Peter K. Schott Yale School of Management & NBER The Impact of U.S. Trade Liberalization with China on U.S. Manufacturing Workers and Firms Peter K. Schott Yale School of Management & NBER
Source: Pierce and Schott (2012) Motivation Source: Pierce and Schott (2012)
Motivation
Motivation
Motivation
Outline Worker adjustment New research using matched employer-employee data Firm adjustment New research using firm level data Other adjustments Health, politics Overview: global firms
U.S. Manufacturing Employment After PNTR
Post-War U.S. Manufacturing Employment -2.9 mill over 3 years
Outline US-China Trade Policy Census Data Old Results New Results
US NTR and Non-NTR Tariffs NTR = Normal Trade Relations Synonym for “Most Favored Nation” (MFN) The US has two basic tariff schedules NTR tariffs : for WTO members; generally low Non-NTR tariffs : for non-market economies; generally high; set by Smoot-Hawley (1930) How does China fit into these categories?
Non-NTR g Renewable NTR g Permanent NTR 2001 (December) China enters WTO Non-NTR PNTR Annual renewals of MFN status were uncertain, particularly in the 1990s, creating uncertainty for U.S. and Chinese firms 1980 (February) China granted temporary NTR status Requires annual approval by President and Congress 2000 (October) U.S. Congress grants China Permanent NTR status, U.S. and Chinese firms have incentive to move production to China
The “NTR Gap” i=industry Define the difference between the non-NTR and NTR rates as NTR Gapi = Non-NTR Tariffi – NTR Tariffi Calculated using tariff-line (HS8) data from Feenstra et al. (2003) Use tariffs for 1999, the year before PNTR Gap for industry i is the mean across tariff lines in industry i
Distribution of NTR Gap 89 percent of the variation in the NTR gap across industries arises from variation in non-NTR rates, which were set in 1930 The higher the NTR gap, the greater the industry’s exposure to potential tariff increases before PNTR Mean: 0.32 Std: 0.15
Simple View of the Data Divide U.S. import and manufacturing industries into two groups: Most-exposed: NTR gap above median Least-exposed: NTR gap below median Plot their trajectory before and after PNTR
Simple View of the Data – Trade Public Census Trade Data Imports from China in the more-exposed products jump after PNTR Most Exposed Least Exposed
Simple View of the Data – Trade Public Census Trade Data Imports from China in the more-exposed products jump after PNTR This trend is not present in imports from rest-of-world (ROW) China ROW
Simple View of the Data – Employment Public NBER-CES Data Before PNTR: high- and low-gap industries follow roughly parallel trends
Simple View of the Data – Employment Public NBER-CES Data Before PNTR: high- and low-gap industries follow roughly parallel trends Post PNTR: employment falls more sharply among high-gap industries Least Exposed Most Exposed
Related Research Employment and trade liberalization Lots of papers China: Autor et al. (2013, 2016, 2017); Bloom et al. (2015) Investment under uncertainty Trade: Handley (2014); Handley and Limao (2014a, 2014b) “Jobless” recoveries Manufacturing: Faberman (2012) Supply-chain linkages US manufacturing: Ellison, Glaeser and Kerr (2010) Trade: Baldwin and Venables (2013)
Census Data LBD CM LFTTD Longitudinal Business Databse Employment of all U.S. establishments Annual data but limited information Concordance issue Census of Manufactures Employment + other attributes for all manufacturing establishments, 1992(5)2007 Lots of information but low frequency Transaction-level US import data Value U.S. importer and foreign exporter ID’s
Part 1: Outline US-China Trade Policy Data Old results Employment New results Conclusion
Empirical Strategy i=industry; t=year DID specification Regress industry-year employment from 1990-2007 on interaction of NTR Gap and indicator for post-PNTR period (year>2000)
Empirical Strategy i=industry; t=year DID specification Regress industry-year employment from 1990-2007 on interaction of NTR Gap and indicator for post-PNTR period (year>2000) Add covariates that account for Industry characteristics Other policy changes (MFA, China’s entry into WTO, etc.) Time-Varying E.g., MFA, NTR rate, etc. Time-Invariant Initial K/L, Changes in Chinese import tariffs, etc. Industry and year FE.
