Broadband Adoption, Productivity Dynamics and Labour Demand: Is There a Skill Bias in Italy?
Emanuela Ciapanna and Fabrizio Colonna
Bank of Italy
November 30, 2018
Introduction
Research Question
How does broadband (high speed) internet affect firm’s productivity labour productivity of skilled vs. unskilled workers (skill workforce composition)
Total Factor Productivity
Introduction
Research Question
How does broadband (high speed) internet affect firm’s productivity labour productivity of skilled vs. unskilled workers (skill workforce composition)
Total Factor Productivity
Introduction
The paper in a nutshell
What is BB? High Speed internet Technology (> 2Mbs) introduced in Italy by Telecom Italia in 1999 and rapidly covered virtually whole country;
Internet speed varies significantly across locations, even within the same municipality, due to technical characteristics
Exploit speedspatial heterogeneity to identify impact of internet speed connection on firms’ productivity
Micro data to estimate a nested CES Production Function;
Accounting for substituability/complementarity b/w skilled and unskilled labour
Introduction
The paper in a nutshell
What is BB? High Speed internet Technology (> 2Mbs) introduced in Italy by Telecom Italia in 1999 and rapidly covered virtually whole country;
Internet speed varies significantly across locations, even within the same municipality, due to technical characteristics
Exploit speedspatial heterogeneity to identify impact of internet speed connection on firms’ productivity
Micro data to estimate a nested CES Production Function;
Accounting for substituability/complementarity b/w skilled and unskilled labour
Introduction
The paper in a nutshell
What is BB? High Speed internet Technology (> 2Mbs) introduced in Italy by Telecom Italia in 1999 and rapidly covered virtually whole country;
Internet speed varies significantly across locations, even within the same municipality, due to technical characteristics
Exploit speedspatial heterogeneity to identify impact of internet speed connection on firms’ productivity
Micro data to estimate a nested CES Production Function;
Accounting for substituability/complementarity b/w skilled and unskilled labour
Introduction
The paper in a nutshell
What is BB? High Speed internet Technology (> 2Mbs) introduced in Italy by Telecom Italia in 1999 and rapidly covered virtually whole country;
Internet speed varies significantly across locations, even within the same municipality, due to technical characteristics
Exploit speedspatial heterogeneity to identify impact of internet speed connection on firms’ productivity
Micro data to estimate a nested CES Production Function;
Accounting for substituability/complementarity b/w skilled and unskilled labour
Introduction
The paper in a nutshell
What is BB? High Speed internet Technology (> 2Mbs) introduced in Italy by Telecom Italia in 1999 and rapidly covered virtually whole country;
Internet speed varies significantly across locations, even within the same municipality, due to technical characteristics
Exploit speedspatial heterogeneity to identify impact of internet speed connection on firms’ productivity
Micro data to estimate a nested CES Production Function;
Accounting for substituability/complementarity b/w skilled and unskilled labour
Introduction
Preview of the results
High-speed internet firms:
Immediately
higher productivity of skilled workers higher share of skilled labor
In the long run
Higher levels of TFP Larger average firms’ size
Introduction
Preview of the results
High-speed internet firms:
Immediately
higher productivity of skilled workers
higher share of skilled labor In the long run
