On the Persistence of the China Shock

Local Labor Supply

"... Figure 7 reports estimation results for population headcounts. In Figure 7a, we find negative but insignificant impacts of trade exposure on the size of the working-age population, though the point estimates grow as the time interval lengthens. The next three panels of the figure make clear that these negative effects are driven by one age group, those 25 to 39 years old. For adults ages 18 to 24 and 40 to 64, the impact of greater import competition on population headcounts is negative but small and highly imprecisely estimated for all time differences. The finding of no net impact of trade exposure on CZ populations is consistent with Faber et al. (2019), while the greater responsiveness of younger adults echoes Bound and Holzer (2000), who show that less-experienced workers are more relatively more mobile in the face of adverse shocks. For the 2000-2019 time difference, the coefficient estimate of􀀀5:89 (t-value=􀀀2:52) indicates that when comparing CZs at the 25th and 75th percentiles of trade exposure, the more-exposed CZ would have a􀀀3:89 (􀀀5:89 ) percentage-point larger decrease in headcounts for this age group. For comparison, the 75th- 25th percentile difference in population growth across CZs for individuals ages 25 to 39 over 2000 to 2019 was 20:96 (= 16:08 + 4:84) percentage points. Cross-CZ spillovers transmitted via net migration appear to be modest and concentrated on a narrow age group....The impact of trade exposure on the total working-age population is negative for both the 2000-2010 ( =􀀀2:42, t-value =􀀀1:92) and 2000-2019 ( =􀀀3:47, t-value=􀀀1:84) time differences, and marginally significant in each case. When we examine the impact on the native-born population, we find a smaller and much less precisely estimated effect: the impact coefficient for 2000 to 2019 is􀀀0:75 (t-value =􀀀0:79). This contrasts with impacts on the foreignborn, for which greater import competition significantly reduces population headcounts over 2000 to 2019 ( =􀀀 6:79, t-value =􀀀2:05). When looking at age subgroups within these populations, we see negative significant impacts of import competition on native-born workers ages 25 to 39 ( =􀀀5:03, t-value =􀀀2:29), which aligns with the results in Figure 7, and on foreign-born workers ages 40 to 64 ( =􀀀12:70, t-value =􀀀2:08). Whereas native-born labor-supply responses are strongest for younger workers, for the foreign-born they are strongest for older workers…”

"... Estimation results appear in Figure 8, where outcomes are in terms of log income and transfers relative to the total population of a CZ, except for total labor compensation, which is relative to wage and salary employment.39 The impact of exposure to import competition on personal income per capita in Figure 8a is negative at all time horizons and precisely estimated for most time differences from 2000 to 2014 forward. Negative effects reach their peak for the 2000 to 2015 time difference, for which the impact coefficient is􀀀3:64 (t-value =􀀀2:55)…”

