individual and policy responses to occupational decline

25
Individual and Policy Responses to Occupational Decline Georg Graetz Department of Economics at Uppsala University DG ECFIN Annual Research Conference, 19 Nov 2018 Copyright rests with the author. All rights reserved

Upload: others

Post on 01-Nov-2021

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Individual and Policy Responses to Occupational Decline

Individual and Policy Responses to OccupationalDecline

Georg Graetz

Department of Economics at Uppsala University

DG ECFIN Annual Research Conference, 19 Nov 2018

Copyright rests with the author. All rights reserved

Page 2: Individual and Policy Responses to Occupational Decline

Outline

Introduction

New evidence on individual-level responses to occupational decline

Forecasting occupational decline

Page 3: Individual and Policy Responses to Occupational Decline

250+ years of economic growth thanks to technology

Sustained growth in average living standards

I Not always shared equally within generations

I Not enjoyed by all generations equally

Structural changes (smooth?)

I Industry: decline of agriculture and manufacturing

I Occupation: decline of e.g. typists, human calculators,

switchboard operators, textile machine operators

Individual experiences can be traumatic

I Plant closures

Page 4: Individual and Policy Responses to Occupational Decline

Policy responses

Market failures

I Missing insurance market: permanent shocks to specific

human capital → re-training subsidies

I If mainly general human capital, case is less strong

I Public good of forecasting occupational employment growth

Redistribution

I Skill-biased technological change → income tax

I Capital-biased technological change → wage subsidies, BIG,

asset redistribution

Should governments act preemptively? (E.g. changes to school

curriculum, fields of study capacity)

Page 5: Individual and Policy Responses to Occupational Decline

This presentation

1. New evidence on individual-level adjustment to occupational

decline

I Study Swedish workers who in 1985 started out in occupations

that subsequently declined sharply

I Earnings losses are mild on average, but substantial for

low-wage (within-occupation) workers

I Re-training programs mainly used by those with largest losses

2. Thoughts on forecasting occupational employment

growth

I “This time is different” type arguments hard to evaluate

almost by construction

I Existing approaches to forecasting

I A suggestion for doing even better

Page 6: Individual and Policy Responses to Occupational Decline

Outline

Introduction

New evidence on individual-level responses to occupational decline

Forecasting occupational decline

Page 7: Individual and Policy Responses to Occupational Decline

Outline

Introduction

New evidence on individual-level responses to occupational decline

Forecasting occupational decline

Page 8: Individual and Policy Responses to Occupational Decline

Occupational decline in Sweden

Work in progress with Edin, Evans, Hernnas, & Michaels

Identify sharply declining occupations

Collect predictors of occupational employment growth, including

from BLS → ‘surprise’ declines

Follow population of Swedish workers 1985-2013

Research question How do the careers of workers starting out

(1985) in a subsequently declining occupation differ from those of

similar workers in non-declining occupations?

Page 9: Individual and Policy Responses to Occupational Decline

Occupational decline in Sweden: findings

1. When comparing similar workers, find moderate losses—5

percent of mean cumulative earnings

2. When comparing observationally similar workers in

similar occupations, see small losses—2 percent of mean

cumulative earnings

3. Workers in declining occupations are less likely to remain

4. Middle-aged workers (in 1985) in declining occupations retire

slightly earlier (zero difference for older workers)

5. Heterogeneity: workers in bottom third (within occupation)

do suffer—lose 8 percent of mean earnings

6. Higher incidence of unemployment & re-training for low-rank

workers

Page 10: Individual and Policy Responses to Occupational Decline

Diverging earnings of workers starting out in subsequentlydeclining occupations

-.1-.0

8-.0

6-.0

4-.0

20

.02

1985 1990 1995 2000 2005 20102013Year

Individual controls

-.1-.0

8-.0

6-.0

4-.0

20

.02

1985 1990 1995 2000 2005 20102013Year

Individual, occupation, industry controls

Cumulative earnings, divided by mean

Page 11: Individual and Policy Responses to Occupational Decline

Occupational decline in Sweden: adjustment mechanisms

We identify a large occupational demand shock—occupations that

declined by 25 percent or more

Reduced likelihood of staying in the initial occupation is one

particular adjustment mechanism—but notice that mobility is high

across all occupations, anyway

Reduced inflow into declining occupations also important—as is

common with gradual changes

Public spending on training was high in the 1990s and early 2000s,

but that does not explain that losses are mild on average

Page 12: Individual and Policy Responses to Occupational Decline

Public spending on training: Sweden and US

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Public spending on training, % of GDP (source: OECD)

Sweden

US

Page 13: Individual and Policy Responses to Occupational Decline

Outline

Introduction

New evidence on individual-level responses to occupational decline

Forecasting occupational decline

Page 14: Individual and Policy Responses to Occupational Decline

Different types of forecasting problems

Accurate predictions

despite wrong theory

Accurate predictions

only because of theory

Somewhat accurate

predictions, tough

challenges to theory

(feedback,

non-linearity)

Forecasting occupational employment growth may yet be a

different problem

Page 15: Individual and Policy Responses to Occupational Decline

Forecasting technological change

Amara’s law

We tend to overestimate the effect of a technology in the

short run and underestimate the effect in the long run.