Higher levels of TFP Larger average firms’ size
Introduction
Preview of the results
High-speed internet firms:
Immediately
higher productivity of skilled workers higher share of skilled labor
In the long run
Higher levels of TFP Larger average firms’ size
Introduction
Preview of the results
High-speed internet firms:
Immediately
higher productivity of skilled workers higher share of skilled labor
In the long run
Higher levels of TFP Larger average firms’ size
Introduction
Preview of the results
High-speed internet firms:
Immediately
higher productivity of skilled workers higher share of skilled labor
In the long run
Higher levels of TFP
Larger average firms’ size
Introduction
Preview of the results
High-speed internet firms:
Immediately
higher productivity of skilled workers higher share of skilled labor
In the long run
Higher levels of TFP Larger average firms’ size
Outline
1 Reference Literature
2 Data and Identification Strategy
3 Stylized facts on labor demand
4 Empirical Model: firms’ performance
5 Results
6 Concluding Remarks and way forward
Reference Literature
1 Effects of ICT in general on labor market outcomes (Acemoglu and Autor, 2011; Autor and Dorn,2013; Michaels, Natraj, and Van Reenen, 2014)
2 Specific effects of internet BB on economic performance and labor market outcomes (Czernich et al., 2011; Akerman et al., 2015)
3 Empirical Strategy: IV distance from the Exchange (Falck et al., 2014;Billari et al., 2017) and CES production function (Kasahara et al., 2013)
Reference Literature
1 Effects of ICT in general on labor market outcomes (Acemoglu and Autor, 2011; Autor and Dorn,2013; Michaels, Natraj, and Van Reenen, 2014)
2 Specific effects of internet BB on economic performance and labor market outcomes (Czernich et al., 2011; Akerman et al., 2015)
3 Empirical Strategy: IV distance from the Exchange (Falck et al., 2014;Billari et al., 2017) and CES production function (Kasahara et al., 2013)
Reference Literature
1 Effects of ICT in general on labor market outcomes (Acemoglu and Autor, 2011; Autor and Dorn,2013; Michaels, Natraj, and Van Reenen, 2014)
2 Specific effects of internet BB on economic performance and labor market outcomes (Czernich et al., 2011; Akerman et al., 2015)
3 Empirical Strategy: IV distance from the Exchange (Falck et al., 2014;Billari et al., 2017) and CES production function (Kasahara et al., 2013)
Data
Invind Dataset, representative sample of 20+ italian firms (1995-2017):
Balance sheet Information (Cerved)
Workforce Information, administrative data (INPS)
Unique Dataset of Top Speed available at firm’s address (Address point geocoding) provided by TIM
Data
Invind Dataset, representative sample of 20+ italian firms (1995-2017):
Balance sheet Information (Cerved)
Workforce Information, administrative data (INPS)
Unique Dataset of Top Speed available at firm’s address (Address point geocoding) provided by TIM
Data
Invind Dataset, representative sample of 20+ italian firms (1995-2017):
Balance sheet Information (Cerved)
Workforce Information, administrative data (INPS)
Unique Dataset of Top Speed available at firm’s address (Address point geocoding) provided by TIM
Broadband ADSL Technology
Broadband ADSL Technology
Copper degradation→Firm’s speed depends on
Broadband ADSL Technology
Copper degradation→Firm’s speed depends on
1 on phone exchange speed
Broadband ADSL Technology
Copper degradation→Firm’s speed depends on
1 on phone exchange speed
Broadband xDSL Technology
Figure:The speed capacity-distance trade-off
Data
Data
(a)Local Exchanges, source: TIM (b)Firms, source: INPS-Invind, 2014 Figure:Distribution of local telephone exchanges and sample firms, Italy.
