The Impact of Automation on the Unemployed Maarten Goos 1 Emilie - - PowerPoint PPT Presentation

The Impact of Automation on the Unemployed

Maarten Goos 1 Emilie Rademakers 2 Anna Salomons 1 Bert Willekens 2

1Utrecht University 2University of Leuven

May 29, 2018 Work in progress

1 / 39

Introduction

Table of Contents

1

Introduction

2

Data

3

Job search in labor markets with task overlap Empirical approach Estimates without task overlap across jobs Estimates with task overlap across jobs Robustness tests

4

The impact of automation on the unemployed

5

Conclusion

2 / 39

Introduction

Modeling automation

Katz&Murphy (92): Technological progress is factor-augmenting such that labor demand increases (under realistic parameter values). Autor&Acemoglu (11): Technological progress is task-replacing such that the automation of labor tasks leads to a decrease in labor demand and workers reallocate to different tasks based on comparative advantage. Acemoglu&Restrepo(16;18a,b): There are several countervailing effects that increase labor demand, especially the creation of new labor tasks.

3 / 39

Introduction

Adjustment costs from automation?

If automation changes the demand for jobs and tasks, the reallocation

• f workers to new jobs and tasks can be a complex and slow

process. Effects are visible in recent studies that focus on the adjustment of local labor markets to negative shocks in labor demand (e.g. Acemoglu&Restrepo (17); Autor, Dorn&Hanson (15)). However, little is known about a mismatch between workers’ task competencies and automation leading to slowdown in the adjustment of employment and wages and in productivity gains.

4 / 39

Introduction

Results preview

We build on Manning&Petrongolo’s (17) geographic search model to illustrate search in markets for detailed jobs that are linked by their task contents. We assume a negative shock to routine-task vacancies to capture automation. We find significantly longer unemployment durations for job seekers with routine-task competencies because an unemployed job seeker’s labor market is restricted to jobs for which she has all or most of the required task competencies.

5 / 39

Data

Table of Contents

1

Introduction

2

Data

3

Job search in labor markets with task overlap Empirical approach Estimates without task overlap across jobs Estimates with task overlap across jobs Robustness tests

4

The impact of automation on the unemployed

5

Conclusion

6 / 39

Data

Data

Main VDAB sample from an online job platform introduced by VDAB (Flemish PES) containing information about: Unemployed job seekers.

details

Job vacancies.

details

8 cross-sections from every quarter of 2013-2014. Auxiliary Social security records of unemployed job seekers in the VDAB sample. Bi-weekly (un)employment spells for 2010-2015. Gender, nationality, location, education.

7 / 39

Data

Measuring tasks

Unemployed job seekers have to complete a task-competency profile by indicating one or more occupation-experience cells (i.e. “jobs”) Use ROME-v3 (comparable to e.g. US O*NET):

3 experiences x 676 occupations or 2028 jobs which can be aggregated into ISCO88 occupation groups. ROME-v3 links 3489 tasks to these jobs.

Vacancies are linked to ROME-v3 by employers or VDAB.

8 / 39

Table 1: Shares of job seekers and vacancies across occupation groups

% first listed % all listed % vacancies (1) (2) (3) 01: armed forces 0.03 0.03 0.04 11: legislators and senior officials 0.18 0.22 0.43 12: corporate managers 5.56 5.55 18.97 13: general managers 0.08 0.10 0.11 21: physical, mathematical and engineering science professionals 0.83 0.82 3.38 22: life science and health professionals 0.18 0.17 0.49 23: teaching professionals 2.96 3.11 2.13 24: other professionals 4.95 4.54 5.05 31: physical and engineering science associate professionals 2.17 2.13 7.69 32: life science and health associate professionals 1.68 1.56 1.67 33: teaching associate professionals 0.00 0.01 0.01 34: other associate professionals 8.91 8.54 15.46 41: office clerks 9.43 9.28 5.92 42: customer services clerks 2.93 3.31 1.88 51: personal and protective services workers 7.82 7.94 3.69 52: models, salespersons and demonstrators 8.14 8.24 5.30 61: market-oriented skilled agricultural and fishery workers 1.35 1.41 0.35 71: extraction and building trades workers 5.21 5.26 5.08 72: metal, machinery and related trades workers 3.26 3.07 6.16 73: precision, handicraft, printing and related trades workers 0.59 0.52 0.19 74: other craft and related trades workers 0.86 0.87 1.11 81: stationary-plant and related operators 0.29 0.31 0.32 82: machine operators and assemblers 3.18 3.37 2.99 83: drivers and mobile-plant operators 5.53 5.94 3.55 91: sales and services elementary occupations 7.56 8.81 6.40 92: agricultural, fishery and related labourers 0.64 0.70 0.10 93: labourers in mining, construction, manufacturing and transport 15.69 14.21 1.52 N occupation-experience cells 1158 1460 877 N sample 17 493 17 493 11 228 N platform 229 535 229 535 70 407

