Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter.
Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.
OCR for page 173
8
Retirement Process in Japan:
New Evidence from the Japanese Study
on Aging and Retirement1
(JSTAR)
Hidehiko Ichimura and Satoshi Shimizutani
O
ne of the most distinct characteristics of the Japanese labor market
of the elderly is “late retirement,” compared to the other countries
of the Organisation for Economic Co-operation and Development
(OECD). The data on effective retirement age, which is most frequently
quoted for an international comparison, show that the average effec-
tive retirement age for Japanese males is 69.5 years and for females is
66.5 years. These are the oldest ages among developed countries (Organ -
isation for Economic Co-operation and Development, 2008).
Clearly, this measure alone is insufficient to capture decisions about
retirement. At least three limitations are pointed out in the literature. First,
the definition of retirement depends on subjective perceptions that may
differ across individuals (Lazear, 1986; Lumsdaine and Mitchell, 1999). For
example, several studies have revealed that the timing of retirement does
not coincide with the decision to leave the labor force or to receive pen -
sion benefits (e.g., Banks and Smith, 2006, for the United Kingdom, and
Shimizutani, 2011, for Japan). Second, individuals may not retire at once
1 Thisstudy was prepared for the Conference on Policy Research and Data Need to Meet
the Challenges and Opportunities of Population Aging in Asia, New Delhi, on March 14−15,
2011. We are grateful to the Indian National Science Academy, the host of the conference,
and other sponsoring organizations. We also thank Asako Jufuku, Hirokazu Matsuyama,
and Yuta Kikuchi for excellent research assistance. In addition, we thank the referees and
Daigo Nakata, as well as participants at the conference for their constructive comments.
Ichimura thanks the Japan Society for the Promotion of Science for its support through the
Basic Research Grant.
173
OCR for page 174
174 AGING IN ASIA
but gradually, and the process of retirement may take some time. In addi -
tion, retirement may not be an absorbing state (Banks and Smith, 2006).
Third, the retirement decision may be jointly made by a married couple
(Gustman and Steinmeier, 2009). If this is the case, retirement behavior
needs to be considered as an outcome of intrahousehold decision-making,
in addition to a variety of factors including socioeconomic, health, and
other circumstances.
In this chapter, we will describe Japanese workers’ retirement pro-
cesses using the Japanese Study on Aging and Retirement (JSTAR). JSTAR,
for the first time, provides publicly available panel data on individuals
who were between the ages of 50–75 in 2007. To our knowledge, this
study is the first to explore the retirement process in Japan using panel
data. Thus, the contribution of this study is to provide new evidence on
the process uncovered by JSTAR.
While research on retirement in Japan has been accumulated, the
studies are limited in two ways.2 First, the studies use cross-sectional data,
which makes it impossible to uncover a retirement “process.” Second, the
studies use data sets with a very limited variety of variables. In particular
do they contain family demographics, such as spouses’ work status or if
they have elderly or other dependents?3
JSTAR, a sister survey of the Health and Retirement Study (HRS),
English Longitudinal Survey on Ageing (ELSA), and Survey on Health,
Ageing and Retirement in Europe (SHARE), overcomes those two obsta -
2 Research carried out in Japanese workers’ retirement behavior is largely limited to two
areas: the labor supply effect of social security earnings test and the effect of mandatory
retirement on the transition from a primary job to a secondary job.
