Non-Tariff Barriers in the U.S.-China Trade War

Layer 1: Overview

Chen, Hsieh, and Song study the use of unofficial non-tariff barriers (NTBs) by China during the U.S.-China trade war of 2018–2019 and in the first year of the Phase 1 purchase agreement (2020). The central motivation is that much prior analysis of the trade war focused on announced tariff hikes, yet abundant anecdotal evidence — permit requirements for U.S. pet food, pest-inspection orders on U.S. apples and lumber, changes to pig-feed formulas reducing soybean content — points to a parallel, opaque regulatory channel. The critical puzzle the paper highlights is that China’s purchases of U.S. goods rose by 156 percent between 2019 and 2020 without any reduction in tariffs, which is only explicable if NTBs were used in reverse to favour U.S. exporters during the Phase 1 period.

The paper uses Chinese customs administrative data from 2015 to July 2020, covering 946 HS-6 products aggregated by state-owned versus non-state importer and by source country. Tariff data are constructed from official Customs Tariff Commission documents listing each round of retaliatory hikes beginning April 2018. The empirical strategy proceeds in three steps. First, demand (elasticity of substitution across source countries, epsilon) and supply (gamma) elasticities are estimated by regressing changes in import quantities and CIF prices on changes in tariff rates, using product-country fixed effects so identification comes from within-product, cross-country variation in tariff changes. The identifying assumption — that tariff changes across countries are orthogonal to NTB changes and foreign supply shifts — is validated empirically. The estimated demand elasticity is epsilon = 3.36 for agriculture and 2.34 for manufacturing; supply elasticities of 42 (agriculture) and 71 (manufacturing) imply near-horizontal foreign supply curves, so essentially all the incidence of Chinese trade barriers falls on Chinese consumers.

Second, NTBs are inferred as a residual: the change in U.S. import quantities relative to imports from other countries of the same HS-6 product, after netting out the estimated price and tariff effect. A normalisation sets the import-weighted average NTB change on non-U.S. source countries to zero, so the residual is attributed to U.S.-specific barriers. This procedure is run separately for non-state and state importers. The tariff-equivalent of NTBs on U.S. agricultural products faced by non-state importers rose by 0.73 log points between 2017 and 2019, while NTBs on state importers were essentially unchanged (Table 4). The weighted average NTB increase for agriculture was 0.60 log points, compared to a tariff increase of 17 percentage points (from 7.5% to 24.5%). For manufactured goods, average NTBs rose by only 0.16 log points versus a tariff increase of 9 percentage points (5.6% to 14.6%). NTBs were highly concentrated: the tariff equivalent rose by 1.0 log points for oil seeds, 1.5 log points for cereals, and 1.1 log points for ores, slag and ash. The variance of tariff-adjusted import growth across HS-6 products increased 18-fold from 0.296 (2015–2017) to 5.31 (2017–2019), and controlling for state versus non-state ownership accounts for 38% of that increase.

Third, welfare effects are computed using a three-nest CES model (HS-6 products, importer firms, source countries). Tariffs harm welfare via dispersion of tariff rates across source countries; NTBs harm welfare via both the mean and dispersion of NTBs across source countries, firm types, and products, and also because — unlike tariffs — NTBs generate no fiscal revenue. The total welfare loss to China in 2019 relative to 2017 is estimated at $40 billion, of which 92% is attributable to NTBs rather than tariffs (Table 7). For agricultural products alone, NTBs account for 86% of the $12.7 billion welfare loss; for manufacturing they account for 94.1% of the $27.2 billion loss. Crucially, for a given dollar reduction in U.S. imports, NTBs impose approximately six times the welfare cost of equivalent tariff hikes (the Figure 2 text says “five times”), because NTBs (i) generate no revenue and (ii) create misallocation by applying to some importers (non-state) but not others (state-owned). By 2020 China’s welfare loss relative to 2017 widened further to $48.11 billion, as NTB reversals in agriculture were partial and manufacturing NTBs were not reversed at all. The paper also documents that the Chinese government’s choice of instrument was strategic: tariff hikes were smaller in sectors with a larger pre-war state importer share, while NTB hikes on non-state importers were larger in those same sectors, consistent with a government pursuing dual objectives of punishing U.S. exporters while protecting state-firm profits.