Table 1: Shares of job seekers and vacancies across occupation groups

% first listed % all listed % vacancies (1) (2) (3) 01: armed forces 0.03 0.03 0.04 11: legislators and senior officials 0.18 0.22 0.43 12: corporate managers 5.56 5.55 18.97 13: general managers 0.08 0.10 0.11 21: physical, mathematical and engineering science professionals 0.83 0.82 3.38 22: life science and health professionals 0.18 0.17 0.49 23: teaching professionals 2.96 3.11 2.13 24: other professionals 4.95 4.54 5.05 31: physical and engineering science associate professionals 2.17 2.13 7.69 32: life science and health associate professionals 1.68 1.56 1.67 33: teaching associate professionals 0.00 0.01 0.01 34: other associate professionals 8.91 8.54 15.46 41: office clerks 9.43 9.28 5.92 42: customer services clerks 2.93 3.31 1.88 51: personal and protective services workers 7.82 7.94 3.69 52: models, salespersons and demonstrators 8.14 8.24 5.30 61: market-oriented skilled agricultural and fishery workers 1.35 1.41 0.35 71: extraction and building trades workers 5.21 5.26 5.08 72: metal, machinery and related trades workers 3.26 3.07 6.16 73: precision, handicraft, printing and related trades workers 0.59 0.52 0.19 74: other craft and related trades workers 0.86 0.87 1.11 81: stationary-plant and related operators 0.29 0.31 0.32 82: machine operators and assemblers 3.18 3.37 2.99 83: drivers and mobile-plant operators 5.53 5.94 3.55 91: sales and services elementary occupations 7.56 8.81 6.40 92: agricultural, fishery and related labourers 0.64 0.70 0.10 93: labourers in mining, construction, manufacturing and transport 15.69 14.21 1.52 N occupation-experience cells 1158 1460 877 N sample 17 493 17 493 11 228 N platform 229 535 229 535 70 407

Figure 1: Number of jobs in which the same task occurs

Data

Examples of tasks

Examples of most frequent tasks:

Basic maintenance and repair of machines or other equipment. Coordinating a team. Registration and dissemination of information related to production processes.

Examples of least frequent tasks:

Dismounting the equipment, structures (cells, ...) and hydraulic, pneumatic and electrical circuits of an aircraft. Developing the hospital policy for nursing. Performing a medical examination of the animal and assessing the therapeutic needs (medication, surgery, ...).

12 / 39

Table 2: Defining task overlap

(1) (2)

• ccupation-exp cell

task overlap = 8/ 11 Production worker, < 2 years (total n.

• f tasks=8 )
• add. tasks Packer, >5 years

ISCO88=93 ISCO88=93

• 1. Logging activity data (number of

pieces,...)

• 9. Check the products upon re-

ceipt, when completing the order

• r upon shipment
• 2. Transporting the products or waste

to the storage, shipping or recycling zone 10. Labelling the product, branding and checking the infor- mation (expiration date,...)

• 3. Providing the workstation with ma-

terials and products or checking the stock

• 11. Preventive or corrective ba-

sic maintenance of machines or equipment

• 4. Clearing and cleaning the work area

(materials, fittings,...) 5. Packaging products according to characteristics, orders and mode of transport 6. Fitting, assembling and attach- ment of pieces. Check that the as- sembly has been correct (use, view)

• 7. Monitoring the flow and progress of

products on a production or transport line 8. Detect and locate visible defect and sort them accordingly (surface, color,...)

