Sie sind auf Seite 1von 51

lOMoARcPSD

Lecture notes Microeconomics of Household Behaviour,


lecture 7
Microeconomics of Household Behaviour (Rijksuniversiteit Groningen)

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

The retirement decision: lecture 7


Coordinator: prof. dr. Rob Alessie
email: r.j.m.alessie@rug.nl
oce: DUI 727 tel: 050-3637240
Lecturer: Viola Angelini email: v.angelini@rug.nl
oce: DUI 854 tel: 050-3633852
February 24, 2014

Literature
Lecture notes
Kapteyn, A., and K. De Vos (1998). Social security and labor-force participation in the Netherlands. The American Economic Review, 88(2),
164-167.

http://www.jstor.org/discover/10.2307/116912?uid=3738736&uid=2&uid=4&s
Stock, J. H., and D. A. Wise (1990): Pensions, the Option Value of Work,
and Retirement, Econometrica, 58, 1151-1180.
see Nestor
Belloni, M., and R. Alessie (2009), The importance of nancial incentives
on retirement choices: New evidence for Italy, Labour Economics, Vol 16(5),
Pages 578-588.

http://www.sciencedirect.com.proxy-ub.rug.nl/science/article/pii/S0927
Belloni, M., and R. Alessie (2013), Retirement choices in Italy: what an
option value model tells us, forthcoming in: Oxford Bulletin of Economics
and Statistics

Verspreiden niet toegestaan | Gedownload door Celine


1 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

AIM:
Studying models and approaches that analyze the decision to retire
Theoretical background: labor supply and life cycle models (see lecture
notes: Labour Supply)
Most economic models of retirement come from empirical literature: studying a particular pension system, policy eects of changes in the system
Degree of complication: models are dicult to solve
Consequence: none of the approaches in the literature is complete
Theory important for understanding the underlying mechanisms

Verspreiden niet toegestaan | Gedownload door Celine


2 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

This class
The option value of retirement: basic idea
Financial incentives of retirement: Kapteyn & De Vos (1998)
Example reduced form model: Belloni and Alessie (2009)
The option value model of Stock and Wise (1990)

Verspreiden niet toegestaan | Gedownload door Celine


3 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

The option value of retirement: basic idea


Individual is employed at date (or age) .
Decision to retire at date s .
What are the implications of retirement?
(i) Retirement is an absorbing state, i.e. a retired person does not return
to to the labor force.
(ii) Implications for income: retirement benet: Bt (s) for t s: income
at date t if someone retired at date s.
Note: pension benet systems are usually such that Bt (s + 1) Bt (s)
for t s + 1.
(iii) No labor income received anymore, full leisure: Lt = 1 for t s.
(iv) Wealth accumulation (denote Bt (0) the non-labour income while working):
At = (1 + rt )At1 + Bt (0) + wt (1 Lt ) cet , if t s 1
At = (1 + rt )At1 + Bt (s) crt , if s t T

(1)

Note: the nancial consequences of retirement, determined by the retirement system, are an important determinant of the retirement decision.
(v) Utility maximization becomes:
max

cet ,crt ,Lt

s1

t=

u(cet , Lt , t) +

u(crt , 1, t)

t=s

subject to the asset accumulation constraints (1).


Note: the utility maximization function is formulated conditional on s.
Denote the solution by V (s).
But the age of retirement, s, is a decision variable itself.
The decision to retire then is made by choosing the retirement age s that
has the largest value V (s):
max V (s) with s Tm .
s

with Tm the statutory retirement age (e.g. 65).


Verspreiden niet toegestaan | Gedownload door Celine
4 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

The solution gives the optimal retirement age s from the perspective of
someone at age .
Decision rule:
Retire at age if s =
Stay on the job if s > (and solve the problem again at age + 1)
Option value of retirement = Vtau (s ) V ( ): value of keeping the option
open to retire later If retirement is irreversible, this option is lost once the
decision to retire has been taken.
Continue working if Vtau (s ) V ( ) > 0. Then the option to retire later
has larger utility than the option to retire now.
Option value model by Stock and Wise.

