Stanford Graduate School of BusinessUnderstanding Geoeconomics: How Global Economics Shape International Relations
Stanford Graduate School of Business
Defining Geoeconomics 0:00
Geoeconomics describes how hegemonic countries like the United States and China use their existing trade and financial relationships to exert power abroad, whether for geopolitical or economic reasons, sometimes bullying a government or a private firm. The speaker notes this topic drew little attention from economists until recently, and it raises core questions: whether this power is real, where it comes from, how it gets used, what optimal policy looks like for a country wielding it, and what it means for welfare, especially for the rest of the world. Unlike the zero-sum, mercantilist view where a fixed pie is simply fought over, the economic approach treats the pie as something that can grow or shrink depending on how power is exerted, so everyone could end up better or worse off.
National Security and Measurement Questions 2:00
The lecture also aims to define national security and a national security externality precisely enough to measure, and to ask whether there is a tradeoff between the standard gains from trade, such as specialization and external economies of scale, and economic dependency. If such a tradeoff exists, the question becomes how optimal policy should manage it. The speaker also flags an interest in policy counterfactuals, such as estimating how much the US economy would shrink and which sectors would suffer if China cut off rare earth exports, and whether that would cause a mild recession, no recession, or something much worse.
Lecture Roadmap 3:01
The session is structured to first give background from outside economics, particularly political science, since most PhD students have not been exposed to it. It then moves into formal theory, followed by two empirical approaches: a sufficient statistics method familiar from trade classes that combines a model with hard data, and a second, less familiar approach that examines unstructured text to identify who is exerting pressure on whom.
Political Science Concepts of Power 4:01
Robert Dahl's 1957 definition holds that A has power over B if A can get B to do something B would otherwise not want to do, a notion political scientists call relational power because it affects the relationship between two parties while taking the surrounding environment as fixed. Susan Strange's concept of structural power instead describes the ability to shape the entire environment, for instance by setting the rules of the game or establishing norms. The speaker's own contribution reframes this structural idea in terms of influencing general equilibrium variables, essentially being large enough to move the whole system, and stresses that political science treatments of this topic have often stayed informal, so today's goal is to make the concepts precise enough for theory and data.
Hirschman and the Herfindahl Index 7:31
The speaker recommends a book by Albert Hirschman about Nazi Germany using its economy to exert influence over the rest of Europe before World War II. Hirschman, later known for other work and even a Netflix documentary called Transatlantic, wrote this as a young economist trying to get a Berkeley faculty job, rejecting both mercantilist zero-sum thinking and the free-trade assumption that trade never creates imbalances. His book produced what is now called the Herfindahl index, originally built to measure concentration of trading partners and resulting power imbalances between sovereigns, though today it is mostly known for measuring firm concentration and markups.
Three Modeling Tools 9:01
The framework being built rests on three tools from economic theory. The first defines power as the gap between a target's inside option and outside option, essentially their participation constraint. The second treats the exercise of power as imposing costly wedges, or deviations from what an agent would privately choose, drawing on decades of public finance and micro theory for solving optimal policy problems. The third captures structural power as the ability to manipulate general equilibrium externalities, distinguishing relational, or micro-level, power from macro-level, structural power.
The ASML Example 11:32
The running example is ASML, the Dutch lithography firm that makes advanced semiconductor manufacturing machines, sourcing from US suppliers and selling to Chinese customers. The US government wants ASML to stop selling these machines to China, but since ASML is not a US domestic firm, the US cannot simply regulate or tax it directly. Instead the US threatened to invoke the foreign direct product rule, which could add ASML to the entity list, barring any US person from doing business with it, effectively a death sentence for the firm. This threat lowers the value of ASML's outside option, and the firm's participation constraint then determines whether it complies with the costly demand rather than lose all US business.
