The number sat on my screen like a splinter I couldn't stop touching. Three percentage points. For every single basis point the Federal Reserve lifted deposit rates, the claim went, stablecoin growth over a ten-week window fell by roughly three percentage points. The instrumental-variable estimate was even more aggressive โ triple the OLS result, so nine.
I read it in eleven different newsletters over four days. Not one of them flinched.
Here is the thing about spending your formative years auditing reentrancy bugs: you develop an involuntary reflex toward numbers that refuse to reconcile. When I pulled apart TheDAO's contract in the autumn of 2016, the vulnerability was not hiding behind clever obfuscation. It sat in plain sight โ a recursive call that drained a balance before the balance updated. The exploit was loud. The consensus was quiet. That gap, between what the data says and what the crowd repeats, is where I have made my living ever since. Searching for truth in the noise of the network.
So let me be precise about what BIS Working Paper #1382 actually delivers, and what it does not. Because this paper matters โ and the way it is being repeated matters more.
Context: what the paper says, and what arrived alongside it
The paper is by Rashad Ahmed and Iรฑaki Aldasoro, published under the Bank for International Settlements' working paper series, number 1382. Strip away the econometric dressing and the central claim is almost humble: bank deposit rates function as sticky gravity on stablecoin adoption. When deposits pay more, stablecoins grow more slowly. The authors are explicit that they have demonstrated correlation, not causation. They have not shown that stablecoin growth actually reduced bank deposits or lending.
The headline writers were considerably less careful.
The timing is what makes this more than an academic curiosity. The GENIUS Act โ the Guiding and Establishing National Innovation for U.S. Stablecoins Act โ was signed in July 2025. Section 4(a)(11) prohibits payment stablecoin issuers from paying interest or yield to holders. That prohibition takes effect on January 18, 2027. In parallel, the distribution layer has been sprinting: Visa Direct moved stablecoin settlement into commercial production, and Mastercard's Global Money Platform reached 130 countries. Somewhere in the reporting, a figure surfaced โ roughly $400 billion in stablecoin payment volume across the first eight months of a year the source material cannot quite pin down. I flagged that ambiguity the moment I saw it. It is not a footnote. It is a tell.
I have spent twenty-five years watching this industry build cathedrals of infrastructure and then act surprised when nobody walks in. The supply-side narrative is seductive precisely because it is easy to photograph: new rails, new logos, new press releases. Build the rails and adoption follows. Build faster rails and adoption follows faster. What the BIS paper does โ almost alone among institutional research โ is say out loud that the rails were never the bottleneck. Demand was. Demand still is.
This is a working paper. It has not been through formal peer review. Treat it as a consultation draft, not a verdict. But treat the direction of its argument as serious, because the direction is right even where the decimals are not.
Core: the five things this paper gets right, and the one number it gets catastrophically wrong
Start with what the distribution-layer story actually is, because the industry has been telling it wrong.
When Visa Direct goes into commercial production for stablecoin settlement, and when Mastercard's platform reaches 130 countries, the instinct is to call this a technical breakthrough. It is not. The stablecoin architecture has barely changed since 2014. What changed is the last mile โ whether a merchant's bank will accept the settlement, whether a card network will route it, whether the compliance department will sign off. That is a distribution problem, not a protocol problem. And here is the part almost nobody says plainly: Visa Direct and Mastercard GMP do not elevate stablecoins into front-end money. They demote them into a back-end settlement rail. The end user may complete an entire purchase without ever knowing a blockchain was involved. That is a commercial triumph and a definitional catastrophe, because it means "stablecoin payment volume" and "people using stablecoins as money" are now two different numbers wearing the same label.
Which brings me to the $400 billion.
I have watched this industry inflate on-chain metrics for a decade. Liquidity mining taught an entire generation that yield is something a protocol mails you for showing up โ and that TVL is a number you rent, not a number you earn. When the incentives stop, the "users" vanish, because they were never users. They were mercenaries responding to a subsidy. The same discipline has to be applied to payment volume. Does the $400 billion include on-chain internal transfers? Exchange-to-exchange rebalancing? Market-maker shuffling? Bot traffic? If it does โ and the reporting gives us no statistical definition to work with โ then the real terminal payment figure for goods and services could be a fraction of the headline. This is not cynicism. It is basic measurement hygiene. A number without a methodology is a rumor with a comma in it.
