A Framework for Geoeconomics

Christopher Clayton (Yale SOM), Matteo Maggiori (Stanford GSB), Jesse Schreger (Columbia). Published version: Econometrica 94(1), January 2026, pp. 105–136. Two talks used: Maggiori at the Hoover Institution economics seminar, Stanford, 10 January 2024 (chaired by John Taylor, Schreger on Zoom, no discussant — the room interrupts throughout; 90 minutes), and Maggiori’s ABFER 13th Annual Conference Master Class, Singapore, May 2026 (two hours, moderated Q&A by Andrew Rose), which covers the framework and the follow-up measurement work. Slide images are frames from those videos; Figures 1–2 and the equations are cropped from the PDF; page references are the journal’s.

The button problem

Khrushchev said you can make anything strategic, even a button, because a soldier needs buttons or he has to hold his trousers up and then what does he do with his rifle. Maggiori opened the Hoover talk with that line, because it is the problem the paper is trying to solve. “Strategic sector,” “economic coercion,” “weaponized interdependence,” “choke point,” “dependency” — every one of these words is used constantly by governments and almost never defined, and the reason they are never defined is that defining them is in the interest of nobody who wants a subsidy. The paper’s contribution is not a new empirical fact. It is a set of definitions, built out of three very standard pieces of economic theory, such that you can argue about whether China’s Belt and Road is extractive, or whether Nvidia’s chips are strategic, in terms that at least mean the same thing to both sides of the argument.

The three pieces are limited enforceability with trigger punishments (Abreu–Pearce–Stacchetti, 1986/1990), input-output amplification (the Leontief inverse, as in Baqaee–Farhi), and externalities in the Greenwald–Stiglitz style, meaning a vector of aggregates that everyone takes as given but that everyone’s actions jointly determine. The claim is that once you have those three, geoeconomic power is not mysterious. It is a participation constraint.

Stealing, trust, and why a joint threat is worth more than two separate ones

Start with a firm in sector $i$ buying an input from the firms in sector $j$. Each period the buyer places an order, the suppliers accept or reject, the goods are delivered, and then the buyer decides whether to pay or to steal — where “steal” means any deviation from a deal that the law can’t fully fix: not repaying a loan, expropriating a mine, defaulting on a bond, walking away from a contract in a jurisdiction where enforcement is thin. If the buyer steals, the suppliers recover only a fraction $1-\theta$ of what they’re owed, and they never trust that buyer again: any future order is rejected. That is the whole source of discipline. The buyer stays honest only because the continuation value of the relationship exceeds the one-off gain from stealing. In the paper’s notation, with $\nu_i(\mathcal{J}_i)$ the value of having all your suppliers’ trust and $\nu_i(\mathcal{J}_i\setminus S)$ the value after losing the ones in S:

This is eq. (1), the incentive-compatibility constraint for stealing from the set $S$. The left side is the cash you’d pocket; the right side is the discounted loss of whatever you can no longer produce without those inputs. Two things follow immediately, and Maggiori made both of them in the Hoover talk before anyone could object. First, the threat is only worth anything if you can’t substitute: “threatening you with something that you can easily buy somewhere else from somebody else just isn’t valuable in equilibrium.” Controlling one variety of oil gives you nothing, because next period the buyer just orders from someone else; controlling all of them might shut down production entirely. Second, the value of trust is computable: it is a counterfactual loss from losing a set of inputs, which is exactly the kind of object trade theory has been computing from expenditure shares and elasticities since Arkolakis–Costinot–Rodríguez-Clare.

Now the move. Suppose the suppliers in $j$ and the suppliers in $k$ agree that if the buyer steals from either of them, both cut it off. Figure 1 draws it.

Figure 1, paper p. 111: “Triggers, action sets, and incentive compatibility constraints.” Panel (a) has separate triggers and three IC constraints; panel (b) has a joint trigger and only the joint one survives.

With separate triggers the buyer faces three constraints: don’t steal from $j$, don’t steal from $k$, don’t steal from both. With a joint trigger, stealing from only one is dominated — you’d be punished by both anyway — so only the joint constraint remains, and that constraint is looser in the relevant sense: the buyer can now be trusted with larger orders from each supplier, because the punishment for any deviation is the loss of everything. This is Bernheim–Whinston multi-market contact and Holmström–Milgrom multitasking wearing a new coat: bundling punishments provides higher-powered incentives. Lemma 1 says the general version of this is just a partition of your supplier set, and Proposition 1 says a hegemon should always choose the coarsest feasible partition — the “maximal joint threat” — because it weakly expands what every targeted firm can do, and you can always ask for more once you’ve given more.

