Empirical trade / master's thesis

When do critical-mineral export controls bind?

China's licensing regimes, supplier switching, and import pressure.

Mario Diez Martínez · June 2026 · In progress

Do importers replace Chinese supply after a licensing control, reduce total imports, or pay a segmented price?

A legal announcement is not automatically a binding trade restriction. Product coverage, licensing practice, Chinese market power, and outside replacement capacity all shape the response.

The thesis builds mineral-specific evidence rather than forcing graphite, antimony, bismuth, tungsten, and related products into one pooled effect. The estimates below are preliminary and diagnostic.

01 / Baseline

A source-specific trade response

The baseline PPML specification compares bilateral imports from China and other suppliers around the relevant control date while absorbing importer-product-source, importer-time, product-time, and source-time shocks.

E [M_{i p s t} | X] = \exp {β D_{p s t} + α_{i p s} + δ_{i t} + λ_{p t} + μ_{s t}}

(1)

D identifies a controlled product imported from China after the policy takes effect. Separate models then test whether non-China supply rises, total imports compress, or China and outside suppliers face different unit-value changes.

02 / Framework

Exposure is not the same as pressure

The thesis starts from importer-product exposure to China, then asks how much of that exposed demand can plausibly move elsewhere. Pre-policy rest-of-world sourcing captures revealed importer access; product-level production outside China captures external availability.

Exposure

E_{i p} = \frac{M_{i p, pre}^{C}}{M_{i p, pre}^{C} + M_{i p, pre}^{R}}

Replacement feasibility

K_{i p} = O_{i p}^{cap} \times A_{p}^{prod}

Switching pressure

P_{i p}^{S} = E_{i p} \times K_{i p}

Compression pressure

P_{i p}^{C} = E_{i p} \times (1 - K_{i p})

The identity $P^{S} + P^{C} = E$ splits China exposure into two possible adjustment margins. These are diagnostic gradients, not independent structural parameters.

01Legal coverageProduct scope and implementation date

02Pre-policy exposureImporter dependence on Chinese supply

03Replacement feasibilityOutside access × external production

04ASwitchingChina down · rest of world up

04BCompressionTotal imports down or price wedge up

Figure 01 The theoretical route from a licensing rule to observable trade margins. Legal treatment alone is insufficient: the same control can produce switching, compression, price segmentation, or no detected HS-level effect.

03 / Design

Four pieces of evidence

01
Audit legal scope and map controlled material to observable HS products.
02
Measure pre-policy dependence on China and non-China replacement capacity.
03
Estimate China-source, rest-of-world, and total-import responses separately.
04
Use event-time and price-wedge estimates as diagnostics, with product-level pre-trend checks.

04 / Preliminary results

No single mineral story

Product	China source	ROW switching	Total imports	Price wedge
Natural graphite	-30.1*	+34.2***	-16.5**	-6.7*
Antimony metal	-86.7***	+40.5***	+28.7	+46.0***
Bismuth	-74.4***	+19.6	-13.4	+83.7***
Tungstates	-15.2	+16.1	+44.1***	-11.5**

Approximate percentage effects from product-specific PPML and price specifications. Stars: * p<0.10, ** p<0.05, *** p<0.01. Samples and identifying variation differ across columns, so entries should not be added mechanically.

Indexed import values for antimony oxides from China, the rest of the world, and all sources around the October 2024 import-effect date. — Figure 02 Descriptive sourcing for HS 282580, antimony oxides. China-source imports weaken while rest-of-world sourcing rises around the October 2024 import-effect date. The HS code includes material outside the controlled purity threshold, so this is reallocation evidence rather than a clean causal estimate.

05 / Reading

Bindingness is product-specific

Large China-source declines for antimony metal and bismuth coexist with different adjustment margins. Antimony shows strong outside-supplier growth and a positive price wedge; graphite combines source switching with total-import compression. Tungstates do not show the same China-source decline.

Pooled event studies currently fail some pre-trend tests, and broad HS codes can mix controlled and uncontrolled material. The thesis therefore treats timing plots as diagnostics and makes its strongest claims at the product level.

06 / Tools

Tools and sources

PythonpandasNumPyPyFixeststatsmodelsPPMLUN Comtrade APICEPII GeoDistEurostatBloombergWindLaTeX

The repository contains code, documentation, selected aggregates, and explicit replication boundaries. Raw trade panels and licensed Bloomberg or Wind data are not redistributed.