Tariff Incidence, Supply-Chain Reallocation, and Domestic Propagation

An official-data research system for measuring how the 2018–2019 U.S. Section 301 actions changed importer costs, customs unit values, quantities, sourcing, and downstream exposure.

International tradeApplied econometricsInput-output economics

Research question and relevance

Question: How did U.S. product-level tariffs on imports from China pass through to importers, reshape sourcing, and propagate through domestic input-output linkages?

A tariff can protect an industry's output while raising the cost of its imported inputs. The project therefore reports importer incidence, sourcing changes, output protection, and input-cost exposure as separate channels rather than collapsing them into one net number.

The statutory setting covers all four Section 301 actions in the research window. New tariff episodes enter through configuration, while announcement dates and effective dates remain distinct throughout the pipeline.

Data, sample, and provenance

The official panel contains 923,440 HS10 × country × month observations from January 2017 through February 2020. Inputs trace to the U.S. Census Bureau, Federal Register and USTR, the U.S. International Trade Commission, the Bureau of Economic Analysis, and the Bureau of Labor Statistics.

LayerPublication treatment
Source, tests, configuration, manifests, and narrative reportsPublished on GitHub and mirrored in the Dataset.
Official legal excerpts used as test fixturesPublished with source identification and parser validation.
Large raw, staged, normalized, and analytical filesExcluded from GitHub; reconstructed from documented official sources.
Synthetic estimator validationRetained and labeled SYNTHETIC_PIPELINE_VALIDATION; never used for empirical claims.

Every result carries a run stamp recording the data period, configuration hash, Git commit, and provenance. Missing values are not silently converted to zero, and customs unit values are never described as transaction prices.

Methodology and evidence licensing

  1. Parse each operative Federal Register annex and reconcile the extracted tariff lines against the count stated in the notice.
  2. Build a point-in-time HS10 tariff schedule with separate announcement and effective dates, including day-weighting for mid-month changes.
  3. Construct the country-product-month panel and preserve customs, duty, freight, quantity, and concentration concepts separately.
  4. Estimate one stacked sub-experiment per tariff wave, using never-treated products as controls so already-treated units do not contaminate later waves.
  5. License interpretation outcome by outcome using pre-trend, date-placebo, leave-one-out, and window-sensitivity diagnostics.
  6. Map product exposure through BEA input-output tables without netting output protection against imported-input costs.

Structural counterfactuals are labeled model-implied. No welfare number is produced because the project does not identify a domestic substitution nest.

What the current official-data run supports

ResultEvidence statusInterpretation
Section 301 schedule: 818, 279, and 5,745 stated lines reconciled exactlyValidated legal parsingChapter 98/99 machinery, partial statutory lines, and one derived truncated code are handled explicitly.
Customs unit value: +0.023, bounded within 0.072 in absolute valuePrecise near-null; date placebo passesNo detectable exporter absorption over the research window. This is the behavioral evidence behind the pass-through conclusion.
Duty-inclusive landed unit value: +0.149Clean stacked designClose to the mechanical log increase implied by the 15.3% value-weighted additional duty; the measure contains the duty by construction and is not independent pass-through evidence.
Quantity: −0.350QualifiedThe pre-period is noisy and the date placebo fails, so the estimate is not given an unqualified causal reading.
Adjusted third-country replacement ratio: 0.11Qualified sourcing counterfactualRoughly a ninth of the estimated China shortfall reappeared from sampled alternative suppliers; chapter, partner, and counterfactual limits remain material.
144 of 402 BEA detail industries exposed on both channelsDescriptive exposureOutput protection and imported-input costs coexist and are reported separately.

Reproduction and citation

Run make setup and make reproduce-sample. The official trade-data build additionally requires a user-supplied Census API key. The repository requires Python 3.12 and records source access, data definitions, design decisions, failed hypotheses, and known limitations alongside the code.

Cite the repository release, pinned Dataset revision, and every original government source used in a reproduced run.

Limitations

  • Product exclusions are intention-to-treat because exclusion grants are finer than published import statistics; the observable gap is bounded rather than guessed away.
  • The quantity outcome fails the date placebo and requires qualified interpretation.
  • The sourcing analysis covers 10 chapters and eight alternative partners; customs data cannot distinguish relocated production from rerouting or origin misdeclaration.
  • The USITC baseline HTS source is a later vintage than the 2018 tariff lists.
  • The structural model covers sourcing across foreign suppliers only and produces no welfare estimate.