Now the part that is genuinely new, and genuinely important: the reserve interest question.
USDC and USDT do not have the token economics we are trained to interrogate. There are no team unlocks, no investor vesting cliffs, no emissions schedules to model. The real token economics of a fiat-backed stablecoin is a single question โ who owns the interest on the reserves. Short-term Treasuries, overnight repo, bank deposits: that reserve book throws off yield, and the GENIUS Act's payment prohibition is, in economic substance, a forced redistribution of that yield. Interest that could have flowed back to holders is legally assigned to the issuer and the distribution layer. Section 4(a)(11) does not merely ban a feature. It converts stablecoin issuance from a service business into a spread business.
That distinction is everything. A service business competes on quality. A spread business competes on the gap between what your assets earn and what your liabilities cost. And a spread business is hostage to the interest rate cycle in a way a service business never is.
Follow the arithmetic and the issuer's bind becomes visible. In a high-rate world, bank deposits pay real money, so stablecoin growth slows โ that is the paper's core finding, and it is directionally sound. In a low-rate world, deposits pay nothing, stablecoins become relatively attractive, growth accelerates โ but the reserve book now earns less per dollar, so unit revenue compresses even as the float expands. The issuer cannot simultaneously maximize scale growth and unit margin. The rate cycle guarantees that one leg of the stool is always wobbling. This is the inference the paper never states outright, and it is the most investable sentence in the entire document.
Then there is the migration of value capture, which is where I want to bring in an old scar.
I spent years watching Cosmos build the most elegant interoperability layer this industry has produced โ IBC is a genuine engineering achievement โ and then watch ATOM capture almost none of the value that flowed across it. Technical elegance and value capture are not the same variable. They are not even correlated. So when I look at the GENIUS Act's prohibition, I do not see a stablecoin story. I see a value-capture migration. The ban constrains the issuer. It says nothing about third-party platforms. An exchange or a wallet can still run a rewards program, offer fee rebates, hand out points, structure a cashback scheme. The yield does not disappear. It relocates โ from the issuance layer to the distribution layer. Coinbase keeps a tool that Circle just lost. The issuer degrades toward a low-margin clearing channel; the distributor inherits the customer relationship and the retention lever.
I have a specific reason to take this seriously. In 2024, working with two Asian asset managers on a white paper about narrative-driven ESG integration for crypto funds, I spent months inside the language of compliance. What I learned there is that regulators do not write prohibitions in a vacuum. Every clause is a negotiated boundary, and the boundary is drawn around whoever has the loudest lobby. Section 4(a)(11) reads as neutral consumer protection. It functions as protection of the bank deposit franchise. Those are not the same thing, and the difference is where the next four years of rulemaking will be fought.
Which leads to the counterintuitive gift buried in the ban.
By prohibiting interest, the GENIUS Act removes the "expectation of profits" element from the Howey test. Walk the four prongs: money invested, yes. Common enterprise, weak โ the holder and issuer sit in a custodial or creditor relationship, not a joint venture. Expectation of profit derived from the efforts of others โ no, because the law forbids the profit. The prohibition that cripples the issuer's competitiveness is also the provision that hardens the stablecoin's non-security classification. The regulatory cost and the regulatory blessing are the same clause, viewed from opposite ends of the table. Almost nobody covering this paper has noticed that. It is the kind of detail that only surfaces when you have read enough offering memoranda to know how lawyers think.
Now the number. The one that made me put down my coffee.
The paper's headline elasticity โ three percentage points of ten-week growth per basis point of deposit rate, tripled to nine under the instrumental-variable specification โ does not survive contact with arithmetic. Take it literally. A 100 basis point hike would shave 300 percentage points off growth. The 2022โ2023 hiking cycle totaled roughly 525 basis points. That implies a growth collapse of about 1,575 percentage points. Growth rates do not go to minus fifteen hundred. They go to minus thirty, minus fifty, and then they stop, because a percentage cannot fall below negative one hundred without changing its meaning.