Alternate Explanations? Must match timing and pattern across industries We account for: Changes in Chinese policy Lower import tariffs Elimination of export licensing requirements Elimination of production subsides Reduced barriers to foreign investment Union Resistance in the US Popped US tech bubble Rising Chinese competitiveness End of Textile and Clothing Quotas
Baseline Results (LBD)
Baseline Results (LBD) Industries with higher NTR gaps experience larger employment declines following PNTR
Baseline Results (LBD) Industries with higher NTR gaps experience larger employment declines following PNTR Economic significance: for each industry, compute the implied effect of PNTR (vs the pre-period) relative to a hypothetical industry with a zero NTR gap
Baseline Results (LBD) Not much action in terms of the other covariates, except for skill intensity, changes in Chinese import tariffs and the MFA phase-out
Timing (LBD) The NTR gap should be correlated with employment after PNTR but not before
Timing (LBD) Plot the 95% confidence interval of the annual DID coefficients for two specifications: no additional controls versus baseline controls
Input-Output Linkages Use input-output tables to compute Upstream exposure via supplier industries Downstream exposure via customer industries Findings Higher downstream exposure associated with employment loss and plant exit Indirect evidence of trade-induced supply-chain disruptions (Baldwin and Venables 2013)
Outline US-China Trade Policy Data Old results Employment Importing New results Conclusion
U.S. Imports (LFTTD) h=product, c=country, t=year Compare products with varying NTR gaps (first difference) before and after PNTR (second difference) and across source countries (third difference) for the years 1992 to 2007 where Ohct represents one of Import value Number of U.S. importers Number of Chinese exporters Number of importer-exporter pairs Fixed Effects Time-Varying Exchange rate and NTR rate Interactions needed to identify q
U.S. Imports (LFTTD) h=product, c=country, t=year Compare products with varying NTR gaps (first difference) before and after PNTR (second difference) and across source countries (third difference) for the years 1992 to 2007
Chinese Exports (NBS) h=product, c=country, t=year Similar to U.S. imports but using Chinese transaction-level trade data from the China National Bureau of Statistics (NBS) Firm x six-digit HS x year Can observe export “type” General Processing & Assembly Can observe firm ownership type SOE Private domestic Private foreign
Chinese Exports (NBS) h=product, c=country, t=year All three types of exports experience relative growth following PNTR
Chinese Exports (NBS) h=product, c=country, t=year All three types of exports experience relative growth following PNTR Notes: Table reports results of product-country-year level generalized triple difference-in-differences OLS regressions of log Chinese export value on noted triple difference-in-differece term as well as product (h), country (c) and year (t) fixed effects. Results for fixed effects as well as all other terms required to identify the triple difference-in-differences term of interest are suppressed. Data span 2000 to 2005. Robust standard errors adjusted for clustering at the product level are displayed below each coefficient. All exports are defined as the sum of general and processing & assembly (P&A) exports. All firms are the union of state-owned enterprises, privately owned domestic firms and privately owned foreign firms. Superscripts ***, ** and * represent statistical significance at the 1, 5 and 10 percent levels.
Chinese Exports (NBS) h=product, c=country, t=year All three types of exports experience relative growth following PNTR Relative growth occurs among all three firm types, including “foreign” which may indicate within-firm offshoring by US firms Notes: Table reports results of product-country-year level generalized triple difference-in-differences OLS regressions of log Chinese export value on noted triple difference-in-differece term as well as product (h), country (c) and year (t) fixed effects. Results for fixed effects as well as all other terms required to identify the triple difference-in-differences term of interest are suppressed. Data span 2000 to 2005. Robust standard errors adjusted for clustering at the product level are displayed below each coefficient. All exports are defined as the sum of general and processing & assembly (P&A) exports. All firms are the union of state-owned enterprises, privately owned domestic firms and privately owned foreign firms. Superscripts ***, ** and * represent statistical significance at the 1, 5 and 10 percent levels.