Figure 2: Heatmap of task overlap across jobs

Data

Summary so far

Collected data for workers’ task competencies and job vacancies and defined task overlap across jobs. However, this is not informative about the importance of jobs and their task overlap for labor market outcomes. Given an unemployed job seeker’s task competencies, relate her job finding probability to tightness in job cells for which she has all or some of the required task competencies.

15 / 39

Job search in labor markets with task overlap

Table of Contents

1

Introduction

2

Data

3

Job search in labor markets with task overlap Empirical approach Estimates without task overlap across jobs Estimates with task overlap across jobs Robustness tests

4

The impact of automation on the unemployed

5

Conclusion

16 / 39

Job search in labor markets with task overlap Empirical approach

Empirical approach

Assume the following Weibull-hazard function for unemployed job seeker i to find a job: hi = kτ k−1

i

exp

• α1 ln( ˜

Vi) + α2 ln( ˜ Ui) + X

′

i δ

• Where:

kτk−1

i

captures duration dependence (if k < 1 it is negative). ln( ˜ Vi/ ˜ Ui) is the log of labor market tightness for i. α1 > 0 and α2 < 0 capture search externalities. Xi are person characteristics.

17 / 39

Job search in labor markets with task overlap Estimates without task overlap across jobs

Specification and estimates without task overlap

Assume the hazard function is: hi = kτ k−1

i

exp

• α1 ln (V all

i

) + α2 ln (Uall

i ) + X

′

i δ

• Where:

V all

i

and Uall

i

is the sum of vacancies and other job seekers across all jobs for which i has all of the required task competencies. Estimate odds ratio’s exp(α1) > 1 and exp(α2) < 1 using maximum likelihood accounting for right-censored unemployment cells.

18 / 39

Table 4: Hazard rates without task overlap across jobs

(1) (2) VARIABLES

• dds ratio
• dds ratio

lnV all

i

1.126*** 1.134*** (0.004) (0.004) lnUall

i

0.925*** 0.925*** (0.003) (0.003) Female 0.971*** (0.009) Belgian nationality: acquired 0.888*** (0.010) Foreign, EU nationality 0.822*** (0.014) Foreign, other nationality 0.915*** (0.014) Part-time 0.757*** (0.012) Part-time or full-time 0.915*** (0.009) Constant 0.433*** 0.398*** (0.001) (0.011) Observations 133,440 130,928 Location FE NO YES Time FE NO YES Weibull k 0.403 0.412

Estimated impacts of vacancies (a) Hazard Function (b) Survival Function

Job search in labor markets with task overlap Estimates with task overlap across jobs

Specification with task overlap

Assume the hazard function is: hi =kτ k−1

i

exp

 α1 ln (V all

i

+

4

• j=1

γjV some

ij

) +α2 ln (Uall

i

+

4

• j=1

βjUsome

ij

) + X

′

i δ

 

Where:

V all

i

and Uall

i

are defined as before. V some

i

and Usome

i

are vacancies and other job seekers in jobs for which i has some of the required task competencies. E.g. j=1: >80%, j=2: 60-79%, j=3: 30-59%, j=4: <30%.

21 / 39

Job search in labor markets with task overlap Estimates with task overlap across jobs

Estimates with task overlap

Assume the hazard function is: hi =kτ k−1

i

exp

 α1 ln (V all

i

+

4

• j=1

γjV some

ij

) +α2 ln (Uall

i

+

4

• j=1

βjUsome

ij

) + X

′

i δ

 

Where:

α1 > 0 and α2 < 0 capture search externalities when i’s labor market consists of jobs for which she has all or some of the required task competencies. γj > 0 and βj > 0 capture the importance of jobs with which there is some task overlap relative to jobs for which i has all the required task competencies.