3 Some existing surveys are often used in analysis of aging in Japan. The National Survey of
Family Income and Expenditure collects data every five years on a wide variety of economic
variables and family demographics but less information on health. The Comprehensive
Survey of People’s Living Conditions is implemented every three years with small-scale
surveys in between years to collect rich information on health, family, and some economic
variables. The Survey on Employment of the Elderly focused on working conditions and
experience of the elderly between 55 and 69 but ended in 2004. Those surveys are large but
cross-sectional. On the other hand, there are three panel data sets on elderly people. The
National Long-run Panel Survey on the Life and Health of the Elderly started in 1987 and
collects data every three years, which is a Japanese version of AHEAD. Together with the
Nihon University Japanese Longitudinal Study of Aging, these surveys provide detailed
information on the health status of the elderly aged 60 (or 65) and older and less information
on economic status. Thus, the retirement process is not captured well. Lastly, the Ministry
of Health, Labour, and Welfare started a panel survey of the senior population (Chukonen
Jyudan Chosa), tracking individuals in their 50s in 2007 every two years. The sample
size is larger than that of JSTAR with nationwide regions, but the information is insufficient
to capture precise amounts of pension income or medical/long-term care expenses. It also
lacks data on previous working experiences or future expectations. In most cases, microdata
are not accessible or only limitedly accessible.
OCR for page 175
175
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
cles. JSTAR contains a variety of variables comparable to those in HRS/
ELSA/SHARE and intends to address a variety of socioeconomic issues
related to the aging population, with an emphasis on both interdiscipli-
narity and international comparability (see Ichimura, Hashimoto, and
Shimizutani, 2009).
MEASUREMENT OF RETIREMENT
Retirement depends on definition. The definitions include an affir-
mative answer to a question regarding retirement status: “Are you cur-
rently retired?” as well as a state in which the individual is out of the
labor force with the intention of remaining out permanently, and a state
in which the individual receives some of his or her income as pension
benefits (Lazear, 1986).4 We explore retirement behavior using the three
measures by examining the first wave (baseline) of JSTAR in this section.
The sample in the baseline are those who are aged 50–75 and randomly
chosen from household registration after regional stratification in each
of the five municipalities in 2007.5 The sample size is more than 4,000,
excluding those who did not provide information on work status from
the total sample size of about 4,200.
Figures 8-1a and 8-1b illustrate nonworking status and its decomposi-
tion for males and females separately.6 For males, the proportion of the
nonworking very gradually increases from less than 5% at age 50 to about
8% at age 59, but the share jumps at age 60 to about 17% and increases
along with age in the 60s. However, the nonworking proportion is still
only slightly above 60% at around age 70. Most nonworkers are accounted
for by retirement, but only slightly above 50% classify themselves as
retired at age 70. The results differ from those by Banks and Smith (2006),
4 Lazear (1986) includes further definitions such as (1) a state the individual has reduced
his/her hours substantially from some lifetime average and intends to maintain hours at or
below the current level, (2) a state that the individual appears on some company’s retire -
ment roll, and (3) a state that the individual receives a primary social security payment. We
will refer to (1) below.
5 Note that JSTAR does not employ a probabilistic national sampling but has an emphasis
on securing a larger number of samples in the same socioeconomic environment.
6 JSTAR asked respondents and spouses, if any, to choose one among the following choices
when asked about their current working status: (1) currently working, (2) leave of absence,
(3) not currently working, (4) don’t know, and (5) refuse to answer. Respondents who choose
(1) or (2) are “working” and those who choose other choices are further asked whether they
are searching for a job currently or plan to search in the future. If the answer is affirmative,
they are categorized as “unemployed.” The respondents who are neither explicitly work -
ing nor unemployed are further divided into retired, homemakers, or medically treated. As
explained above, these questions are also asked for the spouses, but we only use the data
on the respondents in this chapter.
OCR for page 176
176 AGING IN ASIA
A. Male
B. Female
FIGURE 8-1 The proportion of males and females who are nonworking.
SOURCE: Data from JSTAR (2007).
