Trade Integration and Political Radicalization: German Evidence from the Rise of Eastern Manufacturing Competition Christian Dippel (UCLA & NBER), Robert Gold (IfW Kiel), Stephan Heblich (Bristol) November 2014 1 / 31
Motivation 2 / 31
Trade Shocks and Radicalization I Recent success of far-right parties in Europe mostly driven by financial crisis, but there is an underlying trend that (a) started in the late 1980s and (b) has been attributed mostly to globalization A striking change in the political party systems of many established democracies in recent years has been the rise to electoral and political prominence of right-wing populist parties. Popular support for these parties is associated with the job insecurity attributed to deepening international economic integration, or economic globalization. (Source: Mughan et. al. Economic globalization, job insecurity and the populist reaction, Electoral Studies, 2003) Example: Austria s Freedom Party s vote-share went from 5% in 1983 to 27% in 1999 3 / 31
Trade Shocks and Radicalization II In Germany, there are three far-right parties - NPD, DVU, Reps NPD declaration (Dec 2000): The described state of affairs is unbearable, and political opposition to it unavoidable. This is opposition against globalization. Sociologists list these themes (Stoess 2010) as uniting far-right parties/voters since the 1990s: nationstate-less predator capitalism global dictatorship of big money future instead of globalization driven by zionist lobby 4 / 31
This Paper This paper examines the effect of globalization in trade, specifically increasing import competition from Asian and Eastern European manufacturing on voting for far-right parties in Germany 1987 2009 In a nutshell: Find a very robust effect The effect appears to work entirely through labor market channels At the mean, the effect is small, but this is because Germany largely benefitted from globlization For a country like France, the effect could be sizeable We also conduct a placebo analysis of East Germany from 1990 1998 5 / 31
Framework Eastern Europe and China were the main sources of manufacturing competition for Germany Use mapping of industry j imports into region i (Autor Dorn Hanson 2013): Import-Shock i = j=1 L ij L j M j L i Similar to the shift-share empirical approach (Bartik 1991, Notowidigdo 2013). But sector-j imports are more exogenous than the L j used in standard shift-share May nonetheless be confounded by domestic sectoral demand: Therefore instrument M j with other countries imports from same source 6 / 31
Trade Shocks Spatial dispersion of the import-shock: 7 / 31
Estimation Framework Main outcome: FarRight it = α 1 + β 1 NetImports Git + X itγ 1 + τ 1 tr + ɛ it instrumented with: NetImports Git = α 2 + β 2 TradeShock Oit + X itγ 2 + τ 2 tr + ɛ i To unpack job-insecurity channel: LaborMarket it = α 3 + β 3 NetImports Git + X itγ 3 + τ 3 tr + ɛ it 8 / 31
Local Labor Markets The local labor market literature: if people are immobile, local (i.e., sub-national) labor markets may not clear in response to negative labor demand shocks (Topel [1986], Glaeser and Gyourko [2005]) We focus on Germany 1987-2009, where there is a multi-party system and three established far-right parties DVU, NPD, Reps regularly monitored by the national intelligence services (Verfassungsschutz). Unbalanced panel of stacked first differences: West 87 98, 98 09, East 98 09 Separate analysis of East 1990 98 9 / 31
Data I Election Data: Information on national elections for 12, 000 municipalities for elections {1987, 1990, 1994, 1998, 2002, 2005, 2009} From hardcopies: Voting for all parties, including those with a vote share < 5% Employment Data: Sectoral Employment on the municipality level (place of residence) from the 1987 population census and for later years the IAB-Establishment History Panel. Info on population of employees subject to social security. Data also include information on population and structure of workforce (age, sex, nationality, qualification, sector). 10 / 31