"... Figure 9 presents estimates in which we divide the sample of CZs into groups based on whether the share of the college-educated in their working-age populations was above or below the population weighted national median in 2000. There are 336 CZs in the former group and 386 in the latter group.We consider four outcomes: the manufacturing employment-population ratio, the total wage and salary employment-population ratio, the log working-age population, and log personal income per capita. To keep the time horizon the same across outcomes, we evaluate time differences from 2001 to 2002 to 2001 to 2019. To control the false discovery rate when evaluating differences in coefficient estimates across sample splits, we compute and display minimal Benjamin-Hochberg q-values based on the number hypotheses being tested. Although both higher and lower-educated CZs experienced declines in manufacturing employment, the negative impacts of trade exposure on overall employment in Figure 5 are concentrated in CZs with relatively few college-educated workers, as shown in Figures 9a and 9b. These results are consistent with those in Bloom et al. (2019), who study the 1992 to 2012 time period. For the 2001 to 2019 horizon, the first two figures show that in less-educated CZs, a one percentage point increase in import penetration over 2000 to 2012 predicts a 1:74 (t-value =􀀀4:07) percentage point decrease in the manufacturing employment-population ratio and a 2:47 (t-value =􀀀3:59) percentage point decrease in the total wage and salary employment-to-population ratio. Across all CZs (see Figure 5), trade-induced changes in manufacturing and total employment-population ratios are very similar at long horizons, indicating that employment impacts on non-manufacturing employment are null. For CZs with relatively few college workers, the substantially larger impact on total employment than on manufacturing employment reveals a negative impact of trade exposure on non-manufacturing employment, as shown in Appendix Figure A17. Despite manufacturing job losses being compounded by non-manufacturing losses in these CZs, there is no effect of trade exposure on the log working-age population (Figure 9c). It thus appears that more trade-impacted CZs with fewer college-educated workers did not experience differential out-migration, though they did experience larger declines in personal income per capita, as shown in Figure 9d. Since economically-motivated migration would tend to mitigate the local disemployment effects of trade exposure, one interpretation of these results is that the lack of migration is both symptom and cause of the slow adjustment process. In CZs with more-educated working-age populations, the pattern of adjustment is qualitatively different. Impacts of trade exposure on manufacturing employment (Figure 9a) are negative, but somewhat smaller and less precisely estimated. Impacts on the total wage and salary employment population ratio are small and imprecise for short time differences and then become large, positive, and marginally significant for long time differences. For the 2001-2016 time difference and beyond, we easily reject that impact coefficients for wage and salary employment-population are the same in more- versus less-educated CZs. This trade-induced increase in the total employment-population ratio must imply a corresponding positive impact of trade exposure on the non-manufacturing population ratio, as seen in Appendix Figure A17. As critically, more-educated CZs adjusted to adverse trade shocks in part through net outmigration. For all time differences except 2001 to 2006, we reject that impact coefficients on the working-age population are equal for more versus less-educated CZs. At the 2001 to 2019 time difference in more-educated CZs, a one percentage point increase in import penetration is predicted to cause a 9:13 (t-value =􀀀3:31) percentage point decrease in the working-age population, or an annual population decline of approximately one-half a percentage point. The impact coefficient on log non-manufacturing employment at the 2001 to 2019 time difference of 6:04 (t-value = 2:37, Appendix Figure A17b) implies that more of the increase in the non-manufacturing employment population ratio occurred through the out-migration of labor rather than through greater job growth. Like less-educated CZs, more-educated CZs also see negative impacts of trade exposure on personal income per capita (Figure 9d). Distinct from less-educated CZs, these impacts reach their maximum negative value for the 2001 to 2012 time difference and then diminish over time, becoming close to zero for the 2001 to 2016 horizon and beyond…”

David Autor, David Dorn and Gordon Hanson, "On the Persistence of the China Shock," Brookings Papers On Economic Activity, September 9, 2021, https://www.brookings.edu/wp-content/uploads/2021/09/On-the-Persistence-of-the-China-Shock_Conf-Draft.pdf

David Autor returns to the China stock and finds the initial impact of the China shock has been persistent though stopped intensifying a decade ago, "...We evaluate the duration of the China trade shock and its impact on a wide range of outcomes, building on analyses in Autor et al. (2013a) and Acemoglu et al. (2016). This shock plateaued in 2010, enabling study of its effects for nearly a decade past its culmination. Adverse impacts of import competition on manufacturing employment, employment-population ratios, and income per capita in more trade-exposed U.S. commuting zones are present out to 2019. Reductions in population headcounts, which indicate net out-migration, register only for foreignborn workers and the native-born 25-39 years old, implying that exit from work is a primary means of adjustment. More negatively affected regions see larger increases in the uptake of government transfers, but these transfers primarily take the form of Social Security and Medicare benefits and replace only a small fraction of lost personal income. Adverse outcomes are more acute in regions that initially had fewer college-educated workers and were more industrially specialized. Impacts are qualitatively—but not quantitatively—similar to those caused by the decline of employment in coal production since the 1980s, indicating that the China trade shock holds lessons for other episodes of localized job loss. Import competition from China induced shocks to income per capita across local labor markets that are much larger than the spatial heterogeneity of income effects predicted by standard quantitative trade models. Even using higher-end estimates of the consumer benefits of rising trade with China, a substantial fraction of commuting zones appears to have suffered absolute declines in real average incomes...."

