Optimal Tariffs with Geopolitical Alignment

John Sturm Becko, Gene Grossman and Elhanan Helpman. Presented by Becko in Session III of the 4th Kiel-CEPR Conference on Geoeconomics, Sciences Po, Paris, 31 October 2025, with Alberto Martin discussing. Paper: the July 31, 2025 draft (NBER Working Paper 34108).

A tariff is a price list for friendship

The classic theory of the optimal tariff, from Mill and Bickerdike, has nothing to do with geopolitics. A large country taxes imports because doing so pushes down the world price of what it buys; the gain in terms of trade is set against the loss from trading less, and there is a tariff that balances the two. Becko, Grossman and Helpman add one ingredient: the large country also likes having allies, and it can offer friends a better deal than strangers. A preferential trade agreement is the carrot; the tariff everyone else faces is the stick; and once you allow both, the stick is no longer just a terms-of-trade instrument. It is the penalty for declining the carrot, and a hegemon that wants more friends has a reason to make the penalty larger than economics alone would justify. The paper’s central result is exactly that: with geopolitical concerns active, the optimal MFN tariff exceeds the Mill-Bickerdike level. Martin’s two-sentence proof in his discussion is the one to remember. Start at the terms-of-trade optimum; raising the tariff from there costs you only a second-order loss in terms of trade but buys a first-order gain in alignment, so you raise it. Without geopolitics, one flat tariff for everyone; with it, zero for the countries that join and a high rate for the ones that don’t, and small countries sort themselves by how much they mind aligning.

The formula that carries the whole quantitative story (eq. 10 in the paper) is the Mill-Bickerdike inverse-elasticity rule with one extra bracket:

Here $\varepsilon_{e(q_n)}$ is the export-supply elasticity of the non-aligned countries, $\beta$ is what the hegemon gets from one more ally, $S(q,1) - S(q_n,\tau^*)$ is the import surplus it forgoes when that ally’s goods stop paying the tariff, $y/c$ is the ally’s endowment over the hegemon’s per-capita demand for it, and $\lambda(\eta^*)$ is the hazard rate of the alignment-cost distribution at the marginal ally: the density of countries sitting right on the line, divided by the mass still outside. Everything in the calibration turns on that last term. A hegemon whose neighbours are bunched just outside the door will pay a lot to open it a crack; one whose neighbours are either already inside or nowhere near will not.

The machinery

The economics are deliberately spare: an Armington world with a homogeneous numeraire, each country endowed with its own variety, quasi-linear preferences, competitive markets. Export subsidies are ruled out (GATT Article XVI), export taxes too (the U.S. Constitution, China’s accession terms), so the only choices are whether to offer PTAs and what MFN rate to set. The geopolitics are reduced-form and honest about it: a hegemon gets a non-economic benefit proportional to the number of small countries aligned with it; in a bipolar world each great power also pays a cost for every country aligned with its rival; and each small country draws a “valence shock,” the net utility of aligning, which can be positive (security, legitimacy, admiration) or negative (lost autonomy, exposure to the rival), as in Keohane’s old typology. Two readings of GATT Article XXIV matter. Under the strict one, a PTA must be a free trade agreement with zero tariffs, and the hegemon offers one only when the marginal value of an ally exceeds a threshold. Under the lenient one, closer to recent practice, the hegemon can offer agreements of varying generosity, and then it always offers some: the Mill-Bickerdike rate to countries that would join anyway, the least generous rate each remaining country will accept, zero to the marginal joiner, and a higher MFN rate on the non-aligned than under the strict rule. Discrimination buys more allies and punishes the holdouts harder.


With two great powers the same forces appear in more complicated form: each sets its MFN rate taking the other’s as given but internalizing how small countries move among three camps, so there are three margins and three hazard rates instead of one. The theoretical question is whether the rivals’ tariffs are strategic complements or substitutes, and the theory says both are possible. A higher rival tariff makes the non-aligned more dependent on your market, which raises your terms-of-trade motive; but it also changes how many countries sit at the margin of aligning with you relative to how many pay your MFN rate, and that can go either way. The calibration settles it, and the answer is that it barely matters: the interaction is tiny, with China’s tariff a modest strategic substitute for the U.S. tariff and the U.S. tariff a very slight complement for China’s. (Martin’s summary that the tariffs “may actually be strategic substitutes” captures the American half of that.)

Anyway, the calibration

Alignment is UN General Assembly voting: you count as aligned with a great power if you are in the top quartile of voting similarity with it, following Gopinath et al. The price of a vote comes from Dreher et al.’s study of the Security Council, which found that Council members who consistently vote with the United States receive 42 percent more U.S. aid during their tenure than members who do not. That is a premium for voting the right way while you sit on the Council, not a premium for sitting on it, and footnote 37 turns it into a constant: 42 percent of aid running at 6.7 percent of GDP is 2.8 percent of GDP, which is taken to be what it costs to move a country of median similarity (28.2 percent) the 22.2 points up to the 50.4 percent swing line. The worked example is Malaysia, which voted with the U.S. on 23.7 percent of resolutions; bringing it to 50.4 percent (Japan’s and San Marino’s score, as it happens) would cost roughly 3.4 percent of Malaysian GDP, about 13.5 billion dollars a year, a number Becko says the international-relations people in Grossman’s and Helpman’s address books found about right. The great powers’ taste for allies is inferred from military spending (3.4 percent of GDP for the U.S. and 1.7 for China in 2023), on the theory that armies exist because of the countries that aren’t your allies. In the 1990s the U.S. bloc was 48.5 percent of world GDP and the swing state was North Macedonia; now the bloc is 32.6 percent, China’s allies are 5.4, Japan sits on the U.S. margin, Sudan and Lesotho on China’s, and India, Brazil and Russia are non-aligned.