Data II Trade Data: international trade statistics from the UN Commodity Trade Statistics Database (Comtrade). Information on trade volumes (in $) detailed by commodities between over 170 partner countries. Data allow distinction between 1,029 SITC product codes that can be mapped ( concorded ) into SIC industry codes. Instrument change in Germany s sector-j imports, M j, with that of 8 other, similar countries 11 / 31
TradeShock Oit NetImports Git (Left Panel) NetImports Git FarRight it (Right Panel) 12 / 31
Core Result: FarRight it on NetImports Git Outcome: Change Far-Right Voteshare it (1) (2) (3) (4) IV IV IV IV Change NetImports Git 0.119*** 0.109*** 0.114*** 0.101** (3.406) (2.973) (2.831) (2.425) Share Manuf-Employment t-1 0.015*** 0.018*** 0.017*** 0.019** (3.687) (3.436) (3.047) (2.505) (5) (6) (7) (8) OLS OLS OLS OLS Change NetImports Git 0.038* 0.038 0.036 0.009 (1.700) (1.608) (1.462) (0.419) R-squared 0.801 0.804 0.804 0.841 Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). All specifications inlude period specific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 13 / 31
First Stage Outcome: Change NetImports Git (1) (2) (3) (4) FS FS FS FS TradeShock Oit 0.248*** 0.251*** 0.241*** 0.240*** (8.221) (8.514) (7.966) (8.172) Share Manuf-Employment t-1-0.035*** -0.035*** -0.027*** -0.017** (5.074) (5.466) (3.683) (2.104) R-squared 0.478 0.517 0.52 0.535 Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). All specifications inlude periodspecific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 14 / 31
Replication of ADH: ManufEmpl it on NetImports Git Outcome: Change Share Manuf-Employment it (1) (2) (3) (4) IV IV IV IV Change NetImports Git -0.464** -0.731*** -0.752*** -0.678*** (2.089) (3.785) (3.591) (3.360) Share Manuf-Employment t-1-0.177*** -0.115*** -0.113*** -0.129*** (9.404) (4.077) (4.099) (4.495) (5) (6) (7) (8) OLS OLS OLS OLS Change NetImports Git -0.550*** -0.566*** -0.561*** -0.489*** (3.594) (3.743) (3.645) (3.415) R-squared 0.499 0.441 0.443 0.196 Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). All specifications inlude period specific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 15 / 31
log(employment) it on NetImports Git Outcome: Change in Log(Employment) (1) (2) (3) (4) (5) (6) (7) (8) IV IV IV IV OLS OLS OLS OLS Change NetImports Git -0.023*** -0.026*** -0.025*** -0.023*** -0.013*** -0.012*** -0.010** -0.010** (2.849) (3.496) (3.275) (3.325) (3.052) (2.794) (2.393) (2.399) Employm-share in manufacturing -1 Pop-share foreign-born -1 Employm-share among women -1 Unemployment-share -1 Pop-share working age -1-0.004*** -0.002* -0.002** -0.001-0.003*** -0.001-0.002** -0.001* (5.575) (1.829) (2.157) (1.479) (6.029) (1.363) (2.149) (1.676) -0.009*** -0.009*** -0.009*** -0.008** -0.008** -0.008** (2.615) (2.634) (2.617) (2.080) (2.164) (2.084) -0.044*** -0.043*** -0.005-0.044*** -0.043*** -0.004 (7.084) (6.892) (0.568) (7.093) (6.757) (0.455) -0.006*** -0.006*** (5.139) (5.062) -0.010*** -0.012*** (2.774) (3.050) Observations 730 730 730 730 730 730 730 730 R-squared 0.345 0.442 0.447 0.524 Note: The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states that are Kreise (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). 16 / 31
Steps FarRight it Have shown: FarRight it NetImportsGit NetImportsGit TradeShockOit TradeShockOit and: NetImports Git TradeShockOit ManufEmpl NetImports Git it TradeShockOit Now, unpack these: FarRight it ManufEmpl it NetImportsGit TradeShockOit ManufEmpl it 17 / 31
