Do old dogs teach tricks to puppies or chase them away?Evidence from Hungarian plant-level data on the complementarity between young and old workerswork in progress

Zsombor Cseres-GergelyInstitute of Economics, CERSHAS, Budapest, Hungary

http://www.mtakti.hu

WPEG Annual Conference 2013, July 25-26, University of Sheffield

Financial support provided by the OTKA grant No. 101803

Motivation

• Ageing – fiscal pressure through many channels. Important one: pension and inactivity instead of gainful work.

• EU: active ageing policies in place. Employment of the 55+ is rising. Predicted and likely connection.

• At the same time: youth unemployment rising from 17 (2006) to 22.5 (2012) per cent in the EU, trend does not seem to decline. Serious effects of lack of integration.

• Little knowledge about the connection between older and youth employment. Empirical evidence mixed.

• Practically no results based on up to date methods and data.

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

2/14

Main labour market indicators for the EU

Employment rate Unemployment rate

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

3/14

Evidence in the literature

• Theory: Layard, Nickell, and Jackman (1991). No effect should be present due to wage-adjustment (intro. of early retirement)

• Simple x-country panel regressions on aggregate data: Herbertsson (2001). No negative effect of early retirement on youth employment.

• Jousten et al. (2008): refined time-series evidence using retirement incentives with early retirement. No effect on youth.

• Gruber-Wise (2010): Follow-up research to early-retirement. No effect

But:

• Skans (2005): Negative effect on the young with regional data in Sveden

• Grant-Hamermesh (1980): older women crowd out younger men in the US

• Cseres-Gergely (2013): weak crowding out effect for the public sector in Hungary, no wage effects

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

4/14

Motivation and indicators for Hungary

Employment rate Unemployment rate Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

Increase of pensionable age to 62 from 55, 59 (female, male). Effect on activity (Kátay-Nobilis, 2009). Significant crowding out from aggregate regressions (Cseres-Gergely, 2013)

5/14

Naïve aggregate regressions

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

6/14

Results for the public sector

• Motivate estimates through a dynamic labour budgeting model, no “real” parameters of interest.

• Estimate differenced equations with number of youth employment on the LHS, lagged number of employment and wages for other age groups on the RHS. A-B dynamic panel. Fairly robust negative effects, around -0.4, -0.1, -0.1 in the 1., 2. and 3. year, resp. Weaker for experience groups.

• Estimate similar equations with entry on the LHS: effects are weaker, not sig. – conclusion: exit.

• Extended Mincerian wage equation including share of older workers and share-experience interaction. Effect of older sig. negative, but for all other groups.

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

7/14

General second order approximation to a production function with curvature (Christensen, Jorgenson, and Lau 1973):

The private sector: some theory

Need a relationship between factors respecting optimal factor choice:

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

where

The elasticity of complementarity – the effect the change in one factor on the martinal product of the other.Second derivative! If CD, no effect on MP, optimal factor use depends on relative prices only, not on the use of other factors.

8/14

Estimating equation and considerations

Production function form + many factory = collinearity+endogeneity. On competitive input markets:

Estimation in share-form: need price/expenditure data and assumptions of efficient operation and input market structure. Then

where Z is a vector of other factors shifting the marginal product and ε is an unobservable stochastic component, orthogonal to observables.

Use 3SLS with symmetry constraints. Omit the K equation, use homogeneity.

Here the elasticity of complementarity is

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

9/14

Data• Matched Wage-National Tax Authority (MWN) data of the Data Bank

of the Institute of Economics (IE), CERSHAS. Panel for firms.• A random sample of all firms with double-entry bookkeeping having at

least 10 employees and an additional subsample of micro-enterprises. Employees are randomly sampled based on their date of birth. No public sector, only larger firms, only legal employment.

• Individual level data on plant characteristics as well as on workers’ wages and characteristics, along with basic accounting data, such as revenue, turnover or amortisation.

