Racing With or Against the Machine? Evidence from Europe

Racing With or Against the Machine?

Evidence from Europe

Terry Gregory^1,2 Anna Salomons³ Ulrich Zierahn^2,4

1IZA Bonn

2ZEW Mannheim

3Utrecht University

4CESifo Research Network

International Conference

The Socio-Economic Impact of Technological Change 30 November 2018, INAPP, Rome

Motivation

Der Spiegel, 2016

“47% of jobs are at risk”

Der Spiegel, 1978

“80% of jobs will be lost”

Motivation

Der Spiegel, 2016

“47% of jobs are at risk”

Der Spiegel, 1978

“80% of jobs will be lost”

Introduction

I False dichotomy (Acemoglu/Restrepo 2018)

I Alarmists in public press: “end of work”

I Technology only as complement to skilled labor

I Skill-Biased Technological Change (SBTC)

I No technological unemployment

I Rising wage inequality

I Routine Replacing Technological Change (RRTC)

I Computer capital replaces workers in routine tasks (Autor et al. 2003)

I Employment polarization – rooted in older sociological debate

I Declining labor demand possible in theory (several new papers)

I Whether labor races with or agains the machine is an empirical question

Introduction

I False dichotomy (Acemoglu/Restrepo 2018)

I Alarmists in public press: “end of work”

I Technology only as complement to skilled labor

I Skill-Biased Technological Change (SBTC)

I No technological unemployment

I Rising wage inequality

I Routine Replacing Technological Change (RRTC)

I Computer capital replaces workers in routine tasks (Autor et al. 2003)

I Employment polarization – rooted in older sociological debate

I Declining labor demand possible in theory (several new papers)

I Whether labor races with or agains the machine is an empirical question

Introduction

I False dichotomy (Acemoglu/Restrepo 2018)

I Alarmists in public press: “end of work”

I Technology only as complement to skilled labor

I Skill-Biased Technological Change (SBTC)

I No technological unemployment

I Rising wage inequality

I Routine Replacing Technological Change (RRTC)

I Computer capital replaces workers in routine tasks (Autor et al.

2003)

I Employment polarization – rooted in older sociological debate

I Declining labor demand possible in theory (several new papers)

I Whether labor races with or agains the machine is an empirical question

Introduction

I False dichotomy (Acemoglu/Restrepo 2018)

I Alarmists in public press: “end of work”

I Technology only as complement to skilled labor

I Skill-Biased Technological Change (SBTC)

I No technological unemployment

I Rising wage inequality

I Routine Replacing Technological Change (RRTC)

I Computer capital replaces workers in routine tasks (Autor et al.

2003)

I Employment polarization – rooted in older sociological debate

I Declining labor demand possible in theory (several new papers)

I Whether labor races with or agains the machine is an empirical question

Our Paper (1)

Research Question

I What is the net employment effect of RRTC (in Europe)?

What we do:

1. Develop task-based theoretical framework capturing the main mechanisms

I substitution of labor by capital

I product demand effect in tradables

I local demand spillovers to non-tradables

2. Estimate key parameters to analyze RRTC’s net employment effect (EU LFS data, 1999-2010, 238 regions in 27 EU countries) 3. Decompose total effect into three mechanisms

Our Paper (1)

Research Question

I What is the net employment effect of RRTC (in Europe)?

What we do:

1. Develop task-based theoretical framework capturing the main mechanisms

I substitution of labor by capital

I product demand effect in tradables

I local demand spillovers to non-tradables

2. Estimate key parameters to analyze RRTC’s net employment effect (EU LFS data, 1999-2010, 238 regions in 27 EU countries) 3. Decompose total effect into three mechanisms

Our Paper (1)

Research Question

I What is the net employment effect of RRTC (in Europe)?

What we do:

1. Develop task-based theoretical framework capturing the main mechanisms

I substitution of labor by capital

I product demand effect in tradables

I local demand spillovers to non-tradables

2. Estimate key parameters to analyze RRTC’s net employment effect (EU LFS data, 1999-2010, 238 regions in 27 EU countries)

3. Decompose total effect into three mechanisms

Our Paper (1)

Research Question

I What is the net employment effect of RRTC (in Europe)?

