Job Polarization, Skill Mismatch, and the Great Recession€¦ · 1.job polarization accounts for...

Job Polarization, Skill Mismatch, andthe Great Recession

Riccardo ZagoNew York University

6 December 2018

ECB/CEPR Labour Market Workshop

Job Polarization

Riccardo Zago (NYU) Introduction 6 December 2018 1

Job Polarization & the Great Recession

Unemp.

Vertical Downgrade & the Great Recession

Details High-Skilled vs. Low-Skilled

This Paper

- This paper is the first to show that

- the decline of routine employment

- the change in skill-demand across jobs

explain together

- deterioration of skills-to-job match quality → “Skill Mismatch”

- longer unemployment spells

- sluggish labor mobility

Riccardo Zago (NYU) Contribution 6 December 2018 4

Theoretical Mechanism

- A model with endogenous mapping of skills to jobs

- skill-heterogeneous workers

- job-specific technologies and endogenous skill-requirements

- skill-dependent job opportunities and multiple jobs search

- Asymmetric technology shocks and labor market frictions affect

- workers’ job opportunities and mobility

- the process of sorting skills with jobs

↓Skills-to-Job Mismatch

- Routine Biased Technical Change drives Job Polarization

Structural Estimation

- Estimation to match only employment dynamics between 2005 and 2015

- The model accounts well for the reallocation patterns of

- high-skilled workers

- low-skilled workers

- The aggregate predictions of the model are also true within local-labormarkets

Main Results

1. job polarization accounts for the rise in skill mismatch

2. skill mismatch dynamics differ across workers when the market polarizes

3. higher skills attenuate the wage loss from mismatch

4. changes in skill-demand across jobs and frictions explain 38% of the shift-outof the Beveridge Curve

Riccardo Zago (NYU) Results 6 December 2018 7

Policy Relevance

- Inefficiency in labor factor allocations due to frictions

- longer unemployment spells for the low-skilled

- welfare loss due to job polarization

- The central planner

- reduces low-skilled unemployment

- attenuates job-polarization

- reduces skill mismatch by 1/3

Riccardo Zago (NYU) Policy Implications 6 December 2018 8

Outline

1. THE MODEL

- Technologies and Jobs

- Workers and Job-Search

- Equilibria

2. QUANTITATIVE ASSESSMENT

- Estimation to match occupational dynamics between 2005 and 2015

- Comparison of the implied allocation patterns of HS and LS with the data

- Model implications for welfare, matching efficiency and wages

Riccardo Zago (NYU) Outline 6 December 2018 9

The Model

Riccardo Zago (NYU) The Model 6 December 2018 10

RBTC and Temporary Shocks

- Assume abstract and manual technology to follow this

za,t = za + σaεt ; zm,t = zm + σmεt

- Assume routine technology to follow this

zr ,t =

zr ,0(1 + gzr )

t + σr εt for t ∈ [0,T ]

zr ,T + σr εt for t > T

- The technological shock ε follows an AR(1) process:

εt+1 = ρεt + νt+1

and ν being a random shock out of a standard-normal distribution.

- σj governs the the job-specific intensity of the shock (similar to Lilien ’82)

Production and Skill Requirements

- Workers differ in their skill-level x

- Technology zj and skills x are mixed as follows:

y(x ; za) = zaxλa ; y(x ; zr ) = zrx

λr ; y(x ; zm) = zm

- The value of production is

J(x ; zj) = y(x ; zj)−w(x ; zj)+βEs ′j (x)(1−δ)J(x ; z ′j )+[1−s ′j (x)(1−δ)]V (z ′j )

s ′j (x) = s(x , e′j ) = Pr(x ≥ e′j )

- Firms choose the minimum requirement ej to ensure a non-negative J:

J(ej ; zj) = 0

- Countercyclical Skill Requirements: if zj ↓ ⇒ ej ↑Riccardo Zago (NYU) The Model 6 December 2018 12

Vacancy Posting

- Firms posts vacancies vj for j = a, r ,m following this rule

V (zj) = −cj + βEp(θj)J(x , z ′j ) + [1− p(θj)]V (z ′j )

p(θj) = ψjθ−αj

θj =vjuj

=n. of vacancies for market j

n. of qualified unemp. workers for market j

- Free entry condition: V (zj) = 0, ∀t

Employment Opportunities and Unemployment

- Skills x are drown from a U[0,1] pdf

x=0 !!