Verspreiden niet toegestaan | Gedownload door Celine


5 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Financial Incentives to Retire and Alternative Exit routes


Empirical models of retirement:
Many models for retirement are developed from the empirical/econometric
literature
Theory and empirics:
All models in the literature solve only part of the puzzle, e.g. No
private savings, or only public pension system modeled.
Why? (i) Data limitations, (ii) Large degree of complexity in modeling the retirement decision: future, uncertainty, dynamic programming required.
Empirical models:
Structural models: try to implement the economic theory in the
empirical model. Option value model, dynamic programming models. Suitable for doing policy simulations.
Reduced form models: measuring the impact of explanatory
variables on the decision to retire. E.g. specify a Probit model
for the retirement probability, or a hazard rate model. Suitable
for nding out which factors inuence (or: are correlated with)
retirement.

Verspreiden niet toegestaan | Gedownload door Celine


6 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Figure 1:

2.1

Kapteyn en De Vos (1998)

In both structural and reduced form models of retirement the nancial


consequences of retirement play an important role: (see also the Option
Value).
Two concerns in the development of labour market participation of the
elderly:
1. Increase in percentage of population older than 65:
1950: 8 percent; 1995: 13 percent; 2050: 21 percent, see gure 1.
2. Low participation of workers in age range 55-65.
ad 1: Aging problem
a) Retirement of the baby-boom generation
b) Increase in life expectancy due increase in the quality of health care,
see gure 2.
c) Decreasing birth rates, see gure 3.

Verspreiden niet toegestaan | Gedownload door Celine


7 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Figure 2:

Figure 3: Births in the Netherlands (Van Ewijk et al. 2000)

Verspreiden niet toegestaan | Gedownload door Celine


8 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

In addition:
Due to longer period of education young workers enter the labour market at later age
Decrease in labour-market participation of older workers
Ratio of people with age 65+ to people with age 20-64 (grijze druk) will
increase from 22 % now to 43 % in 2040.
Consequence:
Increase in expenditures on Social Security (AOW), the Dutch state
pension
Social Security is nanced as pay-as-you-go
Properties:

50 % of minimum wage for each spouse in a couple


70 % of minimum wage for singles
Supplement for single parents with dependent child 40 %
Independent of earnings history, not means tested, mostly received
on top of employer pension plan.

Verspreiden niet toegestaan | Gedownload door Celine


9 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

ad 2: Low participation of workers in age range 55-65.


Development labor force participation of older workers, see table above
Financial incentives to withdraw from the labor market before age 65?
Various exit routes:
(a) Disability Insurance, WAO (DI)
(b) Unemployment Insurance, WW (UI)
(c) Early Retirement, VUT (ER)
Ad (a). Disability Insurance (WAO):
Level: 70 % previous wage
Eligible if disabled according to medical examination
Consensus between worker and employer: convenient for both, loose
interpretation of eligibility for a long time
Eligibility rules now have been sharpened: more part-time disabled,
change of job
PEMBA: premium dierentiation

Verspreiden niet toegestaan | Gedownload door Celine


10 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Ad (b). Unemployment Insurance


70 % of previous wage (gross)
Eligibility and duration depend on labor market history, job tenure
Maximum duration: 5 years. E.g. someone aged 60 will receive it till
65.
From age 57 on: no obligation to apply for job. (changed now)
Ad (c). Early Retirement
Employer provided: dierence per employer/sector (collective bargaining agreement
Level ranges from 70 to 80 % previous wage
Average age is 60 (depending on employer, sector)
Rights to ER are lost is one stops working before
Due to some policy reform, the ER exit route has become less attractive (a reform in 2006 is especially important)

Verspreiden niet toegestaan | Gedownload door Celine


11 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

How to quantify the financial incentives to withdraw from the labour


force?
Denition Social Security Wealth (SSW ) :
SSW = actuarially discounted sum of [current and future benets minus
current and future contributions]
In formula: SSW for a worker of age a who evaluates retirement at age h
(h a):

(s)Bh (s)
if h = a

s=h+1
SSWa,h =

(s)Bh (s)
(s)c(s) if h > a

s=h+1

s=a+1

where
:= Life span
Bh (s):= Pension benet received at age s if one retires at age h
c(s):= Contributions paid at age s
(s):= Discount factor at age s, which includes a real interest rate and
conditional survival probabilities
Factors that inuence SSW (by working another year, from t to t+1):
Premium contributions for UI, ER, SS and DI. negative inuence of
working.
The benet payment forgone of working at age t. Negative inuence
working.
Early retirement benet can be claimed if one continues working for
a year at age 59 if ER age is 60: positive impact on SSW at age 59.
Positive inuence of working another year
Additional pension rights accumulated if one works for another year.
Positive.