Layer 2: Deep Dive

What is the core identification strategy and its key assumption?

The demand elasticity (epsilon) and supply elasticity (gamma) are estimated from a system of two equations: the change in log import quantity and the change in log CIF price, both regressed on the change in log tariff rates, with product-country fixed effects and year fixed effects. The identifying assumption is that tariff changes across source countries are orthogonal to NTB changes and foreign supply shifts — i.e., China’s retaliatory tariff schedule was not systematically targeted at products where NTBs were also rising or where foreign supply conditions were deteriorating. The authors validate this assumption in two ways: (1) Appendix Figure A2 shows near-zero correlation between imputed NTB changes and tariff changes across HS-6 product-country pairs (OLS coefficient 0.014); (2) Appendix Figure A3 shows near-zero correlation between pre-war import growth (2015–2017) and post-war tariff changes (OLS coefficient -0.02), arguing against correlated foreign supply trends.

How exactly are NTBs measured and what normalization is required?

NTBs are inferred as a structural residual. From the CES demand function, the change in non-state imports of a U.S. product relative to the same product from another source country equals minus epsilon times the relative change in tariff-inclusive CIF price, minus epsilon times the relative NTB. Given estimated epsilon and data on prices and tariffs, the relative NTB (U.S. vs. other countries) is identified. To convert this into the absolute NTB on U.S. goods, the paper normalizes the import-expenditure-weighted average NTB change on all non-U.S. source countries to zero. State-importer NTBs are then backed out from the ratio of state to non-state import growth for U.S. products, using equation (7), which relies on the elasticity of substitution between state and non-state firm types (eta = 3, borrowed from Khandelwal, Schott and Wei 2013).

What are the main threats to identification and how are they addressed?

Three threats are discussed. (1) Quality or supply changes specific to U.S. products: if imputed NTBs reflect deteriorating U.S. product quality rather than Chinese regulatory barriers, U.S. exports to non-China markets should also fall for the same HS-6 products. Appendix Figure A1 shows no such correlation (OLS slope 0.016, SE 0.007), confirming NTBs are China-specific. (2) Endogenous targeting of tariffs toward products also receiving NTBs (violating the orthogonality assumption): Appendix Figure A2 directly shows near-zero correlation. (3) Correlated pre-trends: Appendix Figure A3 shows no correlation between 2015–2017 import growth and 2017–2019 tariff changes, so pre-existing trends do not appear to have driven the targeting of tariffs.

What heterogeneity across firm ownership is documented?

NTBs fell almost entirely on non-state importers of U.S. agricultural products. Non-state NTBs rose by 0.73 log points (2017–2019) while state NTBs were essentially unchanged (Table 4, column 3 vs. column 4). The state share of Chinese agricultural imports from the U.S. roughly doubled from 19.3% in 2017 to 39.8% in 2019 (Table 2), before returning to ~20% in 2020. For imports from the rest of the world, the state share remained stable at ~20% throughout. In manufacturing, state-importer NTBs declined slightly (-0.066) while non-state NTBs rose modestly (0.023). The divergence between state and non-state importers accounts for 38% of the 18-fold increase in variance of tariff-adjusted import growth.

What product-level heterogeneity is found in the use of NTBs vs. tariffs?

NTBs were highly product-concentrated compared to tariffs. Table 5 shows the largest NTB increases in oil seeds (+1.006 log points), cereals (+1.492), and food industry residues (+0.688), all products where the U.S. held large pre-war import shares. For manufactured goods, the largest NTB increases occurred in ores, slag and ash (+1.106) and vehicles (+0.366). By contrast, tariff hikes were distributed more broadly across products. Table 9 shows that, across HS-6 products, (a) tariff increases were significantly smaller for products with a higher pre-war state importer share (OLS coefficient -0.202) and (b) non-state importer NTB increases were significantly larger for those same products (OLS coefficient +4.431). Both patterns hold when controlling for the U.S. import share in total imports of the product.

What is the welfare framework and what are its scope conditions?