R02177
Figure 8-1
OCR for page 177
177
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
which reveal that nonworking status and retirement is identical for people
aged 65 and older in the United Kingdom. For females, the proportion of
those nonworking is higher than that for males and increases with age
after 50, compared to 60 for males. At a closer look, the proportion starts
at about 12% at age 50 and increases to about 40% at age 60. It continues
to increase in the 60s, reaching 70% at age 70. In contrast to males, a larger
fraction of women’s nonworking status is accounted for by homemaking,
not by retirement. We should note that this must be women who are now
no longer working and describe themselves as “homemakers” rather
than “retired,” although they are retired in the sense of having left the
labor force as they reach traditional retirement ages. Those patterns do
not differ much across different educational attainment either for males
or females (results are omitted to save space).7
Figures 8-2a-d present the distribution of actual and expected retire-
ment age in the first wave. We use the term “actual retirement age” for
those who have already retired to differentiate from “expected retirement
age” referring to those who have yet not retired.8 For males, the left panel
of the figure shows twin peaks in the histogram of actual retirement age,
and the mode (25%) is found at age 60, followed by age 65 (15%). In
contrast, the right panel shows that the age to retire in the future is con-
centrated at age 65, followed by age 70 and age 60. While omitted to save
space, the distribution of actual retirement age is homogeneous across
different levels of educational attainment, while that of expected retire -
ment age is later for lower educational attainment. The largest fraction is
observed at age 70 among those who completed junior high school only.
For females, the largest fraction in distribution of actual retirement
age (left panel) is observed at age 60, which is also the case for males but
the distribution is flatter, implying the distribution has a single peak at
age 60. In contrast, the largest fraction in expected retirement age (right
panel) is found at age 65, identical with the case for males, but the second
peak is found at age 60 in contrast to age 70 for males. When decomposing
by educational attainment, females’ expected retirement age is later at age
70 for lower educational attainment.
In sum, the most frequently observed retirement age for those who
have already retired is age 60 for both sexes, followed by age 65 for males.
The most popular retirement age for those who are expecting to retire
7 The proportions of nonworking persons from the Labor Force Survey are 6.9% (39.2%)
for those aged 55–59, 25.6% (57.8%) for those aged 60–64, and 51.5% (74.2%) for those aged
65–75 for males (females).
8 A very small portion of the respondents had retired before reaching age 50, and they are
omitted from the figures. The sample size is 438 (797) for males and 57 (450) for females for
actual (expected) retirement age. Seven respondents answered in a range (i.e., I expect to
retire between age A and age B) and are excluded.
OCR for page 178
178 AGING IN ASIA
A. Male, actual B. Male, expected
.25
.4
.2
.3
.15
Fraction
.2
.1
.1
.05
0
0
50 55 60 65 70 75
50 60 70 80 90 100
Age Age
C. Female, actual D. Female, expected
.3
.2
.15
.2
Fraction
.1
.1
.05
0
0
50 60 70 80
50 55 60 65 70 75
Age Age
FIGURE 8-2 The distribution of retirement age.
SOURCE: Data from JSTAR (2007).
is age 65 for both sexes, followed by age 70 for males and by age 60 for
females.9 The distribution of actual retirement age does not differ much
across educational attainment for both sexes, but the expected retirement
age tends to be later for those with lower education.
Of course, these patterns may be a reflection of employment institu -
tions, such as the start year to receive pension benefits. Thus, we turn to
an examination of the distribution of the age to receive a pension.
The public pension programR02177 consists of three programs: the
in Japan
Figure 8-2
Employees’ Pension Insurance (EPI, Kosei Nenkin) whose pensioners
parts a, b, c, and d combined to fit on one page
vectors, editable
9Rust (1989) found “twin peaks” in the retirement ages for older Americans who file for
Social Security benefits using the Retirement History Survey (RHS) in the 1970s. The two
peaks are observed at age 62 when the individual is eligible to receive a reduced benefit and
at age 65 when the individual is eligible to full benefits. Lumsdaine and Mitchell (1999) argue
that the two marked peaks remain after controlling for pension income available at those ages.