What we do:

1. Develop task-based theoretical framework capturing the main mechanisms

I substitution of labor by capital

I product demand effect in tradables

I local demand spillovers to non-tradables

2. Estimate key parameters to analyze RRTC’s net employment effect (EU LFS data, 1999-2010, 238 regions in 27 EU countries) 3. Decompose total effect into three mechanisms

Our Paper (2)

Contributions

I First estimate of RRTC’s net employment effect

I Complementing work on specific technologies (robots:

Acemoglu/Restrepo 2017, Chiaccio et al. 2018, Dauth et al. 2017, Graetz/Michaels 2018)

I Complementing work on TFP increases (Autor/Salomons 2018)

I Quantifying the mechanisms

I No consensus among reduced-form estimates

I Bridge btw. reduced-form estimates and theory

Preview of Results

I Net positive employment effect of RRTC in Europe from 1999-2010

I Strong substitution effects are overcompensated by product demand effects

I Total effect depends on where increasing capital income accrues

Our Paper (2)

Contributions

I First estimate of RRTC’s net employment effect

I Complementing work on specific technologies (robots:

Acemoglu/Restrepo 2017, Chiaccio et al. 2018, Dauth et al. 2017, Graetz/Michaels 2018)

I Complementing work on TFP increases (Autor/Salomons 2018)

I Quantifying the mechanisms

I No consensus among reduced-form estimates

I Bridge btw. reduced-form estimates and theory

Preview of Results

I Net positive employment effect of RRTC in Europe from 1999-2010

I Strong substitution effects are overcompensated by product demand effects

I Total effect depends on where increasing capital income accrues

Our Paper (2)

Contributions

I First estimate of RRTC’s net employment effect

I Complementing work on specific technologies (robots:

Acemoglu/Restrepo 2017, Chiaccio et al. 2018, Dauth et al. 2017, Graetz/Michaels 2018)

I Complementing work on TFP increases (Autor/Salomons 2018)

I Quantifying the mechanisms

I No consensus among reduced-form estimates

I Bridge btw. reduced-form estimates and theory

Preview of Results

I Net positive employment effect of RRTC in Europe from 1999-2010

I Strong substitution effects are overcompensated by product demand effects

I Total effect depends on where increasing capital income accrues

Our Paper (2)

Contributions

I First estimate of RRTC’s net employment effect

I Complementing work on specific technologies (robots:

Acemoglu/Restrepo 2017, Chiaccio et al. 2018, Dauth et al. 2017, Graetz/Michaels 2018)

I Complementing work on TFP increases (Autor/Salomons 2018)

I Quantifying the mechanisms

I No consensus among reduced-form estimates

I Bridge btw. reduced-form estimates and theory

Preview of Results

I Net positive employment effect of RRTC in Europe from 1999-2010

I Strong substitution effects are overcompensated by product demand effects

I Total effect depends on where increasing capital income accrues

Our Paper (2)

Contributions

I First estimate of RRTC’s net employment effect

I Complementing work on specific technologies (robots:

Acemoglu/Restrepo 2017, Chiaccio et al. 2018, Dauth et al. 2017, Graetz/Michaels 2018)

I Complementing work on TFP increases (Autor/Salomons 2018)

I Quantifying the mechanisms

I No consensus among reduced-form estimates

I Bridge btw. reduced-form estimates and theory

Preview of Results

I Net positive employment effect of RRTC in Europe from 1999-2010

I Strong substitution effects are overcompensated by product demand effects

I Total effect depends on where increasing capital income accrues

Parametric Framework

I Regional modeling approach

I Regions i = 1, 2, . . . , I

I Two sectors: non-tradable and tradable

I Local spillovers between sectors

I Exploit regional variation in exposure to RRTC ^map

I Routine Replacing Technological Change (RRTC)

I Declining costs of capital in routine relative to non-routine tasks

I RRTC only in tradable sector ^evidence

I Non-tradable sector affected indirectly through local spillovers

I Capturing within-EU27-trade (∼70% of total EU trade)

I Taking into account labor supply responses

Parametric Framework

I Regional modeling approach

I Regions i = 1, 2, . . . , I

I Two sectors: non-tradable and tradable

I Local spillovers between sectors

I Exploit regional variation in exposure to RRTC ^map

I Routine Replacing Technological Change (RRTC)

I Declining costs of capital in routine relative to non-routine tasks

I RRTC only in tradable sector ^evidence

I Non-tradable sector affected indirectly through local spillovers

I Capturing within-EU27-trade (∼70% of total EU trade)

I Taking into account labor supply responses

Parametric Framework

I Regional modeling approach

I Regions i = 1, 2, . . . , I

I Two sectors: non-tradable and tradable

I Local spillovers between sectors

I Exploit regional variation in exposure to RRTC ^map

I Routine Replacing Technological Change (RRTC)

I Declining costs of capital in routine relative to non-routine tasks

I RRTC only in tradable sector ^evidence

I Non-tradable sector affected indirectly through local spillovers

I Capturing within-EU27-trade (∼70% of total EU trade)

