China’s share of OECD manufactures imports, 2000 → 2016

Autor, Dorn & Hanson’s China Syndrome identifies local labour market effects of rising Chinese import competition in the US, using commuting-zone-level ΔIPW and an instrument built from Chinese exports to other rich economies. We cannot reproduce that research design without commuting-zone employment data. What we can reproduce is the companion cross-country fact that motivates it: how much of each OECD country’s manufactures-import base is now Chinese.

Published result

Autor, Dorn & Hanson (2013, AER) build a commuting-zone (CZ) exposure measure ΔIPW_j,t = Σ_k (L_j,k/L_j) · (ΔM_k,t/L_k), where the outer weight is industry k’s pre-shock employment share in CZ j, and ΔM_k,t/L_k is the change in imports from China to the US in industry k divided by initial US workers in that industry. They instrument US ΔIPW with an other-high-income-country version (Chinese exports to Germany, Japan, etc., per US-industry worker), to purge domestic demand shocks. Their estimate: a $1,000 per-worker rise in Chinese import exposure over 1990–2007 causes roughly a 0.55 percentage-point decline in the manufacturing-employment share of the working-age population in the average CZ. Follow-up work (Acemoglu, Autor, Dorn, Hanson & Price 2016, JoLE) scales this to 2.0–2.4 million manufacturing jobs lost nationally over 1999–2011.

What we replicate — and what we cannot

We are emphatic that this page is not ΔIPW. ΔIPW has workers in the denominator; our measure has import dollars in the denominator. ΔIPW uses CZ-level employment weights; our measure uses country-level import aggregates. ΔIPW leverages an IV from other-rich-country Chinese trade to identify causal effects on US labour markets; we estimate no causal effect and run no regression — we simply compute, for each OECD economy, the share of total manufactures imports that come from China in 2000 and 2016. That is a descriptive exposure metric, analogous in spirit to the raw ADH exposure before instrumentation and before CZ weighting, but nothing more. Anyone using this page to attach an employment-displacement number to any OECD country would be misreading it.

What we can show, BACI-only, is the cross-country pattern the ADH paper points at before building its CZ measure: China’s share of each OECD economy’s manufactures imports roughly tripled in sixteen years, on average.

Annual time path · OECD-38 mean China-share, year-by-year

ADH’s Figure 1 plots the US-only annual import-penetration ratio across 1991-2007 and anchors the WTO-accession date (2001-12-11) as the breakpoint. The cross-OECD analog — mean China-share across the same 38 economies, computed every year from the start of the BACI partitions — is what turns the static 2000 → 2016 endpoint comparison in Figure 1 into a dynamic event study around China’s WTO accession. If the WTO date is the structural break that ADH’s identification leans on, the cross-OECD mean should bend visibly upward in 2002 and continue accelerating until the 2008 financial crisis.

First-stage IV diagnostic (country-level analog)

ADH identify exogenous Chinese-import exposure by instrumenting US-industry imports with Chinese exports to other rich economies (Table 3, Panel A). Their first-stage Kleibergen-Paap Wald F is roughly 154 at the industry level, well above the Stock-Yogo 10-percent critical value of 16.38 (single endogenous regressor). We cannot reproduce that industry-level first stage without a worker-denominator. We can, however, build a country-level analog: for each OECD economy i, regress its own 2000→2016 ΔChina-penetration on the peer mean Δ across the other 37 OECD economies. If peer exposure is a valid proxy for the supply-side shock, the slope should be positive, large, and precisely estimated.

Placebo falsification · non-China emerging exporters

ADH’s identification relies on a China-specific supply shock, dated to the 2001 WTO accession and its post-MFA apparel quota phase-out. A natural falsification is to rerun the same 2000→2016 ΔOECD-penetration calculation with other plausible low-cost exporters in place of China. If the secular ΔOECD-penetration is driven by a generic emerging-markets force (commodity supercycle, global wage convergence, containerisation maturing), India, Vietnam, Brazil and Indonesia should each deliver a mean Δ of comparable magnitude to China’s. If instead the ADH “China shock” framing is correct, China’s Δ should tower over every placebo.

Heterogeneity · low- vs high-baseline exposure OECD

ADH Section IV.E splits their CZ sample by pre-shock manufacturing intensity and shows the employment response concentrated in high-exposure CZs. The cross-country analog: split the OECD-38 at the median 2000 China-penetration ratio into a “low-exposure” bottom half (countries that started the WTO era with very little Chinese content in their manufactures import base) and a “high-exposure” top half. The ADH mechanism is a supply-sideChina shock, so both groups should see level gains in Chinese share; heterogeneity in Δ tells us whether the shock hit lagging or leading economies harder.

Industry-exposure heterogeneity · HS Section decomposition

ADH Table 2 decomposes the aggregate China-shock by industry (apparel, electronics, furniture, footwear) and shows the penetration move is overwhelmingly concentrated in labour-intensive manufactures. BACI’s bilateral flows have no product dimension in this workbench’s partitioning, but country_year_product exposes China’s world export mix by HS6. Aggregated to HS Section, the ΔChina-share of world exports 2000 → 2016 tells us which industries drove the ADH shock: Section XI (textiles) and Section XVI (machinery & electronics) should lead, Sections VI (chemicals) and XVII (vehicles) should lag. A shock that is uniformly distributed across industries could not generate the within-US labour-market reallocation that ADH document.

Numerical comparison

Why it might differ

Four reasons the numbers on this page are not comparable to ADH’s point estimates. First, denominator: ADH uses US manufacturing employment by industry; we use each country’s total manufactures imports. The ratio differs by orders of magnitude and has a different economic meaning — exposure per worker versus share of trade. Second, geography: ADH’s unit of analysis is the US commuting zone, because the labour-market friction of interest operates locally; we aggregate to whole countries, which averages over within-country heterogeneity that is central to the paper’s story. Third, identification: ADH instruments US Chinese exposure with other-country Chinese exposure, to strip domestic-demand endogeneity; we report raw bilateral shares, which conflate China’s supply shock with each destination’s own cyclical demand. Fourth, sample period: ADH’s headline specification covers 1991–2007; we use 2000–2016, which ends after WTO accession (a clean ADH start) and catches the shock at its mature phase.

What the BACI cross-country pattern does reproduce is the direction and broad magnitude of exposure growth: China’s share of OECD manufactures imports rises sharply and monotonically in essentially every OECD economy across the 2000–2016 window. That is the aggregate fact that ADH’s CZ-level regressions translate into a causal employment estimate the BACI data alone cannot deliver.

BibTeX

@article{autor_dorn_hanson_2013,
  author  = {Autor, David H. and Dorn, David and Hanson, Gordon H.},
  title   = {The China Syndrome: Local Labor Market Effects of Import Competition in the United States},
  journal = {American Economic Review},
  volume  = {103},
  number  = {6},
  pages   = {2121--2168},
  year    = {2013},
  doi     = {10.1257/aer.103.6.2121}
}

For the production-side story of the same shock, see /china-shock. For the 2018-tariff reallocation narrative, see the US-China trade war page. Return to the replication gallery.