22 / 39

Job search in labor markets with task overlap Estimates with task overlap across jobs

Estimates with task overlap: Linearization

If γj = 0 and βj = 0 the estimating equation is non-linear. Define Vi ≡ V all

i

+ V some

i1

and Uit ≡ Uall

i

+ Usome

i1

. Then α1 ln( ˜ Vi) + α2 ln( ˜ Ui) ≈ α1ln(Vi) + α2ln(Ui) + α1 1 − γ1 γ1 V all

i

Vi + α2 1 − β1 β1 Uall

i

Ui +α1

4

• j=2

γj γ1 V some

itj

Vit + α2

4

• j=2

βj β1 Usome

itj

Uit

23 / 39

Table 5: Hazard rates with task overlap across jobs

controls (1) (2) VARIABLES

• dds ratio
• dds ratio

lnVi 1.120*** 1.128*** (0.005) (0.005) lnUi 0.927*** 0.927*** (0.004) (0.004) V all

i

/Vi 1.235*** 1.266*** (0.031) (0.032) Uall

i /Ui

0.830*** 0.814*** (0.020) (0.020) V some

i2

/Vi 1.007 1.006 (0.005) (0.006) V some

i3

/Vi 1.000 1.001 (0.000) (0.000) V some

i4

/Vi 1.000** 1.000* (0.000) (0.000) Usome

i2

/Ui 1.014 1.015 (0.022) (0.022) Usome

i3

/Ui 1.002 1.002 (0.003) (0.003) Usome

i4

/Ui 1.000 0.999* (0.000) (0.000) Constant 0.420*** 0.411*** (0.011) (0.001) Observations 134,793 132,257 Controls NO YES Weibull k 0.403 0.411

Table 6: Imputed values for γj and βj

γj βj Specification: (1) (2) (1) (2) j = 1 0.3501*** 0.3388*** 0.2907*** 0.2695*** (0.0284) (0.0255) (0.0281) (0.0248) j = 2 0.0217 0.0177

• 0.0540
• 0.0525

(0.0163) (0.0156) (0.0815) (0.0782) j = 3 0.0013 0.0015

• 0.0082
• 0.0085

(0.0013) (0.0012) (0.0102) (0.0095) j = 4

• 0.0006**
• 0.0005*

0.0017 0.0020* (0.0003) (0.0003) (0.0011) (0.0011)

Job search in labor markets with task overlap Robustness tests

Robustness tests

Including education does not explain away the importance of (task

• verlap across) jobs.

estimates

Excluding task overlap with higher experience levels within

• ccupations driven by the additive structure of tasks between

experience levels does not affect our results.

estimates

No linearisation by estimating the effect of tightness in overlapping markets separately does not affect our results.

estimates 26 / 39

Summary of point estimates

The impact of automation on the unemployed

Table of Contents

1

Introduction

2

Data

3

Job search in labor markets with task overlap Empirical approach Estimates without task overlap across jobs Estimates with task overlap across jobs Robustness tests

4

The impact of automation on the unemployed

5

Conclusion

28 / 39

The impact of automation on the unemployed

The impact of automation on the unemployed

An unemployed job seeker’s labor market is restricted to those jobs for which she as have all or most of the required task competencies. If automation is replacing routine-labor tasks, unemployed durations for job seekers with routine-task compentencies are expected to increase by more. Predict the impact of automation on the unemployed by:

Defining routine tasks.

details

Assuming a negative shock in routine-task vacancies to capture automation.

details 29 / 39

Figure 5: Mean share of routine tasks by ISCO88 2-digit group

The impact of automation on the unemployed

Predicting changes in hazard rates

We had that ˆ

α1 ln( ˜ Vi) + ˆ α2 ln( ˜ Ui) ≈ 0.13ln(Vi) − 0.07ln(Ui) + 0.27V all

i

Vi − 0.19Uall

i

Ui +0.01V some

i2

Vi + 0.00V some

i3

Vi − 0.00V some

i4

Vi +0.02Usome

i2

Ui + 0.00Usome

i3

Ui − 0.00Usome

i4

Ui Assume a negative shock in routine-task vacancies to predict the change in the hazard rate for each unemployed job seeker. Unemployed job seekers with routine-task competencies will see a larger decrease in their hazard rates compared to a single market scenario.

31 / 39

The impact of automation on the unemployed

Single market scenario

Consider a single market scenario where tightness across job markets is equalized: ˆ α1 ln( ˜ Vi) + ˆ α2 ln( ˜ Ui) = 0.13 ln(V ) − 0.07 ln(U) This would be the case if unemployed job seekers are perfectly mobile across all jobs irrespective of their task competencies. Tightness is no longer individual specific and the impact of the shock is shared equally among all unemployed job seekers.