I Taking into account labor supply responses

Parametric Framework

I Regional modeling approach

I Regions i = 1, 2, . . . , I

I Two sectors: non-tradable and tradable

I Local spillovers between sectors

I Exploit regional variation in exposure to RRTC ^map

I Routine Replacing Technological Change (RRTC)

I Declining costs of capital in routine relative to non-routine tasks

I RRTC only in tradable sector ^evidence

I Non-tradable sector affected indirectly through local spillovers

I Capturing within-EU27-trade (∼70% of total EU trade)

I Taking into account labor supply responses

Production of Tradables

regional production (Y_i^g)

T2

T1 ... Tj

N₁^g K1 N₂^g K2 N_j^g Kj

Inputs (CD) Tasks (CES)

η η

I Firms combine tasks Tj to produce output Y_i^g

I Tasks T_j differ in their routine intensity

I Tasks require labor N_j^g and capital Kj – Labor Demand

Substitution effects

Decreasing capital costs lead to a substitution of labor by capital in routine tasks (in tradable sector) & a shift in production of tradables towards capital-intensive routine tasks;labor demand &

Production of Tradables

regional production (Y_i^g)

T2

T1 ... Tj

N₁^g K1 N₂^g K2 N_j^g Kj

Inputs (CD) Tasks (CES)

η η

I Firms combine tasks Tj to produce output Y_i^g

I Tasks T_j differ in their routine intensity

I Tasks require labor N_j^g and capital Kj – Labor Demand

Substitution effects

Decreasing capital costs lead to a substitution of labor by capital in routine tasks (in tradable sector) & a shift in production of tradables towards capital-intensive routine tasks;labor demand &

Consumption

utility (U)

Cg Cs

c2^g ... c_i^g c1^g

Regions (CES) Sectors (CD)

σ

I Households consume tradables (C_g) and non-tradables (C_s)

I Non-tradables are homogeneous and consumed locally

I Tradables are heterogeneous: each region i produces a bundle c_i^g –

Product Demand

Product demand effect

Falling capital costs lead to cheaper products, which leads to additional product demand in tradables;labor demand %

Consumption

utility (U)

Cg Cs

c2^g ... c_i^g c1^g

Regions (CES) Sectors (CD)

σ

I Households consume tradables (C_g) and non-tradables (C_s)

I Non-tradables are homogeneous and consumed locally

I Tradables are heterogeneous: each region i produces a bundle c_i^g –

Product Demand

Product demand effect

Falling capital costs lead to cheaper products, which leads to additional product demand in tradables;labor demand %

Non-Tradable Sector

I Firms produce using different types of labor (occupations)

I Labor demand depends on local income

I Non-tradable sector income (labor income only)

I Tradable sector income (labor income and firm profits)

I RRTC affects labor demand indirectly via disposable income in tradables Labor Demand

I Firm owners consume locally relax assumption

Product demand multiplier effect

Increasing demand and production leads to higher household incomes, which is partly spent on local non-tradables; demand for local

non-tradables %;labor demand %

Non-Tradable Sector

I Firms produce using different types of labor (occupations)

I Labor demand depends on local income

I Non-tradable sector income (labor income only)

I Tradable sector income (labor income and firm profits)

I RRTC affects labor demand indirectly via disposable income in tradables Labor Demand

I Firm owners consume locally relax assumption

Product demand multiplier effect

Increasing demand and production leads to higher household incomes, which is partly spent on local non-tradables; demand for local

non-tradables %;labor demand %

Labor Supply

Lij = ¯Lijw_ij^ (1)

Stylized approach following Acemoglu/Restrepo (2017)

I Wage elasticity of labor supply 

I Region-occupation wages respond to employment changes

I Identical to labor demand approach if  → ∞

Use macro-elasticity  from Chetty et al. (2011) to capture

I Extensive margin of labor supply

I Wage rigidities

Rising wages partly absorb demand effects (down-scaling)

Labor Supply

Lij = ¯Lijw_ij^ (1)

Stylized approach following Acemoglu/Restrepo (2017)

I Wage elasticity of labor supply 

I Region-occupation wages respond to employment changes

I Identical to labor demand approach if  → ∞

Use macro-elasticity  from Chetty et al. (2011) to capture

I Extensive margin of labor supply

I Wage rigidities

Rising wages partly absorb demand effects (down-scaling)

Labor Supply

Lij = ¯Lijw_ij^ (1)

Stylized approach following Acemoglu/Restrepo (2017)

I Wage elasticity of labor supply 

I Region-occupation wages respond to employment changes

I Identical to labor demand approach if  → ∞

Use macro-elasticity  from Chetty et al. (2011) to capture

I Extensive margin of labor supply

I Wage rigidities

Rising wages partly absorb demand effects (down-scaling)

