Stan de Ruijter

Shift-share (Bartik) designs measure a region's exposure to industry shocks by weighting national shocks by its own-industry employment shares, implicitly assuming each shock affects only its own industry's workers. This assumption does not hold when industries are linked through a production network: a shock to one industry propagates to its suppliers and customers, so regional outcomes respond to both the direct shock and its network-propagated consequences, and the standard Bartik loads an omitted network-exposure term. We develop a closed-form diagnostic for this contamination, φ = tr(QP)/tr(Q) ≥ 1 (for the un-residualized co-location matrix), which depends only on the share and propagation matrices. This diagnostic is a design baseline rather than a bound: under instrumental variables the realized contamination φ^IV also reflects the shock covariance, and in the application below it is roughly twice as large. The correction is straightforward—premultiply the shocks by the Leontief inverse before weighting by shares—and it nests the standard Bartik and requires only a public input–output matrix. We apply the framework to the Autor et al. (2013) China Shock, where the standard 2SLS coefficient is 2.40 times the corrected one; the two nonetheless deliver nearly identical population-weighted aggregate predicted impacts. This cancellation holds under an exact, falsifiable condition: the regional network multiplier must be uncorrelated with instrument exposure, a property of the (Q, P, w, g) design that the China Shock nearly satisfies but fails in its supply-chain-dense subregions. The cancellation concerns the aggregation of partial-equilibrium predicted values rather than general-equilibrium impact. What the correction changes is geography. Of the 100 commuting zones most exposed under the standard measure, 34% are not among the corrected top 100, and on held-out geographies the corrected exposure raises out-of-sample predictive fit from 0.080 to 0.262. We interpret this gain as a check on predictive stability rather than as causal validation of the correction.

This paper analyzes and contrasts the impact of increased global trade with China on the US and Germany during the “China shock” period from 2000 to 2014. We introduce novel decompositions of overall output and labor market effects into direct effects and indirect effects operating upstream from focal industries, while further distinguishing between domestic and international channels. To construct counterfactuals in which Chinese trade flows remain at their 2000 levels, we employ a computable general equilibrium (CGE) model. We find that in both the US and Germany, upstream effects raise output and more than offset the direct effects of increased Chinese trade. Further decomposing upstream effects into domestic and international components, we show that both are quantitatively important in both countries, particularly in manufacturing, with domestic upstream effects playing a central role for service industries. Labor demand responses are highly heterogeneous across worker types: male manufacturing workers experience disproportionately negative effects in the US, while comparable workers see gains in Germany. At the same time, upstream linkages generate absolute increases in labor demand for service workers in both countries.