Verspreiden niet toegestaan | Gedownload door Celine


12 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Marginal Incentive (MI) measures


1. Accrual rate
ACCa = SSWa,a+1 SSWa,a
2. Implicit tax rate a
a =

ACCa
wa

where wa denotes labor earnings.


3. Peak value P Va
P Va = max(SSWa,h SSWa,a )
h

4. Option value
Kapteyn en De Vos (1998) compute the values of SSWa,h , AAa and a for
various cases.
Constant real benets after 1995
Mortality rates independent between worker and spouse
Real interest rate is 3 percent
Tax system remains the same from 1995 on
Wage development same as minimum wage
Men with median earnings, born in 1930 (a = 54, spouse without
income
Entitled to Early Retirement (ER) benets: from age 60 on

Verspreiden niet toegestaan | Gedownload door Celine


13 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Case A. Worker who is participating in Early Retirement scheme.


Results, see gure 2.1.
Comments, (see Panel A of table below)
ER Benets 80 % of previous earnings, after tax 90 % (replacement
rate)
Age 55-58: a > 0, but a < 1: it pays to keep on working
Age 59: one additional year of work leads to eligibility of early retirement (VUT): a = 3.777 < 0: subsidy to continue working
Age 60-64: decrease in SSW is large than the wage if an additional
year of work is added strong incentive to stop working at age 60.
Case B. Worker with the possibility to enter DI (WAO).
Entitled to DI benets before age 65
Replacement rate: 70 percent; after tax: 80 percent
results panel B of table below
a > 1: for all ages 55-64 strong incentive to exit workforce by WAO
if possible.
Overall conclusion: various social insurance systems provide strong nancial incentives to withdraw from the labour market at an age below
65.

Verspreiden niet toegestaan | Gedownload door Celine


14 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Verspreiden niet toegestaan | Gedownload door Celine


15 van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

The Importance of Financial Incentives on


Retirement Choices: New Evidence for Italy
Michele Belloni
CeRP - Collegio Carlo Alberto

Rob Alessie
University of Groningen, Netspar and Tinbergen Institute

ICEEE 2009 Congress


Ancona,30th January 2009
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Overview

Aim: use a panel dataset covering the period 1985-2001 to analyze


the impact of financial incentives provided by the social security
system on retirement choices in Italy

Motivation: a good understanding of the role of financial incentives


is crucial for the design of effective reforms which aim to increase
the average retirement age

Method: following Gruber and Wise (2004), we estimate a probit


model in which the probability for a worker to retire is explained by
financial incentives, in addition to personal characteristics.
Identification exploits the time variation in financial incentives which
stems from a series of social security reforms introduced in Italy in
the 1990s

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

The basic idea: forward-looking worker


At age a he computes:
I

his current social security wealth: SSWa

his SSW associated with retiring at alternative future ages:


SSWa+1 ,SSWa+2 ...

MI variables: ACCa = SSWa+1 SSWa or


PVa = max(SSWh SSWa ) h = a + 1, . . . , R.

two key variables in the model: SSW and MI (ACC or PV)


I

the higher her SSWa the higher her retirement probability (income
effect);

the higher her ACCa (or PVa ) the lower her retirement probability
(substitution effect);

The model is dynamic: worker reconsiders retirement next year if he did


not retire at age a, since new information affecting his expectations are
available. Retirement is an absorbing state.
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Literature: main references and contribution 1


I

I
I

Gruber and Wise (2004): collects 12 recent country studies,


common econometric approach. General finding: strong causal
effect of financial incentives on retirement
Brugiavini and Peracchi (2004): contribution for Italy in the book.
Weak evidence for SSW and MI
Other studies for Italy (Brugiavini and Peracchi, 2003 and 2007;
Ranzani 2006) confirm the weak evidence
Is Italy a puzzling exception?

previous studies exploit various versions of the INPS-O1M archive, which does
not provide info on seniority. Seniority:
I
I
I

is crucial to define eligibility and to compute pensions in Italy


needed to be imputed, under strong hp on the unobserved part of the
career; consequence is possible measurement error in SSW
is observed in our data

to what extent does having better data help explain differences in results
across countries?
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Literature: main references and contribution 2


Most studies on retirement (e.g. Gruber and Wise, 2004) do not account
for individual unobserved heterogeneity:
I