Slide at 00:33:50: “What it really does in this particular example, it transforms the problem from having three ICs to only having the joint IC… I’m giving you higher power incentives by telling you if you deviate on anything, I will punish you all across the board.”

The hegemon, in this paper, is simply the one country that can coordinate these joint triggers — across its own sectors and across the foreign firms one step downstream of them. That’s it. That’s the definition. The US can tell a European bank that if it finances Iranian trade, it loses dollar clearing and access to US technology inputs and its American customers; no individual American supplier could credibly do that alone, because each one sees the bank paying its own bills on time and has no private reason to cut it off. Maggiori’s nicer example, from the Hoover talk: Italy builds infrastructure in Africa and gets expropriated the next day; China builds the same infrastructure and doesn’t, because “the threats that China makes are to suspend lots of other economic activities that are very valuable to that country for any one deviation. So, what they’re doing is essentially cross-collateralization.”

Power is the slack in a participation constraint

Here is the part that makes the framework useful rather than merely tidy. The hegemon offers each firm it can reach a take-it-or-leave-it contract: a joint threat (a gift — it relaxes your incentive constraint), plus demands — a transfer $T_i$ to the hegemon’s consumer, and a set of revenue-neutral wedges $\tau_{ij}$ on your input purchases, which are the workhorse instrument of the macro-prudential literature and which can be specialized to tariffs, quantity restrictions, export controls, or “stop buying from Huawei.” The firm accepts if and only if its value under the contract beats its value without it:

That’s eq. (3), the participation constraint, with $\Gamma_i=\{\mathcal{S}'_i,\mathcal{T}_i,\tau_i\}$ the offered contract and $\mathcal{S}_i$ the firm’s original, un-coordinated set of threats. The value on the left is eq. (2): stage profits, minus the transfer, minus the distortion from the wedges, plus the continuation value, subject to the (now joint) IC. The firm’s decision is voluntary. Nobody is invading anyone. The hegemon “creates slack in the participation constraint by proposing a joint threat, and then can use that slack to demand costly actions.” The slack is what the paper calls Micro-Power: the maximum private cost you can impose on a target before it walks, holding prices and aggregates fixed. If you supply rare earths and the target has no alternative, the slack is huge. If you supply water to a Nordic country, it’s zero.

Slide at 00:46:20: “So the participation constraint here tells me how much power do I have over you at the beginning. In some sense, this is micro power.”

This is Dahl’s 1957 definition of power from political science — A has power over B to the extent A can get B to do something B wouldn’t otherwise do — turned into a Lagrange multiplier. And it has an immediate corollary the paper keeps returning to: a blunt threat (“do what I say or I cut you off”) is inefficient. It worsens the target’s outside option rather than improving its inside option, which tightens the target’s incentive constraints and shrinks what you can extract. The better play is always to offer something valuable and then charge for it. One Hoover participant (unnamed in the captions) pushed on exactly this — why not allow escrow, third-party adjudication, other punishments? — and the answer was that for the relationships the paper cares about (sovereign lending, a mine in Africa, political concessions that nobody wants written down), $\theta$ is the relevant parameter and no escrow fixes it, and that most threats governments like to make are, in equilibrium, worthless precisely because of substitution.

Macro-power: asking people to do things for reasons they don’t see

If that were all, the hegemon would just be a monopolist with unusually well-coordinated enforcement, extracting transfers from foreigners up to their participation constraints and leaving its own firms alone (transfers from domestic firms are a wash for the hegemon’s consumer but still tighten the firms’ IC, so they’re pure deadweight). In the simplest environment — constant prices, no externalities — that is exactly what Supplemental Appendix Proposition 8 delivers, as the paper notes at p. 119: all foreign sectors neutral, no wedges, all the Micro-Power spent extracting transfers from foreign firms until their participation constraints bind, and nothing taken from domestic firms (that last part is Proposition 3 itself).