OCR for page 179
179
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
are private-sector employees; the Mutual Aid Insurance (MAI; Kyosai
Nenkin) covering employees in the public sector and private schools;
and the National Pension Insurance (NPI; Kokumin Nenkin) whose pen-
sioners are not covered by EPI or MAI.10 NPI has a flat-rate benefit only,
and the normal eligibility age is 65 for both sexes. The minimum years
of contribution is 25 years, and the monthly benefit for the fully insured
(with 40 years of contribution) is about 66,000 yen per month (about US
$800). The NPI program allows a 10-year window in claiming benefits.
Individuals who claim benefits between ages 60 and 64 undergo benefit
reduction, and individuals who claim benefits between ages 66 and 70
enjoy benefit rewards.11
The EPI program consists of flat-rate and wage-proportional compo -
nents. The flat-rate component has the same contribution-benefit structure
as NPI and the wage-proportional component depends on age, months
of contributions, and a benefit multiplier that differs across gender and
birthday. The normal eligibility ages for both components of EPI are set at
age 65, but EPI beneficiaries are also entitled to receive a “special benefit”
before age 65 that is close to formal benefits in most cases. The normal
eligibility ages for special benefits differs between males and females and
between flat-rate and wage-proportional components. As of 2011, the
eligibility age for the wage-proportional component is 60 for both sexes,
not allowing earlier or later claiming. Meanwhile, the eligibility age for
the flat-rate component has gradually risen since 2001, and it was 63 for
males and 61 for females in 2007. EPI beneficiaries were able to enjoy
earlier claiming of the flat-rate component of a special benefit for males
aged 60–62 and females aged 60 in 2007. One can delay either the flat-rate
or wage-proportional component. (See the detail formula in Shimizutani
and Oshio, 2011.) In contrast with some European countries that have
high take-up rates, the disability program participation is still low and
the effect on labor force participation is very limited in Japan. The main
reason is the strict eligibility rules, although major revisions to the dis -
ability program have slightly expanded the eligibility for these programs
(Oshio and Shimizutani, 2011).
Together with the social security program, the employment policies
for the elderly have been reformed, focusing on extension of mandatory
retirement age. In 2004, the Employment Measures Law was revised to
include an obligatory clause that requires firms to raise the mandatory
10 In terms of the number of pensioners, EPI and NPI contributed to the total by slightly
less than one-half respectively, and MAI occupies the remaining small portion.
11 For those who were born after April 2, 1941, the actuarial reduction rate before age 65
is 0.5% per month and the actuarial credit rate after age 65 is 0.7% per month (Shimizutani
and Oshio, 2011).
OCR for page 180
180 AGING IN ASIA
retirement age to 65 or above by 2013 or to completely abolish it. The pro-
portion of firms with mandatory retirement steadily increased to above
90% in the mid-1990s, and the most dominant retirement age is now 60.
Some firms have indeed started extending it further to age 65 (Oshio,
Oishi, and Shimizutani, 2011).
Figures 8-3a and 8-3b depict the distribution of age to start receiv-
ing any type of public pension benefits. The sample is confined to those
who have received any benefits. For both sexes, close to one-half of the
respondents started to receive pension benefits at age 60. The second larg-
est fraction is found at age 65: one-quarter for males and more than 30%
for females. This observation reflects the eligible ages to receive public
pension benefits.
That the proportion of those who started to receive pension benefits
at age 65 is larger for females reflects the fact that a larger fraction than
males are NPI pensioners. By educational attainment, females who are
junior high school graduates represent the largest proportion at age 65,
followed by age 60, which also is a reflection that a larger proportion of
NPI pensioners are females rather than males. The distribution of males
is not changed across educational level.
The observation in this section is that age 60 is a specific age in Japan
to retire, probably because it is the age at which people become eligible to
receive pension benefits. Because the eligible age for EPI pension benefits
is now in transition from 60 to 65, it is natural that the expected retirement
age is changing to age 65 for the yet-to-be-retired group. However, we
should keep in mind that the proportion of people working exceeds more
than 30% at age 70 and some portion of the elderly keep working in their
later age. In other words, the institutional reason is an important factor to
account for retirement behavior but cannot completely explain labor sup -
ply behavior of the elderly.12 This is what we examine in the next section.
