Digitalization and the Future of Work
Macroeconomic consequences for tomorrow‘s employment,
unemployment and wages
The Future of Work
Social Situation Monitor / Research Seminar
Brussels
Melanie Arntz, Terry Gregory and Ulrich Zierahn
Centre for European Economic Research (ZEW), Mannheim
A „Jobless Future“?
Der Spiegel,17.4.1979
Der Spiegel,3.9.2016
“According to this report,
47% of all employees in the
USA work in occupations
that likely can be
automated within the next
10 to 20 years.”
“The experts are divided in
two camps. Some claim
that the flood is rising
rapidly and destroys 80
percent of jobs in 20 years.
The others think that this
result will only be achieved
later.”
Research Questions
1.
How many jobs are automatable?
Arntz/Gregory/Zierahn (2017): Revisiting the Risk of Automation, Economics Letters 159: 157-160.
2.
What are the employment effects of digitization and automation?
Arntz/Gregory/Zierahn, ongoing research project “Digitalization and the Future of Work”,
http://www.zew.de/PJ1312-1
Automation Potentials: Methodology
Occupation-based approach (e.g. Frey/Osborne, 2016)
•
Ask experts on automation potentials of selected occupations
•
Estimate statistical model: automation potential = f(occupational tasks)
Extrapolate automation potentials for all occupations
using model and occupation-level task data
Job-based approach (Arntz/Gregory/Zierahn, 2017)
•
Use existing occupation-level automation potentials
•
Estimate statistical model: automation potential = f(job-level tasks)
•
Multiple imputation approach due to occupation- vs. job-level
Extrapolate automation potentials for all jobs
using model and job-level task data
As comparison: extrapolate automation potentials for all occupations
Automation Potentials: Results
0
0.01
0.02
0.03
0
20
40
60
80
100
es
tim
ated
k
er
nel
d
en
sity
job-level approach
occupation-level approach
Jobs with high
automation potential
38%
Automation Potentials: Interpretation
•
Occupation-level approaches overestimate automation potentials
•
Why? Many workers specialize in non-automatable niches
•
Automation potential ≠ employment effects
•
Slow diffusion of technologies
•
Flexibility of workers
Employment Effects: Mechanisms
Lower labor
demand
Rising
competitiveness
Higher labor
demand
Technology
adoption
Machines replace
workers
Demand for
technologies
Higher labor
demand
Labor supply
Employment, unemployment, wages
Capital vs. labor
costs
Employment Effects: Methodology
• Machines complement/substitute
workers
Task Change
• Relative Prices
• Income Effect
• Capital Production
Product Demand
• Adjusting wages compensate
employment responses
Wage Adjustment
• Workers move to growing segments
Mobility
Structural Model: Mechanisms
• Technology Adoption
• Production, Costs
Establishment Survey
• (Un-)Employment
• Wages
• Occupational Mobility
Social Security Records
• Production
• Prices, Costs
Socio-Economic Accounts
• International Flows
• Sectoral Flows
WIOD
Data
Empirical
Estimation
Employment Effects: Preliminary Results
-0.8% -0.6% -0.4% -0.2% 0.0% 0.2% 0.4% 0.6% 0.8%
Metal Industry - Manual Routine
Electronics - Analytical
Yearly Growth Rate of Employment
-1.0%
-0.5%
0.0%
0.5%
1.0%
1.5%
Analytical
Interactive
Cognitive-Routine
Manual-Routine
Manual-Non-Routine
Yearly Growth Rate of Employment
-1.0%
-0.5%
0.0%
0.5%
1.0%
1.5%
Agriculture, Mining
Chemical Industry
Metal Industry
Electronics, Automotive Industry
Other Manufacturing
Hospitality Industry
Construction Industry
Trade, Maintanence
Transport and Communication
Business-Related Services
Energy/Water Supply
Education, Health, Social Welfare
Public Administration
Yearly Growth Rate of Employment
Conclusions
•
Occupation-level studies overestimate automation potentials
•
Automation potentials ≠ employment effects
•
Slow diffusion of technologies
•
Flexibility of workers
•
Macroeconomic adjustment mechanisms
•
Small net aggregate employment effects of technological change
•
Large restructuring (occupations, industries) due to technological change
•
Macroeconomic adjustment mechanisms play an important role
•
Key question is not how many jobs, but which jobs we will have
•
Are workers able to fill these jobs?