Different speeds for different tasks
Table:Example: BB capacity needs for a 50 employees firm in hospitality as of 2018
Task Speed (Mbps)
Web 8
email 5
Video conference 10 Share file services 4
Cloud storage 5
Cloud Office productivity 3
HD Imaging 2.5
Remote/ virtual Desktop 6.4
Total 51.9 Mbps Download/45.2 Upload
More BB capacity for more complexity
Table:Example: BB-enabled Product and Process Innovation, Manufacturing Product Innovation
Low complexity High complexity Crowdsourcing 3D-Digital modeling Collaboration tools Digital simulation
Process Innovation
Low complexity High complexity Customer engagement Cloud based sales tools
Social media Customer/Supplier integration
Marketing Smart factory
Empirical strategy: Baseline Equation
BB speed and labor demand outcomes
Yit =βspeedit+γXit+ςt+ηit
where
speedit maximum BB speed at address point Xit controls and province and sector FE ςt is time FE
ηit firm-specific shock iid
Potential bias sources
1 Reverse causality: larger/more productive firms have higher BB speed
=⇒Look for exogenous supply shifter
IV is distance of facility from closest local exchange
Unique Dataset of geolocal position of the universe of exchanges provided by TIM
2 Omitted variable bias: firms having higher speed BB enjoy large network economies
=⇒Control for average local speed
Control for number of exchanges in a 3-km radius
Potential bias sources
1 Reverse causality: larger/more productive firms have higher BB speed
=⇒Look for exogenous supply shifter
IV is distance of facility from closest local exchange
Unique Dataset of geolocal position of the universe of exchanges provided by TIM
2 Omitted variable bias: firms having higher speed BB enjoy large network economies
=⇒Control for average local speed
Control for number of exchanges in a 3-km radius
Potential bias sources
1 Reverse causality: larger/more productive firms have higher BB speed
=⇒Look for exogenous supply shifter
IV is distance of facility from closest local exchange
Unique Dataset of geolocal position of the universe of exchanges provided by TIM
2 Omitted variable bias: firms having higher speed BB enjoy large network economies
=⇒Control for average local speed
Control for number of exchanges in a 3-km radius
Potential bias sources
1 Reverse causality: larger/more productive firms have higher BB speed
=⇒Look for exogenous supply shifter
IV is distance of facility from closest local exchange
Unique Dataset of geolocal position of the universe of exchanges provided by TIM
2 Omitted variable bias: firms having higher speed BB enjoy large network economies
=⇒Control for average local speed
Control for number of exchanges in a 3-km radius
Potential bias sources
1 Reverse causality: larger/more productive firms have higher BB speed
=⇒Look for exogenous supply shifter
IV is distance of facility from closest local exchange
Unique Dataset of geolocal position of the universe of exchanges provided by TIM
2 Omitted variable bias: firms having higher speed BB enjoy large network economies
=⇒Control for average local speed
Control for number of exchanges in a 3-km radius
Potential bias sources
1 Reverse causality: larger/more productive firms have higher BB speed
=⇒Look for exogenous supply shifter
IV is distance of facility from closest local exchange
Unique Dataset of geolocal position of the universe of exchanges provided by TIM
2 Omitted variable bias: firms having higher speed BB enjoy large network economies
=⇒Control for average local speed
Control for number of exchanges in a 3-km radius
Potential bias sources
1 Reverse causality: larger/more productive firms have higher BB speed
=⇒Look for exogenous supply shifter
IV is distance of facility from closest local exchange
Unique Dataset of geolocal position of the universe of exchanges provided by TIM
2 Omitted variable bias: firms having higher speed BB enjoy large network economies
=⇒Control for average local speed
Control for number of exchanges in a 3-km radius
IV: Identification Assumption
Exclusion restriction: distance from exchange influences firms’ outcome ONLY through BB capacity
Local exchange in Italy date back to 1945 as part of telephone network (BB was born in 1999)