Strategic Sectors and Substitutability 16:00
Micro power is defined as the maximum cost of a demanded action that a firm will accept before preferring its outside option, and this depends heavily on substitutability. A single variety of oil is not very strategic because close substitutes exist elsewhere, but if one supplier like OPEC controls the entire supply, the outside option becomes far inferior and oil becomes strategic. The hegemon's real gain, though, often comes from manipulating the aggregate environment across many agents at once, illustrated by a social media platform that becomes powerful once everyone is persuaded to join it and drop competitors, since each person's private calculation ends up reinforcing the platform's dominance for everyone else too.
Government Intervention and Strategic Screening 19:00
Because strategic-sector manipulation involves a large player deliberately twisting the equilibrium in its favor, rather than a simple coordination failure among symmetric agents, the case for government intervention is stronger here than in typical financial regulation, where fire-sale externalities arise from agents not internalizing shared strategies. An example is a foreign country gaining control of port contracts along the US and Latin American western seaboard, where each individual contract might look harmless but the cumulative effect could threaten national security, justifying government screening of strategic sectors.
Setting Up the Formal Model 21:02
The model includes multiple countries, each with productive sectors treated as distinct even if similar across borders, such as Russian oil versus US oil, along with local factors and firms that buy intermediate inputs and use local factors, with aggregate variables built into production functions to capture externalities such as economies of scale or strategic complementarities. A representative consumer maximizes utility from consumption, optionally including arbitrary geopolitical preferences such as disliking another country's military buildup, and finances consumption through factor income and domestic firm profits, with markets required to clear for both goods and factors.
The Hegemon's Threats and Domestic Policy Game 26:00
In the model's middle stage, an exogenously designated hegemon can threaten foreign entities with loss of access to inputs it controls, such as semiconductors, rare earths, or oil, forcing them to accept wedges in their decisions or pay transfers, with the hegemon holding all the bargaining power through take-it-or-leave-it offers. Before that stage, every country sets its own domestic wedges, such as industrial policy, export controls, or tariffs, with no domestic participation constraints, since countries are assumed to be fully in control of their own firms. These domestic choices are made anticipating that the hegemon will later apply pressure, so countries shape their economies in advance to withstand future coercion.
Inside and outside options formalized 28:32
Each foreign firm weighs an inside option, producing with the full set of inputs but accepting costly actions and making transfers to the hegemon, against an outside option, producing with only a subset of inputs but facing no transfers and no imposed actions. On the inside option, firms maximize profits while perceiving wedges accepted from the hegemon and wedges imposed earlier by their own government. The hegemon's welfare includes its firms' profits, its factor payments, and the transfers it extracts from the rest of the world, while a bullied country's welfare collapses to the value of its outside option once transfers are set to make the participation constraint bind.
A concrete example with bankers 31:30
To get closed-form results, the hegemon is modeled as the United States with a single sector of bankers producing financial services from labor. The rest of the world consists of identical small economies, each with a domestic banking sector and an intermediary that blends the local variety with the US one into a composite good, illustrated with payment or messaging services. Two externalities are built in: domestic use of the local sector gets more productive as it scales up, and use of the global technology gets more productive the more other intermediaries everywhere else also use it. These are captured through a CES aggregator with productivity terms that depend on average usage across countries, with the strength of the global externality governed by a parameter called Kai.
The primal approach as a tool 34:32
To solve this, a global planner benchmark is introduced first, one who directly dictates quantities and maximizes total welfare, internalizing both externalities. Comparing the planner's first-order conditions to those of a private intermediary facing wedges lets you solve for the wedges that would make decentralized behavior match the planner's choices, a technique called the primal approach. The result is intuitive: the planner subsidizes both the domestic and global technologies in proportion to the strength of their externalities. This tool requires convex optimization and complete instruments, meaning the planner can directly command quantities.
A world without a hegemon 41:00
As a second benchmark, imagine no hegemon exists and every country sets domestic policy in a Nash game, taking others' choices as given, with infinitely many small open economies. Each government fully subsidizes its own domestic technology, exactly as the planner would, because that externality stays entirely within its borders. But it neither taxes nor subsidizes the global technology, since it perceives itself as too small to affect it. The general lesson is that externalities crossing borders get under-addressed in a Nash equilibrium: positive ones are under-subsidized, negative ones are under-taxed, because each country only sees part of the effect.