![Calibration headline slide: alignment considerations raise the US tariff by more than 100 percent in the unipolar calibration and more than 30 percent in the bipolar one](figures/fig_00-20-05.jpg ‘Slide at 00:20:05: “the alignment considerations are sort of first order. They’”’"’re as important as … the terms of trade considerations." The paper’"’"’s own figures behind the rounding: 22.6 against 9.7 percent as a hegemon; 12.4 for the U.S. and 7.0 for China in the bipolar Nash equilibrium, about 31 percent above the no-geopolitics benchmark.’)

The results, in the paper’s numbers rather than the slide’s rounding. As a lone hegemon calibrated to 1997, the United States sets an MFN tariff of 22.6 percent; without geopolitics it would set 9.7 percent and offer no FTAs at all. Nearly three-fifths of the tariff is geopolitical. The stick-and-carrot regime attracts allies worth 52.7 percent of world GDP against 50.7 under free trade, and because most trade then enters duty-free, the average applied tariff is only 4.1 percent. In the bipolar Nash equilibrium calibrated to 2023 the U.S. sets 12.4 percent and China 7.0, geopolitics accounting for about a third of the first and a sixth of the second, and the pair are roughly 31 percent higher than they would be without alignment concerns. Under the lenient reading of Article XXIV the hegemon’s MFN rate is higher still.

The surprising thing

The sign is not surprising; anyone would have guessed that liking allies raises the tariff on non-allies. What is surprising is that the model says the U.S. tariff should have fallen between the two calibrations, from 22.6 percent to 12.4, and that China’s arrival has little to do with it. Replace China with small countries drawn from the 2020s distribution and the hegemon’s tariff is 14.4 percent (against 9.2 without geopolitics), barely above the bipolar figure. What changed is the shape of the distribution the hazard rate is taken from.

![Figure 4: histogram of 1995-98 alignment costs with the hegemon, with a three-component Gaussian mixture fit; a single tall hump just inside the alignment margin](figures/pdf_p33_figure-4.png ‘Figure 4, paper p. 32: the 1990s hazard rate. A unimodal pile of countries just inside the U.S. alignment margin, which is why the hegemon’"’"’s optimal tariff more than doubles.’)

![Figure 5: 2021-24 scatter of alignment costs with the U.S. (horizontal) and China (vertical), two tight negatively correlated clusters, Japan on the U.S. margin and Sudan on China’s](figures/pdf_p34_figure-5.png ‘Figure 5, paper p. 33: the 2020s. Two clusters, strongly negatively correlated, and nobody near the diagonal between the powers; Japan on the U.S. margin, Sudan on China’"’"’s.’)

In 1995-98 the alignment costs are unimodal, with a tall hump of countries just inside the margin, so a slightly larger price gap pulls in a lot of GDP. By 2021-24 the picture is bimodal: one cluster already aligned with the United States, one near China’s margin, almost nothing near the swing line and nothing at all on the diagonal where a country would be indifferent between the two powers. The countries you might buy are already bought; the rest are not for sale at any tariff. The paper’s own summary is that the distribution of alignment costs across small countries, “more than just the intensity of great powers’ preferences,” drives the geopolitical tariff. Polarisation makes trade policy a worse instrument of statecraft, not a better one.

As China grows

![Figure 9: China grows at the small countries’ expense holding the U.S. share fixed; China’s tariff rises near-linearly, the U.S. tariff declines, average MFN tariff is U-shaped, trade volume falls at an accelerating pace](figures/pdf_p39_figure-9.png ‘Figure 9, paper p. 38: China grows at the small countries’"’" expense with the U.S. share held fixed. Left, China’"’"’s tariff rises almost linearly while the U.S. tariff drifts down; right, the average MFN tariff is U-shaped with its trough near a Chinese share of 9.4 percent, and trade relative to free trade falls from about 0.97 to 0.88 and accelerates.’)

The second result is about the retreat from globalisation, and here too the mechanism is not the one you’d guess. China’s share of world GDP went from 3.1 percent in 1997 to 17 percent in 2024 while the American share held steady, so the simulation grows China at the small countries’ expense. China’s tariff rises almost linearly with its size, mostly for market-power reasons, since its taste for allies is modest. The U.S. tariff declines: as the small countries shed weight they are worth less as allies. The average MFN tariff is U-shaped, falling until China reaches about 9.4 percent of world GDP because trade is shifting toward the low-tariff power, then rising; the authors call that composition effect “somewhat mechanical” and prefer to look at trade volume, which falls steadily and at an accelerating pace. (If China instead grew at America’s expense, an appendix exercise, the decline is much more gradual, because the tariff-imposing share of the world isn’t growing.) So the model’s story of deglobalisation is not a tariff war between the powers, whose tariffs hardly respond to each other, but a second tariff-setter absorbing the countries that used to trade freely.