Partial Reduced Form I Ideally, would like to estimate: FarRight it = α 4 +β 4 NetImports Git +λ 8 ManufEmpl it +X itγ 4 +τ 4 tr +ɛ it But this has two endogenous regressors. Instead do: FarRight it = α 8 +β 8 TradeShock Oit +λ 8 ManufEmpl it +X itγ 8 +τ 8 tr +ɛ it 18 / 31
Partial Reduced Form II Outcome: Change Far-Right Voteshare it (1) (2) (3) (4) Reduced Form, "Total Effect" TradeShock Oit 0.122*** 0.098** 0.096** 0.094* (3.038) (2.304) (2.082) (1.968) (5) (6) (7) (8) Reduced Form "Partial Effect," Control for Change Manuf.-Share TradeShock Oit 0.106*** 0.053 0.054 0.062 (2.673) (1.328) (1.202) (1.338) Change Share Manuf-Employment it -0.050*** -0.068*** -0.068*** -0.059*** (3.977) (4.254) (4.234) (3.812) Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). All specifications inlude period specific region fixed effects. *** 19 / 31
Partial Reduced Form III Outcome: Change Far-Right Voteshare it (1) (2) (3) (4) Reduced Form, "Total Effect" TradeShock Oit 0.122*** 0.098** 0.096** 0.094* (3.038) (2.304) (2.082) (1.968) (5) (6) (7) (8) Reduced Form "Partial Effect," Control for Change Total Employment TradeShock Oit 0.090** 0.046 0.05 0.045 (2.422) (1.142) (1.093) (1.027) Change in Log(Employment) -1.306*** -1.714*** -1.721*** -2.035*** (4.057) (4.631) (4.635) (5.284) period-fe * region FE Yes Yes Yes Yes Observations 730 730 730 730 Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). All specifications inlude period specific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 20 / 31
Mechanisms 2 Instead, try to estimate this system in IV with two instruments NetImports Git = α 5 + β 5 TradeShock Oit + λ 5 Z it + X itγ 5 + τtr 5 + ɛ it, ManufEmpl it = α 6 + β 6 TradeShock Oit + λ 6 Z it + X itγ 6 + τtr 6 + ɛ it Equations (7) (9) FarRight it 4 NetImportsGit 5 TradeShockOit 4 6 5 ManufEmpl it 6 ManufEmpl it 1 21 / 31
2 instrumented regressors I Outcome: Change Far-Right Voteshare (1) (2) (3) (4) IV, with both endogenous regressors instrumented Change NetImports Git 0.080** -0.003 0.005 0.011 (2.114) (0.058) (0.090) (0.226) Change Share Manuf-Employment it -0.087*** -0.151*** -0.146*** -0.133** (3.477) (2.623) (2.595) (2.528) First Stage, Outcome: Change Share Manuf-Employment it TradeShock Oit -0.322-0.658*** -0.612** -0.547** (1.237) (2.796) (2.537) (2.498) Employm-share in manufacturing -1-0.159*** -0.090*** -0.098*** -0.124*** (7.029) (3.294) (3.503) (4.118) Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East. All s.e. clustered at the level of the IAB's 50 commuting zones "Arbeitsmarktregion". All specifications inlude period specific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 22 / 31
2 instrumented regressors II Outcome: Change in Far-Right Voteshare (1) (2) (3) (4) IV, with both endogenous regressors instrumented Change NetImports Git 0.027 0.211 0.122 0.191 (0.797) (1.245) (0.923) (1.063) Change in Log(Employment) 4.034*** 12.235** 9.536** 13.018** (3.723) (2.151) (2.379) (1.989) First Stage, Outcome: Change in Log(Employment) TradeShock Oit 0.024** 0.031*** 0.027*** 0.024*** (2.273) (3.181) (2.747) (2.765) 0.003*** 0.001 0.001 0.001 Employm-share in manufacturing -1 (5.171) (0.771) (1.480) (1.110) Observations 730 730 730 730 Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East. All s.e. clustered at the level of the IAB's 50 commuting zones "Arbeitsmarktregion". All specifications inlude period specific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 23 / 31
Three Staged Least Squares The previous exercise identifies a LATE but not the one we want. Its main purpose is to see if we can impose: Equations (10) (12) FarRight it NetImportsGit 9 TradeShockOit 7 8 ManufEmpl it FarRight it = α 7 + β 7 ManufEmpl it + X itγ 7 + τtr 7 + ɛ it (1) ManufEmpl it = α 8 + β 8 NetImports Git + X itγ 8 + τtr 8 + ɛ(2) it NetImports Git = α 9 + β 9 TradeShock Oit + X itγ 9 + τtr 9 + ɛ(3) it 24 / 31