The hegemon's offense and the coercion gap 44:01
Visually, the intermediary's value is the area between a flat marginal revenue curve and a rising marginal cost curve; blocking access to foreign inputs shifts marginal cost, shrinking that area at the outside option. The hegemon always makes the participation constraint bind, keeping the blue outside-option area for the country and extracting the rest as transfers. Because the hegemon cares about the gap between inside and outside options rather than the level itself, it can be worth destroying value on the equilibrium path if that pushes the outside option down even faster, widening the gap it can extract, an idea likened to a drug dealer making customers dependent.
Solving the offense and the defense 49:01
Solving formally, the hegemon internalizes that more usage of its technology makes it more attractive to everyone, so it pushes usage of its own system up to exactly the level a planner would choose, since this only benefits the inside option. But unlike the planner, it taxes the domestic alternative to destroy countries' outside options, which is inefficient overall but optimal for the hegemon. Anticipating this, each country's best defense is to subsidize its domestic alternative to build up its own outside option, mirroring real-world moves by Europe or China to reduce dependency on the US. In this particular setup the result is stark: countries impose an infinite tariff on the global technology, producing full fragmentation, so the hegemon ends up with no power at all because no country engages with it, raising the question of whether the hegemon would do better committing to limited bullying instead.
Rethinking multilateral organizations 58:02
You can think of the IMF and the WTO not as neutral planners but as deals a hegemon offers that constrain its own actions to make participation attractive. The hegemon says, in effect, that you can join its sphere of influence, accept some bullying, but still keep meaningful value, because it is voluntarily limiting how much it extracts from you. Formally, if the hegemon commits to taking only a fraction of the inside-option value as a transfer, there is a middle range where the deal holds together, you retain real power, and both sides are better off than under full extraction or full withdrawal. Much of the last few years, the speaker argues, reflects a drastic rethink of that original offer.
Trade gains versus security 1:01:01
The core lesson from these models is that there is a deep, structural trade-off between the gains from trade and economic security. Specialization creates the efficiency gains economists prize, but that same specialization also creates a poor outside option if a trading partner can hold you up and refuse to sell. The two forces are not separate; they come from the same source, the hold-up problem, which is why current global rethinking about "de-risking" often means deliberately giving up some trade gains for more security.
The fragmentation doom loop 1:03:02
Because each country sets its anti-coercion policy taking everyone else's policy as given, pulling back from a shared technology makes that technology less attractive to remaining participants, who then pull back further, prompting the first country to withdraw even more. This amplification can leave everyone in an equilibrium that is overly secure and has sacrificed more trade gains than necessary, when coordinated policy could have done better.
Building power in the general model 1:04:00
Stripping away simplifying assumptions, the full model contains standard trade-policy terms, effects on firm profits, and terms-of-trade manipulation, alongside a genuinely new term called power building. Power building captures how changing one activity ripples through the whole equilibrium and shifts the gap between everyone else's inside and outside options, which is precisely the leverage a hegemon extracts. A semiconductor export restriction, for instance, matters not for its own sake but because it shrinks a downstream military application through the input-output chain.
Estimating the outside option 1:11:31
The empirical section uses a sufficient-statistics approach, borrowed from trade theory's autarky-cost calculations, to infer the unobserved outside option, such as what happens if a firm loses access to Chinese rare earths, from data on the observed inside option.
A nested CES production structure 1:16:04
Using nested constant-elasticity-of-substitution baskets, foreign varieties within a sector are combined, then blended with domestic production, then aggregated across sectors, with finance kept as a separate, nearly indispensable basket. Domestic economy size matters enormously for security, and foreign trade data, gathered at customs for centuries, is far better measured than domestic input shares, which the speaker calls the hardest and most neglected part of the calculation.