Check it against the tape. From March 2022 to mid-2023, the Fed raised rates by roughly 525 basis points. Over that same window, total stablecoin market capitalization fell from around $180 billion to roughly $125 billion โ a drawdown in the neighborhood of thirty percent, and I would want DefiLlama's panel to confirm the exact figures before I put them in a model. Thirty percent. Not fifteen hundred percentage points. The gap between the paper's stated coefficient and observable reality is several orders of magnitude.
There are only a few honest explanations. The original unit was probably basis points, not percentage points, and the transcription lost the distinction. Or the coefficient corresponds to a 100 basis point move producing roughly a three point change in growth. Or the standardization and log-transformation were stripped out somewhere between the PDF and the newsletter. Any of these is plausible. None of them is what got repeated.
So here is my instruction, and I want it to be unambiguous: do not feed "three percentage points per basis point" or "nine percentage points per basis point" into any model, any memo, any pitch deck. Go back to BIS Working Paper 1382 and read the units yourself. The directional claim โ higher deposit rates suppress stablecoin growth โ is consistent with economic intuition and worth keeping. The magnitude is not usable until verified.
The methodology deserves the same scrutiny, and it does not fully survive it either. The instrumental variable is "the federal funds rate crossing three percent" as a proxy for deposit repricing. But the federal funds rate also moves risk-asset valuations, crypto risk appetite, and DeFi yields all at once. That violates the exclusion restriction โ the instrument is supposed to affect stablecoin growth only through the deposit-rate channel, and it plainly does not. An IV built on a variable that touches everything is an IV that proves nothing cleanly. And the dependent variable โ ten-week growth regressed on a one basis point change โ has a dimensional mismatch that should have raised eyebrows before it raised headlines.
The authors, to their credit, say all of this. They admit correlation, not causation. That admission is the most honest sentence in the paper and the one most likely to be deleted in transmission. The title says banks can slow stablecoin adoption just by raising deposit rates. The body says we found a correlation. Those two statements are not the same claim, and the title travels faster than the caveat. A causal headline bolted onto a correlational body is not a summary. It is a distortion with a citation.
One more structural risk that the paper never touches, and that my security background will not let me leave alone. A fiat-backed stablecoin is a centralized IOU. Its trust anchor is not a consensus mechanism; it is a bank, a custodian, an auditor, and a legal framework. That works beautifully in normal times and fails in a specific, ugly way in abnormal ones: issuer freeze, custodian bank failure, a legal injunction. Compare it to the thing it is supposedly replacing. A US bank deposit carries FDIC insurance up to $250,000 per depositor per bank. A stablecoin holder's legal claim on the reserve, in most jurisdictions, is not clearly defined โ priority in bankruptcy, segregation of assets, redemption rights under stress. That is not a technical footnote. It is the difference between a claim you can enforce and a claim you hope for. The distribution layer makes stablecoins convenient. It does not make them insured.
Step back and look at the ecosystem dependency, because it explains why the issuer's position is structurally weak. A fiat-backed stablecoin sits in the middle of a chain it does not control. Upstream: the Federal Reserve's rate path, the supply of short-term Treasuries, the credit of the custodian bank. Downstream: DeFi protocols that use it as base collateral, exchanges that use it as a quote currency, payment networks that route it. The issuer controls neither end. It is a converter โ turning off-chain reserve assets into on-chain liquidity โ and a converter's margin is set by the spread between two prices it does not set. When I map dependency graphs in security audits, I look for the node with the most inbound edges and the fewest outbound controls. In the stablecoin graph, that node is the issuer. Everyone depends on it; it depends on everyone; it controls almost nothing.