Individuals might differ in their preferences (e.g. for leisure) for


reasons not observed by the econometrician
possible dynamic sample selection bias: work-loving individuals stay
longer in the sample

Individual effects and financial incentive variables can be correlated


(Chan and Stevens, 2004; Coile and Gruber, 2000): e.g. work-loving
individuals tend to have higher SSW and low retirement probability
I

Wooldridge (2002) describes how to account for a correlation


between individual effects and r.h.s. vars in probit models

we extend the random effects probit model explaining the retirement


probability by including, as extra regressors, financial incentives in
the first year in which each worker is observed. As a result, financial
incentive parameters are identified exploiting within-individual
variation

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Institutional framework: overview


I

we model the rules of private sector employees: INPS-FPLD fund

two main exit routes o.l.f.: old-age and seniority pensions

Several laws and three main reforms, in 1992, 1995 and 1997, affected
eligibility (see table) and DB formula

Old-Age pension
Year
1985-1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002-2003
2004-2005
2006-2007
> 2007

age
males
females
60
60
61
61
62
63
63
64
65
65
65
65
65
65

55
55
56
56
57
58
58
59
60
60
60
60
60
60

Seniority pension

seniority

age+seniority

seniority only

15
16
16
17
17
18
18
19
19
20
20
20
20
20

52+35
52+35
54+35
55+35
55+35
56+35
57+35
57+35
58+35
58+35

35
35
35
35
36
36
36
37
37
37
37
38
39
40

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Data and sample selection


Data: WHIP + additional INPS pension file
I

WHIP: random sample drawn from INPS archive. Unbalanced panel:


1/90 of private sector, non agricultural workforce in 1985-2001.
I

O1M files: data on working careers (wages, weeks worked,


occupation). Exploited in previous studies
complementary files: data on spells of self-employment (artisans and
traders), atypical work, unemployment, and mobility
main weaknesses: no public, nor agricultural sectors; few individual
characteristics (gender, date and region of birth)

additional INPS pension file: data on pensions paid (dates of


payments, amounts, type of pension) and seniority

Sample selection:
workers aged 50-70, cohorts 1935-1941
workers with long interruptions in the career or contributions mainly
to other schemes automatically excluded
I retirement through disability, unemployment or mobility is
considered involuntary
I employee is defined retired if permanently leaves the O1M archive
Verspreiden and
niet toegestaan
| Gedownload
Celine van Essen
(eline.van.der.rest@hotmail.com)
does not then
work asdoor
a self-employed
or atypical
worker
I
I

lOMoARcPSD

Financial incentives
definition:

SSWa,h

P
s=h+1

(s)Bh (s)
(s)Bh (s)

s=h+1

if h = a
h
P

0 (s)c(s)

if h > a

s=a+1

ACCa

SSWa,a+1 SSWa,a

PVa

max(SSWa,h SSWa,a )

h = a + 1, . . . , R

computation:
I

expected wages: prediction from AR(1) fixed-effects model for log-wages


estimated on O1M data

individuals know the pension formulae and hold static expectations

(s) is evaluated allowing for variation by age, gender, region and cohort
in conditional survival probabilities

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

The retirement model


We consider the following latent variable retirement model for worker i in
period t (t = 1, . . . , T )
yit = wit0 + ci + it

(1)

where we observe
yit = 1

if yit > 0

yit = 0

otherwise

(retires)

due to the absorbing state assumption, yit = 0 implies yis = 0 for each s < t
20
wit0 = SSWit 1 + MIit 2 + x10
it 1 + xi 2

key parameters: 1 expected positive, 2 expected negative

x1it includes time-varying variables: age, expected and current wage,


pensionable earnings, occupation, sector and year dummies; x2i includes
time-constant variables: cohort and area of birth dummies

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

we make the following standard assumption for it


it |wit , ci NID(0, 1)
equation (1) contains an individual effect ci which captures unobserved
differences across individuals in, for example, taste for work. The individual
effects and financial incentive variables can be correlated.
Following Wooldridge(2002) we model it as follows:
ci = z0i1 + i

2
ci |wit , NID(z0i1 ,
)

where zi1 (SSWi1 , MIi1 , x1i1 )0 . measures the correlation between financial
incentives and the individual unobserved effect (no clear economic
interpretations can be given to its estimates)
The likelihood contribution of individual i who retires in period T is:
"T
#
Y 

