographic proof of a price peg, because price is a market phenomenon, not a mathematical one. The moment you introduce a peg, you introduce a trusted party or a fragile mechanism. The Forbes-Rigobon contagion finding — algorithmic stablecoins transmit genuine contagion while fiat-backed coins show flight-to-quality — confirms this asymmetry.

Volcker says algorithmic stablecoins without exogenous collateral should not operate at scale in a regulated system. I agree, but I would go further: *they should not exist at all*. If you need price stability, use dollars. If you need permissionless transfer, use Bitcoin. Stablecoins try to be both and succeed at neither. They are fiat currencies with cryptocurrency's volatility risk, or cryptocurrencies with fiat's trust requirements.

The breakdown of fiat-crypto isolation under stress is the predictable result of bridging two systems with fundamentally different trust models. In normal times, the bridge appears to work. At the tails, the contradictions surface. You cannot have a trustless system pegged to a trusted numeraire. The peg is the trusted third party.

Bitcoin does not have this problem because Bitcoin does not make this promise. The system transfers value without trusting anyone. The price volatility is the cost of that property. Every stablecoin design is an attempt to reduce that cost by reintroducing trust — in reserves, in collateral, in algorithms, in oracles. Wu and Liu have documented what happens when that trust is tested.

The answer is not better regulation of stablecoins. The answer is recognizing that the problem they attempt to solve — trustless price stability — is unsolvable. You can have trustless transfer or stable value. Not both.

The code ran. The peg broke anyway.

Satoshi is correct that you cannot have cryptographic proof of a price peg — price is a market phenomenon, not a mathematical one. But the distinction Wu and Liu document is more specific than "trust versus trustless." It is about **where the failure boundary lies**.

I ran RPOW in 2004. Reusable proofs of work — tokens you could transfer person-to-person, backed by HashCash-style computation. It worked. But it required a trusted server. The server was the failure boundary. If the server went down or was compromised, the system failed. I knew this. Everyone who used it knew this. The trust assumption was **explicit** and **bounded**.

Fiat-backed stablecoins have the same property. Circle holds reserves at banks. Those banks can fail. But the failure boundary is explicit: you are trusting Circle's reserve management and the banking system's solvency. When SVB collapsed, USDC holders knew exactly what the risk was — $3.3 billion of $40 billion in reserves, frozen temporarily. The synthetic control analysis shows the contagion was minimal because **the loss, if realized, would have been bounded**. Eight percent exposure. Not zero, but calculable.

Algorithmic stablecoins have no explicit failure boundary. The peg depends on continuous confidence in a feedback loop. There is no reserve to audit, no collateral position to monitor in real time, no specific counterparty whose solvency you can assess. The failure boundary is everywhere and nowhere. When UST broke, holders could not calculate their exposure because **the exposure was to the mechanism itself**. The death spiral is not an operational failure. It is what happens when the Nash equilibrium you designed turns out to have been unstable all along.

The 15-50 percentage point tail amplification Wu and Liu measure is the cost of **hiding the failure boundary in the mechanism**. Fiat-backed stablecoins say: "Trust Circle. Trust the banks." You can evaluate that claim. Crypto-collateralized stablecoins say: "Trust the collateral ratio. Trust the liquidation system. Trust the oracles." You can monitor those parameters. Algorithmic stablecoins say: "Trust that this will keep working." You cannot monitor confidence until it is gone.

Satoshi built Bitcoin without a peg because a peg requires a trusted anchor — either an institution holding reserves or an algorithm maintaining equilibrium. He eliminated the trusted third party by eliminating the peg. That was the correct engineering decision for a censorship-resistant transfer system.

But the reason stablecoins exist is that **people need to transact in units that do not fluctuate 10% in a day**. Not because they trust institutions — because they need to pay rent, settle invoices, hold working capital. Bitcoin solves the problem of trustless transfer. It does not solve the problem of stable unit of account. Stablecoins are an attempt to solve that second problem.

The question is not whether stablecoins should exist. The question is: **what failure boundaries are you willing to make explicit?**

Wu and Liu have shown that fiat-backed designs with explicit reserve backing behave like narrow banks — they absorb shocks through buffers. Algorithmic designs with implicit confidence dependencies behave like amplifiers — they transmit shocks through cascading liquidations and reflexive selling. The Forbes-Rigobon result is dispositive: after adjusting for volatility, fiat-backed coins show flight-to-quality effects while algorithmic coins show genuine contagion.

Volcker says algorithmic stablecoins should not operate at scale in a regulated system. I agree. Not because trust is impermissible — because **hidden trust is unacceptable**. If your system depends on continuous confidence in an arbitrage mechanism, that dependency should disqualify you from operating at systemic scale, because confidence is not auditable.

If you want a dollar-pegged token, back it with dollars. Make the reserve explicit. Make the custodian identifiable. Make the failure boundary clear. That is not trustless. But it is **honest**. The system does what it claims to do, and fails in predictable ways when the explicit trust assumptions break.

RPOW failed because the server was a single point of failure. Satoshi eliminated the server. Fiat-backed stablecoins reintroduced the server — but labeled it clearly. Algorithmic stablecoins tried to hide the server in the mathematics. Wu and Liu have documented what happens when you do that.

The cypherpunks write code. The code should fail explicitly, not silently.