- For given ea and er , a worker with skill x knows his job-opportunity setΩ(x) = j : ej ≤ x

- The value of unemployment is

U(x ; z) = b + βE ∑

j∈Ω(x)

q(θj)N(x ; z ′j ) +[1−

∑j∈Ω(x)

q(θj)]U(x ; z ′)

with z = [za, zr , zm], a vector of all technologies currently available in thejob-opportunity set

Employment Value and Dynamics

- The value of employment is

N(x ; zj) = w(x ; zj)+βEs ′j (x)[(1−δ)N(x ; z ′j )+δU(x ; z ′)]+[1−s ′j (x)]U(x ; z ′)

- The dynamic for the stock of employment in job j is

n′j = sj(1− δ)nj + ujq(θj)

- For an increase in requirements in j , the factor sj(1− δ) falls such that it

I amplifies job destruction dynamics

I exposes also highly-ranked worker to displacement (differently from Mortensenand Pissarides ’94)

I increases individual employment uncertainty (as in Ravn and Sterk ’15...buthere endogenously)

Wage Equation

- Under Nash Bargaining:

w(x ; zj) = (1− η)b + ηy(x ; zj) + η ∑

j∈Ω(x)

with the value of the out-side option that varies over time and across workersdue to:

I changes in θj

I changes in |Ω(x)|

I changes in both θj and |Ω(x)|

Two Alternative Equilibria

Skill-Separating Equilibrium: PAM of Skills and Technology

AbstractProduction

𝑚(𝜃!)

Routine

Production

Manual

Production

Riccardo Zago (NYU) Skill-Separating Equilibrium 6 December 2018 17

Two Alternative Equilibria

Skill-Pooling Equilibrium: Skill Mismatch

AbstractProduction

𝑚(𝜃!)

Routine

Production

Manual

Production

Riccardo Zago (NYU) Skill-Pooling Equilibrium 6 December 2018 18

Skill-Pooling Equilibrium

Definition

Contingent to technology, a skill-pooling equilibrium is a vectorθj , nj ,w(x , zj), ej , uj∞t=0 for any j = a, r ,m and x ∈ [0, 1] satisfyingsimultaneously the job creation condition, the minimum requirementcondition, the wage equation, employment and unemployment dynamics.

Existence Condition

A skill-pooling equilibrium exists in the routine submarket iff the surplus fromthe match S(x , zr ) ≥ 0 for all x ∈ [ea, 1]; a skill-pooling equilibrium exists inthe manual submarket iff S(x , zm) ≥ 0 for all x ∈ [er , 1].

Riccardo Zago (NYU) Skill-Pooling Equilibrium 6 December 2018 19

QuantitativeAssessment

Riccardo Zago (NYU) Quantitative Assessment 6 December 2018 20

Bringing the Model to the Data

- Use CPS classification of educational attainments as a sufficient statistics forthe distribution of skills

- Define two major skill groups (ILO)

I High Skilled (HS): bachelor, master, phd

I Low Skilled (LS): 11th Grade, high-school diploma, 2 years of college,vocational degree

- Build (quarterly) series for HS and LS employment in each occupation (onlyfull time, non-self employed workers; codes for farming, fishing, forestry andmilitary occupations excluded)

- GOAL: estimate the model to match occupational employmentdynamics from 2005 to 2015 and check reallocation patterns for HSand LS workers

Riccardo Zago (NYU) Structural Estimation 6 December 2018 21

Structural Estimation via SMM

- Preset Parameters: β, b, δ, η, α, zr ,0, gLS Appendix

- Two Step Estimation

I 1st Step: characterize the economy at an initial point (2005q1)

