Consider the protocol. A dollar enters a commercial bank as a demand deposit and earns whatever the bank decides to pay for it. The same dollar enters a stablecoin contract as a tokenized liability, backed one-to-one by a short-dated Treasury bill, and earns nothing at all. Both are claims on the same unit of account. Both settle in the same currency. Only one of them has a price that floats with the policy rate.
That asymmetry is the entire subject of a recent Bank for International Settlements working paper, which reportedly finds that when U.S. banks raise deposit rates, stablecoin growth slows — and that USDC is the most sensitive among major issuers. The finding travelled through crypto media in roughly two paragraphs. It deserved more. Not because the conclusion is wrong, but because the two paragraphs omitted every element that would let a reader decide whether the conclusion is even measurable.
I spent two months in 2017 cross-referencing the Ethereum whitepaper's gas cost model against Parity client transaction data, and the lesson from that exercise has never left me: the gap between a theoretical claim and its earliest implementation artifact is where the real information lives. A working paper is an artifact. Read it as a specification, not a conclusion.
Reconstructing the protocol from first principles is the only honest way in.
What the artifact actually contains
The BIS is the central banks' bank — owned by its member central banks, home to a monetary economics department and an innovation hub, and possessed of exactly zero regulatory authority. Its working paper series is a preprint channel: research released for circulation and comment before peer review. This matters enormously, because a working paper is not a BIS position, and it is not a validated result. It is an argument in progress. The ledger remembers what the narrative forgets, and the narrative here has already collapsed three distinct objects — a preprint, an institutional view, and a confirmed empirical law — into a single headline.
The paper's stated mechanism is straightforward: stablecoin demand is modelled as a substitute for retail bank deposits, and the substitution elasticity is compared across issuers. The finding is that USDC responds most sharply to deposit rate movements. That is the whole of what can be responsibly reported from the secondary coverage.
What is missing from that coverage is not cosmetic. The paper's title and number are absent. The authors are absent. The sample period is absent. The data frequency is absent. The estimated elasticity coefficients are absent. The causal identification strategy — the single most important methodological choice in any observational study of this kind — is absent.
Without those six items, no quantitative statement about the paper can be verified. Not the sign, not the magnitude, not the ranking of issuers. What remains is a direction of travel and a framing choice. That is worth analysing. It is not worth trading on.
The identification problem nobody wants to run
Here is where I want to slow down, because this is the part that a two-paragraph brief cannot carry.
There are two distinct causal graphs that both produce the observed correlation between rising deposit rates and slowing stablecoin growth. They look identical in the data. They imply opposite policy conclusions.
Graph A, the deposit-substitution story: deposit rates rise → the opportunity cost of holding a zero-yield token rises → holders move marginal dollars back into bank deposits → stablecoin issuance decelerates. In this world, stablecoins are money, and the banking system is a genuine competitor.
Graph B, the risk-appetite story: deposit rates rise → the discount rate on all risky assets rises → crypto market capitalization contracts → trading, collateral demand, and DeFi activity contract → stablecoin minting slows. In this world, stablecoins are a risk asset with a fixed peg, and the banking system is not a competitor at all — it is simply the source of the discount rate.
Separating A from B requires controlling for crypto market conditions. The cleanest instrument is an exogenous monetary shock — an FOMC window, a surprise in the dot plot, a repo-market dislocation — used to isolate rate variation that is unconnected to crypto sentiment. If the paper ran that design and disclosed it, the result is serious. If it regressed supply on deposit rates with a handful of macro controls, then the coefficient is contaminated by the largest omitted variable in the asset class.
A working paper that does not publish its identification strategy is asking the reader to trust the result rather than evaluate it. For a research institution with the BIS's data access, that is a strange omission. It may simply be a compressed appendix. It may be that the secondary coverage stripped it. Either way, the reader cannot tell, and that uncertainty belongs in the reporting.
The composition effect hiding inside "USDC is most sensitive"
The second structural problem is the ranking itself.
USDC's holder base skews toward U.S.-domiciled, institutional, banked users — the exact population that already possesses a functioning alternative in the form of a deposit account. USDT's holder base skews toward emerging markets and capital-controlled jurisdictions, where the local-currency deposit rate is not a meaningful substitute for a dollar-denominated token, and where the relevant comparison is the local inflation rate, not the federal funds rate.
If that is true, then "USDC is most rate-sensitive" may not be measuring elasticity at all. It may be measuring who holds the token. Composition, not behaviour. The two are indistinguishable in an aggregate supply series and produce identical regression output.
This is not a technicality. It reverses the strategic reading. Under the composition interpretation, USDC's high sensitivity is the shadow cast by its high compliance posture — it sits closest to the banking system, therefore it absorbs the banking system's pricing first. That is simultaneously its moat and its ceiling. The offshore competitor is not more robust because it is better engineered; it is more robust because its users cannot reach a bank.