Empirical Implementation

(A) Estimating labor demand (tradable sector):

log N_ijt^g =β0+ β1log Y_it^g+ β2log c_it^I + β3Rj× t + β4log wit+ θt+ υij+ ijt (2) (B) Estimating product demand:

log Y_it^g= δ0+ δ1log c_it^I + δ2log MPt+ νi+ εit (3)

data sources RTI market potential instruments

→ obtain parameters(1 − η)(1 − κ)γR= β3,η= β2,σ= −δ1, andη^N_w= −β4

(C) Decomposing total labor demand change (simplified):

∆Nit=(1 − η)(1 − κ)γR

| {z }

A















J

X

j =1

RjN_ijt^g + η 1 −ηR_it^IN_it^g

| {z }

B

− σ

1 −ηR_it^IN_it^g

| {z }

C

− σ

1 −ηR_it^IN_it^s

| {z }

D















→ substitution effect (AxB), product demand effect (AxC), multiplier (AxD)

Empirical Implementation

(A) Estimating labor demand (tradable sector):

log N_ijt^g =β0+ β1log Y_it^g+ β2log c_it^I + β3Rj× t + β4log wit+ θt+ υij+ ijt (2) (B) Estimating product demand:

log Y_it^g= δ0+ δ1log c_it^I + δ2log MPt+ νi+ εit (3)

data sources RTI market potential instruments

→ obtain parameters(1 − η)(1 − κ)γR= β3,η= β2,σ= −δ1, andη^N_w= −β4

(C) Decomposing total labor demand change (simplified):

∆Nit=(1 − η)(1 − κ)γR

| {z }

A















J

X

j =1

RjN_ijt^g + η 1 −ηR_it^IN_it^g

| {z }

B

− σ

1 −ηR_it^IN_it^g

| {z }

C

− σ

1 −ηR_it^IN_it^s

| {z }

D















→ substitution effect (AxB), product demand effect (AxC), multiplier (AxD)

Empirical Implementation

(A) Estimating labor demand (tradable sector):

log N_ijt^g =β0+ β1log Y_it^g+ β2log c_it^I + β3Rj× t + β4log wit+ θt+ υij+ ijt (2) (B) Estimating product demand:

log Y_it^g= δ0+ δ1log c_it^I + δ2log MPt+ νi+ εit (3)

data sources RTI market potential instruments

→ obtain parameters(1 − η)(1 − κ)γR= β3,η= β2,σ= −δ1, andη^N_w= −β4

(C) Decomposing total labor demand change (simplified):

∆Nit=(1 − η)(1 − κ)γR

| {z }

A















J

X

j =1

RjN_ijt^g + η 1 −ηR_it^IN_it^g

| {z }

B

− σ

1 −ηR_it^IN_it^g

| {z }

C

− σ

1 −ηR_it^IN_it^s

| {z }

D















→ substitution effect (AxB), product demand effect (AxC), multiplier (AxD)

Parameter Estimates

Parameter Description Estimate

(1 − η)(1 − κ)γR routinization coefficient (×100) -1.743***

(0.081)

η substitution elasticity between tasks 0.285**

(0.103)

−[(1 − κ) + κη] wage elasticity of labor demand -0.509**

(0.052)

κ labor share 0.689***

(0.136) γR annual log routine-replacing capital price change (×100) -7.833*

(4.441) σ substitution elasticity between bundles of tradables 0.862***

(0.166) Notes: Standard errors reported in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

labor demand labor demand: 1st stage labor demand: business cycle product demand product demand: 1st stage product demand: business cycle

Predicted European Employment Change, 1999-2010

-4.60 -1.64

3.97

1.41 5.64

2.01 5.00

1.79

-4-20246Million Jobs

Substitution Product Demand Spillover Net Effect

Labor Demand Employment

I Large substitution, overcompensated by product demand effects

I Effects partly absorbed by rising wages

I 1.79 Million jobs correspond to ∼7.7% of actual employment increase

employment in Europe robustness actual vs. predicted

Robustness to Parameter Estimates

Mean Std dev 5th pctile 95th pctile Labor demand change (in millions of jobs)

Substitution -4.71 0.83 -6.23 -3.61

Product Demand 4.08 1.07 2.57 6.00

Spillover 5.80 1.51 3.65 8.53

Net Effect 5.17 2.12 2.02 8.92

Employment change (in millions of jobs)

Substitution -1.68 0.33 -2.30 -1.24

Product Demand 1.44 0.30 1.01 1.98

Spillover 1.89 0.54 1.13 2.88

Net Effect 1.64 0.79 0.43 3.00

Notes: Distribution of predicted effects obtained by bootstrapping predictions with 10,000 draws. Bootstrap clustered by region-occupation for labor demand parameter estimates; and by region for the product demand parameter estimate.

