Career Progression and Formal
versus on the Job Training
Public Policy and life-cycle decisions
C. Meghir
UCL and IFS
with J. Adda, C. Dustmann, J.-M. Robin
Presentation prepared for WISE conference in XIAMEN
Go to data
Introduction
Empirical Public Finance and Dynamic Models
• Designing a tax and welfare benefit system in a 2nd best
context requires inputs from empirically estimated
parameters
• The literature tends to focus on estimation of commodity
and labour supply elasticities, which are the basic variables
that respond to taxes and benefits.
• However, individuals are likely to change behaviour in a
number of different ways faced with a new regime of taxes
and benefits.
Introduction
Empirical Public Finance and Dynamic Models
• In particular we have in mind the implications for Human
capital accumulation of programmes such as
– Unemployment Insurance
– Earned Income Tax Credit
• Such programmes will have long run dynamic effects on
– education choice
– job mobility and job choice.
Introduction
Empirical Public Finance and Dynamic Models
• The specific context we will consider is the choice by young Germans
to follow apprenticeship training and the subsequent path of wages and
employment.
• Our model will explain how individuals choose between the two
alternative career paths and will quantify the returns to the career
• We will then use this framework to simulate the impact of programmes
such as the EITC, which are not currently operating.
• Our emphasis will be on the effects of these programmes on human
capital accumulation.
Introduction
The Policy issue
• Welfare to work policies such as EITC may have
deeper implications than just affecting
employment
• They change educational incentives
• They may affect job search behaviour
The Approach
To achieve this we need:
1.
A model of the education choice
•
2.
This allows us to associate incentives from welfare programmes
with earlier choices
This model must be linked to career choices and job
mobility
•
This allows us to study how wage growth may be affected
Overall aims of the paper
1.
To examine the relative benefits of formal versus
informal on-the-job training.
•
•
Returns and their source
Labour market Flexibility
2.
To model wage formation and identify the sources of
wage growth.
3.
To create a framework where the implications of welfare
policies can be analysed in a life-cycle context with all
their dynamic implications.
More specific
questions relating to apprenticeship
• About 80% of Germans who stop school at 16 follow an
apprenticeship training programme. The remaining 20%
remain formally unskilled.
• How valuable or otherwise is a formal vocational system?
Does it reduce flexibility?
• How does it compare with less structured on-the-job
training that typically occurs in the labour market?
• How important are economic incentives for the choice of
education?
Why a model and not just a field experiment
or a natural experiment?
• The policy issue we are concerned with is longer term. It
relates to the behaviour of future cohorts.
• Even with a longer term perspective in mind we cannot
envisage the existence of a control group
• We need to infer the impact of the policy from analogous
effects implied by variations in the current environment:
How do changing incentives affect education choice?
• To achieve our aim we need to specify structure and use
observational data.
Introduction
Data
• We use administrative data from German Social Security
records
• These contain a calendar of all jobs with start and end date
excluding civil service job
• They report the average daily wage either over the year or
part thereof when a job change occurs. No averaging
across jobs.
• They also report periods in apprenticeship because these
are jobs – albeit very low paid
Introduction
Some relevant literature
• Becker (Human Capital), Willis and Rosen JPE 1979
• Keane and Wolpin JPE `97 (Education, work and occupational
choices)
• Eckstein and Wolpin Econometrica `99 (education, work and
attainment)
• Chris Taber Restud
• Altonji and Shakotko `90
• Topel JPE `91 and Topel and Ward QJE
• Dustmann and Meghir REStud 2005
• Heckman, Lochner and Cossa (1999)
.25
.2
5
.15
0
3
.05
.1
4
Log wage difference
4.5
3.5
0
20
40
60
time since entry on labor market (quarter) ...
Wage In apprenticeship
Skilled wage
80
Unskilled wage
Log wage difference
The Model
Overview
• Describes choices from the point where statutory education
ends up until mid career.
• We take those who end formal education at 16.
• Utility is linear in earnings – so effectively individuals
would maximise lifetime income except for the fact that
they like leisure
• In our context liquidity constraints and uncertainty are not
a factor because of linearity.
• Wages are match specific – So part of wage growth comes
from looking for better jobs.
Choices
• Individuals decide whether they will attend
apprenticeship or not.
• They trade off future returns with current
cost – Lost earnings and utility costs
• Individuals know the distribution of future
shocks.
Choices
• At the start of their career individuals are assumed
to receive with certainty a job offer and an offer
for apprenticeship.
• The former involves a wage and a non-pecuniary
benefit. The latter a cost shock as well as a wage
and benefit.
• We do not model the wait from school to any of
the two state above.
Choices
• Once in a firm individuals make a choice to change jobs or
not when they receive an alternative offer. [Job arrival Rate
for the employed]
• Given the shock to the match specific effect individuals
may choose to move to unemployment.
• They can only choose to go back to work if they gent an
offer [Job arrival Rate for the unemployed]
• Transitions to unemployment can also take place because
of job destruction. [Job Destruction Rate]
The Source of Dynamics
• Education choices now affect job opportunities in the
future
• The returns to experience and tenure
• The potential difference in the job arrival rates for the
employed and the unemployed.
• The stochastic structure.
The environment and sources of uncertainty
• We consider a stationary economy
• There are aggregate shocks to productivity.
• A random cost of education (how hard you find it to
learn)
• Random arrival of job offers.
• The match specific effect on wages
• Match specific effects evolve as a random walk
• In every period a job may be destroyed exogenously
The wage equation
• Wages
ln wift 0 

