China Trade Shock

Definition

The China trade shock refers to the large, rapid increase in U.S. imports from China following China's market liberalization and World Trade Organization (WTO) accession (formally 2001, but accelerating through the 1990s). China's share of world manufacturing exports rose from 2.3% in 1991 to 18.8% in 2013; its share of world manufacturing value added rose from 4.1% to 24.0%. As an empirical tool, it is exploited by Autor, Dorn, and Hanson (2013a, 2016, 2019) as an instrument for local labor demand shocks: commuting zones (CZs) with heavier employment in import-competing manufacturing industries experienced larger, plausibly exogenous job losses, enabling causal identification of trade's effects on labor markets and downstream social outcomes.

Key Ideas

Chinese import penetration in the U.S. rose ~0.95 pp 1990–2000 and a further ~1.15 pp per decade 2000–2014; 60% of the rise accrued to male-intensive industries.
Variation across the 722 mainland U.S. CZs comes from differences in initial industry mix (1990), not from anything specific to a CZ's outcomes — the standard Bartik logic.
The instrument for endogeneity uses contemporaneous Chinese export growth to 8 other high-income countries (Australia, Denmark, Finland, Germany, Japan, New Zealand, Spain, Switzerland): cross-industry correlation = 0.92; each comparison country saw import growth in 343–385 of 397 four-digit industries.
China's trade surge was internally driven — Mao-era suppression kept China far inside its production frontier (GDP rank fell from 59th to 134th in Penn World Tables 1952–1978); Deng's reforms and WTO accession (privatization + direct trading rights for private firms) unlocked latent comparative advantage. This internal origin validates the natural experiment.
Gender-specific shocks are constructed by weighting industry-level penetration by each industry's initial male (or female) employment share within the CZ, producing male-shock and female-shock measures ( $\rho = 0.80$ ) with independent effects on corresponding gender outcomes.

How It Works

The Bartik Measure

For each CZ $i$ and period $\tau$ :

$\Delta IP_{i\tau} = \sum_j \frac{L_{ij,90}}{L_{i,90}} \cdot \Delta IP_{j\tau}^{cu}$

where $\Delta IP_{j\tau}^{cu}$ is the growth in Chinese import penetration in U.S. industry $j$ (imports / initial absorption) and $L_{ij,90}/L_{i,90}$ is industry $j$ 's share of CZ $i$ 's 1990 employment.

The Instrument

Chinese imports to 8 comparison countries replace U.S. imports, and employment shares are lagged 10 years:

$\Delta IP_{i\tau}^{co} = \sum_j \frac{L_{ij,80}}{L_{i,80}} \cdot \Delta IP_{j\tau}^{co}$

The exclusion restriction: what drives the common component of Chinese export growth across high-income countries is Chinese productivity and trade cost improvements — not correlated demand shocks specific to U.S. CZs.

Formal Gravity Framework (ADH 2016, Eqs. 1–3)

If trade has a gravity structure (Eaton-Kortum 2002), total demand by the US economy for traded output from region $i$ is:

$X_i = \sum_k \frac{A_{ik}\tau_{ik}^{-\theta}}{\Phi_k} E_k \tag{1}$

Totally differentiating and using $\hat{x} = dx/x$ , the log change in region $i$ 's output is:

$\hat{X}_i = \sum_k \phi_{ik}\hat{E}_k - \theta\hat{w}_i + \sum_k \phi_{ik}\hat{A}_k - \sum_k \phi_{ik}\sum_{i' \neq c}\rho_{i'k}\hat{A}_{i'k} - \sum_k \phi_{ik}\rho_{ck}\hat{A}_{ck} \tag{2}$

where $\phi_{ik} = X_{ik}/X_i$ is industry $k$ 's share of region $i$ 's total sales. The rightmost term captures the China shock treatment and can be rewritten as:

$\sum_k \phi_{ik}\rho_{ck}\hat{A}_{ck} = \sum_k \phi_{ik}\left[\frac{X_{ck}\hat{A}_{ck}}{E_k}\right] \tag{3}$

This is the weighted average exposure of region $i$ to import penetration mandated by changes in China's productive capacity. Equation 2 is the reduced-form specification; the $\phi_{ik}$ weights (industry share of CZ sales) summarize pre-shock specialization and capture variation in regional exposure to China's supply-driven export growth.