Lower-Bound Predicted European Employment Change

-4.60 -1.64

3.97

1.41 3.14

1.69 2.51

1.47

-4-2024Million Jobs

Substitution Product Demand Spillover Net Effect

Labor Demand Employment

Racing With Rather Than Against the Machine

Key Results:

1. Positive net employment effect from RRTC in Europe

2. Product demand effects overcompensate substantial substitution 3. Size of spillovers depend on where gains accrue

Labor was racing with the machine

I Labor demand estimates exceed employment estimates ⇒ more jobs would have been created had labor supply been more elastic

I Focusing on substitution (potentials) alone is insufficient – compensating effects are quantitatively important

I Allocation of gains matters for employment effects – who owns the capital?

Racing With Rather Than Against the Machine

Key Results:

1. Positive net employment effect from RRTC in Europe

2. Product demand effects overcompensate substantial substitution 3. Size of spillovers depend on where gains accrue

Labor was racing with the machine

I Labor demand estimates exceed employment estimates ⇒ more jobs would have been created had labor supply been more elastic

I Focusing on substitution (potentials) alone is insufficient – compensating effects are quantitatively important

I Allocation of gains matters for employment effects – who owns the capital?

Thank you for your attention!

CESifo Working Paper No. 7247

Production of tradables - cont’d

I Firms minimize costs of producing Yi

min C_i=

J

X

j =1

c_it^TT_ij s.t.





J

X

j =1

(β_ijT_ij)^η−1η





η η−1

= Y_i^g (4)

I Regional task demand

Tij= Y_i^gβ^1−η_ij c_i^I w_ij^κr_j^1−κ

!^η

(5)

I falling relative capital costs for routine tasks incentivises firms to start shifting their tradable production towards these tasks

Go Back

Production of tradables - cont’d

I Firms minimize the costs of producing tasks Tij

min w_ijN_ij+ r_jK_ij s.t. T_ij(N_ij, K_ij) = N_ij^κK_ij^1−κ (6)

I Occupational labor demand (tradable sector)

N_ij^g= T_ij r_j wj

κ 1 − κ

1−κ

(7)

I falling capital costs for routine tasks induces firms to substitute capital for human labor in routine tasks

Go Back

Consumption - cont’d

I Households optimize the consumption of their goods bundles

max

" _I X

i =1

(c_i^g)^σ−1σ

#σ−1^σ

s.t.

I

X

i =1

τ_ii⁰p_i^gc_i^g = µI (8)

I Demand for regional tradables

c_i^g =

 τii⁰

p_i^g P^g

−σ

µ I

P^g (9)

I rises with falling relative prices for goods and services ^p

g i

P^g as well as with falling transport costs τ_ii0 to the extent that consumers can switch to tradables from other regions (σ)

I rises with households’ real income _P^Ig and with the share of income spent on these tradables µ.

Go Back

Non-tradable sector - cont’d

I Firms minimize costs of producing non-tradables C_i^s by minimizing the costs of obtaining the labor aggregate L^s_i

min L_iw_i^s =

J

X

j =1

w_ijN_ij s.t. L_i =





J

X

j =1

(β_ijN_ij)^η−1η





η η−1

(10)

I Occupational labor demand (non-tradable sector)

N_ij^s = (1 − µ)β_ij^s1−ηs wj

w_i^s

−η^s

Ii

w_i^s. (11)

where Ii= w_i^sL^s_i + φ1−KY_i^g and φ1−K = p_i^g−PJ

j =1rjKij/Y_i^g

I declines with occupational wages relative to regional wages _w^wjs i to the extent that tasks can be substituted (η^s)

I rises with local income comprising of incomes from tradable and non-tradable sector

Go Back

Data sources

European employment

I European Labour Force Survey (EU LFS)

I 27 countries; time period: 1999-2010

I regional classification: mostly NUTS-2

I occupation classification: 1-digit ISCO-1988

I industry classification: 1-digit NACE rev. 1 Routine Task Intensity (RTI) of occupations

I US DOT 1977, calculation of RTI as in Autor & Dorn (2013), normalized as in Goos et al (2014)

Output, marginal costs

I OECD STAN database

I Output: total production

I Marginal costs: nominal production - nominal net operating surplus real production

Wage data

I Cambridge Econometrics European Regional Database (ERD)

Go Back

Total employment in Europe, 1999-2010

180190200210Employment (millions)

2000 2005 2010

Year

Note: Non-military, non-agricultural employment across 27 European countries.