TRANSITION IN WORKING STATUS BETWEEN
JSTAR FIRST AND SECOND WAVES
This section focuses on the transition of work status, using both the
first and second waves in JSTAR. By doing so, we capture retirement “pro-
cess,” which has been unexplored in Japan. The sample is confined to the
respondents who were interviewed in both waves in the five municipalities.
First, we preview retirement process transition between two years in
terms of the change of work status and hours worked before retirement.
12 Banks and Smith (2006) provide evidence that the proportion of nonworking and retire-
ment jumps to 100% at age 65 in the United Kingdom because of an institutional reason:
pension benefits depend on the last salary.
OCR for page 181
181
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
A. Male
.5
.4
.3
Fraction
.2
.1
0
50 55 60 65 70
Age
B. Female
.4
.3
Fraction
.2 .1
0
50 55 60 65 70
Age
FIGURE 8-3 The distribution of starting age to receive benefits.
SOURCE: Data from JSTAR (2007). R02177
Figure 8-3
parts a and b combined to fit on one page
vector editable
OCR for page 182
182 AGING IN ASIA
The work status and hours worked are measured at the time of inter-
view. Table 8-1 shows the change in work status between the first and
the second wave in three definitions (working/nonworking, employed/
self-employed, and full-time/part-time status) in three age ranges (60–64,
65–69, and 70 and older as of the first wave). In what follows, we call those
who are wage earners and not self-employed “employed” and those who
are working on a regular basis “full-time” workers. For males, the upper
panel of the table shows that the transition probability into “not work -
ing” from “working” increases after age 65 from about 20 to 25%. The
transition probability into “working” from “not working” drops sharply
after age 65 from 17 to 5% and remains the same for the age group 70–75.
For females, the transition probability into “not working” from “work-
ing” increases after age 70 from less than 20 to 27% while the transition
probability into “working” from “not working” gradually drops from
about 8 to 3% from age 60–75. The middle panel shows that there is very
little transition between self-employment and employment status from
age 60–75 for both sexes.
TABLE 8-1 Transition of Work Status Between Two Years
Male Female
2009 2009
Age Working Not Age Working Not
60−64 Working 60–64 Working
2007 Working 80.3% 19.7% 2007 Working 83.0% 17.0%
Not 17.1% 82.9% Not 8.2% 91.8%
Working Working
2009 2009
Age Working Not Age Working Not
65−69 Working 65–69 Working
2007 Working 75.0% 25.0% 2007 Working 85.1% 14.9%
Not 5.4% 94.6% Not 5.6% 94.4%
Working Working
2009 2009
Age Working Not Age Working Not
70−75 Working 70–75 Working
2007 Working 77.5% 22.5% 2007 Working 73.0% 27.0%
Not 5.0% 95.0% Not 2.6% 97.4%
Working Working
continued
OCR for page 183
183
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
TABLE 8-1 Continued
Male Female
2009 2009
Age Employed Self- Age Employed Self-
60−64 Employed 60–64 Employed
2007 Employed 98.0% 2.0% 2007 Employed 98.5% 1.5%
Self- 0.0% 100.0% Self- 1.8% 98.2%
Employed Employed
2009 2009
Age Employed Self- Age Employed Self-
65−69 Employed 65–69 Employed
2007 Employed 100.0% 0.0% 2007 Employed 97.1% 2.9%
Self- 4.1% 95.9% Self- 3.8% 96.2%
Employed Employed
2009 2009
Age Employed Self- Age Employed Self-
70−75 Employed 70–75 Employed
2007 Employed 100.0% 0.0% 2007 Employed 100.0% 0.0%
Self- 0.0% 100.0% Self- 0.0% 100.0%
Employed Employed
Male Female
2009 2009
Age Full time Part time Age Full time Part time
60−64 60–64
2007 Full time 28.6% 71.4% 2007 Full time 42.1% 57.9%
Part time 5.4% 94.6% Part time 2.0% 98.0%
2009 2009
Age Full time Part time Age Full time Part time
65−69 65–69
2007 Full time 40.0% 60.0% 2007 Full time 37.5% 62.5%
Part time 2.0% 98.0% Part time 0.9% 99.1%
2009 2009
Age Full time Part time Age Full time Part time
70−75 70–75
2007 Full time 16.7% 83.3% 2007 Full time 50.0% 50.0%
Part time 0.2% 99.8% Part time 0.1% 99.9%
SOURCE: Data from JSTAR (2007).