they responded to a logic of "Universal service" for telephone Neer their number and location remains fixed over time
exclude firms born after 1999
IV: Identification Assumption
Exclusion restriction: distance from exchange influences firms’ outcome ONLY through BB capacity
Local exchange in Italy date back to 1945 as part of telephone network (BB was born in 1999)
they responded to a logic of "Universal service" for telephone Neer
their number and location remains fixed over time exclude firms born after 1999
IV: Identification Assumption
Exclusion restriction: distance from exchange influences firms’ outcome ONLY through BB capacity
Local exchange in Italy date back to 1945 as part of telephone network (BB was born in 1999)
they responded to a logic of "Universal service" for telephone Neer their number and location remains fixed over time
exclude firms born after 1999
IV: Identification Assumption
Exclusion restriction: distance from exchange influences firms’ outcome ONLY through BB capacity
Local exchange in Italy date back to 1945 as part of telephone network (BB was born in 1999)
they responded to a logic of "Universal service" for telephone Neer their number and location remains fixed over time
exclude firms born after 1999
Distance vs. density
IV Regression
Yit =βtspeedIVi+λdensi+γXit+ςt+νit
where
speedIVi is the first stage fit
densi is BB density, i.e. number of LTEs within a radius of 3Km from facility Xit controls and firm fixed effect
ςt is time FE
νit firm-specific shock iid
IV Regression
Yit =βtspeedIVi+λdensi+γXit+ςt+νit
where
speedIVi is the first stage fit
densi is BB density, i.e. number of LTEs within a radius of 3Km from facility Xit controls and firm fixed effect
ςt is time FE
νit firm-specific shock iid β
First Stage
Table:First Stage
Variable Download Speed
distance -1.567***
(0.053)
dist2 0.204***
(0.012)
Province fix. effect yes
Industry fix. effect yes
R-squared 0.187
F-statistics of instruments 52.7
Obs 2250
***p<0.01,**p<0.05,*p<0.1 Standards Errors in parentheses
First Stage
Labor outcomes
Average Size
Employment skill shares∼contractual qualifications
Blue Collars White Collars Managers
Labor outcomes
Average Size
Employment skill shares∼contractual qualifications
Blue Collars White Collars Managers
Labor outcomes
Average Size
Employment skill shares∼contractual qualifications Blue Collars
White Collars Managers
Labor outcomes
Average Size
Employment skill shares∼contractual qualifications Blue Collars
White Collars
Managers
Labor outcomes
Average Size
Employment skill shares∼contractual qualifications Blue Collars
White Collars Managers
Occupational tasks and professional qualifications
Figure:ISCO tasks by professional qualifications
Labor outcomes: differences between firms closer/farther
than 2 Km from LTE
Employment: Average size
Employment: Skill Shares
MDL of Cobb-Douglas production function with embedded CES aggregators
We extend the theoretical framework of Kasahara et al. (2013)
Yit =LαitlKitαk"ωit
Yit is VA of firm i at year t;
Lit and Kit are labor and capital inputs
"ωit is unobserved TFP
MDL of Cobb-Douglas production function with embedded CES aggregators
Labor is a composite input of skilled LS
and unskilled LU units:
Lit =
θitσ1LSitσ−σ1+ (1− θit)σ1LUit σ−σ1
σ−σ1
θit is relative skilled labor productivity (skilled labor share in the Cobb-Douglas case)
σis the elasticity of substitution between skilled and unskilled labour
2-step Estimation
1 Estimateσandθitfrom correlation b/w labor share and relative min wages
2 Estimateαl,αk andωusing Wooldridge or LP methodology
2-step Estimation
1 Estimateσandθitfrom correlation b/w labor share and relative min wages
2 Estimateαl,αk andωusing Wooldridge or LP methodology
First step
Given wages of skilled and unskilled labor WtS, WtU Rearranging FOC with respect to LU and LS:
ln
LUit LSit
=σln
WitS WitU
+ln
1
− θit
θit
we have
lit=σwit+ln
1
− θit
θit
+uit
uit firm-specific labour demand shock, unforeseen before t and independent of wi,t,θi,t
Industry-level minimum contractual wages
σ: skilled labor shares elasticity to skilled labour wage θit are residual
First step
Given wages of skilled and unskilled labor WtS, WtU Rearranging FOC with respect to LU and LS:
ln
LUit LSit
=σln
WitS WitU
+ln
1
− θit
θit