Gender Decomposition

The overall measure is split into male-industry and female-industry subcomponents by multiplying the Bartik weight by the initial female (or male) employment share within each industry × CZ cell. This produces instruments that affect male and female employment in their respective sectors, enabling clean gender-comparative Instrumental Variables (IV).

Empirical Evidence

Industry-Level Impacts (Acemoglu et al. 2016, Tables 2–3)

Period	Avg annual import exposure (pp)	Avg annual log Δ employment
1991–2011	0.50	−2.71
1991–1999	0.27	−0.30
1999–2007	0.84	−3.62
2007–2011	0.30	−5.73

Two-Stage Least Squares (2SLS) estimate (1991–2011): 1 pp import penetration → −1.30 log pts manufacturing employment (t = 3.2). Ordinary Least Squares (OLS) = −0.81 (downward bias from domestic demand contamination).

Total job losses 1999–2011: 560K direct manufacturing, 985K with input-output linkages, 2.0–2.4M total economy (including CZ demand multiplier).

CZ-Level Impacts (Autor et al. 2013a; Table 4)

Per $1,000 increase in CZ import exposure per worker:

Outcome	Coefficient	SE
Δ Employed in manufacturing (pp)	−0.60***	(0.10)
Δ Employed in non-manufacturing (pp)	−0.18	(0.14)
Δ Unemployed (pp)	+0.22***	(0.06)
Δ Not in Labor Force (NILF) (pp)	+0.55***	(0.15)
Δ Log CZ population	−0.05	(0.75)
Δ Average log weekly wage (log pts)	−0.76***	(0.25)
Δ Annual income per adult (US $) \| −$ 549***	(169)
Δ Transfers per capita (US $) \| +$ 57.7***	(18.4)

No offsetting reallocation. Overall CZ employment-to-population falls at least one-for-one with the manufacturing decline. No significant employment gains in unexposed sectors over a decade.

Transfer Cascade (Figure 7)

Per $1,000 of import exposure, annual government transfer response per capita in CZs (1990–2007):

Transfer category	$ change
Unemployment Insurance + Trade Adjustment Assistance (TAA)	$3.65
Social Security Administration (SSA) disability benefits (SSDI)	$8.40
SSA retirement benefits	$10.00
Other income assistance	$15.04
Government medical (Medicaid/Medicare)	$18.27
Total	$57.73

TAA — the designated federal trade adjustment program — is the smallest category. SSDI absorbs more than twice as much. The $549 income loss is offset by only ~$ 58 in transfers (10%).

Worker-Level Heterogeneity (Autor, Dorn, Hanson, Song 2014)

Using SSA longitudinal earnings records matched to import penetration by 1991 industry:

Workers in trade-exposed 1991 industries accumulate substantially lower earnings 1992–2007, with more job churning and more years on SSDI.
Industry-exposure persistence (Figure 8): Correlation between 1991 industry trade exposure and current-employer trade exposure = 0.43 in 2007; counterfactual (if workers always moved to unexposed industries on first separation) = 0.17. Low-wage workers remain trapped in high-exposure industries even after separating from initial employers.
Wage tercile heterogeneity: Top-tercile workers relocate outside manufacturing with no earnings loss. Bottom-tercile workers relocate within manufacturing, remain in subsequently shocked industries, suffer large lifetime earnings losses.

Why It Matters

Causal Identification of Trade Effects

Before ADH (2013a), the literature struggled to separate trade-driven manufacturing decline from concurrent forces (automation, domestic demand changes). The China shock is large, geographically concentrated by pre-existing industry mix, and plausibly supply-driven — making it a near-ideal instrument. ADH 2013a showed it caused substantial U.S. employment losses; subsequent work extended it to earnings inequality, family structure, mortality, crime, and Disability Insurance (DI) enrollment.