Autor also did an exercise looking at the China shock relative to the collapse in demand for coal and the housing bust and finds that the China trade shock was different, "...In summary, we observe that Chinese import competition, the fall in demand for coal, and the Great Recession each reduced employment and labor incomes as these shocks unfolded. CZs with greater shock exposure did not experience large declines in population in any of these cases, suggesting that migration did not disperse the local impacts of these shocks. CZ more exposed to the Great Recession eventually experienced a substantial recovery of their employment rates, and faster population growth. Those with greater exposure to the China or coal shocks instead endured persistently depressed employment rates and labor income levels, combined with a slow decline in population. The comparison with the coal shock in particular indicates that the China shock’s long-lasting impact on CZ labor market conditions and the sluggish population response to depressed local labor market conditions was not without precedent. However, the large magnitude of the China shock, which had a sizable impact on many local labor markets in the US, sets it apart from shocks such as the decline of the coal sector whose impact was more limited in scope. As the United States prepares for potentially more job loss due to the ongoing energy transformation and expected changes in oil and gas production, the failure of local labor markets to adjustment successfully to the coal and China trade shocks reminds us that the adjustment process is typically slow and sclerotic, unlike the textbook model of frictionless labor market adjustment..."
The basic story, "...The first wave of China shock research found that greater import competition caused localized job loss in manufacturing, declines in earnings for low-wage workers, and greater economic distress across a wide range of outcomes. The primary mechanism of adjustment to trade exposure was exit from work, rather than increased employment in non-manufacturing or migration to other regions. The impacts of rising import competition, originally documented for 1991 through 2007, are manifest out to 2019, nearly two decades after China’s accession to the WTO in 2001 and nine years after the plateau of China’s export surge. More trade-impacted CZs suffered durable increases in joblessness and decreases in personal income per capita that are not close to being offset by government transfers. The resulting economic distress appears to be most acute in local labor markets that lacked abundant supplies of college-educated workers—consistent with the dearth of human capital hypothesis—and that were narrowly specialized in labor-intensive manufacturing—consistent with the reverse agglomeration hypothesis. For CZs such as North Hickory, North Carolina, the consequences of the China trade shock have been profound and long-lasting….”
"...In more trade-exposed regions, we find that negative effects on manufacturing employment continue to build well beyond the culmination of the trade shock itself. Adverse impacts persist out to 2019, nearly a decade after the shock reached peak intensity. Although the trade shock did generate modest net out-migration, this occurred only among the foreign-born and native-born adults ages 25 to 39. Impacts of the trade shock thus appear to be long-lasting and to entail suppressed participation in work, which is evident in declines in wages and salaries and personal income in trade-exposed regions. Income losses are only minimally offset by the increased uptake of government transfers, which also remain elevated to the end of the analytic window. Why has the China trade shock had such enduring consequences? One explanation is that the shock itself never stopped intensifying. We find little support for this idea. China’s spectacular manufacturing export growth slowed dramatically in the last decade, by which point China’s reform driven boom had run its course and the government had begun to rollback reforms (Lardy, 2019; Brandt et al., 2020; Brandt and Lim, 2020). Export growth also did not simply move en masse from China to other nearby countries. Whether we look at the U.S. market presence of China alone or combined with Vietnam and other Southeast Asian nations, to which China has begun to offshore some labor-intensive activities, the trade shock reached peak intensity around 2010 and stabilized thereafter. Although the China trade shock did not unwind, it did plateau, letting us examine its consequences for nearly a decade beyond its full expression...."

"...We conduct preliminary analyses of two other explanations for why trade-exposed labor markets suffered long-lasting hardship. One hypothesis is that many traditional manufacturing regions were poorly positioned to recover from job loss because of a dearth of college-educated workers, who are in high demand by sectors that are expanding nationally (Glaeser et al., 1995; Diamond, 2016; Bloom et al., 2019). A second is that specialization in a narrow set of industries left these regions exposed to industry-specific shocks that, once industry decline initiated, would set in motion a process that is self-reinforcing (e.g., Dix-Carneiro and Kovak, 2017). We find some support for both the dearth-of human- capital and reverse-agglomeration hypotheses. We also discuss other mechanisms that may be at work, though we do not provide a definitive, mono-causal explanation..."