Ed 

Ed i X 
X i , Ed i , 
T 
T i , Ed i , 
i 
i
i 
i
G 
Ed i 
G t ift
• All parameters depend on apprenticeship status.
• Ability is denoted by i. This affects levels of wages, returns
to apprenticeship status (Edi), experience and tenure.
• The match specific heterogeneity is denoted by ift
#
A description of the statistical approach
• The model consists of a set of value functions – Bellman
equations.
• There is a value for:
–
–
–
–
going into education,
working in the current job
Switching to a new job if an offer is available
Not working
• These value functions define the probabilities of observed
events
• The probabilities of all transitions we observe and the
density for observed wages constitute the likelihood
function SKIP
Formal Description of the model
• The term G Ed i Gt represents the price of
human capital which depends on the business
cycle
• The Business cycle evolves according to an AR(1)
which we discretise
G 
G v,
2
v  IID, N 
0, 
v
•  represents the aggregate shock
Formal Description of the model
• The distribution of match specific effects is
defined by
t t1 u t ,
u t  N
0, 2u 
• The initial condition characterises the actual offer
and its distribution depends on education
2
2
0  N 
0, 

and


N

0,


0
• The term 0 represents non-wage benefits of the
job and enters utility
Formal Description of the model
• The instantaneous flow utility from working is given by
RW w
E, G t , Xt , T t , t , 

• When unemployed German workers receive UI as a
proportion of the earnings in their last job. The
replacement rate is 55%.
• We also allow for utility of leisure. The flow utility from
unemployment is given by
RU 
E, X, w1 , 
, U w1 
E, X, 

0
with U=0.55 and  is an iid shock with a normal
distribution.
Value functions and transitions
• The value of unemployment can now be defined by
U 
E, G, X, w1 , R U 
E, X, w1 , 
, 
If you get an offer
If you do not get offer


E, X
E G ,,
U
,max
0
U 
E, G , X, w1 , 
 
W
E, G , X, 0, 0 ,  

1 
E, X

E G ,U 
E, G , X, w1 , 
U
Offer arrival rate when unemployed
Value functions and transitions
• The value of employment can now be defined by
W
E, G, X, T, , w
E, G t , X t , T t , t , 