•
Rising inequality
Contact
Dr. Melanie Arntz
Dr. Terry Gregory
Dr. Ulrich Zierahn
Acting Head
Senior Researcher
Senior Researcher
Labor Markets, Human Resources and Social Policy
Centre for European Economic Research (ZEW)
L7, 1
68161 Mannheim
Tel.: +49 621 1235-159 / -306 / -280
Fax: +49 621 1235-225
Job-Level Approach (1)
•
Method
•
Estimate effect of job-level tasks (PIAAC) on occupation-level automation potentials (expert
assessments)
𝑎𝑎
𝑜𝑜
= �
𝑘𝑘=1
𝐾𝐾
𝛽𝛽
𝑘𝑘
𝑥𝑥
𝑘𝑘𝑘𝑘
+ 𝜖𝜖
𝑘𝑘
•
Predict job-level automation potentials
�𝑎𝑎
𝑘𝑘
= �
𝑘𝑘=1
𝐾𝐾
�
𝛽𝛽
𝑘𝑘
𝑥𝑥
𝑘𝑘𝑘𝑘
•
Predict occupation-level automation potentials
�𝑎𝑎
𝑜𝑜
= �
𝑘𝑘=1
𝐾𝐾
�
𝛽𝛽
𝑘𝑘
�𝑥𝑥
𝑘𝑘𝑜𝑜
•
Explanatory variables
Job-Level Approach (2)
•
Regression Model: 𝑎𝑎
𝑜𝑜𝑜𝑜
= ∑
𝑘𝑘=1
𝐾𝐾
𝛽𝛽
𝑘𝑘
𝑥𝑥
𝑘𝑘𝑘𝑘
+ 𝜖𝜖
𝑘𝑘𝑜𝑜
•
Measurement errors in a
oj
•
Job-level tasks vs. occupation-level automation potentials
•
Multiple-assignment-problem: occupational correspondence tables
•
Solution: Multiple-Imputation Approach (Ibrahim 1990) / EM Algorithm
•
1. Estimate Model
•
2. Re-calculate Weights
𝑤𝑤
𝑘𝑘𝑜𝑜
=
∑
𝑓𝑓( �𝑎𝑎
𝑖𝑖
−𝑎𝑎
𝑜𝑜𝑜𝑜
|𝑥𝑥
𝑖𝑖𝑖𝑖
,𝛽𝛽
𝑖𝑖
)
𝑜𝑜=1
𝐽𝐽
𝑓𝑓( �𝑎𝑎
𝑖𝑖
−𝑎𝑎
𝑜𝑜𝑜𝑜
|𝑥𝑥
𝑖𝑖𝑖𝑖
,𝛽𝛽
𝑖𝑖
)
•
Repeat until weights converge
Regression Results
Coef. Std. Err. p -0.034 0.012 0.006 age group 1: 16-19 0.020 0.039 0.614 0.161 0.038 0.000 0.027 0.038 0.477 0.053 0.039 0.168 0.100 0.038 0.009 0.093 0.039 0.018 0.067 0.039 0.083 0.077 0.041 0.061 0.190 0.041 0.000 education 1: low (ISCED 0, 1, 2)-0.255 0.035 0.000 -0.497 0.039 0.000 -0.002 0.000 0.000 0.001 0.000 0.030 -0.001 0.000 0.003 -0.168 0.013 0.000 firm size 1: 1-10 0.103 0.016 0.000 -0.006 0.022 0.794 0.135 0.013 0.000 -0.479 0.017 0.000 -0.044 0.013 0.001 -0.075 0.014 0.000 -0.270 0.014 0.000 -0.154 0.020 0.000 0.061 0.014 0.000 Model Results 2: 11-1000 3: >1000 responsibility for staff (0: yes; 1: no) educational job requirements (0: ISCED0-4; 1: ISCED 5-6)
payment scheme (0: piece/hourly wage, no wage; 1: monthly/yearly wage)
yearly income (percentile rank) not challanged enough (0: yes; 1: no) more training necessary (0: yes; 1: no) Variable
5: 35-39 6: 40-44 7: 45-49 8: 50-54 gender (0: male, 1: female)
w or ke r c ha ra ct er is ti cs sk ill s jo b ch ar act er is ti cs 9: 55-59 10: 60-65 2: 20-24 3: 25-29 4: 30-34
required job experience (0: <1 year; 1: >1 year)
2: medium (ISCED 3, 4, 5B) 3: high (ISCED 5A, 6) literacy (mean score)
numeracy (mean score) problem solving (mean score) sector (0: private; 1: public, non-profit)
Coef. Std. Err. p 0.806 0.220 0.000 -2.886 0.265 0.000 -4.884 0.386 0.000 2.782 0.235 0.000 0.482 0.255 0.059 -1.785 0.264 0.000 -2.052 0.278 0.000 -0.965 0.245 0.000 -4.522 0.258 0.000 0.237 0.266 0.373 -0.889 0.233 0.000 -1.395 0.285 0.000 -0.709 0.186 0.000 1.085 0.182 0.000 -1.492 0.192 0.000 -4.069 0.369 0.000 -4.670 0.340 0.000 0.261 0.267 0.330 -3.772 1.082 0.000 -0.808 0.203 0.000 -0.752 0.235 0.001 -1.365 0.495 0.006 -1.036 0.265 0.000 -4.515 0.568 0.000 -1.637 0.454 0.000 1.332 0.084 0.000 internet use for work-related info
using programming language using communication software constant
reading manuals writing articles filling forms
calculating shares or percentages complex math or statistics working physically for long using fingers or hands reading instructions
reading professional publications reading books Model Results Variable ta sk s ( sh ar e o f w or ki ng ti m e ( % ), s ee Ar nt z e t a l. 2 01 6 f or d et ai ls ) exchanging information training others presenting selling consulting
planning own activities panning activities of others organizing own schedule influencing
negotiating
solving simple problems solving complex problems