"...China’s post-2010 slowdown is also apparent when we compare its trade performance to the U.S., which is helpful for defining the shock facing U.S. manufacturing industries. A simple method to evaluate China’s relative export success is to use the Balassa (1965) measure of revealed comparative advantage (RCA), which is the ratio of a country’s share of world exports in a particular sector to its share of world exports of all goods.13 In Figure 2, we show the gap between the log China and log U.S. RCA for manufacturing and non-manufacturing. China began with a modest comparative advantage in manufacturing over the U.S. in 1991—as indicated by a China-U.S. manufacturing versus all goods export differential of just 2.0 percentage points in that year. By 2011, its manufacturing-versus-all goods export differential over the U.S. had leaped to 13.1 percentage points, before falling and rising over the ensuing eight years. After 2010, the China-U.S. relative RCA exhibits little trend. By construction, China’s RCA in non-manufacturing compared to the U.S. exhibits approximately the inverse pattern, dropping steadily through the 1990s and 2000s and then remaining at negative levels after 2011. China’s comparative disadvantage in non-manufacturing relative to the U.S. reflects the success of U.S. firms in exporting agricultural products, certain mineral products, and oil and gas, to the rest of the world, as well as market forces in China that have concentrated factors of production in manufacturing at the expense of other sectors. Even though China’s export growth in manufacturing slowed after 2010, it still could have contributed to overall U.S. manufacturing import growth by offshoring production to other low wage countries. Chinese firms have been active in building industrial parks for export production in Southeast Asia, and in Vietnam in particular. To investigate this possibility, in Figure 3 we plot shares of U.S. imports and import penetration in the U.S. market for China alone and for China combined with low-income countries in Southeast Asia. We select these countries—Cambodia, Indonesia, Laos, Myanmar, Philippines, Vietnam—based on their per capita GDP in 2010 being less than that of China. We also add neighboring Bangladesh to this group because in recent decades multinational companies from Asia have expanded apparel factories in the country (Heath and Mobarak, 2015). Figure 3 underscores that China’s exports to the U.S. dwarf those of Southeast Asia. In 2010, China accounted for 23.4% of U.S. manufacturing imports, whereas the Southeast Asian nations accounted for just 2.6%. In 2018, these figures were only modestly changed, at 23.6% and 3.8%, respectively. Although Southeast Asian countries did gain a larger share of U.S. imports, the increase over the 2010 to 2018 period was just 1.2 percentage points. Import penetration, shown in Figure 3b, tells a similar story. Whether we look at China alone or combined with low-income Southeast Asian nations, its U.S. market presence largely stabilized after 2010..."

Evidence, "...This section presents our main empirical results for the impact of trade exposure on U.S. commuting zones. We use equation (2) to estimate how the 2000-2012 trade shock affected CZs over2000 to 2019, focusing on three sets of outcomes. The first are ratios of manufacturing employment, non-manufacturing employment, total wage and salary employment, and unemployment to the working-age population (defined as individuals ages 18 to 64 years old). We measure employment using the Bureau of Economic Analysis’ (BEA) Regional Economic Information System (REIS), unemployment using Local Area Unemployment Statistics (LAUS), and population using the National Vital Statistics System, where we aggregate data from the county to the CZ level. Trade shock impacts on these outcomes reveal the direct consequences on manufacturing employment, and the magnitude of labor reallocations to unemployment, non-employment, and other sectors. A second set of outcomes are population headcounts, overall and by nativity and age-based subgroups, which reveal whether exposure to import competition led to net reductions in the resident population (e.g., due to out-migration). A third set of outcomes are per capita personal income, labor compensation, and government transfers, overall and by program type, which reveal the impact of trade shocks on average income and its components, and allow us to quantify the degree to which government transfers offset income losses caused by trade exposure. We measure income, earnings, and government transfers using county-level data from the REIS. For most series, we use 2000 as the initial year.Summary statistics on outcome and control variables are reported in Appendix Tables A1 and A2; Appendix A.2 contains further details on data sources and variable construction…”