Outline US-China Trade Policy Data Old results New results Margins of firm adjustment
Margins of Adjustment Job creation margins Plant expansion (PE) within continuing firms Plant birth (PB) within continuing firms Firm birth (FB) Job destruction Plant contraction (PC) within continuing firms Plant death (PD) within continuing firms Firm death (FD) Examine change in employment along each margin
Background: U.S. Private Sector JC and JD
Background: U.S. Manufacturing JC and JD
Raw Data: Gross Margins of Adjustment
Raw Data: Gross Margins of Adjustment
Raw Data: Net Margins of Adjustment
Implied Impact of PNTR from DID Regressions Contribution of relatively strong job destruction Helps explain jobless recovery trends in manufacturing noted by Faberman (2012) Contribution of relatively weak job creation Contribution of weak JC rises over time, preventing a “normal” recovery Notes: Job destruction (JD) margins are plant contraction (PC) at continuing firms, plant death (PD) at continuing firms, and firm death (FD). Job creation (JC) margins margins are plant expansion (PE) at continuing firms, plant birth (PB) at continuing firms, and firm birth (FB).
Ongoing Research: Mechanisms
Ongoing Research: Mechanisms
Ongoing Research: Mechanisms Estimated Impact of Interquartile Shift in Industry Exposure to PNTR Notes: Figures display the 95% confidence intervals associated with an interquartile shif in industry exposure to PNTR. Source: authors calculations based on data from the NBER-CES Manufacturing Productivity database.
Outline US-China Trade Policy Data Old results New results Margins of firm adjustment Margins of worker adjustment
Where Did U. S. Manufacturing Workers Go Where Did U.S. Manufacturing Workers Go? Evidence from Matched Employer-Employee Data
Motivation Large literatures examine worker reallocation across industries, firms and regions in response to economic shocks This paper analyzes the earnings and employment trajectories of workers that are directly (industry) and indirectly (county) exposed to a large negative shock induced by a change in U.S. trade policy Manufacturing workers with greater direct exposure experience Lower cumulative earnings Fewer years of employment Non-manufacturing workers with greater indirect exposure exhibit similar outcomes So, why does this matter? Isn’t what we find obvious? Are we trying to recover frictions?
Existing Research “China” Shock Acemoglu et al. (2016): direct and indirect impact of China on industry, regional employment Autor et al. (2014): examine direct and indirect impact on worker long-term earnings and employment trajectory of exposure to Chinese imports using social security records More… Displacement/Mass Layoff Jacobson, LaLonde and Sullivan (1993) Hummels, Jorgenson, Munch and Xiang (2014) Von Wachter and Davis (2011) Acemoglu et al. (2016) look at three effects: direct industry effect; local commuting zone effect; reduction in demand effect. Here, we can look directly at the impact on different groups of workers, i.e., those that are affected by the shock directly via their initial industry of employment and those that are only affected indirectly. Autor et al. (2014): observe age, gender, labor market experience, race, foreign-born but not education. Their region data is not so good, so they don’t have that in the main specs. Look at labor churning and earnings outcomes, but not exposure via region. They can look at self employment
Margins of Worker Adjustment Background Data Sector transitions Earnings
Background: Manufacturing vs Other Non-Farm Employment Post-2000 Declines 2000-3: -17% 2007-10: -18% Created in vox_02.do in pntr/drafts/
Margins of Worker Adjustment Background Data Sector transitions Earnings
Data: Matching Workers and Firms Longitudinal Employer-Household Dynamics (LEHD) Based on state unemployment insurance (UI) records Encompasses 46 states for 2000-11 (excluded: Alabama, Arkansas, New Hampshire, Mississippi, DC) Omits workers not covered by states’ reporting systems, primarily agriculture and federal, military and postal workers Observables Employment History File (EHF) Earnings and employment by longitudinal personal identifiers (PIKs) that map 1:1 to SSN Links workers to establishments and firms; we assign workers to industries and counties based on the establishments where they work Individual Characteristics File (ICF) Age, gender, race, foreign-born Additional links possible (SIPP?)