The interesting part arrives when prices move and production externalities exist. Then each firm, when it decides whether to accept, considers only its own private cost. But the hegemon is contracting with all of them at once, and it knows that their collective actions move the equilibrium. Every firm in Europe weighing whether to drop Huawei thinks “given the equilibrium, what does this cost me”; the US government is thinking “what happens if the entire Eastern seaboard’s ports run on Chinese telecoms.” The paper’s optimal wedge formula is built to exploit exactly this gap:

Eq. (5). $\eta_i$ is the multiplier on firm $i$’s participation constraint — the marginal cost to the hegemon of using up its power over that firm — and $\mathcal{E}_{ij}$ is the hegemon’s total perceived externality from an increase in $i$’s use of input $j$. So the wedge is “how much I want this activity changed, divided by how expensive it is to ask you.” Activities with positive spillovers get subsidized; negative ones get taxed. The first bracketed term is the direct effect (eq. 6): the activity’s effect on the hegemon’s own firms’ profits, on its consumer’s non-economic preferences (the $u_m(z)$ term — national security, disliking a rival’s military, whatever), and on the hegemon’s own future power, via how the activity changes other targets’ inside and outside options. The second term is the one that makes a sector “strategic” in the macro sense: $dz^{*NC}/dz_{ij}$ is a generalized Leontief inverse (Proposition 2) that tracks how changing this one firm’s behaviour propagates through prices and production externalities to sectors the hegemon cannot contract with at all. Proposition 2 is the technical contribution — it folds price-based amplification and externality-based amplification into one matrix — and Maggiori was candid that it is “15 lines of algebra” but that nobody had written it down in that generality.

Slide at 01:00:50: “There’s a whole part of the world economy that I don’t control. I don’t have any threats on them. I don’t even trade with them. But they’re going to respond in equilibrium to the actions I asked out of the part that I control. In principle, I might care more about them than the ones I control.”

So there are two distinct things a sector can be strategic for. Micro-strategic: it lets the hegemon form valuable threats (widely used, high value added, poor substitutes — rare earths, oil, SWIFT). Macro-strategic: demanding costly actions from it moves the world equilibrium in the hegemon’s favour through the Leontief inverse (finance, R&D, IT — things with strategic complementarities and scale economies). The hegemon particularly wants micro-power over macro-strategic sectors, because that is where the gap between the target’s private cost and the hegemon’s social benefit is widest, and the gap is what it is monetizing. Rearranged, $\eta_i=-\mathcal{E}_{ij}/\tau^*_{ij}$: the marginal value of more power over a sector is how much you’d like to control its activities relative to how much you actually do. Maggiori’s aside at Hoover, which he flagged as not in the paper and which the room found the most interesting thing said all afternoon: this is a general structure, not a geopolitical one. Regulatory capture, a dominant firm coordinating an industry, any big player in the same position: “I’m asking people to do individual actions cuz I have some power over them. They accept to do my bidding, but they think that the equilibrium is constant. But the reason why I’m actually asking them is I’m changing the equilibrium, and they’re not noticing that.”

Friends, enemies, and a global planner

The sign of $\mathcal{E}_{ij}$ gives you a theory-based definition of friends and enemies that does not depend on who you are but on how your activity moves the hegemon’s value function: a foreign sector is unfriendly if all its spillovers on the hegemon are negative (tax it), neutral if zero (leave it alone), friendly if positive (subsidize it — this is the paper’s reading of NATO, the hegemon putting positive utility on allies’ defence sectors and optimally not fully exercising its coercive power on them). Nvidia selling to China is unfriendly in that specific activity even though the US wants Nvidia large and successful; the same firm gets a subsidy at home and an export ban abroad, and the formula says why both at once.

Proposition 4 then asks what a global planner with the same instruments and the same constraints would do. Answer: the same maximal joint threats (enforcement is globally valuable — the Kindleberger “hegemonic stability” public good is real in this model), zero transfers (they tighten ICs and are negative-sum), and different wedges, because the planner’s perceived externalities are computed on world welfare rather than the hegemon’s. The hegemon therefore does two things at once: it pushes the world Pareto frontier out by supplying enforcement, and then picks a point inside that frontier by extracting rents and manipulating externalities its own way. Whether any given target is better or worse off than in a world with no hegemon depends on which effect dominates — which is why, as Maggiori put it, “the newspaper version of what’s positive and what’s negative sum doesn’t work that well.”

