In August 2017, Washington announced it was investigating China’s trade practices under Section 301. The tariffs didn’t bite until July 2018 - 322 days later. The political relationship had measurably soured in 2016, a year before the investigation was even announced.

So a firm that moved when the politics turned had roughly two years to act. A firm that waited for the tariffs had days. Both would have described themselves, sincerely, as "watching geopolitical risk."

The variable that separated them was the runway: the gap between the first credible warning and the moment a measure actually binds. Across the policy actions we’ve been able to date fully, that gap runs from 322 days at one extreme to literally zero at the other; some sanctions designations are announced, legally decided, and in force the same afternoon. The median is 52 days.

Why the range exists

Runway varies so widely because a policy-driven rupture isn’t one event with one date. Politics, policy, and the real economy move on separate timetables. The political relationship deteriorates with rhetoric, indices and headlines. Policy follows through public, dateable stages: announcement, legal adoption, implementation. Trade shifts, and jobs later still, if they shift at all. Think of them as three clocks that are usually assumed to be synchronized and rarely are; the runway is what the gaps between them add up to.

The first instalment of this series showed why a single risk number fails in space: a US county’s exposure to Chinese export demand and its exposure to Chinese import competition are essentially unrelated.

The same number fails in time, with the evidence backed by data. 91 bilateral deterioration episodes were documented with trade and employment data to date each stage at annual resolution. One regularity holds across all of them: the economy never moves first. Not once in 91 events does trade or employment lead the political signal. Whatever markets believe they anticipate, it doesn’t show up here. Politics leads or ties in every classifiable case.

Which means the political warning is real, and it comes early. The problem is that it arrives without the two pieces of information you actually need. That’s what the three questions below are for.

Question 1: Will this become an economic event at all?

Here is the finding worth carrying out of this piece: in 11 of the 91 events, the political rupture is visible, but the measured annual trade and local-employment signals do not become large and persistent enough. Very few workers, fewer than one can measure in any county, loses a job over it. Roughly one in eight is political-only in the measured annual data.

Trying to explain those silent cases: perhaps firms are exposed not because they sell to the partner or compete with its imports, but because they rely on its intermediate inputs. I built an exploratory proxy based on the OECD input-output tables for importer-input dependence and reran the full classification as a diagnostic. Under the proxy, 15 events classify as political-only. Adding a plausible missing mechanism made more ruptures look politically loud and economically quiet. While the proxy has real limits as it’s built from national import shares instead of a true bilateral matrix, that does not prove nothing happened, but it makes the “nothing visible in the measured economy” finding harder to dismiss.

The practical consequence is a rule for standing down, but the timing would have to come from a company’s own data - e.g., orders, cancellations, supplier lead times. In this project’s annual public data, if neither trade nor operational indicators move after a defined monitoring window, do not let news volume alone drive escalation. And when the shock does arrive, it can be enormous, rising Chinese import competition explained about one-quarter of the contemporaneous decline in US manufacturing employment in their 1990-2007 setting. This is exactly why firms need a way to separate alarms that enter the business from alarms that remain mostly political.

Question 2: If it arrives, through which door?

The same rupture enters different businesses three different ways. It can hit a firm as a customer shock, if the affected country is an important market and sales there fall. It can hit as a competitive shock, if firms from that country compete in the same US industry and the terms of that competition change. Or it can hit as a supplier shock, if US production depends on inputs from that country and those inputs become more expensive, slower, or harder to get.

Those three channels are often collapsed into one vague word: exposure. But they do not mean the same thing. A tariff on Chinese steel can help a US steelmaker facing import competition and hurt a US machinery firm buying steel as an input. Calling both firms “China exposed” is technically true but not useful. The question is not just how exposed are we. It is exposed how: as seller, rival, or buyer?

Question 3: How much runway is there?

If the event is real and you know your door, the last question is time.

Tariff and trade-preference actions run long. While one can argue Section 301’s 322 days is the extreme, the pattern is general: median 63 days in our sample, with no same-day cases. These instruments move through public legal stages, each one a dated marker. Burundi’s termination from AGOA ran notification (October 2015) → presidential proclamation (December 2015, per the Federal Register) → effective date (January 2016): a roughly 90-day countdown anyone could read.