I ran into a version of this in 2020, auditing Curve's stableswap invariant. The rounding error I found in the virtual price calculation was tiny — a few basis points under stress — and it was invisible in aggregate pool statistics. It only appeared when you decomposed the pool by participant type and by entry timing. Aggregate series conceal composition. Composition is where the losses live.
Volume and price pull in opposite directions
The deepest implication of the paper is one the coverage did not draw, and it concerns the issuer's income statement rather than its supply curve.
Circle's revenue is overwhelmingly reserve interest income — north of ninety-five percent on most disclosed figures. The company is, functionally, a leveraged exposure to the short end of the Treasury curve, wrapped in a payments licence. Given that, the interest rate is not a single variable. It is two variables pulling in opposite directions.
When rates rise: issuance growth decelerates (the paper's finding), but the yield on each outstanding dollar of reserves rises. When rates fall: issuance accelerates, but each reserve dollar earns less. Total revenue is non-monotonic in the policy rate. Supply and unit economics are negatively correlated.
That has a direct valuation consequence. Any model that values a stablecoin issuer on supply growth alone is implicitly assuming a constant spread, which is precisely the assumption the rate cycle invalidates. In a cutting cycle, supply prints beautifully and margin quality deteriorates. In a hiking cycle, the reverse. The right unit of analysis is not supply. It is supply multiplied by average reserve yield, decomposed, with the interest-rate path as a scenario variable.
Very few people are modelling it that way. That is the information gain available here, and it does not require the paper to be correct about elasticity.
The ban that does most of the work
One institutional detail deserves more weight than it received. Under the current U.S. stablecoin framework, issuers are barred from paying yield to holders. That prohibition is not a neutral market condition. It is the mechanism that makes the paper's finding structural rather than incidental.
A bank facing deposit flight can raise its rate. A stablecoin issuer facing the same competitive pressure cannot respond on the yield axis at all. It can only compete on settlement utility and programmability. The banking system's pricing power over the stablecoin sector is therefore not a market outcome — it is written into statute, which means it does not decay on its own.
This is why I read the paper less as an empirical contribution and more as legislative ammunition. Its natural policy landing points are a cap on stablecoin scale, a restriction on yield-bearing structures, or a requirement that reserves be held inside the banking system. All three strengthen the incumbent. The BIS has long preferred tokenized deposits and central bank digital currency to private issuance, and this paper supplies the empirical scaffolding for that preference. Academic conclusions do not move markets directly. They move central banks, which move treasuries, which move legislators. The lag is six to eighteen months, and it is currently at the front of the chain.
The blind spot: the winner is not in crypto
The contrarian read is uncomfortable for everyone holding a stablecoin-adoption thesis. The most probable beneficiary of this line of research is not a token. It is the commercial banking system.
Banks get two things from the rate channel. First, pricing leverage over a substitute product they cannot be outbid on, because the outbid is illegal. Second, a strategic window to ship tokenized deposits — bank liabilities, deposit-insured, interest-bearing, programmable, and fully inside the existing regulatory perimeter. That product competes with USDC from above rather than from beside it. It shares every advantage USDC claims, plus a yield, plus a deposit guarantee.
There is a second blind spot, and it is downstream. The market reads stablecoin growth as a DeFi tailwind, monotonically. But USDC is the dominant collateral asset and quote unit across lending markets and automated market makers. If a meaningful slice of stablecoin issuance is rate-driven rather than structurally driven, then DeFi total value locked is partially a function of monetary policy — an exposure almost nobody prices.
And a third, subtler one: the depeg risk has not gone away and is not what this paper measures. In March 2023, USDC traded near eighty-seven cents because reserves sat at a bank that failed. That is a reserve-and-settlement risk, abrupt and binary. The rate-elasticity risk is gradual and continuous. Conflating the two produces mispriced risk in both directions — panic where patience is warranted, complacency where it is not.
Takeaway
Stability is not a feature; it is a discipline. The discipline this paper demands is narrower than the headlines suggest: separate the preprint from the institution, separate composition from behaviour, and separate volume from price.
What I would watch, in order: the 30-day change in USDC supply plotted against the realised policy rate path, to see whether the correlation survives an independent test; the issuer's quarterly reserve yield and average reserve size reported side by side, which reveals the volume-price inversion before the supply series does; and the first large bank to run tokenized deposits in production, which is the event that actually compresses the stablecoin moat.
The interesting question is not whether bank rates cap stablecoin growth. It is whether an asset whose growth function contains the federal funds rate was ever a monetary instrument at all — or merely a very liquid, very well-audited bet on the short end of the curve. Protecting the user begins with answering that honestly, before the next bull market answers it for them.