Every Capital Control Assumes Households Can't Leave. Stablecoins Turned That Into a Variable.

The Mundell-Fleming trilemma has always been a statement about three things a government can choose. Fix the exchange rate, allow capital to move freely, run your own monetary policy: pick two. Capital mobility sat in that list as a policy setting, something a finance ministry could dial up or down by writing rules that banks then enforced.

A New York Fed staff report published in August 2026 argues that the dial has changed hands. Once households can move dollar value through payment rails that sit outside the domestic banking system, capital mobility stops being something the government sets and becomes something the population produces. The trilemma still holds. What changed is who determines one of its three vertices.

That reframing is the paper's contribution, and it is more consequential than the finding that people buy dollars during a crisis, which nobody disputes. We've focused here on the model's mechanism and its policy predictions, and on what the accompanying blockchain evidence can and cannot support.

Why this argument arrived now

The empirical version of this question was settled a month earlier by someone else. The Bank for International Settlements published a working paper on 21 July 2026 by Boris Hofmann, Aaron Mehrotra and Jan Paulick, drawing on foreign-currency deposit data and stablecoin flows across more than 130 economies. As Central Banking reported, it found that foreign exchange restrictions have little apparent effect on stablecoin dollarisation, in sharp contrast to conventional deposit dollarisation, which capital controls have historically curbed because banks are obliged to enforce domestic rules. The same paper documented strong persistence in both channels, meaning once residents shift into dollars the behaviour is hard to reverse.

The scale behind that finding keeps climbing. The BIS Annual Economic Report 2026 put the stablecoin market at roughly $320 billion by end-May 2026, with 99.4% of fiat-backed stablecoin value pegged to the US dollar, and noted that blocking domestic intermediaries from handling unapproved tokens is likely to remain imperfect as a control. In Latin America, Bitso Business reported an 81% year-over-year rise in stablecoin payment volume in the first half of 2026, and found that USDT and USDC together accounted for 40% of regional crypto purchases in 2025, passing bitcoin for the first time.

So the empirical pattern was established. What was missing was a model explaining why capital controls should fail specifically against this rail, and what a government's best response looks like once they do. That is the gap this paper fills, and it cites the BIS work as complementary evidence.

The mechanism is division, and that is the whole argument

The model's central move fits in a fraction of a line. Effective capital control equals enforcement effort divided by stablecoin adoption.

Enforcement effort is what the state spends on policing the border: supervising banks, restricting exchanges, monitoring transfers. Adoption is the share of households using programmable rails instead. Putting adoption in the denominator says something specific and slightly brutal. Enforcement doesn't fail because it gets weaker. It fails because it gets spread across a shrinking share of the traffic it was designed to police. A fixed enforcement budget buys less control every time another household moves off the rail the enforcement operates on.

The second half of the mechanism decides when that happens. Household adoption rises with what the paper calls flight pressure, the gap between the world real rate and the domestic natural rate, and it rises faster when the token is more programmable. Households draw idiosyncratic adoption costs, and the ones whose costs sit below the benefit switch.

Put the two together and the timing is the problem. Adoption climbs precisely when the rate gap widens, which is precisely when a government most needs its capital controls to hold. The wedge that lets a country run a domestic rate different from the world rate shrinks exactly at the moment it is being leaned on. In the paper's framing, the feasible set of monetary and exchange-rate policy contracts endogenously during stress.

What remains has to go somewhere. The government absorbs the residual pressure through the policy rate, through a currency depreciation, or through some mix, with the split determined by how much it dislikes missing each target. Under a hard peg, all of it surfaces as a rate gap. Under full rate control, all of it surfaces as depreciation. Neither option is available in the quantity a government would want, and both worsen as adoption rises.

Governments resist, then accommodate, and the model says where the switch happens

The most striking result is about government behaviour rather than household behaviour, and it is not the monotonic story you would guess.

Optimal enforcement is hump-shaped in adoption. Below a threshold, a government facing more adoption should spend more on enforcement. Above that threshold, it should spend less, and the more adoption rises the less it should spend. The logic is that enforcement costs are convex while the enforcement technology is being diluted, so past a point each additional unit of effort buys too little control to justify its cost. The paper's language for the two regimes is resistance and accommodation.

That has an uncomfortable implication for anyone reading policy signals. A government that stops fighting stablecoin adoption may not have changed its mind about the risks. Loosening can be the optimal response to having already lost the enforcement argument, and it looks identical from the outside to a genuine liberalisation. Regulatory posture, in this model, is not a reliable signal of regulatory intent.

Two things do move monotonically regardless of which regime a country sits in. Both the interest-rate gap and the currency depreciation rise with adoption. So the non-monotonic enforcement result doesn't soften the conclusion. However a government adjusts its enforcement spending, more adoption always means worse achievable outcomes on rates and the exchange rate.

The evidence is narrower than the theory, and the authors say so

The paper builds a genuinely new dataset. It takes roughly three million historical Ethereum Name Service registrations, tags wallets to countries using flag emoji and script or language signals in the registered names, and joins that to complete ERC-20 transfer records for the nineteen largest USD stablecoins. The result is a wallet-by-event-by-week panel of about 4.5 million observations built around nine crisis events across eight countries between 2021 and 2025, including the Myanmar coup, Russia's sanctions, the Nigerian naira float, Argentina's post-election devaluation, and two waves of the Iranian rial crash.