There is also a feedback loop that the paper's single-direction framing misses entirely. Stablecoin reserves are held in short-term Treasuries and overnight repo. That makes the stablecoin sector a standing source of demand at the short end of the curve. More stablecoin float means more T-bill demand, which nudges short rates lower, which makes stablecoins relatively more attractive, which grows the float. Read it the other way and the same loop runs in reverse during redemptions. The paper models rates affecting stablecoins. It never models stablecoins affecting rates. The loop is probably small relative to the size of the Treasury market โ I would not overstate it โ but it is directionally real, and it means the "gravity" is not a one-way field.
And the external validity problem is bigger than the paper admits. The entire analysis is American. But the growth market for dollar stablecoins is not America โ it is Argentina, Turkey, Nigeria, and everywhere the local currency is a worse store of value than a tokenized dollar. In those economies the relevant comparison is not a 5% Treasury-backed deposit. It is a peso deposit paying 60% against 100% inflation, or no formal deposit account at all. The deposit-rate gravity that the paper measures is a phenomenon of banked, dollarized, high-rate economies. Extrapolate it to the dollarized periphery and the coefficient may invert. A US finding is not a global finding, and the paper never says so.
And the divergence between USDC and USDT is the empirical fingerprint of all of this. USDC's sensitivity runs higher โ the reported figures are 4.8 versus 1.9, though I would verify both before trusting them โ precisely because its user base is the most banked, the most institutional, the most deposit-like. Its compliance strength and its rate fragility are the same property. USDT's users sit further offshore, in economies where the relevant alternative is not a 5% Treasury-backed deposit but a local currency losing value by the month. For them, the question is dollar access, not yield. The most compliant stablecoin is the most rate-fragile stablecoin, because compliance is what makes its holders look like depositors. That is the paradox at the heart of the paper, and it is real.
Contrarian: the competitor nobody in this debate is naming
Everyone is arguing about whether banks can slow stablecoins by raising deposit rates. Almost nobody is asking the question that actually decides the next five years: what happens when banks issue the stablecoin themselves?
Tokenized deposits are the blind spot. A bank-issued deposit token carries the rail efficiency of a stablecoin plus the two things no independent issuer can offer โ it can pay interest, and it may sit inside the deposit insurance perimeter. That is not a competitor. That is a downgrade. If JPMorgan, Citi, and HSBC ship interest-bearing deposit tokens at scale, the stablecoin is squeezed out of precisely the segments where it currently has the strongest claim โ institutional treasury, corporate settlement, large-value transfer โ and pushed toward the unbanked, the offshore, the frontier. The paper does not mention tokenized deposits once. That omission is not a gap in the data. It is a gap in the frame.
The second blind spot is the rate cycle itself. The paper's sample runs through a historically aggressive hiking regime. Its conclusion is a snapshot of high-rate gravity. If the Fed pivots to sustained cuts, deposit rates fall, the gravity weakens, and stablecoin demand improves automatically โ without any protocol changing, any product shipping, any narrative shifting. The paper measures a force that reverses with the weather, and never mentions that weather changes. Read the conclusion as conditional, not structural.
And I would push back on the yield-versus-utility framing itself. It assumes stablecoins and deposits compete on the same axis. They do not. A deposit cannot be composed into a smart contract at 3 a.m., cannot settle a machine-to-machine micropayment, cannot serve as collateral in an automated lending pool, cannot be programmed to release funds on a verified data feed. Those are architectural differences, not pricing differences. The competition is being framed as a rate war when it is partly a capability gap. Where code meets culture, the real value emerges โ and the machine economy is where that value will be priced first.
Takeaway
The BIS paper is worth reading for one reason and one reason only: it drags the interest rate back into a growth equation that the industry had been solving with infrastructure alone. But read it as a direction, not a dial. Verify the units. Question the causality. Watch the January 18, 2027 window, because that is when the prohibition stops being a statute and starts being an enforcement question โ and the enforcement question is whether a rewards program is interest wearing a costume.
The narrative is the asset; the code is the proof. Right now the narrative says stablecoins won the rails. The proof says the rails were never the argument. What decides the next cycle is simpler and harder: can a dollar that cannot pay you compete with a dollar that can, on every axis except the one where it is the only option? I do not think we know the answer yet. I think we are about to find out.