LEHD Caveats Not possible to distinguish between unemployment vs NILF vs movement to out-of-scope state or industry We categorize workers that do not show up in a particular year as “not employed” Workers may have more than one job at a time Follow the LEHD literature and assign worker to job with highest earnings
Attributes of Manufacturing vs Non-Manufacturing Workers 2000 and 2011 In 2000, manufacturing workers are More male More foreign-born Less-educated Older By 2011, manufacturing workers become Even more relatively male Relatively younger (retirements!) 2011 2011
Margins of Worker Adjustment Background Data Sector transitions Earnings
Outline Background Matched employee-employer data Where did the manufacturing workers go? M = manufacturing NM = non-manufacturing NE = not employed Direct and indirect impact of economic shock
Transitions Among M, NM and NE 2000-11
Transitions Among M, NM and NE 2000-11 There are 17.5 million workers in manufacturing (M) in 2011
Transitions Among M, NM and NE 2000-11 There are 17.5 million workers in manufacturing (M) in 2011 In 2007: 33% move to NM 33% remain in M 34% are NE
2000-7 vs 2000-11 Why similar transition to NE for M and NM? Maybe M was different before 2000s? Maybe trade shock operates along industry and region margins? Other info <10 percent of transitions from MgNM take place within firms 87 percent of M workers that are NE in 2007 are NE in 2011 84 percent of NM workers that are NE in 2007 are NE in 2011
Transitions to NE Based on Initial Age All Workers, 2000-11 Transition to NE are higher for those >50 in 2000 Not shown: 87 percent of initial M workers that are NE in 2007 are also NE in 2011 For initial NM workers, it is 84 percent Workers <=50, 2000-11 Workers >50, 2000-11
Where Did Manufacturing Workers Go? First column reports the outflow of workers that were in manufacturing in 2000 Second column reports these outflows as a share of total outflows excluding “public” and “not employed” E.g., 9 percent of year 2000 manufacturing workers that did not stay in manufacturing or go to not employed or public went into construction Third column reports the stock of workers in 2000, net of workers that flowed in from manufacturing E.g., 7.3 percent of workers in 2007 were in construction
Where Did Manufacturing Workers Go? Sorted by 2000-11 shares (final column) First column reports the outflow of workers that were in manufacturing in 2000 Second column reports these outflows as a share of total outflows excluding “public” and “not employed” E.g., 9 percent of year 2000 manufacturing workers that did not stay in manufacturing or go to not employed or public went into construction Third column reports the stock of workers in 2000, net of workers that flowed in from manufacturing E.g., 7.3 percent of workers in 2007 were in construction
Margins of Worker Adjustment Background Data Sector transitions Earnings and employment after PNTR
Economic Shock: PNTR Pierce and Schott (2016) U.S. has two tariff schedules NTR: generally low; for WTO members Non-NTR: generally high, for non-market economies U.S. granted China access to NTR rates starting in 1980, but continued access depended on annual approval by Congress Absent approval, tariffs would spike to non-NTR levels After PNTR in 2000, China’s access to low NTR rates “locked in” PNTR’s reduction in tariff rate uncertainty gave U.S. and Chinese producers greater incentive to locate production in China
Data: Direct and Indirect Exposure to PNTR j=industry ; c=county; L=employment Direct exposure: NTR gap for industry j is Indirect exposure: NTR gap for county c is the employment share weighted average NTR gap of each industry (Ljc/Lc) in 1990 (i.e., 10 years before PNTR)
Distribution of Direct and Indirect Exposure Mean / SD Industry 33 / 14 County 7 / 6 Interquartile Shift Industry 8 County 22 (incl 0s)