Measuring power through expenditure shares 1:21:31
Power is expressed as percentage value-added loss from restricted inputs, built from expenditure shares at each nesting level: how much goes to manufacturing versus finance, how much of a sector's spending is foreign versus domestic, and how much of the foreign share comes from the hegemon. This last figure, like "90 percent of some good comes from China," draws heavy attention but is only one piece of the picture.
Nonlinear losses near full cutoff 1:26:30
Calibrating elasticities from Cobb-Douglas assumptions for finance and Arkolakis-Costinot-Rodríguez-Clare style values elsewhere, the model shows losses become infinite as the import share from a single source approaches one, a pattern already visible in financial services heavily concentrated in the United States and its allies.
When Dependency Becomes Dangerous 1:28:01
The formula for power is highly nonlinear, which means most trade dependencies create almost no leverage at all. Real danger only appears near the extreme corners: when one supplier controls close to 100 percent of a good, or when there is almost no technical way to substitute away from it. This is why sectors like advanced semiconductors, radars, and certain financial services look like genuine choke points, while most ordinary trade relationships do not. The old idea of vulnerability, simply summing up trade concentration across partners, gets refined here into something sharper: concentration matters, but only combined with how hard it is to technologically rearrange supply if you get cut off.
Power Is Not a Zero-Sum Transfer 1:31:01
Because power depends on this nonlinear formula, it is not additive. Losing power does not simply hand the same amount of power to a rival. Singapore is the example: if the US controls Singapore's financial services, that pushes US dominance in finance even closer to total control, which is hugely valuable. But if China took control of Singapore instead, it would not make China a financial superpower, since finance is not China's base of strength. China might still want to dilute US power there, but the amount the US loses is not matched by an equivalent gain for China.
Where the Numbers Come Out 1:32:30
Running the calculations shows that small, open economies with heavy trade exposure, like Singapore, show up as having the largest losses because they have little domestic production to fall back on. For the US-led coalition, much of the power comes from finance, a small sector by output but one that touches nearly everything else and is tightly controlled. China's power comes mostly from manufacturing, which is generally easier to substitute, except for cases like rare earths, which behave more like the tightly controlled, hard-to-replace goods that generate real leverage.
Limits of the Data and Model 1:35:00
These are medium-run estimates, meaning firms are allowed to reoptimize their sourcing after a cutoff, but prices, wages, and exchange rates are held fixed, so the numbers are not a full long-run picture. Finer-grained data would capture narrow choke points better, but it gets noisier fast, especially for domestic production shares and for services, which are poorly tracked since they are rarely tariffed. As data gets disaggregated, substitution elasticities also tend to rise, since a single imported blue shirt is not really irreplaceable. Indirect exposure, such as Chinese content flowing into US imports through Mexico or Vietnam, is another gap current calculations do not fully capture.
Turning to Text and AI 1:39:00
Because many real threats never show up as an actual cutoff, since targets typically comply once threatened, and because demands are often hard to predict or categorize in advance, the researchers turned to text. Earnings calls from CEOs and CFOs, along with independent analyst reports, offer a way to capture pressure that formal trade data misses, since analysts covering a company may speak more freely about political pressure than the CEOs themselves. Large language models, using their cross-attention mechanism to link related words across long stretches of text, let researchers extract structured information at a scale that would be impossible by hand, since one document took a human about 25 minutes to code and there are 1.2 million documents to process.
Building the Classifier and Choosing Models 1:46:00
The team fed detailed prompts, written the way instructions to a human research assistant would be written, into an LLM used purely as a classifier, turning free text into structured variables like whether a firm was affected by a tariff or export control. A key technical choice is between closed-weight models, where the company controls the algorithm and it may change without notice, and open-weight models, which can be downloaded and run locally for reliable, repeatable results and fine-tuning, though at higher computational cost, a cost that has dropped sharply over the past three years.