E, X
E G ,U 
E, G , X , w
E, G, X , T , 
, 
U 
E, G , X , w
E, G, X , T , 
, 


1 
E, X

E, X
E G ,,u ,

W
,max
0

1 
E, X
1 
E, X
E G ,,u max



W
W
E, G , X , T ,  u , 
 
W
E, G , X , 0, 0 ,  
U 
E, G , X , w
E, G, X , T , 
, 
W
E, G , X , T ,  u , 
Value functions and transitions
• The value of employment can now be defined by
W
E, G, X, T, , w
E, G t , X t , T t , t , 


E, X
E G ,U 
E, G , X , w
E, G, X , T , 
, 
Job destruction rate
U 
E, G , X , w
E, G, X , T , 
, 


1 
E, X

E, X
E G ,,u ,

W
,max
0

1 
E, X
1 
E, X
E G ,,u max



W
W
E, G , X , T ,  u , 
 
W
E, G , X , 0, 0 ,  
U 
E, G , X , w
E, G, X , T , 
, 
W
E, G , X , T ,  u , 
Value functions and transitions
• The value of employment can now be defined by
If job is destroyed

W 
E, G, X, T, , w
E, G t , X t , T t , t , 


E, X
E G ,U 
E, G , X , w
E, G, X , T , 
, 
Job destruction rate
U 
E, G , X , w
E, G, X , T , 
, 


1 
E, X

E, X
E G ,,u ,

W
,max
0

1 
E, X
1 
E, X
E G ,,u max



W
W
E, G , X , T ,  u , 
 
W
E, G , X , 0, 0 ,  
U 
E, G , X , w
E, G, X , T , 
, 
W
E, G , X , T ,  u , 
Value functions and transitions
• The value of employment can now be defined by
W
E, G, X, T, , w
E, G t , X t , T t , t , 


E, X
E G ,U 
E, G , X , w
E, G, X , T , 
, 
U 
E, G , X , w
E, G, X , T , 
, 


1 
E, X

E, X
E G ,,u ,

W
,max
0
W
E, G , X , T ,  u , 
 
W
E, G , X , 0, 0 ,  
Job is not destroyed and alternative offer made

1 
E, X
1 
E, X
E G ,,u max



W
U 
E, G , X , w
E, G, X , T , 
, 
W
E, G , X , T ,  u , 
Value functions and transitions
• The value of employment can now be defined by
W
E, G, X, T, , w
E, G t , X t , T t , t , 


E, X
E G ,U 
E, G , X , w
E, G, X , T , 
, 
U 
E, G , X , w
E, G, X , T , 
, 


1 
E, X

E, X
E G ,,u ,

W
,max
0

1 
E, X
1 
E, X
E G ,,u max



W
W
E, G , X , T ,  u , 
 
W
E, G , X , 0, 0 ,  
U 
E, G , X , w
E, G, X , T , 
, 
W
E, G , X , T ,  u , 
Offer arrival rate for the
employed
Value functions and transitions
• The value of employment can now be defined by
W
E, G, X, T, , w
E, G t , X t , T t , t , 


E, X
E G ,U 
E, G , X , w
E, G, X , T , 
, 
U 
E, G , X , w
E, G, X , T , 
, 


1 
E, X

E, X
E G ,,u ,

W
,max
0

1 
E, X
1 
E, X
E G ,,u max



W
Job is not destroyed but no alternative offer made
W
E, G , X , T ,  u , 
 
W
E, G , X , 0, 0 ,  
U 
E, G , X , w
E, G, X , T , 
, 
W
E, G , X , T ,  u , 
Offer arrival rate for the
employed
Value functions and transitions
• The value of work while being an apprentice is
given by
Fraction of unskilled wage earned during apprenticeship
W
G, X, T, , A w
0, G, X, T, 
A
 