Distribution of NTR Gapc Across Counties 51th-75th 76th-100th 0-25th 26th-50th
DID Panel Regression Examine whether outcomes for worker j industry i employed at firm f change after PNTR Sample Panel of workers 1997 to 2007 Focus on workers employed by the same firm during the entire pre-PNTR period, 1997 to 1999. For computational convenience, restrict to 5 percent random but representative sample
DID Panel Regression Initial Manufacturing Workers Quarters Employed Ln(Quarters) Notes: Figure displays the coefficient estimates and 95% error bands from estimation of the number of quarters worked on workers’ exposure to PNTR via their initial county of employment. Second covariate is the difference-in-differences coefficient of interest. Remaining coefficients capture worker and firm characteristics noted in text. Estimation period is 1997 to 2007. Estimation sample is a 5 percent random sample of manufacturing workers employed in the same firm from 1997 to 1999. Initial Manufacturing Workers
DID Panel Regression Initial Manufacturing Workers Quarters Employed Notes: Figure displays the coefficient estimates and 95% error bands from estimation of the number of quarters worked on workers’ exposure to PNTR via their initial industry of employment. Second covariate is the difference-in-differences coefficient of interest. Remaining coefficients capture worker and firm characteristics noted in text. Estimation period is 1997 to 2007. Estimation sample is a 5 percent random sample of manufacturing workers employed in the same firm from 1997 to 1999. Quarters Employed Initial Manufacturing Workers
Need man earnings county x quantile? DID Panel Regression Notes: Figure displays the coefficient estimates and 95% error bands from estimation of log earnings on workers’ exposure to PNTR via their initial county of employment. Second covariate is the difference-in-differences coefficient of interest. Remaining coefficients capture worker and firm characteristics noted in text. Estimation period is 1997 to 2007. Estimation sample is a 5 percent random sample of manufacturing workers employed in the same firm from 1997 to 1999. Need man earnings county x quantile? Ln(Earnings+1) Initial Manufacturing Workers
DID Panel Regression Initial Manufacturing Workers Ln(Earnings+1) Notes: Figure displays the coefficient estimates and 95% error bands from estimation of log earnings on workers’ exposure to PNTR via their initial industry of employment. Second covariate is the difference-in-differences coefficient of interest. Remaining coefficients capture worker and firm characteristics noted in text. Estimation period is 1997 to 2007. Estimation sample is a 5 percent random sample of manufacturing workers employed in the same firm from 1997 to 1999.
Need nonman earnings county x quantile? DID Panel Regression Initial Non-Manufacturing Workers Ln(Earnings) Need nonman earnings county x quantile? Notes: Figure displays the coefficient estimates and 95% error bands from estimation of log earnings on workers’ exposure to PNTR via their initial county of employment. Second covariate is the difference-in-differences coefficient of interest. Remaining coefficients capture worker and firm characteristics noted in text. Estimation period is 1997 to 2007. Estimation sample is a 5 percent random sample of manufacturing workers employed in the same firm from 1997 to 1999.
Conclusions / To-Do List Impact of PNTR “spilled” outside manufacturing Still to do Characterize employment and earnings effects over time Lots of other outcomes to look at
Outline US-China Trade Policy Data Old results New results Margins of firm adjustment Margins of worker adjustment Health effects
Disclaimer Any opinions and conclusions expressed herein are those of the authors and do not necessarily represent the views of the National Center for Health Statistics, the Board of Governors of the Federal Reserve System or its research staff.