Tracing Pressure Through ASML and Beyond 1:50:00
Applying the method to ASML's public filings, the algorithm correctly extracts that the US and Netherlands imposed pressure, China received it, the affected products were extreme and deep ultraviolet lithography tools, and the required action was lowering sales to China, all pulled automatically from free text. Because text comes from firms across the whole supply chain, including customers and even uninvolved third parties, researchers can trace ripple effects, such as Chinese or Indian firms benefiting from cheap Russian oil after Western sanctions. Aggregate trends show export control mentions rising sharply, driven by US pressure on Chinese semiconductors and, more recently, Chinese pressure on the US over rare earths, while sanctions data clearly picks up the Crimea and Ukraine episodes as well as the lesser-known Huawei and ZTE sanctions.
Export controls hit a small target 1:56:30
The US pattern of export controls looks like a small, high fence around a few sectors, echoing Jake Sullivan's description of the strategy toward China: pick a narrow set of critical industries, such as semiconductors and aircraft, and impose very high controls there, while leaving most trade, like autos, untouched. Tariffs look completely different. Mapped the same way, they resemble a Christmas tree, hitting almost every country, including tiny islands with no real economic weight, rather than concentrating pressure on major trading partners who would struggle to find alternatives.
Tariffs work like reverse export bans 1:57:31
A tariff can be understood as a threat not to buy rather than a threat not to sell, effectively a partial or infinite tax on someone else's exports to you. The logic mirrors the export control model: if you are a large enough buyer, cutting off your market forces the seller to slash prices to find new customers. Russia illustrates this, having to sell oil to India and China at roughly a 30 percent discount because those markets are smaller than the Western buyers it lost.
How firms actually respond 2:00:01
Text from company filings shows Chinese firms report real financial pain from US export controls, with far more firms citing negative effects than positive ones, and their main response is ramping up domestic R&D to replace the lost American semiconductor supply. On the US side, Nvidia's filings reveal it developing slightly downgraded chips specifically to stay just under the export control threshold, effectively working around the restriction rather than absorbing the loss.
Tariffs as a tax and a subsidy combined 2:02:00
Tariffs function as two policies at once: a tax on consumers and a subsidy to domestic producers. Firm filings split cleanly along this line. Firms hurt by tariffs report higher input prices and only partly pass those costs on to buyers. Firms helped by tariffs, like a US steel mill with no foreign inputs, benefit because competitors' costs rose, letting them either raise prices and profit margins or hold prices to grab market share, and only these winning firms report plans for new domestic investment.
Limits and quirks of AI-based data 2:06:00
This text-based method has real weaknesses: CEOs speak vaguely about raising prices without giving numbers, so quantifying effects like pass-through rates is hard, and the noise in the data doesn't behave like ordinary measurement error, making it difficult to correct. AI also fails unpredictably, once flooding results with irrelevant hits because "tariff" also means a phone or Netflix fee schedule in Latin countries, a case where a precise definition actually outsmarted the researchers, while elsewhere it struggled with subtler distinctions, like separating export controls from climate or safety-driven trade restrictions.
A wide open research frontier 2:12:01
Generating data from unstructured text and video is a major opportunity for students, comparable to earlier shifts from Robert Shiller's 1980s paper surveys to large-scale online human surveys. Costs have fallen sharply, much of the underlying text, such as congressional records and earnings calls, is public, and meaningful work no longer requires massive computing resources. The instructor describes this as one of the most exciting areas of his career, likely to shape economics for the next twenty years given its bearing on US-China competition and global order.
Open questions still to tackle 2:15:00
Ongoing and future work includes modeling a world with multiple rival powers rather than one dominant hegemon, and examining tension between a firm's own interests and its government's national security goals, using Nvidia as an example of wanting to sell broadly for profit while the US wants it to sell cheaply everywhere except China. Other open questions include whether state control over firms gives China an edge, whether the threat of government seizure discourages innovation, how to build causal rather than purely structural evidence, and how to give policymakers reliable, quantified answers about which industries truly matter for national security instead of accepting every lobbying claim.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