3SLS for a double-iv I Three Stage Least Squares ("Double-IV") (1) (2) (3) (4) Third Stage, Outcome: Change NetImports Git Predicted Change Share Manuf-Employment it -0.238*** -0.173*** -0.176*** -0.155*** (3.683) (3.448) (3.415) (3.002) Second Stage, Outcome: Change Share Manuf-Employment it Predicted Change NetImports Git -0.471*** -0.723*** -0.743*** -0.676*** (4.082) (6.384) (6.121) (5.675) First Stage, Outcome: Change NetImports Git TradeShock Oit 0.248*** 0.251*** 0.241*** 0.240*** (8.221) (8.514) (7.966) (8.172) Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). All specifications inlude period specific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 25 / 31
3SLS for a double-iv II Three Stage Least Squares ("Double-IV") (1) (2) (3) (4) Third Stage, Outcome: Change NetImports Git Predicted Change Log(Employment) -5.798*** -4.616*** -5.343*** -4.953*** (3.102) (2.689) (2.742) (2.587) Second Stage, Outcome: Change in Log(Employment) Predicted Change NetImports Git -0.023*** -0.026*** -0.024*** -0.022*** (4.968) (5.863) (5.149) (5.010) First Stage, Outcome: Change NetImports Git TradeShock Oit 0.248*** 0.251*** 0.241*** 0.240*** (8.221) (8.514) (7.966) (8.172) Note: N = 730. The data is a stacked panel of first differences at the Kreis level. The panel comprises 322 Kreise in the West, observed in 1987 1998 and 1998 2009, and 86 Kreise in the East, observed in 1998 2009. We drop 4 city states (Hamburg, Bremen, Bremerhaven in the West, and Berlin in the East). All s.e. clustered at the level of the IAB's 50 commuting zones ("Arbeitsmarktregion"). All specifications inlude period specific region fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 26 / 31
Empirical Strategy East Germany We omit East Germany 1990 1998, but far-right tendencies have been particularly strong there We construct a comparable trade-integration shock using Balassa s (1965) Revealed Comparative Advantage RCA j RCA WG j = X j WG X WG / X j ROW X ROW East German region i exposure as employment-weighted measure of West Germany (WG) RCA, in 1990: RCA W i = j=1 L ij L i RCA WG j Intuition: Eastern regions exogenous productivity-weighted similarity to Western German industries at reunification. 27 / 31
Core Results in East 1990-1998 Outcome: Change Far-Right Voteshare 1990-1998 (1) (2) (3) (4) RCA i W 0.019** 0.012 0.016 (2.072) (1.144) (1.470) 0.001 0.001 0.001 Employm-share in manufacturing 1989 (0.106) (0.123) (0.143) R-squared 0.100 0.264 0.468 Outcome: Change Share Manuf-Empl 1990--1998 W RCA i -0.178* 0.348*** 0.226*** 0.191*** (-1.948) (5.668) (3.222) (3.101) Employm-share in manufacturing 1989-1.133*** -0.994*** -1.001*** (-12.743) (-12.880) (-12.688) Add. Controls Y Y State-FE Y R-squared 0.047 0.810 0.864 0.884 Note: N = 86 Kreise in the East, in period 1. *** p<0.01, ** p<0.05, * p<0.1. 28 / 31
Eastern Manufacturing I 29 / 31
Trade Shocks Spatial dispersion of the import-shock: 30 / 31
Discussion I Main Take: There is a significant causal effect of import-competition on far-right voting Aggregate effect is small: Germany did not see a right-shift over the full 20 years But Germany was unique in Europe in that its import-competition and export-access growth were offsetting Extension 1: Try to get a back-of-the envelope sense for shocks across OECD cross-section and match changes in other countries voting patterns 31 / 31
Discussion II Extension 2: Try to unpack far-right party supply responses from voter s demand for far-right content In the aggregate, check mainstream parties response (2000 s Kinder statt Inder ) Gather data on right actions (crimes) to gauge cheaptalk East Germany 1990s: Not shown today: Changes in manufacturing on 1989-manufacturing have R-squared of 0.9 This was not associated with rising far-right voteshare. Eastern Germany 1990s was special 32 / 31