Introduction A large literature investigates the effect of economic shocks on health Research in public health (e.g. Case and Deaton 2015) identifies an increase in U.S. mortality among middle-aged whites after 1999 This paper examines the relationship between county-level mortality rates and a plausibly exogenous change in U.S. trade policy – U.S. granting of Permanent Normal Trade Relations (PNTR) to China in 2000 – that differentially exposed U.S. counties to increased competition from China
Motivation: Case and Deaton (2015) Case and Deaton (2015) Figure 2 Crude Death Rates, Whites 45-54
Extending Case-Deaton (2015) Back in Time Accidental Poisoning Suicide Alcohol-Related Liver Disease (ALRD)
Extending Case-Deaton (2015) Back in Time Accidental Poisoning Suicide Alcohol-Related Liver Disease (ALRD)
Extending Case-Deaton (2015) Back in Time Accidental Poisoning Suicide Alcohol-Related Liver Disease (ALRD)
Introduction A large literature investigates the effect of economic shocks on health Public health researchers (e.g. Case and Deaton 2015) have identified an increase in mortality among middle-aged whites in the U.S. after 1999 This paper examines county-level relationship between mortality and the U.S. granting of Permanent Normal Trade Relations (PNTR) to China in Oct 2000 Two identification strategies Difference in Differences: do counties more exposed to the trade liberalization (first difference) experience greater changes in mortality after it was implemented (second difference)? Instrumental Variables: does mortality rise with the unemployment rate, using exposure to PNTR as an instrument? Note: this is not a welfare analysis of PNTR This paper examines the relationship between county-level mortality rates and a plausibly exogenous change in U.S. trade policy – U.S. granting of Permanent Normal Trade Relations (PNTR) to China in 2000 – that differentially exposed U.S. counties to increased competition from China
Preview of Results Mortality: Exposure to PNTR associated with higher mortality from “deaths of despair”, in particular suicide and accidental poisoning Implied impacts are strongest for whites, especially white males, and tail off as workers age Mechanisms: Exposure to PNTR associated with persistently higher unemployment, lower labor force participation and lower per capita personal income
Literature Business cycle, mass layoffs, plant closings g Mortality Cyclicality: Ruhm (2000, 2015), Stevens et al. (forthcoming) Layoffs: Jacobson et al. (1993), Sullivan and von Wachter (2009), Browning and Heinesen (2012) China/low-wage countries g U.S. labor market Imports: Bernard et al. (2006), Ebenstein et al. (2011), Autor et al. (2013), PNTR: Pierce and Schott (2016), Handley and Limao (2016), Feng, Li and Swenson (2016) Import competition g crime, public goods, health, marriage, fertility Dix-Carneiro et al. (2015), Feler and Senses (2015), McManus and Schaur (2015), Autor, Dorn and Hanson (2015)
Outline Data Mortality rates Trade policy DID empirical strategy and results 2SLS empirical strategy Robustness Conclusion
Data: County-Level Mortality Rates CDC microdata containing all U.S. death certificates filed between 1990 and 2013 Observe age, gender, race, county of residence, year of death and underlying cause of death Match year x county x gender x race x age bin death counts to NCI’s SEER population estimates to compute death rates Rates conventionally expressed per 100,000 population Two types of county-level death rates Crude: just divide deaths by population Age-adjusted: weighted average crude death rate across age categories, using the year 2000 overall population shares as weights for all counties; (baseline results use 5-year age bins)
Age-Adjusted Death Rates, 2000 Population-Weighted Averages Across Counties Source: CDC.
Distribution of All-Cause Death Rates Over Time Population-weighted averages across counties in year 2000 are Overall 858 Male 1047 Female 719 f_ar_all.emf Figures, section A.2
Data: Trade Policy
U.S. Trade Policy NTR = Normal Trade Relations Synonym for “Most Favored Nation” (MFN) The US has two basic tariff schedules NTR tariffs : for WTO members; generally low Non-NTR tariffs : for non-market economies; generally high; set by Smoot-Hawley (1930) How does China fit into these categories?
