Autor Dorn and Hanson 2016 — The China Shock Learning from Labor-Market Adjustment to Large Changes in Trade

Summary

A comprehensive review article synthesizing the Autor, Dorn, and Hanson (ADH) research program on China's trade shock and its labor-market consequences. The paper formalizes the Bartik-IV identification strategy in a gravity-model framework (Eaton-Kortum 2002), presents industry-level and commuting zone (CZ)-level empirical results, and documents the complete cascade of government transfer responses. The central finding — that trade-exposed CZs show persistent, unreversed employment declines with no offsetting reallocation — demolishes the pre-2000 consensus that US labor markets adjust smoothly to trade shocks. The paper also synthesizes worker-level Social Security Administration (SSA) data showing that low-wage workers remain trapped in trade-exposed industries (industry correlation $0.43$ vs. $0.17$ counterfactual), while high-wage workers exit cleanly, and provides a welfare assessment showing near-zero net gains in the immediate aftermath.

Key Claims

China's world manufacturing export share rose from $2.3\%$ (1991) to $18.8\%$ (2013); value-added share rose from $4.1\%$ to $24.0\%$ . Growth was internally driven by Mao-era suppression of comparative advantage, not by US conditions — validating the natural experiment.
Formal gravity framework (Eq. 2): Log change in region $i$ 's output = exposure to US demand shocks + endogenous wage change + national productivity change + competing-country supply change + China supply shock (the treatment). The China shock term simplifies to $\sum_k \varphi_{ik} (X_{ck} \hat{A}_{ck} / E_k)$ , where $\varphi_{ik}$ = industry $k$ share of region $i$ 's sales. This is the Bartik measure.
Instrument validity (Table 1): Cross-industry correlation of Chinese export growth between US and 8 comparison countries $= 0.92$ ; every comparison country saw import growth in $343$ – $385$ of $397$ four-digit manufacturing industries, confirming the supply-side (not demand-side) origin of the shock.
Industry-level (Acemoglu et al. 2016, Table 3): $1$ percentage point (pp) rise in import penetration $\to$ $-1.30$ log pts manufacturing employment (two-stage least squares (2SLS), $t = 3.2$ ). Ordinary least squares (OLS) estimate $-0.81$ . The coefficient is stable across 1991–1999 and 1999–2007 subperiods.
Job loss magnitudes: $560{,}000$ direct manufacturing jobs (Acemoglu et al. 2016), $985{,}000$ with input-output linkages, $2.0$ – $2.4$ million total economy (including CZ-level demand multiplier), 1999–2011.
CZ-level outcomes (Table 4), per $\$1{,}000$ import exposure per worker: Manufacturing employment $-0.60$ pp; non-manufacturing employment $-0.18$ pp (not significant); unemployment $+0.22$ pp; not in labor force (NILF) $+0.55$ pp; log weekly wages $-0.76$ log pts; annual income per adult $-\$549.3$ ; transfers per capita $+\$57.7$ .
No offsetting reallocation: Overall CZ employment-to-population falls at least one-for-one with manufacturing decline. No significant employment gains in unexposed sectors. Population mobility is modest (log population $-0.05$ , not significant).
Transfer cascade (Figure 7), per $\$1{,}000$ import exposure: Trade Adjustment Assistance (TAA) (unemployment insurance (UI)+TAA benefits) $= \$3.65$ ; Social Security Disability Insurance (SSDI) $= \$8.40$ ; SSA retirement $= \$10.00$ ; other income assistance $= \$15.04$ ; government medical $= \$18.27$ ; total $= \$57.73$ . TAA — the designated trade policy tool — is essentially irrelevant in dollar terms; SSDI absorbs more than twice as much.
Wage impacts concentrated at bottom: Using quantile IV (Chetverikov et al. 2016), wage effects concentrated in bottom four wage deciles; present outside manufacturing, not just within it.
Worker-level trapping (Autor, Dorn, Hanson, Song 2014): Using SSA longitudinal earnings records, workers in trade-exposed 1991 industries accumulate substantially lower earnings 1992–2007. Industry correlation between initial and current employer $= 0.43$ in 2007 vs. $0.17$ counterfactual (if workers always exited trade-exposed industries on first separation) — low-wage workers remain in high-exposure industries even after separating.
Heterogeneity: High-wage workers (top earnings tercile) react to trade exposure by relocating outside manufacturing with no earnings loss. Low-wage workers (bottom tercile) relocate within manufacturing, remain in subsequently shocked industries, suffer large lifetime earnings losses.
Welfare assessment: Caliendo, Dvorkin, Parro (2015) dynamic general equilibrium (GE) model: net welfare gains $\approx 0$ immediately post-shock; ultimate gains realized only after full regional reallocation. Galle, Rodríguez-Clare, Yi (2015): geographic standard deviation (SD) of welfare gains $= 2\times$ national mean. The $\$549$ income loss offset by only $\$58$ in transfers ( $\approx 10\%$ ).
Pre-2000 consensus overturned on all three pillars: (1) trade does reduce manufacturing employment significantly; (2) workers do not readily relocate — adjustment is stunningly slow, lasting a decade+; (3) effects are locally concentrated, not nationally diffused through factor prices.
DI connection: Figure 7 explicitly shows SSDI rising with import exposure, consistent with Autor and Duggan (2003): displaced manufacturing workers apply for SSDI as permanent exit from the labor force. This provides upstream causal identification for disability insurance (DI) growth separate from the health channel.

Concepts Introduced or Extended

Entities Mentioned

Quotes

"Adjustment in local labor markets is remarkably slow, with wages and labor-force participation rates remaining depressed and unemployment rates remaining elevated for at least a full decade after the China trade shock commences."

"For every extra $100 in local import exposure per worker transfer receipts rise by approximately $6 per capita — but also their relative magnitudes across categories. TAA... is effectively inconsequential in local adjustment to trade shocks."

"Trade-exposed CZs experience a reduction in annual household wage and salary income per adult of $549, whereas per capita transfer income rises by approximately $58, thereby offsetting just a small portion of the earnings loss."

My Take

This is the definitive synthesis of the ADH China shock program and the clearest single-paper statement of why the pre-2000 trade consensus failed. The gravity-framework formalization (Eqs. 1–3) is the key methodological contribution beyond ADH 2013: it shows exactly which confounds the Bartik IV addresses and why the exclusion restriction is satisfied. The Figure 7 transfer cascade is striking — it shows that the US government's policy response to trade-displaced workers operates almost entirely through programs designed for disability and retirement, not through TAA (the program nominally designed for trade adjustment). This is important for understanding DI growth: the China shock didn't just increase DI applications through health deterioration; it caused workers to use DI as a permanent income replacement when trade eliminated their manufacturing jobs. The welfare assessment (near-zero short-run gains) is the most uncomfortable finding for standard trade theory, but the paper is careful to note this is short-run — long-run gains exist once reallocation occurs. The unresolved question is how long "long-run" takes, and whether it exceeds most workers' working lifetimes.