we have
lit=σwit+ln
1
− θit
θit
+uit
uit firm-specific labour demand shock, unforeseen before t and independent of wi,t,θi,t
Industry-level minimum contractual wages
σ: skilled labor shares elasticity to skilled labour wage θit are residual
First step
Given wages of skilled and unskilled labor WtS, WtU Rearranging FOC with respect to LU and LS:
ln
LUit LSit
=σln
WitS WitU
+ln
1
− θit
θit
we have
lit=σwit+ln
1
− θit
θit
+uit
uit firm-specific labour demand shock, unforeseen before t and independent of wi,t,θi,t
Industry-level minimum contractual wages σ
θit are residual
First step
Given wages of skilled and unskilled labor WtS, WtU Rearranging FOC with respect to LU and LS:
ln
LUit LSit
=σln
WitS WitU
+ln
1
− θit
θit
we have
lit=σwit+ln
1
− θit
θit
+uit
uit firm-specific labour demand shock, unforeseen before t and independent of wi,t,θi,t
Industry-level minimum contractual wages
σ: skilled labor shares elasticity to skilled labour wage
First step
Given
ln
LUit LSit
=σln
WitS WitU
+ln
1
− θit
θit
we have
lit=σwit+ln
1
− θit
θit
+uit
θit =βi+βt+βstspeedIV,i+βdtdensi+vit
vit firm-specific shock to TFP, unforeseen before t and independent of wi,t,θi,t
Impact of speed on (relatively) skilled labour productivity
First step
Given
ln
LUit LSit
=σln
WitS WitU
+ln
1
− θit
θit
we have
lit=σwit+ln
1
− θit
θit
+uit
θit =βi+βt+βstspeedIV,i+βdtdensi+vit
vit firm-specific shock to TFP, unforeseen before t and independent of wi,t,θi,t
Impact of speed on (relatively) skilled labour productivity
2nd Step
Pluggingσ,ˆ θˆin
Lit=
θitσ1LSitσ−σ1 + (1− θ
σ1
it )LUit σ−σ1
σ−σ1
and taking logs:
yit=αkkit+αllit+ωit
ωit=ξt+ξi+γωωit−1+γstspeedIV,i+γdtdensi+vit where
ξt,ξi are time and firm TFP shock,
speedit is available BB speed of firm i in year t,
uit firm-specific shock to TFP, unforeseen before t and independent ofξt
Results
Baseline High Speed year>2000 High Speed
& t>2000
σ 0.56 0.56 0.56 0.56
αl 0.45 0.03 0.02 0.12***
αk 0.13 0.05 0.02 0.01
Results
Results
Gains from High Speed
(2000-2007) Enhance higher skilled labour productivity Firms adjust skilled labour ratio
(2008-) Efficiency gain
Firms with Low-Speed had to adjust skill ratio by cutting employment Firms with High-Speed experienced TFP gains and kept employment level
Cobb Douglas vs. CES
Baseline High Speed year>2000 High Speed
& t>2000
CES
σ 0.56 0.56 0.56 0.56
αl 0.45 0.03 0.02 0.12***
αk 0.13 0.05 0.02 0.01
CD
σ 1 1 1 1
αl 0.46 -0.01 -0.11*** 0.02
αk 0.15 0.02 0.04 -0.05
Cobb Douglas vs. CES
Cobb Douglas vs. CES
CES vs Cobb-Douglas
With higher skilled/unskilled complementarity:
Output elasticity not fixed, depend on relative input ratio
Complementarity increases output elasticity to relative scarce input CD doesn’t fully disentagle output contribution of TFP and skill ratio
Concluding remarks and way forward
Effects of different BB capacity (not only availability of connection) among Italian firms on labor demand and firm performance
novel and rich data-set
technology-based IV: minimum distance firm-LTE
BB as a GPT leads to skill biased technological change (change in work organization, ICT adoption, process-digitization)
higher share of skilled labor higher TFP
next step: exogenous labor supply shifter to investigate different workers’
behavior
Thanks for the attention!
Introduction and Motivation
Once upon a time...
Return
Figure:Test of the first Italian automatic telephone exchange of Prati di Castello, Rome
Introduction and Motivation
Once upon a time...
’20s-’30s: Division of the telephone service into five zones, first duplex telephone, radio link experiments (STIPEL)
1940-’45: Heavy damage to infrastructure due to bombing: significant fall in subscriptions
1945-’50: Adoption of coaxial cables, radio links andnew generation of automatic telephone exchanges
1952: every Italian town has access to the national telephone network (Universal service principle)
1991: Activation of ISDN network for integrated digital data and voice transmission via telephone lines
1999: Introduction of ADSL technology for fast Internet connections
[
label=hs] Return
Employment: Hirings, by skill
Employment: Shares of Blue Collar on Hirings
Employment: Separation Rates
[
label=sd] Return
Figure:The speed capacity-distance trade-off