Employment Outcomes

"... Figure 5 shows the cumulative impact of trade exposure for progressively longer time differences (1 to 18 years)....The results for manufacturing employment in Figure 5a show a negative effect of trade exposure that builds quickly and then stabilizes.... As one moves past 2012, the negative effect of greater import competition on manufacturing employment continues to grow, even though the trade shock itself no longer appears to be intensifying. Given a decadalized increase in import penetration over 2000 to 2012 of 0:89 percentage points (Table 1),the implied reduction in the manufacturing employment share over 2001 to 2019 is􀀀1:59 (=􀀀1:790:89) percentage points, relative to the overall 2001-2019 change in the manufacturing employment share of􀀀2:68 percentage points (Appendix Table A1). Alternatively, when comparing CZs at the 25th and 75th percentiles of trade exposure, the latter would be predicted to have a reduction in its manufacturing employment share that is 1:18 (=􀀀1:79 ) percentage points larger than in the former over 2001 to 2019. This compares to the 25th-75th percentile differential change in the manufacturing-employment-share of􀀀2:17 (=􀀀3:79 + 1:62) percentage points for the same period. Exposure to import competition from China thus appears to account for a large share of net manufacturing job loss in US commuting zones after 2000.29 Those losing jobs in manufacturing may move into other sectors, into unemployment, or out of the labor force. Simultaneously, they may move to another commuting zone.30 In Figure 5b, we consider the impact of exposure to import competition on the CZ share of the working-age population employed in non-manufacturing industries. Impact coefficient estimates are close to zero at all time intervals and are imprecisely estimated (e.g., the impact coefficient for the 2001-2019 period is 0:05, t-value= 0:06). We see no evidence that on net non-manufacturing sectors absorbed local workers released from manufacturing due to the China trade shock.31 The wide standard error bounds in Figure 5b are suggestive of heterogeneity across CZs in how non-manufacturing absorbs displaced manufacturing workers, a possibility we explore in section 4.4. The fall in manufacturing employment and absence of offsetting job gains in non-manufacturing imply a decline in total wage and salary employment in trade-exposed CZs, confirmed by Figure 5c. Decreases in the employment-population ratio are comparable in magnitude to those for manufacturing employment. The effect reaches its maximum value for the 2001 to 2017 period: the impact coefficient of􀀀1:91 (t-value=􀀀2:70) compares to􀀀1:88 (t-value=􀀀5:89) for manufacturing employment at the same horizon. The impact attenuates modestly late in the period, ending at􀀀1:74 (t-value=􀀀2:35) over 2001 to 2019, which is close to the manufacturing impact at this horizon. At long time horizons, much of the net absorption of manufacturing job loss is through increases in nonemployment. Amior and Manning (2018) argue that changes in the local employment-population ratio summarize changes in local average real income. By this logic, the China trade shock would have substantially altered the distribution of well-being across CZs, a possibility we explore in more detail below through our analysis of personal income per capita. Figure 6a maps the actual change in wage and salary employment-population ratios across CZs for the 2001 to 2019 time period. While the national employment rate barely changed over this period (+0.19 percentage points), the map reveals considerable variation across space, with lower employment growth in parts of the South, Midwest and Northeast, and faster employment growth in the Great Plains and some of the Western and Northeastern coastal areas. In Figure 6b, we map the implied impact of the China trade shock (as measured in (1) for 2000 to 2012) on changes in employment-population ratios for the same 2001 to 2019 time period. A visual comparison between Figure 6’s two panels shows a striking correlation, as many of the CZs that lost employment overall (in panel a) were also more adversely affected by the trade shock (in panel b).This correlation suggests that the China shock had an important influence on the differential employment growth across U.S. regions in the last two decades. The top 5% of CZs in terms of implied reductions in employment-population ratios, which are listed in Appendix Table A4, include a preponderance of locations that in 2000 were relatively highly specialized in manufacturing and had relatively few college-educated workers. In 2000, 33 of the 38 most exposed CZs had a manufacturing employment share above 25 percent, relative to the population-weighted median of 15.4%, and 33 of 38 had a college-educated share of the working-age population below 20 percent, relative to the national population-weighted median of 23.4%. The distinctiveness of trade-impacted commuting zones motivates the heterogeneity analysis we undertake in section 4.4..."