- use na,2005, nr,2005, nm,2005, ShareHSa,2005, ShareHSr,2005, ShareHSm,2005, ShareHSu,2005,

wHSr,2005

wHSa,2005

,wHSm,2005

wHSa,2005

,wLSr,2005

wLSa,2005

,wLSm,2005

wLSa,2005

- back-up za, zm, cj , ψj , λa, λr , γ Appendix

I 2nd Step: let the economy move on the RBTC trend and shock it to generatethe dynamics observed from the Great Recession (2005q1 to 2015q4)

- use long-run gnr , ∆na,GR , ∆nr,GR , ∆nm,GR , Corr(ut , ut−1) Appendix

- back-up gzr , σj , ρ

Parameter Description Value

Technologyza Tech. in abstract jobs 1.09

zm Tech. in manual jobs 0.68

Labor Marketca Vacancy posting cost in abstract 0.02

cr Vacancy posting cost in routine 0.04

cm Vacancy posting cost in manual 0.05

ψa Matching efficiency in abstract 0.79

ψr Matching efficiency in routine 0.68

ψm Matching efficiency in manual 0.46

Skillsλa Return to skills in abstract 1.02

λr Return to skills in routine 0.49

γ Lowest skill for HS workers 0.71

Dynamicsgr Growth of routine tech. -9.81×10−5

σa Std. for tech. shock in a 0.040

σr Std. for tech. shock in r 0.051

σm Std. for tech. shock in m 0.017

ρ Persistency of the shock 0.91

Loss Function Moments (model vs. data) Skill Returns (data) Skill-Pooling Existence

Job Polarization and Jobless Recovery

0 5 10 15 20 25 30 35 40 45Time

0.52Routine Employment

nDatar

nModelr

0 5 10 15 20 25 30 35 40 45Time

0.11Unemployment

uModel

- The model generates too high unemployment in the long-run. Why?Rise in non-participation rate after the Great Recession

Change σ (employment) Change grz

Riccardo Zago (NYU) Model’s Prediction 6 December 2018 24

Employment Dynamics by Skill Group

0 5 10 15 20 25 30 35 40 45Time

-0.005

0.03" Share of HS in A (

DataModel

0 5 10 15 20 25 30 35 40 45Time

-0.025

-0.015

-0.005

0.01" Share of HS in R (

DataModel

0 5 10 15 20 25 30 35 40 45Time

10 #10-3 " Share of HS in M (

DataModel

0 5 10 15 20 25 30 35 40 45Time

-0.005

0.02" Share of LS in A (

DataModel

0 5 10 15 20 25 30 35 40 45Time

0.01" Share of LS in R (

DataModel

0 5 10 15 20 25 30 35 40 45Time

0.04" Share of LS in M (

DataModel

- Temporary reversal in emp. shares for HS workers- Permanent change in emp. shares for LS workers

Change σ (sorting) Data (state-level) E → U (aggregate) Other Dynamics

Riccardo Zago (NYU) Model’s Prediction 6 December 2018 25

Employment Mismatch

Social PlannerFollowing Bhattacharya and Bunzel ’03, assume a social planner maximizes totalexpected output and total value of “leisure” at the net vacancy costs:

maxθj ,ej ,n′j

E∞∑t=0

βtyana + yrnr + ymnm + b(1− na − nr − nm)−∑j

cjθjuj

s.t. n′j = s(1− δ)nj + ujq(θj)

y(x ; zj)U[x≥ej ]dx ⇒ average output in j

Riccardo Zago (NYU) Employment Mismatch 6 December 2018 26

Employment Mismatch

Polarization (Planner) On ψ

Riccardo Zago (NYU) Employment Mismatch 6 December 2018 27

The Shift of the Beveridge Curve

Planner Frictions vs. Shocks

Riccardo Zago (NYU) The Beveridge Curve 6 December 2018 28

The Wage Ladder10

Abstra

Routin

Major Groups Job Ladder (data)

Abstra

Routin

LS HS 95% CI

Major Groups Job Ladder (model)

Wage loss is bounded for HS workers when moving down the ladder

Riccardo Zago (NYU) The Wage Ladder 6 December 2018 29

Conclusion

- The change in the occupational structure and in skill-demand across jobsexplain the rise in skill mismatch:

- mismatch dynamics differ across skill-groups

- the wage-loss from mismatch is bounded for high-skilled

- Job polarization is associated with specific reallocation patterns

- A central planner reduces skill mismatch and the process of polarization

- Changes in skill-demand across jobs and frictions explain the deterioration ofaggregate matching efficiency

Riccardo Zago (NYU) Conclusion 6 December 2018 30

APPENDIX

Riccardo Zago (NYU) Appendix 6 December 2018 31

Job Polarization

- Since the 80s, routine employment is falling along with wages

- Jobs grouped by task (Acemoglu and Autor ’11) :

I abstract: Management, Professionals, and Related jobs

I routine: Production and Clerical jobs

I manual: Food prep and service, personal/child care, recreation and hospitalityjobs

- Job Polarization is driven by:

I Routine Biased Technical Change (RBTC): robotics, IT innovations, etc.

I International Trade: imports of “routine” products (e.g. cloths from China),offshoring, etc.

Unemployment

Vertical Downgrade & the Great Recession

Polarization and Mismatch

0 5 10 15 20 25 30 35 40 45Time

nDatar

nModelr

0 5 10 15 20 25 30Periods After Shock

Emp. Mismatch Dynamics

DataEconomy

Flows from Unemp. to Emp.

Job-Finding and Skill-Requirements

0 10 20 30 40 50Time

0.73Job-Finding Ratio in A (HS v. LS)

0 10 20 30 40 50Time

0.138Job-Finding Ratio in R (HS v. LS)

0 10 20 30 40 50Time

0.116Job-Finding Ratio in M (HS v. LS)

0 10 20 30 40 50Time

0.688Requirements in A (ea)

0 10 20 30 40 50Time

0.15Requirements in R (er)

0 10 20 30 40 50Time

1Requirements in M (em)

Up-skilling (data) Model

From Unemployment to Abstract Jobs

From Unemployment to Routine Jobs

From Unemployment to Manual Jobs

Back Model

Returns to Education over the Cycle

HS Diploma

Some C

ollege

Vocati

Master Phd

Beefore Recession Recovery After

Schooling and Wages in Abstract Jobs

HS Diploma

Some C

ollege

Vocati

Master Phd

Schooling and Wages in Routine Jobs

HS Diploma

Some C

ollege

Vocati

Master Phd

Schooling and Wages in Manual Jobs

The long-run decline in routine employment Back

log(Routine Emp.) log(Routine Emp.) (FITTED VALUE)

log(Routine Emp.)t = β*time + εt

- γ is the skill level in the interval [0, 1] that splits the skill distribution in twosubgroups: high-skilled (∀x ≥ γ), low-skilled (∀x < γ)

- The share of low-skilled population decline at a rate gLS = −0.1%. Back

log(LS Share) log(LS Share) (FITTED VALUE)

log(LS Share.)t = β*time + εt

Job-specific Surplus over time

Surplus in A

Surplus

xTime 050

Surplus

Surplus in R

50Time x050

Surplus in M

Surplus

xTime 500

Table: Preset Parameters

Parameter Description Value

β Discount factor (quarterly) 0.95

b Value of leisure 0.40

δ Separation rate 0.10

η Employer bargaining power 0.50

α Matching elasticity 0.50

gLS Growth of LS pop. Share −1.1× 10−3

zr ,0 Technology in routine jobs 1

Table: Targeted moments and model moments

Moment Data Model

na in 2005 0.285 0.286

nr in 2005 0.512 0.510

nm in 2005 0.152 0.153

HS Share of na in 2005 0.660 0.675

HS Share of nr in 2005 0.154 0.152

HS Share of nm in 2005 0.102 0.105

HS Share of u in 2005 0.12 0.11

wa,HSin 2005 0.683 0.689

wa,HSin 2005 0.572 0.590

wa,LSin 2005 0.810 0.795

wa,LSin 2005 0.603 0.590

nr long-run growth rate −1.5× 10−3 −1.6× 10−3

%∆na during GR -0.68% -0.67%

%∆nr during GR -4.00% -4.01%

%∆nm during GR -0.24% -0.22%

Corr(ut , ut−1) during GR 0.900 0.899

Vertical Downgrade over the Cycle

- Say your company shuts down. What is your next job going to be overthe cycle?