• Observations from 2000 to 2008. Only those with nonzero share of all labour types are used in the estimation. 81,429 observations, 8-11 thousand per year. Need to weight due to linking. Require that L>30, positive costs, capital cost < K. Half lost. Require nonzero L-s. Final sample: around 1000-1300 observation per year, a total of 11924.

• Employment: estimate share from employee sample, get absolute figures from total number of employees.

• Cost shares: total numbers, average wages and firm-level wage-tax as well as amortisation, interest payment, stock of capital.

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

10/14

Stylised facts from microdata

• Share of firms with nonzero values for young and old employees, respectively in 2000: 39% and 32%. In 2008: 38% and 63%. An average of 25% has both nonzero. Sample these.

• Average share of old up from 6 to 12% (7 to 11). Young: down from 11 to 6% (11 to 8). More than population proportion changes.

• Almost no correlation between old and young employment share in the cross-section (-0.03). Low correlation between change of the number of old any young is -0.1, stable over time.

• The young earn some 83% of the prime age, but no correlation with proportion/share of older workers in the cross-section. C

sere

s-G

erg

ely

: E

vid

en

ce o

n o

ld-y

ou

ng

co

mp

lem

en

tarit

y.

WP

EG

an

nu

al c

on

fere

nce

20

13

- U

niv

ers

ity o

f S

he

ffie

ld

11/14

Decomposing changes in youth employment share

The average share of young workers can be multiplicatively decomposed as

where

•N is the total number of employers Mi is the number of employers with nonzero share of Ly.

•lyi=Lyi/Li is the share of Ly within an employer, while is the share of Ly within the population, among those with nonzero share.

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

  lyshareof 

Ly>0ly for 

nonzero2000 0.110 0.433 0.2532004 0.070 0.459 0.1542008 0.064 0.407 0.140

12/14

Response of wages to older workers’ share

• Alternative margin for adjustment: wages. Check response.• Base: Mincerian wage regression. Extend to allow for flexible

age-earnings profiles, effect of share of age-groups and their interaction:

• w: wage, X: demographics, a: age, s: share of group in total employment.

• Parameter of interest: δ1 and δ2 – shifts experience profile.• Actual implementation: include the effect of both share of the

young and old. Use nonparametric profile (indicators).

• Result: larger share of old push down wages, but for everyone.

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

13/14

Estimation results: all time periods

Parameter estimates:

Elasticities of complementarity:

Effects on quantities

(assuming no wage effects):

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

  Ly Lp Lo K

Ly -4,254 -1,695 -0,795 1,047

Lp -1,695 -1,882 -1,053 1,001

Lo -0,795 -1,053 -5,713 1,023

K 1,047 1,001 1,023 -0,584

  Ly Lp Lo K

Ly 0,030 -0,027 -0,004 0,001

Lp -0,027 0,059 -0,032 0,000

Lo -0,004 -0,032 0,035 0,001

K 0,001 0,000 0,001 -0,002

  Ly Lp Lo K

Ly -1,000 -6,575 -0,218 30,631

Lp -0,055 -1,000 -0,040 4,011

Lo -0,119 -2,602 -1,000 19,049

K 0,014 0,227 0,016 -1,000

14/14

Conclusions so far

• Change in youth employment share explained mostly by intensive changes.

• Q-substitutability between all age groups, including the young and the old

• Results stable over time periods• No wage effect on the youth, but a generational effect only.

• Need to: look at heterogeneity and firms in/out zero youth/old share

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

15/14

Thank you for your attention!

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

16

Cse

res-

Ge

rge

ly:

Evi

de

nce

on

old

-yo

un

g c

om

ple

me

nta

rity.

W

PE

G a

nn

ua

l co

nfe

ren

ce 2

01

3 -

Un

ive

rsity

of

Sh

eff

ield

17

Estimation results: selected time periods

  betayy betayo sigmayy sigmayo dLydLo dwydLo

2000 0.036 -0.004 -3.557 -0.431 -0.109 -0.025

2004 0.031 -0.005 -6.641 -0.527 -0.112 -0.035

2008 0.028 -0.003 -3.961 -0.301 -0.084 -0.017