OCR for page 194
194 AGING IN ASIA
A. Top 50th percentile, male
1.0
0.9
0.8
0.7
0.6
50–59
0.5
60–64
0.4
65–69
0.3
70–74
0.2
0.1
0
0 0–30 30–40 40–
Working hours
B. Bottom 50th percentile, male
1.0
0.9
0.8
0.7
0.6
50–59
0.5
60–64
0.4
65–69
0.3
70–74
0.2
0.1
0
0 0–30 30–40 40–
Working hours
FIGURE 8-5 Predicted retirement rate (male).
SOURCE: Data from JSTAR (2007).
R02177
Figure 8-5
parts a and b combined to fit on one page
two uneditable bitmaps, plus vector labels for A and B
OCR for page 195
195
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
than 10%. Males in their 70s have a similar tendency, though the decline
of retirement probability is larger in the lower percentile.
Second, the largest difference is observed for males aged 65–69. Males
in this age category who have higher index values retire with much lower
probability compared to those who have lower index values (21% versus
59% when they work less than 30 hours and about 0% versus 21% when
they work more than 40 hours). The effect of the higher family index is
opposite for this age group compared to males in their 50s. Overall, the
only index that affects the retirement decision in 2009 is the family index.
The health and the socioeconomic indices do not seem to affect the retire-
ment decision with statistical significance.
On the other hand, the health index affected the working hours deci -
sion in 2009. The CES-D measure used to normalize the index and the
grip strength are statistically significant variables in the health index. This
can be seen in the second column of Table 8-3 reporting the results from
Regression 3, which we use to predict working hours in 2009 by age and
hours-worked group below. First, one can see a clear difference between
the age groups. Except for males in their 50s and 70s, on average, working
hours seem to be declining. Second, the working hours of males in their
50s rebound from 0 to about 10 hours, but males above 60 seem to stay
constant at around 0.
The effect of health index values can be seen clearly in Figures 8-6a
and 8-6b. These figures are analogously constructed with Figures 8-5a and
8-5b, except that the vertical axis is the predicted hours worked instead of
the predicted retirement probability. For males in their 50s, the predicted
working hours for those with the low index values and who worked less
than 30 hours per week is about 14 hours per week, whereas for those with
high index values, it is more than 34 hours per week. This amount does not
differ much from those who worked longer hours per week in 2007. Analo-
gous results hold for those in their 70s. Those with lower index values are
predicted to work less hours in 2009 compared to their working hours in
2007, but those with higher index values are predicted to keep working
around the same hours per week with the hours worked per week in 2007.
Compared with males in their 50s and 70s, the difference between high and
low index values are much smaller for males in their 60s.
Table 8-4 reports the estimated coefficients for females in
Regressions 1–3. The third column reports the result of Regression 2,
which explores the factors affecting probability of retirement in 2009 given
the respondent reported being retired in 2007. The result indicates that
those in their 50s with a lower health index and higher family index retire
with higher probability. The effects of the indices are opposite for females
above 70. Those who have a higher health index value and lower family
index value retire with higher probability. However, the effect is not so
OCR for page 196
196 AGING IN ASIA
A. Top 50th percentile, male
50
45
40
35
30
50–59
25
60–64
20
65–69
15
70–74
10
5
0
0 0–30 30–40 40–
Working hours
B. Bottom 50th percentile, male
50
45
40
35
30
50–59
25
60–64
20
65–69
15
70–74
10
5
0
0 0–30 30–40 40–
Working hours
FIGURE 8-6 Predicted working hours (male).