Two applications

The Huawei case (Section 4.1) is the macro-power story in three sectors. Rest-of-world firms use a Chinese technology $H$ that has interoperability spillovers: the more of your neighbours use it, the more productive it is for you. The US dislikes its use for national-security reasons. The US can pressure some of those firms (call them $i$, Western Europe) but not others ($j$). The optimal restriction on i has three pieces: the direct national-security benefit; a term for building power, since each firm that drops $H$ makes $H$ less productive for the rest of its sector and so less attractive to reject the contract for; and a network-amplification term, because $i$ dropping $H$ lowers $j$’s productivity in $H$ too, so $j$ reduces usage on its own, which the US values and which further loosens $i$’s participation constraint. The last bit is the point: the US pushed Europe harder than the direct benefit alone would justify, because once Europe was off the technology, adoption elsewhere would fall without the US having to threaten anyone. “Ultimately, I might get to reach the equilibrium even if I control only a small part of the world economy.”

Figure 2, paper p. 123: “Application: national security externality.” The hegemon ($\kappa$, $\delta$) can threaten $i$ but not $j$; the production externality between $i$ and $j$ is what lets pressure on $i$ reach $j$.

The Belt and Road case (Section 4.2) is the micro-power story with a twist about measurement. China lends to a developing-country sector and also supplies it manufacturing inputs. Assume the loan is completely unenforceable ($\theta=1$) and the manufacturing relationship completely enforceable ($\theta=0$). With separate triggers, lending is limited Eaton–Gersovitz-style by the value of future borrowing alone. With a joint trigger, defaulting on the loan also forfeits the manufacturing relationship, so the borrowing limit rises by the present value of the manufacturing surplus:

(Section 4.2, unnumbered; $b$ the loan, $R$ its gross rate, $f_{ij}$ the manufacturing production function, $\bar T_i$ the transfer.) Trade acts as an endogenous cost of default — Bulow–Rogoff’s seizure of exports, but volunteered by the borrower because it raises its own debt capacity. China takes the surplus as $\bar T_i>0$, which can be a markup on the machinery, an above-market interest rate, or a political concession like not recognizing Taiwan. The cautionary implication, which Maggiori stressed at Hoover and which the paper states flatly: evaluating BRI by the return on the loans is evaluating one leg of a joint threat. “China might be making horrible returns on the loans, but if he’s getting political recognitions or to use naval bases, that might be worth a small fortune.”

Slide at 01:24:35: “Both the sustainability and the actual profits you make out of this might not come from the interest rate on the loans.”

What the rooms pushed on

The Hoover seminar was less a presentation than a 90-minute cross-examination by a room that, per the video description, included Cochrane, Admati, Bulow, Diamond, Judd and Taylor (the captions rarely name who is speaking, so attributions below are hedged), and the objections cluster usefully. On the game theory, Ken Judd (if the captions have the name right; the speaker refers to “Shabina and I,” which fits Judd–Yeltekin–Conklin) argued that building the value function outward from the inside, adding one trusting sector at a time (the “inner approximation,” as he called it in his own work), is not guaranteed to find the full equilibrium set — the APS machinery works by shrinking from the outside and converging to the maximal fixed point. Maggiori’s answer, after some crosstalk, was that they are not claiming to find the best sustainable equilibrium at all: it’s one Markov SPE with permanent exclusion, chosen for tractability, and front-loaded or back-loaded punishments would do better. Someone added that in both theory and practice, punishment tends to be finite and forgiveness follows; the answer was that Poisson re-entry is feasible and just scales the continuation loss. One participant, repeatedly (the captions don’t name him; the Peter Navarro framing is his), wanted to know what any of this had to do with tariffs on Chinese EVs or with “we need to win the strategic economic competition” — the Peter Navarro version of geoeconomics — and Maggiori kept asking for patience until the macro section, where the answer is: the rivalry is in $u_m(z)$ and the externalities, but whether rivalry produces less trade is not a prediction of the model; indirect effects can dominate and bilateral trade in some sectors can rise with geopolitical tension. Another questioner noted the model has no regime-versus-populace distinction, which matters when sanctions target governments; conceded. Several people asked about multiple hegemons, endogenous network formation, and the target’s defensive options. All three are outside the main analysis, which assumes a single hegemon, a fixed network, and small passive targets — though Supplemental Appendix B.2 does carry an extension to competition between multiple hegemons, a few worked setups rather than a general characterization. Maggiori described each as a separate project — multi-hegemon competition “is a game” and much harder; anti-coercion policy for a targeted government is precisely about making domestic firms internalize that they are being picked because the hegemon is exploiting the gap between their private cost and its social benefit. A question on Biden’s executive order to count global benefits in cost-benefit analysis was, diplomatically, parked.