Non-NTR g Renewable NTR g Permanent NTR 2001 (December) China enters WTO Non-NTR PNTR Annual renewals of MFN status were uncertain, particularly in the 1990s, creating uncertainty for U.S. and Chinese firms 1980 (February) China granted temporary NTR status Requires annual approval by President and Congress 2000 (October) U.S. Congress grants China Permanent NTR status, U.S. and Chinese firms have incentive to move production to China
Measuring U.S. Counties’ Exposure to PNTR j=industry ; c=county; L=employment NTR gap for industry j is NTR gap for county c is the employment share weighted average NTR gap of each industry (Ljc/Lc) in 1990 (i.e., 10 years before PNTR)
County Exposure to PNTR Mean 7.3 Std Dev 6.4 25th 2.4 75th 10.6 County SIC codes (in tariff schedule)
County Exposure to PNTR Mean 7.3 Std Dev 6.4 25th 2.4 75th 10.6 County SIC codes (in tariff schedule)
Distribution of NTR Gapc Across Counties 51th-75th 76th-100th 0-25th 26th-50th
Outline Data DID empirical strategy and results 2SLS empirical strategy Robustness Conclusion
DID Empirical Strategy Focus on aggregate “deaths of despair” Suicide Accidental poisoning Alcohol-related liver disease (ARLD) Rationale: Highlighted in Case and Deaton (2015), etc. More likely to respond quickly to change in policy More easily observable (suicide and accidental poisoning) More likely to be accurately recorded on death certificate Concordance of death codes across time is straightforward First look at aggregate deaths of despair Then look at each underlying cause separately
“Post” Difference-in-Differences Specification c=county; t=year Death Ratect = q 1{Post-PNTR} * NTR Gapc + b Xct + g 1{Post-PNTR} * Xc + dc + dt + a + ect DID term: post x county exposure Time-varying county attributes (e.g., NTR Tariff Rate, MFA exposure) Post x time-invariant county attributes (e.g., Change in Chinese import tariffs, education attainment) County and year fixed effects Notes Sample period 1990-2013 Standard errors clustered by county Economic significance: move a county from the 25th to the 75th percentiles of the NTR gap distribution, i.e., q*8.3
County-Level Controls Policy Labor-weighted average U.S. NTR tariff Labor-weighted average exposure to end of MFA/ATC quotas Labor-weighted average exposure to 1997-2002 changes in Chinese import tariffs Labor weighted average exposure to 1996-2005 reduction in Chinese production subsidies County attributes 1990 county median household income Spurious change in healthcare? 1990 county share of population with no college education Spurious change in technology? 1990 county share of population that are veterans Return of Gulf-War veterans?
Baseline Results: Aggregate Deaths of Despair Higher exposure to PNTR associated with higher mortality rates from deaths of despair
Baseline Results: Aggregate Deaths of Despair Higher exposure to PNTR associated with higher mortality rates from deaths of despair Larger declines in Chinese production subsidies associated with lower mortality (greater export access?)
Baseline Results: Aggregate Deaths of Despair Higher exposure to PNTR associated with higher mortality rates from deaths of despair Larger declines in Chinese production subsidies associated with lower mortality (greater export access?) Counties with higher shares of veterans or citizens with less education experience greater mortality in the 2000s versus the 1990s
Baseline Results: Aggregate Deaths of Despair Higher exposure to PNTR associated with higher mortality rates from deaths of despair Higher exposure to PNTR associated with higher mortality rates from deaths of despair Larger declines in Chinese production subsidies associated with lower mortality (greater export access?) Counties with higher shares of veterans or citizens with less education experience greater mortality in the 2000s versus the 1990s Economic significance: interquartile shift in NTR gap distribution associated with a 6-10% increase in mortality rates versus 2000
Baseline Results: Aggregate Deaths of Despair Higher exposure to PNTR associated with higher mortality rates from deaths of despair Higher exposure to PNTR associated with higher mortality rates from deaths of despair Larger declines in Chinese production subsidies associated with lower mortality (greater export access?) Counties with higher shares of veterans or citizens with less education experience greater mortality in the 2000s versus the 1990s Economic significance: interquartile shift in NTR gap distribution associated with a 6-10% increase in mortality rates versus 2000
Aggregate Deaths of Despair: Economic Significance Implied Impact of an Interquartile Shift in Exposure to PNTR Implied impact of an interquartile shift in county explosure to PNTR is an increase in “deaths of despair” mortality of ~1.5 100,000, or ~7 percent of average rate across counties in 2000.
Aggregate Deaths of Despair by Race and Gender Implied Impact of an Interquartile Shift in Exposure to PNTR 9% increase vs 2000 13% increase vs 2000
Why Whites, White Males? Share of Males, Whites and White Males in Manufacturing Employment, 1999 Source: www.bls.gov.