The headline estimates are real but modest. In crisis weeks, wallets tagged to the crisis country receive about 0.127 log points more in stablecoins and are 1.8 percentage points more likely to receive any, both statistically significant. Sending activity moves by similar magnitudes but is not significant. Against a clean pre-event baseline, the two weeks before the crisis show nothing for any outcome, receipts jump in the crisis week itself, and the probability of sending rises two weeks later.

The authors are careful about what this supports. They state plainly that they interpret the results as validating the model's adoption assumption rather than as an independent causal finding about stablecoins, and they note twice that because a wallet only enters the panel if it receives at least one stablecoin somewhere in the 53-week window, the receipt measure captures the timing of receipts among eventual receivers rather than unconditional adoption. That is an unusually candid framing and it should be taken at face value.

Two further limits are worth naming, because they bear on how far the evidence travels.

The data is Ethereum only. Retail stablecoin activity in several of the countries studied, Nigeria, Argentina and Turkey among them, runs heavily on other chains where fees are lower, and Tron in particular carries a large share of retail USDT. A panel restricted to ERC-20 transfers is likely observing crypto-native holders with ENS names rather than the households the model describes, which are the ones who would need to leave the banking system for the enforcement mechanism to dilute.

The placebo also does less work than it appears to. The paper checks Wrapped Bitcoin and finds no comparable crisis-week effect, concluding that wallets don't accumulate non-stablecoin crypto in response to a crisis. But WBTC is an Ethereum-side wrapper used mostly inside DeFi, and it is not how anyone fleeing a currency would hold bitcoin. A null result there tells you something about DeFi collateral behaviour and rather less about whether the crisis response is specific to stablecoins.

None of this undermines the model. It does mean the empirical section is a plausibility check on one assumption, as advertised, rather than a measurement of how much capital mobility stablecoins have actually added.

What follows for each group

Central banks in emerging economies. The model implies that the value of enforcement depends on where a country already sits on the adoption curve, and that the same enforcement budget produces different results in two countries with identical rules. It also implies that the moment to assess enforcement capacity is before a stress episode, since the mechanism bites hardest when flight pressure is already high.

Banks and payment providers. The dilution runs through the domestic banking system's share of cross-border flow. Institutions that operate on both rails occupy the position where the leakage is observable, which is also the position supervisors are most likely to ask about.

Issuers. Programmability enters the model as a parameter that steepens adoption's response to flight pressure. That places issuer design choices, not just issuance volume, inside a monetary sovereignty debate that issuers have generally treated as somebody else's argument.

Policymakers and legislators. The resist-then-accommodate result suggests jurisdictions will diverge based on where their adoption already stands rather than on stated policy preferences, which is one reason we maintain the free Stablecoin Regulation Tracker covering licensing regimes and supervisory guidance across 200+ markets.

The vertex changed owner

The paper's own summary is that stablecoins tighten the trilemma. The more useful way to put it is that they moved one of its corners out of the government's hands.

Under the classical framework, a country choosing monetary autonomy under a peg accepts the administrative cost of capital controls and gets the autonomy. Under this one, it accepts the cost and gets whatever autonomy its own households permit. Nothing in the impossible trinity breaks. The set of feasible combinations just stops being a policy choice and starts being an equilibrium, and it contracts on its own during exactly the episodes when a government would most want it to hold.

That is the theoretical counterpart to what the BIS documented empirically when it found that dollar stablecoins slip past the controls that bite on bank deposits. One caveat sits between the model and the world, though. Household migration to a stablecoin rail is only as easy as the on-ramp and off-ramp at either end, and those remain the slow and expensive part of the journey, as a central bank found when it sent 200 USDC around the world and discovered the blockchain was the cheapest leg. Enforcement that can no longer reach the transfer can still reach the conversion. How long that stays true is the open question, and the model suggests it is the one worth watching.

Source: Stablecoins Meet the Mundell-Fleming Trilemma, Pablo D. Azar, Maryam Farboodi and Nish D. Sinha, Federal Reserve Bank of New York Staff Reports no. 1202, August 2026. The report presents preliminary findings and its views are the authors' own rather than those of the Federal Reserve Bank of New York or the Federal Reserve System.

Additional data: BIS Working Paper No 1370 by Boris Hofmann, Aaron Mehrotra and Jan Paulick, published 21 July 2026, as reported by Central Banking. Market size and dollar-peg share from the BIS Annual Economic Report 2026 as reported by crypto.news and CryptoDaily. Latin American payment volume and regional purchase share from Bitso Business as reported by crypto.news.

Methodology: We focused on the model's mechanism, its predictions for government enforcement, and the scope of the accompanying empirical evidence, rather than reproducing the paper's derivations. Assessments of the dataset's chain coverage and placebo design are our own.

This article is for informational purposes only and does not constitute financial, investment, or legal advice.