I We observe a net job creation of 23 million jobs over 1999-2010

Map Go Back

European regional employment growth, 1999-2010

(2.0,28.2]

(0.4,2.0]

(-0.7,0.4]

(-2.6,-0.7]

[-21.1,-2.6]

No data

Go Back

Routine Task Intensity of European Employment

-.08-.06-.04-.020.02Routine Task Intensity Index

2000 2005 2010

Year

(0.07,0.27]

(0.03,0.07]

(-0.02,0.03]

(-0.09,-0.02]

[-0.25,-0.09]

No data

List of Regions Go Back

Most and least routine regions by country

Country Most routine Least routine

AT Vorarlberg Tyrol

BE Prov. Antwerpen Brussels Capital Region

CH Eastern Switzerland Z¨urich

CZ Central Moravia Prague

DE Bremen Berlin

DK Northern Denmark Capital Region (incl. Copenhagen)

ES La Rioja Extremadura

FI Western Finland Eastern Finland

FR Upper-Normandy ˆIle de France (incl. Paris)

GR Attica (incl. Athens) Thessaly

HU Western Transdanubia Central Hungary (incl Budapest) IE Border Midland & Western IE (incl. Dublin) Southern & Eastern Ireland

IT Lombardia Calabria

NL Limburg North-Holland (incl. Amsterdam)

NO Trøndelag Oslo & Akershus

PL Silesia Province Podlaskie Province

PT North Portugal Algarve

RO Central Romania North-Eastern Romania

SE Sm˚aland & the islands Stockholm

SI Eastern Slovenia Western Slovenia

SK Western Slovakia Bratislava Region

UK Northern Ireland Kent

Note: Routine intensity of employment obtained by weighting occupational RTI with 1999 occupational employment shares. Islands and small autonomous regions excluded.

Go Back

Tradability and ICT intensity of European industries

NACE Industry Classification Gini ICT-intensity

Level ∆

(1) (2) (3)

C Mining and quarrying Tradable 0.54 2.70 11.03

D Manufacturing Tradable 0.37 2.39 1.93

E Electricity, gas and water supply Tradable 0.27 5.65 4.09

F Construction Non-Tradable 0.16 0.45 0.26

G Wholesale and retail trade; repair of motor Non-Tradable 0.15 1.96 2.39 vehicles, motorcycles and personal and

household goods

H Hotels and restaurants Non-Tradable 0.21 0.42 0.28

I Transport, storage and communications Tradable 0.34 7.32 5.09

J Financial intermediation Tradable 0.30 9.51 11.56

K Real estate, renting and business activities Tradable 0.37 4.07 5.16 L Public administration and defense; compulsory Non-Tradable 0.10 0.95 1.49

social security

M Education Non-Tradable 0.10 0.72 1.13

N Health and social work Non-Tradable 0.10 0.67 1.79

O Other community, social and personal services Non-Tradable 0.10 1.58 1.99 activities

P Activities of private households as employers Non-Tradable 0.10 0.00 0.00 Note: Industries classified with NACE revision 1. Agriculture, Hunting and Forestry (NACE A);

Fishing (NACE B); and Extraterritorial Organisations and Bodies (NACE Q) have been excluded from the dataset. Change in ICT-intensity measured as change ICT capital services per hour worked, 1995 reference, over 1999-2007. Level of ICT-intensity measured in 1999.

Go Back

Occupational Routine Task Intensity (RTI) index

ISCO Occupation RTI

1 Legislators, senior officials and managers -0.94

2 Professionals -1.01

3 Technicians and associate professionals -0.28

4 Clerks 2.01

5 Service workers and shop and market sales workers -0.75

7 Craft and related trades workers 0.38

8 Plant and machine operators and assemblers 0.48

9 Elementary occupations 0.10

Notes: RTI standardized to have a zero mean and unit standard deviation across occupations. Armed forces (ISCO 6) and farming professionals (ISCO 0) have been excluded from the dataset.

Go Back

Construction of market potential

A region’s market potential = sum of GDP across all trading partners of the region, lowered by the trading costs to these partners.

Constructed as follows:

I Use trade flows between German regions to estimate an index of trade costs (based on a trade flow specification of product demand);

I Estimate the relationship between distance and trade costs for this sample;

I Use these estimates to calculate trade costs between all European regions.