AE
G ,u ,,max
0
W
G , X , T ,  u , 
A
   
W

G
, X , 0, 0 ,  
A

1 
E G,u W
G , X , T ,  u , 
A
A
Alternative job arrival rate while training
Constructing the Likelihood function
• Each individual has a history of transitions and wage
observations:
–
–
–
–
Work in the same form as before (w)
move from employment to unemployment
Move from unemployment to employment (w)
move to a new firm (w)
• Each transition has a probability conditional on the state
variables, which include unobserved heterogeneity
Value functions and transitions
• Apprenticeship is chosen after integrating over all
future possibilities based on the decision rule
0
1



W

G,
0,
0,

,





Z,





Z,


G


W

E 0, G, 0, 0, 

0
0, 
0
0
A
iid cost shock
One off utility cost of training
Value of apprenticeship
Value of working as an unskilled worker
Exogenous variation and identification
• 20 cohorts in 11 regions
• Industries not uniformly distributed across country – this
makes different parts of the country vulnerable to different
shocks.
• Factor price equalisation ensures wages are independent of
local shocks
• Demand for labor and hence apprentices varies as a result
of these shocks. Thus these vary differentially by region
and cohort
Exogenous variation and identification
• Key identifying assumptions:
– Distribution of unobservables is the same across
cohorts
– Mobility for apprenticeship is costly (but not
necessarily prohibitively so)
– Factor price equalisation across regions
The Data
• German Administrative data
• 1% of all Social Security Records
• Men who do not work in the civil service
• All jobs since entry in the labour market are recorded with
exact transition dates
• Average daily wages within firm for year.
• Bottom and top coded – Not relevant in practice for this
educational group
The Data
• Period Covered 1975-1995.
• Only entry cohorts whose career start is observed
are used.
• we have 27525 individuals. We use 1400.
• The average age at first observation is 16.7.
• The oldest individual in our data is 35 years old.
.15
.1
-.05
0
.05
Wage Growth
0
20
40
60
time since entry on labor market (quarter)
80
Wage Growth - Skilled Workers
Stayers
.1
0
.05
-.05
Wage Growth
.15
.2
Movers
0
20
40
60
time since entry on labor market (quarter)
Wage Growth - Unskilled Workers
Movers
Stayers
80
Estimation
• Maximum likelihood.
• Assume problem stationary and solve using value function
iterations
• Use Gaussian quadrature for integrals
• Use only 1400 individuals picked randomly, for feasibility
purposes
• Skip to returns
• Skip to Simulations
Unobserved Heterogeneity
• We use a discrete bivariate distribution.
• Overall four types of individuals
• There are two types as far as the utility costs of
education are concerned
• There are two types for wage levels / productivity
• We allow these to be correlated, which gives rise
to the endogeneity of education
Unobserved Heterogeneity
• There is no initial conditions problem since we
observe all individuals as they enter the labour
market
• We assume that unobservables are independent of
region of birth and of time/cohort.
The fit of the Model
Goodness of Fit, Average Experience and Tenure over Time by Education
Mean Experience Apprentices
Mean Experience Non Apprentices
15
15
Observed
Predicted
10
Experience
Experience
10
5
5
Observed
Predicted
0
0
0
5
10
15
0
5
Time (Years)
10
Mean Tenure Apprentices
Mean Tenure Non Apprentices
12
12
Observed
Observed
Predicted
10
Predicted
10
8
Tenure
8
Tenure
15
Time (Years)
6
6
4
4
2
2
0
0
0
5
10
Time (Years)
15
0
5
10
Time (Years)
15
Employment Apprentices
1
0.8
% Employed
0.6
0.4
0.2
Observed
Predicted
0
0
5
10
Years
15
Empl oyment Non Apprenti ces
1
0.8
% Employed
0.6
0.4
0.2
Observed
Predicted
0
0
5
10
Years
15
Goodness of Fit, Observed and Predicted Log Wage, by Education
Apprentices
Non Apprentices
5
5
4.5
4.5
4
4
3.5
3.5
3
3
0
5
10
15
Time (Years)
20
0
5
10
15
Time (Years)
20
Results
Low wage types
High wage types
Results
Parameter
Qualified Apprentices
Standard dev. of innovation to match specific effect (
u)
Standard dev. of initial match specific effect (
0)
Non-Apprentices
0.086
(6e-5)
0.285
(0.003)
0.34
(0.005)
0.019
(0.002)
0.029
(0.002)
0.106
(0.004)
0.094
(0.006)
0.234
(0.009)
0.225
(0.006)
Quarterly job destruction rate ()
Quarterly offer arrival rate when employed ( 
W)
Quarterly offer arrival rate when unemployed ( 
U)
Note: asymptotic standard errors in parenthesis. When only one parameter estimate and its
standard error are presented in a row this parameter is restricted to be the same across
the two groups
Results
30
Average Treatment on Treated
Average Treatment Effect
% Wage Return to Apprenticeship
25
OLS Return
20
15
10
5
0
5
10
15
Years
20
Results
Returns
r 
E G,,W