Sanctions designations run to zero by design. Three of the five OFAC-style actions in our sample were announced, decided, and effective the same day. There is no runway to use, only preparation done in advance.

So the response has to be matched to the regime:

With 300+ days, structural moves are on the table: re-source suppliers, shift production, renegotiate long-term contracts, exit or enter markets. This is what Section 301 offered, but only to firms that started at the announcement. The runway begins at the first public stage, and our data says waiting is the norm: across the 29 events I studied in detail with the help of Claude, genuine anticipation, measurable movement before any announcement, appears in just 5. In 12, the largest group, nothing detectable happens until the policy is already in force.

With 50–90 days, the median regime, you’re down to financial and logistical moves: hedge, reprice, build inventory, accelerate shipments ahead of the effective date. Burundi’s countdown was exactly this window.

With same-day instruments, the only defense is what you did last year: entity screening at onboarding, sanctions and force-majeure clauses in contracts, dual-sourcing for any input with a plausibly designatable supplier. If your counterparty profile makes designation conceivable, your runway is already zero.

One warning cuts across all three regimes. In 2000, the United States granted China permanent normal trade relations, a policy that changed no tariff rate and merely removed the possibility that tariffs might someday snap back. A widely cited study of that episode links it to the swiftest period of US manufacturing job losses on record. The removal of uncertainty, in either direction, is itself a binding event. When a long-running question gets a definitive answer, treat the answer date as an implementation date even if nothing in the tariff schedule moves.

One more thing closes the loop on all three questions. Public trade and employment data are annual; your orders, cancellations, supplier lead times, and hiring plans are weekly, if not daily or hourly. Inside the calendar year, a company’s own operating metrics are the only instruments fine enough to tell you which question is currently being answered.

Applying it now

Apply the test to the trade measures currently moving through announcement and comment stages.

Question 1: will this become an economic event at all? the political signal is old news. The relationships in question have been deteriorating for years, so the live issue is whether this round produces measurable economic movement or joins the silent minority.

Question 2: if it arrives, through which door? The answer differs by firm. A portfolio-level “exposure score” will not distinguish the steelmaker from the machinery maker.

Question 3: how much runway is there? The instrument is tariff, which historically means a runway measured in months with publicly dated stages. The firms that treated August 2017 as the starting gun were the only ones that got to use all 322 days.

The next time a rupture leads the front page, resist the reflex to ask how bad it is. Ask the three questions in order. Will this ever become economic at all, or is it one of the one-in-eight that never do? If it’s real, which door does it come through? And what instrument is it, because the instrument sets the runway.

The median gap between warning and wall is 52 days. The range is 322 to zero. Where you sit on that range is knowable on day one, and by the time it’s obvious, most of it is gone.

A note on method and sources

This analysis is based on annual public data, so the timing results should be read as year-level evidence. I first identified bilateral geopolitical deterioration episodes from a dyadic geopolitical-score dataset, using the project’s corrected deterioration measure. I then compared the event year with changes in bilateral trade and employment in US county-industry cells exposed to that partner.

Trade data come from US Census Bureau merchandise trade files, matched from HS product codes to NAICS industries using the Pierce-Schott concordance. Employment comes from the Bureau of Labor Statistics QCEW data, measured at the county by NAICS3 industry level. Exposure was measured before each event, using the pre-event years, and kept separate for export specialization and same-industry import competition. The imported-input channel mentioned in the piece is exploratory and uses an OECD input-output proxy.

A trade or employment response is counted only when it is large relative to the event’s own pre-event variation and persists beyond a single annual movement. That means the analysis is trying to identify visible, sustained adjustment rather than every short-lived change in the data. The findings are descriptive timing evidence without establishing a causal estimate of job losses caused by geopolitics.

The policy-date layer was researched separately for the 29 priority events used in the policy-clock analysis with Claude Code. For each policy-linked episode, the file records the first public announcement date, the legal adoption or decision date, and the implementation or effective date where available. Multi-instrument episodes are kept as separate policy rows. Official sources were preferred, including USTR, the Federal Register, Treasury/OFAC, White House records, State Department materials, and similar government releases; major news sources were used where they helped establish the first public date.

Cover photo by Sean Thoman on Unsplash