Go Back

Instruments

(A) Estimating labor demand (in tradable sector):

I regional production (Y_it^g): regional net capital stock

I regional marginal costs (c_it^I): Bartik (1999) IV that relies on national variation in marginal costs over time (i.e. weight the national industry-specific marginal costs with the initial regional industry shares)

I regional wages: Bartik (1999) IV that relies on national variation in female employment over time (i.e. weight the national female employment share with the initial regional female employment shares)

(B) Estimating product demand:

I market potential (MPt): capital stock of all regions discounted by the transport costs towards these regions (using trade flow data)

I regional marginal costs (c_it^I): see above

Go Back

Labor demand in the tradable sector, controlling for wages

Dependent variable: log employment in tradable sector (in region-occupation-year cells)

FE FE FE-IV FE-IV FE-IV

Full sample Restricted sample with wages with IV wages

(1) (2) (3) (4) (5)

Standardized occupational RTI × timetrend -1.678*** -1.743*** -1.743*** -1.743*** -1.743***

(0.075) (0.092) (0.081) (0.081) (0.081)

Log regional gross production in tradables 0.748*** 0.744*** 0.745***

(0.075) (0.081) (0.082)

Log regional marginal cost index 0.416** 0.285** 0.324*

(0.129) (0.103) (0.164)

Log regional wage in tradables -0.507*** -0.357

(0.052) (0.590)

Number of observations 21,632 11,744 11,744 11,744 11,744

R-squared 0.980 0.982 0.176 0.195 0.193

Notes: European regions, 1999-2010. Models (1) and (2) include region-occupation and region-year fixed effects. Models (3), (4) and (5) are estimated with region-occupation fixed effects and control for a linear timetrend. Standard errors clustered by region reported in parentheses. Coefficients on RTI multiplied by 100.

Go Back

Labor demand in the tradable sector: 1st stage

FE-IV model 1 FE-IV model 2

First stage First stage First stage First stage Output Marginal costs Output Marginal costs

(1) (2) (3) (4)

Log regional net capital stock in tradables 0.536*** -0.013** 0.537*** -0.013**

(0.040) (0.004) (0.043) (0.004)

Log counterfactual regional marginal cost index -0.069 0.896*** 0.038 0.900***

(0.109) (0.027) (0.132) (0.027)

Number of observations 11,744 11,744 11,744 11,744

Sanderson-Windmeijer first-stage F-statistic 183.1 1,606.3 156.1 2,028.2 FE-IV model 3

First stage First stage First stage Output Marginal costs Wages

(5) (6) (7)

Log regional net capital stock in tradables 0.519*** -0.013** 0.003 (0.040) (0.004) (0.025) Log counterfactual regional marginal cost index -0.107 0.896*** -0.222 (0.102) (0.027) (0.174)

Female labor supply shock 0.717*** -0.007 -0.175*

(0.096) (0.011) (0.087)

Number of observations 11,744 11,744 11,744

Sanderson-Windmeijer first-stage F-statistic 26.6 3.1 4.4

Notes: European regions, 1999-2010. All models include region-occupation fixed effects and control for a linear timetrend. Standard errors clustered by region reported in parentheses. Coefficients on RTI multiplied by 100.

Go Back

Labor demand in the tradable sector

Dependent variable: log employment in tradable sector (in region-occupation-year cells)

FE FE FE-IV FE-IV FE-IV

Full sample Restricted sample with wages with IV wages

(1) (2) (3) (4) (5)

Standardized occupational RTI × timetrend -1.675*** -1.741*** -1.741*** -1.741*** -1.741***

(0.074) (0.092) (0.082) (0.082) (0.082)

Standardized occupational RTI × timetrend 0.000 0.000 0.000 0.000 0.000

× trough dummy (0.000) (0.000) (0.000) (0.000) (0.000)

Standardized occupational RTI × timetrend 0.000* 0.000 0.000 0.000 0.000

× peak dummy (0.000) (0.000) (0.000) (0.000) (0.000)

Log regional gross production in tradables 0.738*** 0.737*** 0.737***

(0.071) (0.077) (0.075)

Log regional marginal cost index 0.435** 0.281* 0.321

(0.133) (0.110) (0.165)

Log regional marginal cost index 0.046 0.053 0.051

× trough dummy (0.052) (0.045) (0.042)

Log regional marginal cost index 0.114** 0.158*** 0.146*

× peak dummy (0.042) (0.041) (0.058)

Log regional wage in tradables -0.506*** -0.376

(0.051) (0.505)

Number of observations 21632 11744 11744 11744 11744

R-squared 0.980 0.982 0.179 0.197 0.196

Notes: European regions, 1999-2010. Models (1) and (2) include region-occupation and region-year fixed effects. Model (3), is estimated with region-occupation fixed effects and controls for a linear timetrend. Standard errors clustered by region reported in parentheses. Coefficients on RTI multiplied by 100.