Aggregate Deaths of Despair for Whites, by Age
Disaggregate results for suicide, poisoning and ARLD
Implied Impact Within Deaths of Despair Positive and statistically significant associations except for alcohol related liver disease
Implied Impact Within Deaths of Despair All Races, Genders 4% increase vs 2000 28% increase vs 2000
Implied Impact Within Deaths of Despair White Males and Females Suicide Accidental Poisoning ARLD 5% increase vs 2000 4% increase vs 2000 23% increase vs 2000 37% increase vs 2000 6% increase vs 2000
Prior Trends?
“Annual” Difference in Differences Specification c=county; z=commuting zone; t=year DID term for own county DID term for surrounding counties Time-varying county attributes Initial (time-invariant) county attributes County and year fixed effects
Prior Trends and Timing: Whites 95% Confidence Interval of Estimated DID Coefficients
Outline Data DID empirical strategy and results 2SLS empirical strategy Robustness Conclusion
Comparison with Other Estimates Large literature examining cyclicality of mortality rates Ruhm (2000) finds that a 1sd increase in state unemployment rate is associated with 2.7% increase in suicide rate Here, use PNTR as IV for unemployment rate to compare economic magnitudes
First Stage: PNTR and Labor Market Outcomes DID Post Specification An interquartile shift in exposure to PNTR is associated with 0.4 percent decline in employment 28 percent increase in the unemployment rate 2.7 percent decline in the labor force participation rate 0.5 percent decline in per capita personal income
First Stage: PNTR and Unemployment Rate Annual Panel Specification
NTR Gap as IV for Unemployment Rate 1sd increase in u-rate implies 16% increase in suicide
NTR Gap as IV for Unemployment Rate 1sd increase in u-rate implies 16% increase in suicide 1sd increase in u-rate implies 46% increase in accidental poisoning
NTR Gap as IV for Unemployment Rate 1sd increase in u-rate implies 16% increase in suicide 1sd increase in u-rate implies 46% increase in accidental poisoning Why Do We Find Stronger Impact than Ruhm?
Why Do We Find Stronger Impact that Ruhm? Source: Pierce and Schott (2012)
Outline Data Empirical strategy and results 2SLS empirical strategy Robustness and other causes of death Conclusion
Robustness Exercises 10% increase vs 2000 8% increase vs 2000
Other Causes of Death Baseline “Post” DID Specification Percent of Year 2000 Death Rate
Outline Data Empirical strategy and results 2SLS empirical strategy Robustness and other causes of death Conclusion
Conclusions We examine whether the rise in deaths of despair among middle-age whites is related to a shock to labor markets induced by trade liberalization We find that counties more exposed to import competition from China via the change in policy exhibit increases in both suicides and accidental poisonings, particularly among whites and white males Exposure is also associated with persistent increases in unemployment rates and decreases in labor force participation Next step: use firm-worker level data to examine the labor-market shock more closelyThese trends coincide with persistent increases in the unemployment rate and declines in employment, labor force and personal income
Thanks!
End
appendix
Post-War U.S. Manufacturing Employment
Manufacturing versus Other Non-Farm Employment
Firm Margins of Adjustment Job creation Plant expansion within continuing firms Plant birth within continuing firms Firm birth Job Destruction Plant contraction within continuing firms Plant death within continuing firms Firm death
Net Margins of Adjustment
Gross Margins of Adjustment
Gross Margins of Adjustment
Re-Defining a Manufacturing Firm In the previous slides, a manufacturing firm in year t is defined as a firm with manufacturing activities in year t This definition does not catch firms before or after they switch into or out of manufacturing Broader definition Period: 1977-2012 Define an “ever-manufacturing” firm as a firm that is ever observe to have a manufacturing establishment during this period
Ever-vs Never-Manufacturing Firms 1977 2012 Manufacturing Employment 1987 1997 2007 Non-Manufacturing Employment 1977 2012 1987 1997 2007 Non-Manufacturing Employment 1977 2012 1987 1997 2007