0
0
Li (, |yi , wi , zi1 , i ) =
(2yit 1) wit + zi1 + i
t=1

where i is integrated-out in the standard way


Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Results: males

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Results (other)

policy simulations as in Gruber and Wise (2004): three year


reform increases the average retirement age by up to 1.5 years

females react stronger to financial incentives than males

results robust to alternative assumptions on expected future wages


and specifications for age

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Discussion
Is Italy a puzzling exception? To what extent does having better data
help explain differences in results across countries?
Controlled experiment: suppose we do not observe seniority in the data
1. impute the variable seniority as in previous studies:
I

I
I

from SHIW data compute average age of entry into the l.m. by
gender and occupation for similar cohorts
impute this average to each worker according to her characteristics
assuming a continuous career, compute seniority as the difference
between age and the imputed age of entry into the l.m.

2. compute SSW and MI according to the imputed seniority:


considerable overestimation of SSW
3. estimate a pooled probit model using our dataset and replacing
actual SSW and MI with their imputed counterparts

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Figure:

Country studies in Gruber and Wise


(2004) and this paper: estimated sign for PV
and SSW and quality of the data

I results based on imputed seniority (BA-i)


are qualitatively similar to ITA
I The quality of the SSW and MI variables
differs across countries because of data
limitations. Data requirements may be
country-specific because pension systems
differ. We construct a quality index
I country studies which dispose of good
(??) data, report correct signs for the
PV and SSW coefficients, with the
exception of Belgium and Spain
I In countries with incomplete (?) data,
one generally obtains the wrong sign for
SSW and the correct one for PV
I These findings are in line with those of
the controlled experiment, suggesting
that measurement error in SSW explains
the pattern observed in the table
I caution: to our knowledge, the statistical
properties of the probit model taking into
account the non-classical measurement
error problem have not been formally
derived yet

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Conclusions
I

We quantify the effects of financial incentives on retirement in Italy

we exploit a dataset which - for the first time in Italy - provides info
on seniority
our methodology takes into account correlated unobserved
heterogeneity

Main findings:
I

in accordance with Gruber and Wise (2004), financial incentives have


an effect on retirement, this effect goes in the expected direction

the procedure used in previous Italian studies to impute seniority


leads to a considerable overestimation of that variable and of SSW.
We show that, due to these measurement errors, the estimate of the
SSW coefficient takes the wrong sign

comparison of retirement studies across countries provides prima


facie evidence that a lack of good quality data often leads to
wrongly signed estimates of the SSW coefficient

Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

4. The option value model of Stock and Wise (1990)


A. The option value model: Stock and Wise (1990)
B. Three models of retirement: Computational Complexity versus Predictive Validity
Lumsdaine, Stock an Wise (1992): Next week

27
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

A. The option value model: Stock and Wise (1990)

Stock and Wise (1990) use data on employees of large Fortune 500 firm. Pension
scheme:
- Regular pension benefit from the age of 65 on
- Age 55-64: entitled to Early Retirement (VUT). Reduction of normal pension benefits:
actuarially beneficial (about 3% per year).
- Age 50-54: retirement possible, but based on regular pension benefit spread over 10
additional years: benefit reduced actuarially (about 7% a year)
- From age 62: entitled to Social Security (comparable to AOW)
- From age 65: Social Security subtracted

28
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Type of benefit: Defined benefit, depending on


- age
- tenure, tenure > 10
- final wage
B = k x (#years served) x (final wage)

29
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Retirement:
- absorbing state
- pension scheme determines financial incentives
Financial incentives: retirement leads to
- stop of wage as source of income
- stop of pension accruel
- obtaining income out of pension (if eligible)
Trade-off between these financial consequences:
- At each possible retirement age (50-68) determine the present value of income
At which age is this present value highest?