What Martin and the room did with it

Martin’s discussion began with what he called a cheap shot that is not really cheap: geopolitics here is a vote count, every ally worth the same, whereas an actual United States has many dimensions of strategic interaction at once, containing China, guaranteeing the supply of certain inputs, maybe wanting a foot in the Arctic once the ice goes away, and each of those involves a different set of countries valued differently. A single uniform tariff is a coarse instrument for that. That led to his real question: why tariffs at all? If you want allies, pay them. His guess was that the answer depends on which model of the WTO you are in, that is, how much you are allowed to discriminate with tariffs (if not at all, tariffs are very unappealing), and on the distribution of alignment costs; and that the model could say which powers would reach for which tool, given that China notoriously prefers concessional loans and FDI. He came back to this on the calibration: the paper prices alignment off a literature that says countries voting with the U.S. receive about 40 percent more aid, so it already assumes alliances are bought with transfers, and those transfers are interacting with trade policy whether the model admits it or not. He also asked what the model implies for MFN itself, since discriminating helps you but your rival can discriminate too, so the value of the rules-based system is ambiguous and the model could say who defects first. And he flagged two quantitative points he wanted explained: the implication that the U.S. should have cut tariffs between the unipolar 1990s and the bimodal 2020s, and the fact that once you average the high MFN rate with all the zero-tariff agreements, trade falls only about one to one and a half percent even under strong geopolitical preferences.

![Figure 8: Nash tariffs and their geopolitical components as all four preference parameters scale up, with two FTA-threshold jumps; average MFN and applied tariffs rise and trade volume falls](figures/pdf_p39_figure-8.png ‘Figure 8, paper p. 38: the panel behind Martin’"’"’s “one, one and a half percent.” As geopolitical preferences scale from the point where FTAs first appear to double the baseline, the average MFN tariff rises from about 8.7 to 11 percent while trade relative to free trade falls only from about 0.9415 to 0.930.’)

The trade number is a reading of Figure 8(b), where the trade volume runs from roughly 0.9415 to 0.930 of the free-trade level as preferences scale from just enough to trigger FTAs to double the baseline; the text gives no figure of its own. Martin offered two readings: either geopolitics does not matter much for trade, or, as he preferred, trade is not shrinking but fragmenting into camps, which is what the IMF keeps saying. His self-promotion was substantive: in his work with Broner, Meyer and Trebesch alignment is multilateral and strategic complementarities create multiplicity, so the bimodal distribution the paper takes as given in the 2020s might be endogenous. Once China became big enough to assert itself, countries started aligning with it because others were.

Becko’s answer on transfers was that there is an extension, described in the talk, in which the hegemon can offer a transfer as well as an FTA, and the finding is that it uses both: a small deviation from its optimal tariff has second-order cost to itself and first-order effect on others’ incentives, so even with a more efficient instrument available it does not put the tariff down. The July 2025 draft does not contain that extension; aid enters it only as a calibration input and, in the conclusion, as future work. (The same second-order-versus-first-order argument does appear in the draft, in an appendix footnote, to explain why a discriminating hegemon offers some partners a rate below Mill-Bickerdike.) On MFN, he agreed the value of the rules is unclear and depends on how countries are allowed to compete: if each could tailor its carrot to each country, the stick is always the same threat of autarky while the carrots get sweeter as geopolitics intensifies, so more geopolitics might mean more trade.

The best exchange was about 2025. Rodrigo (presumably Adão; the video description lists an award ceremony that afternoon for Rodrigo Adão and Christopher Clayton) asked whether the model speaks to a United States raising tariffs on its allies, where it looks as though countries are paying to stay in the American sphere rather than being paid to join it. Chris (presumably Clayton, who is also in the paper’s acknowledgments) pushed back on the transfers answer: the reason the hegemon uses both instruments is that the paper lets it pay lump-sum in the geopolitical market while extracting like a monopolist in the goods market; allow non-distortionary side payments in both and you would only ever use the side payments, so the interesting question is how far the tools leak from one market into the other. Becko said he did not have a great response to that. Bob (Staiger, by inference: Becko answers “Rodrigo and Bob,” and the chair introduces Robert Staiger’s paper next) added that the Miran blueprint says explicitly that allies will be deterred from retaliating by the threat of withdrawn security guarantees, which inverts the paper, with security as the cudgel and tariffs as the thing extracted. Becko’s reply was that, thanks to quasi-linearity, “alignment” can be relabelled as anything separable the small country gives up, low tariffs on you, an investment in your country, and the results go through; in a richer general-equilibrium model the two would interact, and he would like to think about it. Which is to say the paper’s carrots and sticks are symmetric enough to describe a world where the hegemon sells protection rather than buying friendship. The model, being quasi-linear, is indifferent between the two. The allies, one suspects, are not.