32 / 39

Figure 6: Cumulative density of predicted changes in hazard rates

Density of predicted changes in expected unemployment durations

The impact of automation on the unemployed

Mobility for job seekers with routine-task competencies

Assume that for unemployment job seekers with routine-task competencies mobility into jobs with task overlap is the same as into jobs for which they have all of the required task competencies. For unemployed job seekers with routine-task competencies, we replace ˆ

α1 ln( ˜ Vi) + ˆ α2 ln( ˜ Ui) =0.13 ln(V target

i

+ 1

n

• j=1

V some

ij

) − 0.07 ln(Utarget

i

+ 1

n

• j=1

Usome

ij

) This has only a limited impact on predicted changes in mobility.

35 / 39

Figure 7: Cumulative density of predicted change in hazard rate

Figure 2: Density of predicted changes in expected unemployment duration

Conclusion

Table of Contents

1

Introduction

2

Data

3

Job search in labor markets with task overlap Empirical approach Estimates without task overlap across jobs Estimates with task overlap across jobs Robustness tests

4

The impact of automation on the unemployed

5

Conclusion

38 / 39

Conclusion

Conclusions

Historical records suggest that adjustment costs for workers from technological progress can be significant, long-lasting and substantially reduced by policies (e.g. “Engle’s pause” in the 19th century). This paper is a simple illustration of the importance of search costs from ongoing automation for unemployed job seekers in markets for detailed jobs that are linked by their task contents. It would be interesting to build formal equilibrium models automation and the creation of new labor tasks in which search takes place in task space to address questions such as optimal search breadth for workers and firms or optimal training policies.

39 / 39

Sampling of unemployed job seekers

Table A1: Sampling of unemployed job seekers from the platform

Strata Sampling Percentage Newly registered in last month 20% 1-2 months unemployed 20% 2-3 months unemployed 20% 3-6 months unemployed 10% 6-12 months unemployed 10% More than 12 months unemployed 5%

40 / 39

Sampling of unemployed job seekers

back

Unemployed job seekers with shorter durations are oversampled from the platform because of stock sampling. The platform itself contains almost all unemployed job seekers:

Platform contains each unemployed job seeker claiming unemployment benefits. VDAB reports that claimants account for 75% of all unemployed job seekers registered in 2013-2014. This is in line with EULFS data showing that 14% of Flemish unemployed job seekers were not registered with VDAB in 2011-2015.

41 / 39

Sampling of job vacancies

Table A2: Sampling of job vacancies

Strata Sampling Percentage Vacancies that closed in the past month 50% All outstanding in the past month 5%

42 / 39

Sampling of job vacancies

back

Figure A1: Vacancy shares at the 3-digit ISCO88 level by 1-digit group

43 / 39

Table 5: Hazard rates with task overlap across jobs

back

(1) VARIABLES

• dds ratio

Female 0.971*** (0.009) Belgian nationality: acquired 0.886*** (0.010) Foreign, EU nationality 0.821*** (0.014) Foreign, other nationality 0.911*** (0.014) Part-time 0.759*** (0.012) Part-time or full-time 0.917*** (0.009) Constant 0.385*** (0.014) Observations 132,257 Weibull k 0.411

Table 5: Hazard rates with task overlap incl. education

back (3) (4) VARIABLES

• dds ratio
• dds ratio

lnVi 1.103*** 1.114*** (0.005) (0.005) lnUi 0.948*** 0.945*** (0.004) (0.004) V all

i

/Vi 1.218*** 1.251*** (0.031) (0.033) Uall

i /Ui

0.817*** 0.805*** (0.020) (0.020) V some

i2

/Vi 1.002 1.003 (0.005) (0.006) V some

i3

/Vi 1.000 1.000 (0.000) (0.000) V some

i4

/Vi 1.000** 1.000* (0.000) (0.000) Usome

i2

/Ui 1.041* 1.036 (0.023) (0.023) Usome

i3

/Ui 1.004 1.004 (0.003) (0.003) Usome

i4

/Ui 1.000 1.000 (0.000) (0.000) Education = 1, High School 1.255*** 1.219*** (0.016) (0.016) Education = 2, College 1.349*** 1.309*** (0.021) (0.021) Constant 0.324*** 0.308*** (0.010) (0.012) Observations 130,428 127,979 Controls NO YES Weibull k 0.405 0.411