- Use Displaced Worker Supplement (DWS) to identify workers that had beenfired for “exogenous” reasons (plant closing, abolished jobs,...)

- For worker i consider:

Pr(Downgradei 6= 0|Xi ) = Φ(δ′sβ + X ′i γ)

I δs is a vector of mutually exclusive dummy variables for state-specificexpansion, recession and recovery periods

I X controls for sex, age, education, experience, marital status, number ofchildren.

The role of σ

0 5 10 15 20 25 30 35 40 45Time

nDatar

nModelr ,<j = 0:03; 8j = fa; r;mg

nModelr ,<j = 0:05; 8j = fa; r;mg

nModelr , <r = 0:05;<a;m = 0

0 5 10 15 20 25 30 35 40 45Time

0.11Unemployment

uModel,<j = 0:03; 8j = fa; r;mguModel,<j = 0:05; 8j = fa; r;mguModel, <r = 0:05;<a;m = 0

The role of σ

0 5 10 15 20 25 30 35 40 45Time

0.05" Share of HS in A (

<j = 0:03; 8j = fa; r;mg<j = 0:05; 8j = fa; r;mg<r = 0:05;<a;m = 0

0 5 10 15 20 25 30 35 40 45Time

-0.045

-0.035

-0.025

-0.015

-0.005

0" Share of HS in R (

0 5 10 15 20 25 30 35 40 45Time

-0.008

-0.006

-0.004

-0.002

0.01" Share of HS in M (

0 5 10 15 20 25 30 35 40 45Time

0.03" Share of LS in A (

0 5 10 15 20 25 30 35 40 45Time

0" Share of LS in R (

0 5 10 15 20 25 30 35 40 45Time

0.05" Share of LS in M (

The role of gzr

0 5 10 15 20 25 30 35 40 45Time

nDatar

nModelr ; gzr

= cgzr

nModelr ; gzr

= 3cgzr

Employment Dynamics across States’ Cycles

∆Emp. Shares,t = βyears + X ′s,tγ + εs,t

0 1 2 3 4 5N. of years after recession

ΔEmp. ShareHS,A

-.8-.6

-.4-.2

ΔEmp. ShareHS,R

ΔEmp. ShareHS,M0

ΔEmp. ShareLS,A

ΔEmp. ShareLS,R

ΔEmp. ShareLS,M

Job Polarization (Planner)

0 5 10 15 20 25 30 35 40 45Time

nDatar

nModelr

0 5 10 15 20 25 30 35 40 45Time

0.11Unemployment

uModel

Emp. Mismatch Policy Relevance

The role of ψ and the Social Planner

Economy , Aj = Aj

P lanner

Economy , Aa = 0:5Aa

Economy , Ar = 0:5Ar

Economy , Am = 0:5Am

Economy , Aj = Aj

P lanner

Economy , Aj = 0:5Aj

Economy , Aj = 0:8Aj

Emp. Mismatch

HS Emp. Mismatch

DataEconomyP lanner

Emp. Mismatch

The Beveridge Curve: Planner vs. Economy

Frictions vs. Shocks

Search Frictions Shock Asymmetry

Up-Skilling X

Shift-out BC X

Job Polarization X

LS Mismatch X+ X

HS Mismatch X X+

Related Literature

1. Job Polarization and Technical Change

- over the cycle: Jaimovich and Siu ’13, Foote and Ryan ’15, Restrepo ’15- in the long-run: Acemoglu and Autor ’11, Autor ’07, Autor and Dorn ’13

2. Skill Mismatch and Inefficiency in Labor Allocation

- cyclical reallocation of skills and efficiency: McLaughlin and Bils ’01,Altiwanger et al. ’15, Carillo-Tudela and Visschers ’13

- vertical displacement and wage loss: Huckfeldt ’16, Krolikowsky ’17, Jarosch’14

- fall in aggregate matching efficiency: Sahin et al. ’14, Barnichon and Figure’11

