An export control can close a sale and it cannot close a dependency

The case for export controls is usually made on one number and judged on the same one: how much less the restricted country buys. That number is the wrong scoreboard, and defending the instrument on it is the surest way to lose the argument about it. My position is that an export control does one thing well, that the thing it does is narrow, and that the narrowness is a property of the instrument rather than a defect in its administration.

Stated plainly, the position is this. A control can stop a named good reaching a named buyer, and that is a real capability worth having. It cannot reduce a dependency, it cannot build a supplier, and the trade series most often cited as evidence that it worked is not a measure of dependency at all. The line falls at the edge of the transaction.

The line that moved

Goods imports from China averaged 36 billion dollars a month in 2020, rose to 45 billion a month by 2022, and stood at 26 billion a month in 2025. The last step alone is 11 billion a month, from 37 in 2024, and the 2025 reading is 10 below where the window opened. These are monthly averages of a bilateral goods series, not annual totals. On any reading, the pair moved, and it moved a long way.

The selling side of the same ledger sits at the lowest of its own six readings. Goods exports to mainland China averaged 13 billion dollars a month in 2021 and again in 2022, 12 in each of the two years that followed, and 9 in 2025. That is one billion a month below 2020, when the figure was 10. A restriction is administered against a buyer and paid for by a seller, and the seller in this ledger is domestic.

None of that is attribution. This data holds no licence, no denial and no control list, so it cannot say how much of either movement came from a control, how much from a tariff, how much from demand, and how much from goods taking a longer route to the same destination. I am not going to claim the fall as a policy result, and the argument does not need me to.

The aggregate did not follow

Set the total against the pair. The balance in goods and services ran minus 53 billion dollars a month in 2020, minus 69 in 2021, minus 75 in 2022, minus 62 in 2023, minus 74 in 2024 and minus 78 in 2025. The last figure is the widest of the six. Against 2022 it is wider by 3 billion a month, over the same years in which the bilateral goods line fell by 19.

Those are different measures and I will not difference them. One is bilateral and covers goods; the other covers every partner and adds services. What the pairing establishes is a direction, and a direction is enough: the pair the instrument names got smaller, and the total the country is actually exposed to did not. Purchases have counterparties, and counterparties can be changed without anything upstream of them changing at all.

Where a capability would show

If controls were doing the work their advocates describe, the evidence would sit in a production series rather than a trade one. Computer and electronic product output, on an index set to 100 in 2017, stood at 117 in 2022 and 126.8 in 2025, a rise of 9.8 index points, and it is 21.1 points above its 2020 reading of 105.7. Manufacturing output, on the same 2017 base, was 98 in 2022 and 97.3 in 2025, a fall of 0.7 points. The computer and electronics line therefore sits 26.8 points above the shared base year and the manufacturing index around it sits 2.7 points below it.

That divergence is worth having and it is worth describing accurately. An index of this kind records how much was made against a base year. It does not say which devices, at which technology generation, on whose equipment, with which imported inputs, or at which stage of manufacture. A country can raise the total while the specific parts a defence programme depends on are still made in one place on the other side of an ocean, and the index would not register it. The series also cannot say why it rose, which cuts against my case as firmly as it cuts against anyone else's.

The objection worth answering

The strongest reply is that I have picked a fight with a claim nobody serious makes. Controls were never meant to shrink a trade balance, the argument goes; they were meant to slow an adversary's access to a specific capability, and on that question a trade series is evidence for nothing in either direction. That is correct, and I accept it without qualification.

It also concedes the whole point. If a bilateral line cannot demonstrate that a control worked, then it cannot be offered as demonstration either, and it is offered constantly, in hearings, in press releases and in the sentence that a programme reduced dependency by some number of billions. Each time it is offered, the case rests on a series that will not bear it. The reckoning comes later, when someone competent notices, and the real argument goes down with the inflated one.

A second objection cuts nearer. A bilateral goods series falls when trade stops and it falls when trade is re-routed, and nothing here separates the two. That is true, and it is why I have stated the movement rather than interpreted it. It is also the second reason the bilateral ledger makes a poor scoreboard: it can be moved by a change of paperwork.

Where the line falls

Three propositions, then, and I would sign all three. A control can deny a specific good to a specific buyer, which is a genuine instrument and should be used where the good is genuinely specific. A control cannot reduce an aggregate exposure, because the aggregate has other counterparties and the instrument acts on a pair. And a control cannot manufacture a substitute, because substitutes come from capital, workers, qualification and time, none of which is contained in a prohibition.

This page carries no chart, and the absence is deliberate. The argument runs across a bilateral goods pair, a total that includes services, and two production indexes on a 2017 base, which is three different measures and one point about how they relate. A single line drawn from any one of them would look like the argument and would be standing in for it. The figures belong in sentences here, where the base of each one can be stated alongside it.

What follows for practice is modest. Judge a control against the thing it denies and the time it buys, name the date by which the substitute is meant to exist, and publish the production series that would show whether it does. Do not judge it against a trade line. Denial is the beginning of an industrial policy, not a result of one, and the country that mistakes the first for the second will keep announcing victories in a series that was never keeping score.