Unknown (?)

Human-level AI is always about 20 years away.

[A convenient prediction to make—work on something that is

relevant, not likely to soon be proven false]

Page 16: Individual and Policy Responses to Occupational Decline

Predictions about human-level AI

Source: Armstrong & Sotala, 2012

Page 17: Individual and Policy Responses to Occupational Decline

Existing approaches to forecasting occupationalemployment growth

BLS outlook

I Produced every two years for hundreds of occupations

I Tries to take into account demographics, consumer demand,

input-output, trade, technology

I Much predictive power even at longer horizons

Frey & Osborne (2017), and followers

I Consulted engineers to identify frontier of automation,

bottlenecks

I Feasibility of automation does not imply an occupation’s

decline, but methodology can be adapted (Arntz et al. 2017)

Page 18: Individual and Policy Responses to Occupational Decline

Any guidance from economic theory?

Discover some fundamental mechanism → not only match past

data, but also obtain accurate out-of-sample predictions

Very hard (impossible?) in the social sciences, but give it a try

anyway. Models of task-biased technological change

I Economic mechanism to explain balanced long-run growth in

the presence of automation

I Economic mechanism to explain decline in middle-wage jobs

Page 19: Individual and Policy Responses to Occupational Decline

Economic model of automation and balanced growth

Replaced tasks

Automated tasks

Tasks performed by capital

Tasks performed by capital

Labor-intensive tasks

Labor-intensive tasks

Capital Labor New tasks

N − 1

N − 1

N − 1 I* N

N

N N

~ I

~ I

~ I

I* = I

I* = I

I* = I

Panel A

Panel B

Panel C

Figure 2. The Task Space and a Representation of the Effect of Introducing New Tasks (Panel B) and Automating Existing Tasks (Panel C )

Source: Acemoglu & Restrepo, 2018

Page 20: Individual and Policy Responses to Occupational Decline

Economic model of automation and job polarization

0.00

0.20

0.40

0.60

0.80

1.00

1.20

0.77 0.67 0.59 0.53 0.48 0.43 0.40Labor share

Task automation and labor share

Fraction of skilled tasksautomated

Fraction of unskilledtasks automated

Based on Feng & Graetz, 2018

Page 21: Individual and Policy Responses to Occupational Decline

Is this time different?

If so, some existing methods will fail (BLS predicts taxi drivers’

employment will grow as fast as average over next ten years—no

mention of driverless cars)

Frey & Osborne approach more robust? Correlation with BLS

forecast is 0.46

Economic theory emphasizes adoption incentives and new tasks,

not a break with past patterns

If want to argue this time is different, would be helpful to provide

model, and make explicit auxiliary predictions with shorter horizon

(e.g. business dynamism seems to be declining)

Page 22: Individual and Policy Responses to Occupational Decline

A new source of forecasts: prediction markets

Take a contract that pays 1 Euro if some specified event E occurs.

Under risk neutrality, price reveals market expectation of

probability of E . For instance

I “Employment of bartenders grows by 2 percent 2016-2026”

(BLS prediction)

With family of contracts, can trace out distribution (alternative

contract types: index, spread)

Many practical issues to be specified (data authority, merging of

occupation codes...), may be addressed by auxiliary contracts

Page 23: Individual and Policy Responses to Occupational Decline

Prediction markets—strengths & limitations

Strengths

I They aggregate information—not so clear how to do this

otherwise (different sources, methods: BLS, FO, theory)

I Incentives for truthful revelation of beliefs

I In practice broadly efficient, hard to manipulate

Limitations

I Can be insufficiently liquid (how to attract traders, especially

for long-term contracts?)

I Contracts may be hard to understand, markets may be

dominated by insiders (not major issues here)

I If prediction markets are so great, why not more common?

Page 24: Individual and Policy Responses to Occupational Decline

Example of a successful prediction market

Figure 5: Macro derivatives are weakly more accurate than survey forecasts.

−400

−200

0

200

400

−400 −200 0 200 400

Non−Farm Payrolls

45

50

55

60

65

45 50 55 60 65

Business Confidence (ISM)

−1

−.5

0

.5

1

1.5

−1 −.5 0 .5 1 1.5

Retail Sales (excluding Autos)

300

320

340

360

380

300 320 340 360 380

Initial Unemployment Claims

Rel

ease

d V

alue

(A

ctua

l)

Forecast Value

Economic Derivatives Mean Forecast Survey: Average Across Forecasters

Source: Snowberg, Wolfers, & Zitzewitz, 2012

Page 25: Individual and Policy Responses to Occupational Decline

Conclusion

Technological change in the past has not benefitted everyone

I Swedish evidence: Losses appear to be modest for those

worst-affected by occupational decline (welfare state?)

I Adjustment mechanisms often decentralized (occupation

switching)

Is this time different and is much preemptive action therefore

required?

I Hard to say

I Prediction markets can help, and governments can encourage

their creation