The ABFER class, two and a half years and one administration later, added the measurement. Maggiori showed the micro-power counterfactual computed with a nested CES calibration and OECD input-output data: the loss to a target from losing access to the hegemon’s share of a sector scales with the target’s expenditure share on the sector, the inverse of the elasticity minus one, and $\log(1-\omega)$ where $\omega$ is the hegemon’s share of the target’s purchases in that sector. The log term is the lesson. Power is non-linear in dominance: controlling 95% of a sector is worth enormously more than controlling 60%, and the first 10% that a competitor peels away destroys most of it. That is why, in his telling, the US’s power runs overwhelmingly through plain-vanilla finance — custody, payments, correspondent banking, where the Western coalition has near-total shares — while China’s runs through manufacturing, where it is a large exporter but faces high substitutability. A 10% Chinese share of cross-border payments would be “an enormous policy mistake” to dismiss as small: “you lose most of your power from losing that first 10%.”

Slide at 01:11:15: percentage loss to each target from being cut off by the US coalition (left) and by China (right), split into goods & services vs finance. “For the US, a lot of power comes from the blue bars. That’s finance.”

The room’s sharpest challenge, from a questioner the captions leave unnamed, was that the framework says a hegemon should be “cuddly” toward allies and cultivate institutions, while the actual hegemon has spent a decade undermining the WTO, leaving the WHO, tearing up TPP, and alienating the allies whose participation makes its financial threats non-linear. Maggiori’s answer was in three parts. The thirty-year story: the world economy shifted toward traded sectors with scale economies and complementarities, exactly the sectors where this power exists, so the game got bigger. The ten-year story: a collapse in US commitment, which in the model is a short-run temptation — if everyone built dependencies on you expecting you to stick to your word, deviating is very profitable until they adjust and fragment — and possibly a deeper force, the arrival of a competitor: “suppose that I tell you now that 10 years from now, most of the value of this relationship is going to have to be shared with China. That might turn me very nasty and extractive in the short run.” And on allies: “the last thing you want to do in this model is to gratuitously annoy your allies… That’s a disastrous policy in this model. That doesn’t mean that there’s not one that doesn’t get pursued.” Steve Davis then pressed on whether the input-output data exist at the granularity the formulas demand, citing Laura Alfaro’s work on rare earths; Maggiori agreed entirely, said the domestic data are worse than the cross-border data because nobody ever had a reason to tax goods that don’t cross a border, and argued that the value of the theory right now is mostly disciplinary — if you claim export controls on a chip will cost China 7% of GDP, the formula tells you what you have to believe about an elasticity nobody has estimated.

The odd thing

The paper’s title is a promise of neutrality, and it is mostly kept: the machinery is a principal-agent problem, and the hegemon is just the principal who happens to be a country. But the one substantive claim that survives every simplification is a little uncomfortable for both sides of the current argument. Enforcement is positive-sum — the hegemon as global policeman really does expand the production frontier, because it lets people trust each other who otherwise couldn’t. And the hegemon will always, optimally, spend that surplus moving the world inside the frontier it just created, because that is what the participation constraint is for. The framework’s advice to a hegemon is to be maximally coordinated and reliably useful, and then charge; its advice to everyone else is to notice that they are being asked to do the cheap thing for the hegemon’s expensive reason. Which is, if you squint, a fairly precise description of what the ABFER text-mining found: US export controls cluster tightly in the handful of sectors the model says are chokepoints, and the targeted Chinese firms’ dominant response is to report doing domestic R&D to build the alternative — raising the elasticity, and eroding the log term, exactly as a target should.

(Maggiori’s own summary of where the empirical measurement stands: “rudimentary.” Of the policy conversation around dependency shares: “looking at shares and interpreting them as dominance is a serious mistake and is one that I would say it’s routine in the press and even in policy circles.” Both rooms laughed at the right places.)