The Pre-2000 Consensus and Its Failure (ADH 2016)

The consensus ca. 2000 rested on three claims, all refuted by the China shock evidence:

Trade had not been a major cause of manufacturing decline or rising inequality — Rejected. ADH 2013a shows large, significant effects; Acemoglu et al. 2016 extends to 2.0–2.4M total jobs.
Workers in trade-impacted regions could readily relocate — Rejected. CZ population barely moves; employment rates remain depressed for 10+ years; individuals tracked in SSA data do not exit the traded sector cleanly.
Any labor-market impacts of trade would diffuse nationally through factor prices (law of one price for skill) — Rejected. Wage effects are concentrated in bottom four deciles and are geographically local, not national; spatial mobility is insufficient to equilibrate wages.

A further mechanism explaining the slow reallocation: the US ran a trade deficit with China throughout the period (China's current account surplus averaged 4.8% of GDP in the 2000s, the US deficit averaged 4.4% of GDP). Under balanced trade, workers displaced from exposed tradables would reallocate to unexposed tradables. With imbalanced trade, they instead flow into nontradables or out of the labor force entirely — delaying the standard reallocation mechanism indefinitely.

Transfer Programs as De Facto Trade Adjustment

The Figure 7 transfer cascade reveals that the US government's response to trade displacement operates almost entirely through programs designed for other purposes: SSDI, Medicaid, income assistance, and early retirement. TAA — the nominally designated instrument — is negligible. This matters for DI growth: exposed workers use SSDI as permanent income replacement, not as a response to health shocks. See Conditional DI Applicants and DI Growth Decomposition.

Welfare Assessment

Short-run (Caliendo et al. 2015 dynamic General Equilibrium (GE)): Net welfare gains ≈ 0 immediately post-shock; ultimate gains realized only once workers reallocate across regions.
Geographic dispersion (Galle et al. 2015): Geographic standard deviation (SD) of welfare gains = 2× national mean; distributional concerns dominate aggregate accounting.
Trade gains exist long-run — China's growth lifted hundreds of millions of Chinese citizens — but the distribution within the US is deeply unequal and adjustment is far slower than standard models predict.

Connection to Conditional DI Applicants

The Autor-Duggan (2003) framework identified declining low-skill wages as the structural driver of DI growth. The China trade shock is the specific, measurable demand-side force that depressed those wages and displaced those workers. The Figure 7 transfer cascade directly confirms this: SSDI take-up increases monotonically with CZ-level import exposure. The China shock thus provides the causal upstream identification for the Conditional DI Applicants mechanism.

Connection to Deaths of Despair

ADH 2019 provides the strongest causal evidence that manufacturing job loss causes the deaths-of-despair mortality cluster. A unit male-industry trade shock raises male mortality from drug and alcohol (D&A) poisoning +60.3, HIV/AIDS +66.5, and homicide +103.0 per 100,000 per decade (gender-specific estimates). Case and Deaton (2015, 2017) documented the pattern; ADH 2019 provides causal identification of the mechanism they hypothesized.

Limitation: Absolute vs. Relative

Trade shocks reduce both male absolute earnings and male earnings relative to women simultaneously, so the ADH design cannot cleanly distinguish Becker (relative economic stature) from Wilson (absolute economic stature) as the mechanism through which marriage and fertility fall.

Open Questions

Would reversing trade pressure (e.g., tariffs, reshoring) undo the family structure effects, or has cumulative disadvantage created persistent equilibria?
How much of the DI application surge in manufacturing-impacted CZs is directly attributable to the China shock vs. the broader wage compression it triggered?
Do the mortality effects generalize to other trade-exposed economies, or are they specific to U.S. institutional features (weaker labor protections, opioid prescribing regime)?
What is the long-run speed of regional labor-market adjustment? Caliendo et al. (2015) show gains require reallocation; but if low-wage workers remain in dying industries (correlation 0.43), how long does the "short-run" last?