The global propagation of the US–China trade war

To name a few, see Amiti et al. (2019), Amiti et al. (2020) and Fajgelbaum et al. (2019) for US consumers and welfare and Amiti et al. (2020) and Handley et al. (2020) for investment and exports of US firms.

These countries were selected based on the availability of monthly product-level export data in the UN Comtrade database and the cross-country input-output matrices in the World Input-Output Database (WIOD). We add a few more countries from alternative sources as well. See Sect. 3 for a detail.

In 2016, the US was the single largest export market for China, accounting for 18.4% of its total exports.

Using the episode of US import restrictions in the 1990s, these papers show two channels through which a trade policy distorts the targeted country’s trade with third countries—“trade diversion” and “trade depression”.

For a review of recent empirical work on production networks, see Carvalho and Tahbaz-Salehi (2019).

Early work using the IO tables to study production networks includes Bems et al. (2011) and Acemoglu et al. (2016). Bems et al. (2011) highlight the importance of vertical linkages in the trade collapse during the Great Recession of 2008-2009 based on a global IO framework. Using the detailed US IO tables, Acemoglu et al. (2016) empirically test the theoretical upstream and downstream network effects of demand and productivity shocks on the US economy.

Even before April 2018, both countries raised their tariffs on one another for specific products, such as solar panels, washing machines and aluminium. However, the breadth and trade volumes associated with these tariffs were relatively small. For a detailed timeline of US–China trade disputes, see Fajgelbaum et al. (2019) and Amiti et al. (2019).

More formally, I run a regression for 12-month log difference in US imports from China (\(\Delta {\log imv}^{US}_{i,t}\)) against US tariff changes (\(\Delta \tau ^{US \rightarrow CN}_{i,t}\)) between January 2017 and September 2019, both of which are aggregated at the 87 Chinese SIC industry level (the level of aggregation in this paper):where \(\alpha _{i}\) and \(\eta _{t}\) are industry and time fixed effects, and the standard error in parenthesis is clustered at the industry level. The growth rate of US imports from China drops by 1.179 percentage points in response to a one percentage point rise in US tariffs on China. The size of the coefficient is in line with prior literature (e.g. -1.52 to -1.42 at the HS 10-digit level in Amiti et al. (2019) and Fajgelbaum et al. (2019)).

Related to the fall in exports of capital goods, the trade war may have affected investment for Chinese firms through various channels. Higher tariffs on Chinese imports and a subsequent fall in US demand may have directly lowered expected returns to capital. Moreover, trade policy uncertainty could also stifle investments of Chinese firms (Handley and Limão 2015). As a related research, Amiti et al. (2020) show that US–China tariff actions through 2018 and 2019 significantly lowered the investment growth rate of listed US firms.

The share of each category in the 2017 total exports to China is 67.3% for intermediates, 14.8% for capital goods, 6.6% for consumption goods and 11.3% for others that are classified into multiple categories or unclassified.

See the Appendix Table 6 for the full list of the countries with the values of their exports to China for 2016.

The remaining 46 industries in non-tradable sectors include energy production, construction and a wide range of private and public services industries that are not subject to tariffs. According to the 2012 imported input matrix for China available in the WIOD, 82.0% of imported intermediates into China are used by tradable industries, while the other 18.0% are used by non-tradable industries.

The World IO table is constructed using both product-level trade flows across countries and national IO tables. For more details on WIOD, refer to Timmer et al. (2015).

For matching between the CSIC 5-digit industries and the HS 6-digit products, we exploit the lookup table between the Korean SIC (KSIC) 5-digit industries and the HS 6-digit products provided by Statistics Korea as a bridge since the KSIC 5-digit industries and the CSIC 5-digit industries are very well aligned.

These firms make up around 8% in total number of entities and account for only about 3% in aggregate sales, which implies that they are quite small firms. Adding these outlier firms yields qualitatively same results, with somewhat larger and equally significant estimates overall.