G,X0,T0, , 
A

E G,,W 
Eu,G,X0,T0, , 
1
Average
Type 1
Type 2
Low Wage
Low Cost
Type 3
Type 4
High Wage
High Cost
Low Cost
High Cost
Return to Apprenticeship at age 15
Average Treatment Effect (ATE)
-1.7 %
5.9 %
2.2 %
-1.2 %
-5.5%
Average Treatment on the Treated (ATTE)
8.4 %
6.7 %
5.4 %
8.8 %
7.1%
ATE, net of utility of education
2.8 %
9.5 %
8.8 %
2.3%
2.3 %
ATE, net of opportunity cost of education
8.8 %
13.1 %
9.4%
9.6 %
5.3 %
Policy Simulations
• Consider introducing an EITC programme
• Employment effects for low paid workers
• Change of Incentives for education
• Change of incentives for Job Mobility
• Parameters
Policy Parameters
0
0
5
Benefits
Distribution of Daily Wage
.005
10
.01
15
.015
Density of Wages and In-Work Benefit Scheme
0
100
200
300
Daily Wage
Density of Wages
Benefits
Effect of Policies on education
% Individuals trained as Apprentices, Type 1
% Individuals trained as Apprentices, Type 2
0.05
% Deviation from Baseline
% Deviation from Baseline
0.05
0
-0.05
-0.1
Flat UB
0
-0.05
-0.1
EITC
Flat UB
% Individuals trained as Apprentices, Type 3
% Individuals trained as Apprentices, Type 4
0.05
% Deviation from Baseline
0.05
% Deviation from Baseline
EITC
0
-0.05
-0.1
Flat UB
EITC
0
-0.05
-0.1
Flat UB
EITC
The Impact of Policy on the take-up of
Results
% Individual in Employment
0.06
Flat Unemployment Benefit
EIT C
0.05
Deviation from Baseline
0.04
0.03
0.02
0.01
0
4
6
8
10
12
14
16
18
20
Time on Labor Market (Years)
Proportion Individuals Working, Compared
Results
Firm-Worker Match Specific Effect
0.005
Flat Unemployment Benefit
EIT C
0
-0.005
% Deviation from Baseline
-0.01
-0.015
-0.02
-0.025
-0.03
-0.035
-0.04
-0.045
4
6
8
10
12
14
16
18
20
Time on Labor Market (Years)
Policy Effect on Firm-Worker Match Specific
Conclusions
• The overall returns to apprenticeship, taking into
account all costs are very low.
• Apprentices earn more
• Apprentices keep more stable jobs.
• However, experienced non-apprentices are more
effective at finding new jobs after a layoff or a
quit – more flexible?
Conclusions
• The decision to join an apprenticeship is quite sensitive to
costs of education and to future returns
• Introducing an EITC type programme overall reduces the
incentive to obtain education
• However it can increase the incentive for lower ability
types because it makes it easier to obtain a job and benefit
from the subsidy
• EITC type programmes compress wages by reducing the
accumulation of search capital.