Table 7: Hazard rates excl. higher experience levels within an occupation

back

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

• dds ratio
• dds ratio
• dds ratio
• dds ratio

lnVi 1.122*** 1.131*** 1.106*** 1.118*** (0.004) (0.005) (0.005) (0.005) lnUi 0.928*** 0.929*** 0.950*** 0.947*** (0.004) (0.004) (0.004) (0.004) V all

i

/Vi 1.282*** 1.312*** 1.296*** 1.320*** (0.037) (0.038) (0.038) (0.039) Uall

i /Ui

0.793*** 0.771*** 0.746*** 0.736*** (0.026) (0.025) (0.025) (0.025) V some

i2

/Vi 1.006 1.005 1.001 1.002 (0.005) (0.006) (0.006) (0.006) V some

i3

/Vi 1.001 1.001 1.001 1.001 (0.000) (0.000) (0.000) (0.000) V some

i4

/Vi 1.000** 1.000** 1.000** 1.000** (0.000) (0.000) (0.000) (0.000) Usome

i2

/Ui 1.002 1.009 1.023 1.025 (0.023) (0.023) (0.020) (0.020) Usome

i3

/Ui 1.001 1.002 1.003 1.003 (0.002) (0.002) (0.002) (0.002) Usome

i4

/Ui 1.000 1.000 1.000 1.000 (0.000) (0.000) (0.000) (0.000) Female 0.971*** 0.962*** (0.009) (0.009) Part-time 0.759*** 0.770*** (0.012) (0.013) Part-time or full-time 0.916*** 0.913*** (0.009) (0.009) Belgian nationality: acquired 0.886*** 0.904*** (0.010) (0.011) Foreign, EU nationality 0.821*** 0.843*** (0.014) (0.015) Foreign, other nationality 0.909*** 0.952*** (0.014) (0.015) High School 1.259*** 1.222*** (0.016) (0.016) College 1.356*** 1.316*** (0.021) (0.021) Constant 0.403*** 0.390*** 0.403*** 0.317*** (0.001) (0.015) (0.011) (0.013) Observations 134,722 132,187 130,361 127,913 Location FE NO YES NO YES Time FE NO YES NO YES Weibull k 0.403 0.411 0.403 0.411

Table 8: Estimates by task-overlap bin

back (1) (2) (3) (4) (5) VARIABLES

• dds ratio
• dds ratio
• dds ratio
• dds ratio
• dds ratio

lnV all

i

1.134*** (0.004) lnUall

i

0.925*** (0.003) lnV some

i1

1.025*** (0.003) lnUsome

i1

1.016*** (0.004) lnV some

i2

1.020*** (0.005) lnUsome

i2

0.972*** (0.005) lnV some

i3

0.994 (0.004) lnUsome

i3

1.002 (0.004) lnV some

i4

0.990** (0.005) lnUsome

i4

1.002 (0.006) Constant 0.398*** 0.416*** 0.424*** 0.387*** 0.404*** (0.011) (0.011) (0.021) (0.012) (0.014) Observations 130,928 96,623 35,223 107,231 122,294 Controls YES YES YES YES YES Weibull k 0.412 0.416 0.424 0.404 0.402

Routine tasks in ROME-V3

back

Define three groups as routine tasks:

Task related to logging of activity data and the sharing of information: e.g. counting and registering pieces produced. Tasks related to assembly line work: e.g. providing workstations with materials or checking the stock. Administrative tasks: e.g. registering and sorting of mail.

About 15% of jobs in ROME-v3 have non-zero shares of routine tasks. Variation across 2-digit ISCO88 occupation groups corresponds to e.g. RTI index used in the literature.

48 / 39

An adverse routine-task biased shock

back

Assume a 25% reduction of vacancies multiplied by the share of routine tasks in a job. For example, if 50% of tasks in a job is routine, the number of vacancies in that job is reduced by 12.5%. The overall size of the shock corresponds to a 1 ppt reduction in the share of routine jobs in 2013-2014. This is the same order of magnitude as the actual ppt decrease in routine employment shares.

49 / 39