Welfare is derived from a three-level CES utility function over HS-6 products (elasticity sigma), importer firms (elasticity eta), and source countries (elasticity epsilon). Tariff revenue is rebated to consumers; NTB costs are not. The welfare cost operates through three channels: (1) tariffs raise dispersion of prices across source countries, reducing welfare with elasticity epsilon; (2) NTBs affect both the mean and the dispersion of import prices, with no offsetting revenue effect; (3) differential NTBs across firm types (state vs. non-state) add a misallocation channel scaled by eta. The framework accounts for expenditure reallocation across source countries within an HS-6 product and across HS-6 products, but not between imported and domestic Chinese goods. This last restriction means welfare losses are likely understated, as the model does not capture the cost of switching from foreign to domestic substitutes.

What are the quantitative welfare results and how do they decompose?

Total welfare loss in 2019 relative to 2017: $40 billion. Agriculture: $12.7 billion (of which tariffs account for $1.7B and average NTBs for an additional $9.3B; differential state/non-state NTBs add a further $1.7B). Manufacturing: $27.2 billion (of which tariffs account for only $1.6B; average NTBs add $23.5B and differential NTBs a further $2.1B). NTBs’ share: 92% of total (86% for agriculture, 94% for manufacturing). By 2020, the overall welfare loss widened to $48.11 billion, because partial NTB reversal in agriculture was more than offset by continued welfare losses from manufacturing NTBs.

Why are NTBs so much more costly per dollar of import reduction than tariffs?

Two mechanisms. First, tariffs generate revenue that is assumed to be rebated to consumers, partially offsetting their welfare cost; NTBs generate no government revenue. Second, because NTBs are unofficial and opaque, they can be and were applied selectively to non-state importers but not to state importers, creating misallocation: within an HS-6 product, some importers face artificially high effective prices while others (state firms) do not, so the aggregate consumption basket becomes inefficient. The welfare elasticity with respect to import value is approximately five to six times larger for NTBs than for tariffs (Figure 2; the abstract states six times, the Figure 2 text states five times — a minor internal discrepancy).

What does the paper show about the Phase 1 purchase agreement (2020)?

In 2020 China agreed to increase purchases of U.S. goods without reducing tariffs. The paper shows this was accomplished by partially reversing NTBs. The average NTB for agricultural products fell from +0.60 log points (2017–2019) to +0.14 log points over the full 2017–2020 period, implying substantial 2020 reversal. This reversal applied exclusively to non-state importer NTBs on agricultural products; state importer NTBs and manufacturing NTBs were not reversed. The U.S. share of Chinese agricultural imports rose from 13.7% in 2019 to 17.2% in 2020 despite unchanged tariffs (Table 1), directly confirming the NTB reversal interpretation. Welfare in 2020 from agricultural imports partly recovered but remained $7.3 billion below 2017 baseline; manufacturing welfare loss persisted, yielding an overall 2020 welfare loss of $48.11 billion.

How does this paper relate to prior work on the U.S.-China trade war?

The paper builds most directly on Fajgelbaum et al. (2019), borrowing their IV procedure to estimate demand and supply elasticities (using tariff variation across source countries as instruments) and replicating their finding of near-horizontal foreign supply curves. It differs in focusing on Chinese consumers rather than American consumers and in measuring NTBs in addition to tariffs. It also extends Khandelwal, Schott and Wei (2013), whose analysis of state-firm export quotas motivated the state/non-state ownership dimension; the current paper inverts the logic to study selective barriers on non-state importers. Benguria and Safdie (2021) similarly find product variation in U.S. exports to China correlated with state ownership, but do not impute NTBs structurally or quantify welfare. Ma, Ning and Xu (2021) and Liu (2020) use Chinese customs data to document tariff effects on imports but do not examine NTBs. Chor and Li (2021) use night-lights data to estimate aggregate tariff exposure effects.

What robustness checks are conducted and what do they show?