Using a balanced panel of firms would alleviate any selection bias associated with entry or exit. A large number of firms are dropped due to the use of the balanced panel.

For instance, the CSIC-industry “Glass and glass products (30055)” includes intermediates (HS 6-digit of 700100, “Cullet and other waste and scrap of glass; glass in the mass”), consumption goods (701322, “Stemware drinking glasses, other than of glass-ceramics: Of lead crystal”) and capital goods (701322, “Signalling glassware and optical elements of glass”).

There are 93 CSIC industries in the tradable sectors: agriculture, mining and manufacturing. Among these, three industries do not export: “Support service to farming, forestry etc.” (05005), “Mining support activity etc.” (11011), “Slaughtering and processing of meat” (13016), and “Other electronic equipment (39090)”.

This US vertical shock measure is conceptually linked to the “upstreamness measure” by Antràs and Chor (2018), and the “vertical specialization measure” by Hummels et al. (2001) as these all capture the forward linkages in productions across countries. The main difference is that our vertical shock measure attempts to identify a relatively short-run upstream effect of US tariffs that is specific to input suppliers to China, whereas Antràs and Chor (2018) capture the cross-country upstreamness in a general equilibrium setting.

For empirical evidence, see Bernard et al. (2008) and Amiti et al. (2014).

Processing trade was introduced by the Chinese government in the 1980s in an effort to boost their competitiveness in global markets (Yu 2014). Chinese firms import all or part of the raw materials and intermediate inputs, and then re-export the finished products after local processing or assembly. Despite their diminishing share, processing exports still accounted for 37.8% of China’s total exports in 2014 (Kang and Liao 2016).

By the same token, we admit that using the US share of China’s total exports as in this paper would possibly overstate the US tariff effects on Chinese producers.

Analogously, $30 (=100*15/(5+15+10+20)) and $20 (=100*10/(5+15+10+20)) of the imports are assigned to the motor-shipbuilding and battery-auto industry pairs, respectively.

Both the vertical impact and trade depression due to US tariffs predict a negative impact on third country exports to China. But while trade depression could take place in any type of the end-use categories, the vertical propagation should occur in intermediates only.

Chinese government cut its MFN tariffs on other trading partners when imposing retaliatory tariffs against the US in 2018 (Bown et al. 2019).

Since some ISIC 2-digit sectors include only one CSIC-industry, we group these sectors with other similar sectors together. Specifically, we combine “Crop and animal production, hunting and related service activities (A01)”, “Forestry and logging (A02)” and “Fishing and aquaculture (A03”) as a single sector group. We further group “Manufacture of paper and paper products (C17)” and “Printing and reproduction of recorded media (C18)” as “C17-18”, “Manufacture of chemicals and chemical products (C21)” and “Manufacture of basic pharmaceutical products and pharmaceutical preparations (C22)” as “C21-22”, “Manufacture of basic metals (C24)” and “Manufacture of fabricated metal products, except machinery and equipment (C25)” as “C24-C25”, and “Manufacture of motor vehicles, trailers and semi-trailers (C29)” and “Manufacture of other transport equipment (C30)” as “C29-C30”. The results using the initial definition of sectors are largely similar, however. We also test the sector-specific second-degree time polynomials to account for potentially nonlinear aspects of the global business cycle. The results are very similar, with almost none of the coefficients for sectoral trends being significant.

This is to account for the possibility of correlated residuals within sectors as our vertical shock measure exploits the IO elements in the WIOD. Alternative clustering—at the country-industry pair—does not make much difference, though.

Note that Table 1 reports different numbers of observations in regressions for each end-use category. This is because certain end-use products are entirely missing for some country-industry pairs. For instance, there is no capital goods in the CSIC industry ‘Paper and paper products (22036)’ for most countries.

This indicates that the fall in exports of capital goods to China could have been far more severe in the absence of the trade diversion.

