Transcript Do Individual Accounts Postpone Retirement?

Do Individual Accounts
Postpone Retirement?
Evidence from Chile
Alejandra C. Edwards and Estelle James
Work disincentives in old
Chilean system
Like other DB countries--Gruber & Wise
 Early pension easy & not decreased on
actuarial basis so taking pension at first
eligible age max PV of lifetime benefits
 Public sector--pensioners couldn’t work
 Private sector—could work but
incremental benefit small (50%,1%,70%)

2
Chilean reform of 1981





Replaced DB with DC—contributions go into
individual accounts, pension depends on
accumulation, on actuarial basis
Cut payroll tax, especially for pensioners
Raised retirement age and made it more
difficult to retire early
Most workers under 50 (born>1931) switched
We expect this to raise lfpr of older workers
through 2 channels:
– Postponed pensioning, so liquidity constrained
workers must work
– Increased work propensities among pensioners
3
1) Postponed pensioning required=>
raises lfpr through liquid income effect





Normal pension set at age 65 and 60 for men
and women respectively
Early pension not permitted until 1987
Tighter conditions for early pension—pension
must be 50% of own-wage and 110% of
minimum pension guarantee
If pension is postponed, it increases on
actuarial basis (50% in 5 years because
larger accumulation, fewer retirement years)
Therefore we expect postponed pensioning,
which should increase work propensities of
4
credit-constrained workers--liquidity effect
2) Reduced work disincentives, espec.
for pensioners (substitution effect)

Payroll tax fell from 23-25% to 12.5% so net
wage rises
 Pensioners exempt from pension payroll tax
 No penalty for work after pension
 These positive effects should be greater for
workers who were younger on date of reform
and should be phased in over time as more
older workers in new system
5
Data set from Greater Santiago Area
Household Surveys 1960-2002

Individual level data on
– Labor force participation
– Demographic characteristics
– Labor and pension income
We link to official macroeconomic data
 Shortcomings:

– Not longitudinal, no retrospective data
– We don’t know retirement age or type pension,
new or old system affiliates or no affiliation
– This leads us to underestimate impact of reform
on new system affiliates
6
Aggregate trends are consistent
with our hypotheses. We observe:





a decrease in pensioning after mid-80’s
an increase in labor force participation of 50+
after mid-eighties1981
higher participation rates among cohorts born
after 1931
these effects are stronger for pensioners
Our strategy: do these trends pre and post
reform remain after adding individual and
macro-economic variables? Answer: they do
7
Pension probabilities were rising,
but fell after reform
pension probabilities by age group over time
0.8
0.7
pension probabilities
0.6
40 - 44
0.5
45 - 49
50 - 54
0.4
55 - 59
60 - 64
0.3
65 - 69
70 - 74
0.2
0.1
0
1957-61
62-66
67-71
72-76
77-81
82-86
87-91
92-96
1997-2002
year observed
8
Male 50+ lfp rates were falling,
go up after the reform
labor forceparticipation ratebyagegroups
1.2
labor forceparticipation rates
1
40- 44
0.8
45- 49
50- 54
0.6
55- 59
60- 64
65- 69
0.4
70- 74
0.2
0
1957-61
62-66
67-71
72-76
77-81
82-86
87-91
92-96
1997-2002
year observed
9
Changes in lfp are more dramatic
for pensioners
Participationrates by age groups, pensioners andnon-pensioners compared
1.2
1
labor force participation rates
NP50 - 54
0.8
NP55 - 59
NP60 - 64
NP65 - 69
0.6
Pen50 - 54
Pen55 - 59
Pen60 - 64
0.4
Pen65 - 69
0.2
0
67-71
72-76
77-81
82-86
87-91
92-96
1997-2002
years observed
10
We use probit analysis to see if these
results hold after controlling for
other variables, :





2 dependent variables: probability of being
pensioned, probability of being in labor force
Individual characteristics: age, education, real
hh income, marital status, # children, spousal
age, etc.
Macro-economic variables: unemployment
rate, real annual GDP growth, deviations from
trend GDP
Pension status and amount
3 reform indicators
11
We model impact of reform in 3
alternative ways:
Change in time trends of pension rates and lfpr
rates for 50+ and 65+ groups pre- and post-reform
(reform effects start in 1987)
Change in cohort trends pre- and post-reform
(1931 birth cohort is the first exposed to reform)
1)
2)

Time and cohort trends are interacted with pension status
to allow for differences in trends
Dummies for groups of individuals exposed to postreform incentives.
3)

We expect an increasing fraction of 50+ individuals to
respond to the new system’s incentives each year, starting
in 1987
12
The data set: men 30+
Series of 40 annual cross-sections from
1960 to 2002 (1963-64 missing).
 Sample representative of Greater
Santiago (most urban 1/3 of country’s
population)

– 95,000 individual cases organized by year
of observation
– 93,000 individual cases organized by birth
cohort starting in 1900
13
Trends in other variables
Unemployment Rate
Median schooling
0.25
0.2
0.15
0.1
0.05
95
98
10
1
98
10
1
92
89
86
83
80
77
74
71
68
65
60
98
10
1
95
92
Hodrick-Prescott filter
95
lGDP
89
86
83
80
77
74
71
68
0
65
60
13
12
11
10
9
8
7
6
5
GDP deviations from trend
6.92551
0.2
0.15
0.1
0.05
92
89
86
83
80
77
74
71
68
60
-0.05
65
0
-0.1
-0.15
5.23164
60
64
68
72
76
80
Year
84
88
Hodrick -Prescott filter and ln(realGDP)
92
96
100
-0.2
14
Explanatory power increases
when co-variates are added, but
reform effect remains

Pension prob higher and work prob lower for
more educated workers over 50
– this increases gap explained by reform

Pension prob rise and lfp falls with UnE
– there is a full cycle after 1987

Adding pensioner status into lfp equation
raises R2, cuts pure age effect, and reduces
negative impact of macro effects
 Virtually entire downward trend in lfp before
reform and upward trend after reform is due
to pensioners
15
Main reform effects, after
controlling for co-variates:

Pension probability, age 50-64
– fell 15 percentage points, cut in half by 2002
– shift to non-pensioner status increased
aggregate work, since lfpr > for non-pensioners

LFP among pensioners, age 50-64
– rose >2% per year, >30 percentage points by
2002, compared to pre-reform trend line
– more than doubled mid-1980’s to 2002
– Smaller effects for older groups, age 65+
 These effects are increasing for retirees who
were younger on date of reform
 All 3 reform indicators give consistent results16
Estimated per year change in lfp
(percentage points) at means
non-pensioners
pensioners
Age 50-64
Before reform
.32
-.92
After reform
-.11
1.47
Net change
-.43
2.39
Age 65+
Before reform
.11
-.22
After reform
.07
.56
Net change
-.04
.78
17
Estimated per year change in lfp
(percentage points)
non-pensioners
pensioners
Age 60 in 1990
Before reform
.22
-1.45
After reform
-.08
2.32
Net change
-.30
3.77
Age 66 in 1990
Before reform
.19
-.50
After reform
.12
1.30
Net change
-.07
1.80
18
Policy implications and future
research

Incentives from shift to DC system have had
positive effects on supply of older workers
 Which is more important—actuarial fairness
or system constraints and taxes?
– Larger drop in pension prob before 65 suggests
that early retirement constraints play major role
– Larger lfpr effect among pensioners suggests that
exemption from payroll tax plays key role

Future research using new longitudinal data:
– Do workers take pension as soon as eligible?
– Has lfpr also increased among old system and nosystem affiliates (or does lfpr of those in new
19
system increase more when we identify them)?