Three main robustness exercises. (1) Falsification test: for products where high NTBs are imputed, U.S. exports to non-China markets do not fall (Appendix Figure A1, slope 0.016, SE 0.007), confirming NTBs are China-specific rather than reflecting U.S.-side supply deterioration. (2) Orthogonality check: Appendix Figure A2 shows near-zero correlation between imputed NTBs and tariff changes across product-country pairs. (3) Alternative country normalization: NTBs are estimated for the four largest non-U.S. exporters to China (Brazil, Canada, Thailand, Australia), assuming barriers on the remaining countries average zero. Brazil, Canada, and Thailand show essentially zero imputed NTB changes 2017–2019, consistent with the identifying normalization. Australia shows a modest NTB increase consistent with documented retaliations after Australia’s 2018 national security law, but far smaller than the U.S. NTB increase. Additionally, Appendix Tables A1-A3 re-run all estimates with alternative parameter values: sigma = 1 (instead of 1.47/1.25) and eta = 5 (instead of 3). All qualitative results survive: NTBs exceed tariffs in magnitude, fall disproportionately on non-state importers, and impose far larger welfare costs per dollar of import reduction.

What are the policy implications and their scope conditions?

The main policy implication is that opaque regulatory tools are an unusually costly instrument of trade retaliation — approximately five to six times more costly per unit of import reduction than equivalent tariffs — because they neither generate revenue nor require the same importer to bear equal costs. If the Chinese government’s objective was to punish U.S. exporters, it chose a particularly self-damaging instrument. A secondary implication concerns the Phase 1 deal: the deal’s purchase commitments were met not through tariff reductions but through NTB reversals, and those reversals were partial, selective (agriculture but not manufacturing; non-state but not state), and left China’s welfare substantially below the 2017 baseline. Scope conditions: the welfare model does not account for import-to-domestic substitution, so welfare costs are likely understated. The elasticity estimates assume CES preferences and a particular nesting structure. The NTB measurement relies on the normalisation that average barriers on non-U.S. sources did not change, which is validated but not directly observable.

What does the paper reveal about the strategic logic of China’s instrument choice?

Section 7 shows that Chinese authorities’ instrument choice is consistent with a dual-objective government: punish U.S. exporters while protecting state-firm profits. Tariffs, which apply uniformly to all importers, harm state firms importing from the U.S. as much as non-state firms. NTBs, being unofficial and selectively enforced, can exempt state importers. Regression evidence (Table 9) confirms: tariff hikes were systematically smaller for products with higher pre-war state importer shares (coefficient -0.202, SE 0.042), while NTB hikes on non-state importers were systematically larger for the same products (coefficient +4.431, SE 0.655). These patterns hold controlling for the U.S. product share in total Chinese imports.

Key Concepts

Non-tariff barrier (NTB): In this paper, unofficial and opaque regulatory measures — health inspections, permit requirements, informal directives to importers — that function as trade barriers but are not publicly disclosed as such and are not uniformly applied to all importing firms. Measured in tariff-equivalent units as the residual change in U.S. import share after controlling for tariff and price effects.

Tariff-equivalent of NTBs: The ad-valorem tariff rate that would produce the same reduction in import demand as the estimated NTB, derived from the structural demand equation. Expressed in log points (e.g., 0.60 log points for average agricultural NTBs in 2017–2019).

Misallocation from selective NTBs: The welfare loss that arises specifically because NTBs are applied to non-state importers but not state importers within the same HS-6 product category. This within-product dispersion of effective prices across firms generates an allocative inefficiency absent when tariffs are used, since tariffs apply uniformly.

Phase 1 purchase agreement: The January 2020 U.S.-China trade deal in which China committed to purchasing specified amounts of U.S. goods in 2020–2021. The paper shows that China fulfilled these commitments by reversing NTBs rather than reducing tariffs, and that the reversal was partial, concentrated in agricultural imports by non-state firms.

Elasticity of substitution across source countries (epsilon): The parameter governing how sensitive Chinese import demand for an HS-6 product from a given country is to that country’s relative price. Estimated at 3.36 for agriculture and 2.34 for manufacturing using tariff variation as an instrument.

State vs. non-state importer: The ownership classification of Chinese importing firms in the customs data. State-owned importers were largely exempt from NTBs during the trade war, while non-state (private) importers bore nearly all of the NTB increases on U.S. agricultural products. This differential application is the central mechanism generating misallocation.

Welfare channel distinction: tariffs vs. NTBs: Tariffs affect welfare only through the dispersion of prices across source countries (revenue is rebated). NTBs affect welfare through both the mean and dispersion of prices across source countries, firm types, and products, with no revenue offset. This structural distinction is why the paper finds NTBs impose approximately five to six times greater welfare cost per dollar of import reduction.