30
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Definitions:
- Ys: wage income at age s (yearly basis)
- Bs(r): pension benefit if worker with age s retires at age r
For a worker of age 50 who retires at age r we have:
Present value of wage (earnings):
50

r 1
50

Y
1 i

(with i = 5%)

Present value pensions (retirement benefits):


50

1
r

1 i
S

(with S end of life)

B (r )

Table 1: figures in US dollars

31
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

32
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

The optional value model for retirement


- Behavioural model, not only financial incentives
- Decision rule for retirement
A. Definition of utility and value
Utility function:

Uw(Ys) : utility for someone who works with income Ys


Ur(Bs(r)) : utility for someone retired with benefit Bs(r)

Note: even if Ys = Bs(r) the utility levels may be different due to differences in
preferences for working and not.
Value for someone with age t who retires at age r :
r 1

Vt ( r ) U w (Ys ) s tU r ( Bs (r ))
s t

sr

s t

with a discount factor measuring time preference. Vt(t) is the value of retiring now.
33
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

B. Decision rule
Uncertain future: base decision on expected value:
EtVt(r)
- Determine the maximum expected present value of retiring in future:
*
*
max EtVt (r ) EtVt (r ) (so maximum at r ).

r {t 1,..., S }

- Determine EtVt(t): value of pensioning now.


- Expected return at time t of postponing retirement to r*:

Gt(r*) = EtVt(r*) EtVt(t)

- Gt(r*) measures the gain from retiring in the future


- Retire NOW if Gt(r*) 0
- Continue working if Gt(r*) > 0

34
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

C. Specification
Uw(Ys) = Ys + s
With a parameter and s a random proxy for unobserved determinant of utility.
Ur(Bs) = (kBs) + s
With k a parameter that determines the difference in the marginal utility of income in
different labour market states:
k > 1: (marginal) utility retirement > (marginal) utility work (at same income)
k = 1: only income itself matters, indifferent between states
k < 1: disutility of retirement

s: random variable.

35
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Interpretation of s and s :
Unobserved (or unobservable) factors that influence utility like
- individual preferences
- health status
- private savings
Expected Values for different retirement ages are compared: Uncertainty:
- In preferences: A stochastic process for s and s is specified (as an AR(1) process,
Normally distributed disturbances, see Stock and Wise for details)
- Income: an equation for income is specified to be able to determine the expected future
value of income, and to incorporate income uncertainty: Income is specified as a
function of Age and Tenure, Dynamic process (see Stock and Wise for details).

36
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

D. The retirement probability can be specified:


Two variants:
1 year model: cross section data:
Probability of retirement of someone aged t = P(Gt(r*) 0)
Multiple years: uses multiple observations for 1 individual:
Probability of retirement at age :
P(Gt(r*) > 0, Gt+1(r*) > 0,, G-1(r*) > 0, G(r*) 0)
(retirement at , no retirement previous years).

37
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

E. Results
Sample of 1500 workers of Fortune 500
- aged 50 or older
- at least 3 years of tenure
Table IV: Multiple year model, 6 variants estimated:
- variant 1: same retirement probability for everyone (simplification, lowerbound on
model fit)
- variant 2: retirement probability the same for everybody with the same age
- variant 4: the complete model
Note:
< 1: Decreasing marginal utility of income, some risk aversion
k > 1: Higher utility of retirement ( at equal income)

38
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

39
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Variant 5: Extension: utility is made dependent on age.

age
k k

55

k1

Why?
- Variant 4: Age only enters through the income process and the pension system.
- Can the income process and the pension system (that determine utility) completely
describe the retirement decision or are there age effects that cannot be described?
- Thus, variant 5 serves as a test of the model
- Outcome: k0 = 0.950, k1 = 4.87 > 1: Utility of retirement rises with age.

40
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

Fit of the model: Table V


Compare observed retirement rate with the predicted
- Model is able to fit the early retirement peak at age 55
- Hard to predict retirement rates above age 64 due to low number of observations.
- There may be a customary retirement age effect not associated with monetary gain at
age 65

41
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

42
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

F. Policy Simulations
Use the model and the estimated parameter values to simulate the effects of changes in
the pension scheme.
a). Increase of early retirement age from 55 to 60.
b). From defined benefit to defined contribution:
ad a). effects: Table VI.
- Early Retirement age of 55: before age 60 65% has already retired
- Early Retirement age of 60: before age 60 only 42% retired
- Note also higher retirement rates before age 55: individuals who before waited with
retirement until age 55 to become eligible dont want to wait until age 60 and retire
earlier than before.
Ad b). Towards defined contribution: no large peaks in retirement any longer: Table VII
- higher retirement rates before and after the age of 55. Actuarially Fair.

43
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

44
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

lOMoARcPSD

C.
45
Verspreiden niet toegestaan | Gedownload door Celine van Essen (eline.van.der.rest@hotmail.com)

Das könnte Ihnen auch gefallen