Go Back

Product demand in the tradable sector

Dependent variable: log regional production of tradables (in region-year cells)

FE FE-IV

(1) (2)

Log market potential 1.275*** 1.416***

(0.097) (0.116)

Log regional marginal cost index -0.653*** -0.862***

(0.130) (0.166)

Number of observations 1,904 1,904

R-squared 0.607 0.604

Notes: European regions, 2001-2010. All models are estimated with region- occupation fixed effects. Standard errors clustered by region reported in paren- theses.

Go Back

Product demand in the tradable sector: 1st stage

Dependent variable: log regional production of tradables (in region-year cells) First stage First stage Market potential Marginal costs

(1) (2)

Log regional net capital stock in tradable sector 1.284*** 0.050**

(0.038) (0.018)

Log counterfactual regional marginal cost index 0.235*** 0.914***

(0.033) (0.022)

Number of observations 1,904 1,904

Sanderson-Windmeijer first-stage F-statistic 751.5 769.1 Notes: European regions, 2001-2010. All models are estimated with region- occupation fixed effects. Standard errors clustered by region reported in parentheses.

Go Back

Product demand in the tradable sector: business cycle

Dependent variable: log regional production of tradables (in region-year cells)

FE FE-IV

(1) (2)

Log market potential 1.182*** 1.289***

(0.091) (0.103) Log industry marginal cost index -0.508*** -0.663***

(0.132) (0.156) Log industry marginal cost index × trough dummy 0.033 0.024

(0.044) (0.046) Log industry marginal cost index × peak dummy 0.047 0.026

(0.040) (0.042)

Number of observations 2,048 2,048

Notes: European regions, 2001-2010. All models are estimated with region- occupation fixed effects. Standard errors clustered by region reported in paren- theses.

Go Back

Actual vs. Predicted Employment Change

Dependent variable: actual regional employment-to-population change

(1) (2) (3) (4)

Predicted regional 0.714*** 0.426*** 0.480*** 0.411***

employment-to-population change (0.125) (0.083) (0.061) (0.072)

Number of observations 238 238 216 216

Sample All regions 5th-95th percentile

Fixed effects None Country None Country

R-squared 0.122 0.853 0.226 0.599

Notes: European regions, 1999-2010 long difference. All models are weighted by the region’s initial employment size in 1999. Models in columns 3 and 4 exclude regions with an actual employment-to-population change below the 5th and above the 95th percentile.

Go Back

Alternative parameter estimates

Parameter Description Estimate

A. Baseline B. Without wages C. Instrumenting wages (1 − η)(1 − κ)γ_R routinization coefficient (×100) -1.743*** -1.743*** -1.743***

(0.081) (0.081) (0.081)

η substitution elasticity between tasks 0.285** 0.416** 0.324*

(0.103) (0.129) (0.164)

−[(1 − κ) + κη] wage elasticity of labor demand -0.507** – 0.357

0.052) (0.590)

κ labor share 0.689*** 0.689*** 0.951

(0.136) (0.136) (1.066)

γ_R annual log routine-replacing -7.833* -9.597*** -52.467

capital price change (×100) (4.441) (2.167) (1150.072)

Notes: Estimates in Panels A, B, and C are obtained from model 4, 3, and 5, respectively, in Table 15. In panel B, κ is assumed to be equal to the estimate obtained from the baseline model (panel A). Standard errors reported in parentheses.

Decomposition: Alt. Labor Demand Estimates Decomposition: Alt. Product Demand Estimates Decomposition: Alt. Labor Supply Elasticity Go Back

Robustness check: alternative labor demand estimates

-1.64 -2.00-1.78

1.41 1.73 1.53

2.01 2.46

2.19 1.79

2.19 1.94

-2-10123Million Jobs

Substitution Product Demand Spillover Net Effect

baseline without wages

instrumenting wages

Go Back

Robustness check: alternative product demand estimates

-1.60 -1.96-1.72

1.46 1.79

1.57 2.08

2.54 2.24

1.94 2.37

2.09

-2-10123Million Jobs

Substitution Product Demand Spillover Net Effect

baseline without wages

instrumenting wages

Go Back

Robustness check: labor supply elasticity

012345Net Effect

0 2 4 6 8 10

epsilon

Employment Labor Demand

Go Back

Lower bound for local spillover

I Our model provides an upper bound for the local spillover since it assumes firm and capital owners spend their additional income from RRTC in the region where the firm operates and where the capital is used in production.

I Alternatively, we can obtain a lower bound by assuming that profits and capital income do not feed back into consumption at all (i.e.

local income is only composed of wage income).

I Deriving the new spillover effect and implementing this empirically, we find a lower bound for the spillover effect of 2.8 million jobs (i.e. a total effect of 1.9 million jobs and a multiplier of 0.32).

Go Back