2. Skill-pooling and Up-skilling

- skill-pooling and requirements: Albrecht and Vroman ’02- counter-cyclical skill requirements: Modestino et al. 15

Riccardo Zago (NYU) Literature 6 December 2018 58

Validation of the Skill-Pooling Equilibrium

Under Nash Bargaining, the value of production is a share of the surplus

J(x ; zj) = (1− η)S(x ; zj)

Under the estimated parameters, the condition for existence of a skill-poolingequilibrium holds. OverTime Back

0 0.2 0.4 0.6 0.8 1x

0.6Surplus in A

S(x; za)er

0 0.2 0.4 0.6 0.8 1x

0.5Surplus in R

S(x; zm)er

0 0.2 0.4 0.6 0.8 1x

0.5Surplus in M

S(x; zm)er

Riccardo Zago (NYU) Literature 6 December 2018 59

Job Polarization, Skill Mismatch, and the Great Recession€¦ · 1.job polarization accounts for...

Documents

nber.orgnber.org/papers/w21376.pdf · NBER WORKING PAPER SERIES MULTIDIMENSIONAL SKILL MISMATCH Fatih Guvenen Burhanettin Kuruscu Satoshi Tanaka David Wiczer Working Paper 21376

The Growth of Low-Skill Service Jobs and the Polarization

The skill mismatch challenge - CEDEFOP

The consequences of education mismatch and skill mismatch

Top DBR mirror replaced with CTF and chiral STF bilayers The CTF (QWP) introduces a pi/2 retardance to compensate the polarization mismatch between

UNIVERSITY OF CALIFORNIA Santa Barbara Polarization control … · 2005-06-22 · [8] Y. Okuno, "Investigation on direct bonding of III-V semiconductor wafers with lattice mismatch

What the Current Labor Says about Job Skill Mismatch

Tackling unemployment while addressing skill mismatch · addressing skill mismatch in the EU-28 Member States. Matching skills to jobs is gaining attention among policy-makers. Earlier

EDUCATION, SKILLS AND SKILL MISMATCH: A REVIEW AND … · The mismatch between workers’ skills and ﬁrms’ skill needs is usually referred to as “skill mismatch”. Skill mismatch

Education and Skills Mismatch in Developing Countries ... · Rationale for mismatch research Persistent concerns in OECD countries (over decades) 1 •Job skill requirements high

Impact of Skill Mismatch on Organization’s … of Skill Mismatch on Organization’s Performance 2 Proceedings of 2nd International Conference on Business Management (ISBN: 978-969-9368-06-6)

Skills forecast and skill mismatch in the EU

Skill Mismatch and Migration in Egypt and Tunisia€¦ · 2 SKILL MISMATCH AND MIGRATION IN EGYPT AND TUNISIA 1 Anda David PSL, Université Paris Dauphine, LEDa, IRD, UMR DIAL david@dial.prd.fr

A Tool for Future Strategies. Turnover Rates Wages Workforce/Employer Connection Skill Mismatch High Growth Occupations

How Can Skill Mismatch be Measured? New Approaches ......methods, data, analyses | Vol. 8(2), 2014, pp. 137-174 DOI: 10.12758/mda.2014.006 How Can Skill Mismatch be Measured? New Approaches

SKILL MISMATCH AND PUBLIC POLICY IN OECD …...ECO/WKP(2015)28 7 SKILL MISMATCH AND PUBLIC POLICY IN OECD COUNTRIES By Müge Adalet McGowan and Dan Andrews1 1. Introduction 1. This

DISABILITY AND SKILL MISMATCH - Cardiff Universityorca.cf.ac.uk/86653/1/Disability and Skill Mismatch FINAL_REVISED.p… · disability reduces an individual’s ability to work and

Skill mismatch The European experience · 2017-05-18 · Employer talent management practices Labour market Institutions Country -level Enterprise level Source: Cedefop (2015) Skill

Education, skills and skill mismatch. A review and some

DIIN PAP SI - IZA Institute of Labor Economicsftp.iza.org/dp10786.pdf · How Useful Is the Concept of Skills Mismatch?1 The term skill mismatch is very broad and can relate to many