When the US vertical shock is scaled by non-intermediate share \((1-\omega ^{CN}_{i,c})\), the coefficient was -0.932 which is insignificant as expected, with the standard error of 0.962.

The absence of market-switching could be due to various frictions on international transactions including upfront entry costs that are likely to be country-specific, as discussed extensively in trade literature.

We confirm that dropping any individual country, sector or industry does not alter the size of the coefficient on the US vertical shock, nor its significance substantially.

In other tariff controls (\(\widetilde{Z_{i}}*D_{t}\)), China’s MFN tariff changes (\(\Delta {Tariff}^{MFN}_{i,t}\)) were not transformed as these would affect third countries’ exports more immediately. Applying the same transformation to the MFN tariffs makes no difference in our results.

One may wonder whether countries with a higher share of intermediates in their exports to China (\(\omega ^{CN}_{i,c}\)), regardless of their exposures to the US vertical shock, experienced a larger fall in their export growth. However, the correlation between the changes in their export growth and their intermediate shares was merely -0.09, and thus, the initial share of intermediate cannot explain much part of the changes in exports.

Once US tariff revenues and other general equilibrium effects are accounted for, they suggest that the losses could have been reduced to 0.04 percent of GDP.

The magnitude of this mechanism could be huge. Alessandria et al. (2010) suggest that inventory adjustments account for up to 20% of the drop in US imports during the trade collapse of 2008-2009. The impact could be largely heterogeneous across industries as well. In their case study on the US auto industry, imports fell by more than twice the sales of imported autos during the same period.

Literature including Handley and Limão (2015) highlights the importance of trade policy uncertainty for trade and investment decisions.

One related hypothesis could be the presence of a global component of fixed export cost. For instance, Mau (2017) argues that firms should pay not only the conventional destination-specific fixed cost but a global (product-specific) fixed cost of exporting which results in economies of scale in serving multiple destinations. This implies that a negative demand shock in one foreign market, particularly a large one (the US), could induce Chinese firms to exit not only from the affected market but also from other destinations due to the increased per destination global fixed cost, which would result in much larger fall in their input demand.

As evidenced by the diagonal elements of most IO tables, a large fraction of inputs produced in each industry are purchased by its own industry. As a study using information other than the IO tables, Handley et al. (2020) exploit the firms’ imports of the same HS 4-digit categories as their export products in evaluating the supply chain effect of the US import tariffs on the US exporters.

Note that these coefficients estimate the effects of a one standard deviation increase in each component of shocks. If we consider the actual changes in each shock component between pre- and post-2018.Q1 periods, the vertical effect from other sectors contributes less than a tenth to the total vertical effect.

The correlation between the original and the inventory-augmented vertical shocks is 0.75 between 2018:Q2-2019:Q3—the period when massive tariff changes occurred.

Defever et al. (2020) show that wholesalers—the key agent holding inventories—play an important role of supplying foreign intermediate inputs and they cover around 21.4% of total intermediate imports for Chinese manufacturing sectors in 2002.

Besides that, the merchandise exports account for over 30% of the Korea’s GDP. This may imply that a large fraction of Korean manufacturing firms engage in exporting—particularly to China as their largest foreign market—and should thus be exposed to the US–China trade war either directly or indirectly.

This magnitude seems somewhat large given that the US vertical effect on Korea’s export growth to China was −3.1 percentage points in the previous section.

Cigna et al. (2022) use monthly product-level US imports from 30 countries between 2016 and 2019 to examine whether US tariffs on China increased US imports from third countries. Similarly to our results, they do not find evidence of a significant trade diversion effect over the time horizon they considered.

The Cobb–Douglas production assumes that country-input varieties are complements. The overall implications of the model remain valid for a more general form of production functions such as in Pakel et al. (2020) under some assumptions when country-input varieties are incomplete substitutes.

For simplicity, we do not consider firm entry into exporting that would incur additional fixed costs.