SOURCE: Data from JSTAR (2007).
R02177
Figure 8-6
parts a and b combined to fit on one page
two uneditable bitmaps, plus vector labels for A and B
text redrawn
OCR for page 197
TABLE 8-4 Female Estimation
Column 1 2 3 4
Working Hours in 2007 > 0 Working Hours in 2007 = 0
Retirement in 2009 Working hours in 2009 Retirement in 2009 Working hours in 2009
Constant –0.127 (0.135) 15.42 (11.95) 0.832 (0.150)*** 1.378 (3.919)
H index (health index) –0.0686 (0.0760) –0.0300 (0.261) –0.152 (0.0692)* 3.800 (1.804)*
F index (family index) –0.155 (0.0915) 0.0012 (0.00038)** 0.0986 (0.0494)* –1.932 (1.473)
E index (economic index) 0.203 (0.0937)* –0.362 (2.063) –0.0264 (0.0350) 3.213 (1.902)
Age 60−64 –0.0765 (0.226) –19.48 (16.85) –0.242 (0.227) 6.538 (5.137)
Age 65−69 –0.151 (0.271) 2.211 (23.79) –0.0950 (0.188) 2.894 (4.231)
Age 70−74 1.090 (0.494)* 13.35 (39.75) 0.194 (0.175) –1.931 (4.141)
0.0699 (0.139) 19.43 (16.55)
H3040 (30 ≤ working hours ≤ 40)
Hm40 (40 < working hours) 0.241 (0.163) 41.63 (15.45)**
Age 60−64 * H3040 0.680 (0.718) –35.55 (24.46)
Age 65−69 * H3040 0.708 (0.554) –0.677 (34.49)
Age 70−74 * H3040 0.201 (0.520) 6.217 (203.9)
Age 60−64 * Hm40 –0.0415 (0.279) 40.96 (25.97)
Age 65−69 * Hm40 0.486 (0.457) 124.4 (50.22)*
Age 70−74 * Hm40 2.027 (1.937) –0.246 (68.71)
H3040 * H index –0.104 (0.0979) 0.110 (0.908)
Hm40 * H index 0.143 (0.106) 0.164 (1.346)
H3040 * F index 0.0871 (0.0784) –0.0021 (0.00052)***
Hm40 * F index 0.164 (0.0992) –0.00029 (0.00051)
H3040 * E index –0.179 (0.119) 2.263 (3.409)
Hm40 * E index –0.232 (0.104)* 6.036 (4.156)
Age 60–64 * H index 0.0880 (0.155) –0.146 (1.208) 0.0397 (0.0735) –2.355 (1.720)
Age 65–69 * H index 0.121 (0.200) 0.0352 (0.352) 0.133 (0.0731) –3.691 (1.868)*
Age 70–74 * H index 0.101 (0.274) –0.0228 (0.298) 0.160 (0.0708)* –3.880 (1.820)*
197
continued
OCR for page 198
TABLE 8-4 Continued
198
Column 1 2 3 4
Working Hours in 2007 > 0 Working Hours in 2007 = 0
Retirement in 2009 Working hours in 2009 Retirement in 2009 Working hours in 2009
Age 60−64 * H3040 * H index 0.992 (0.217)*** –0.191 (1.586)
Age 65−69 * H3040 * H index –0.111 (0.290) –0.128 (1.088)
Age 70−74 * H3040 * H index –0.536 (0.916) –1.661 (14.32)
Age 60−64 * Hm40 * H index –0.306 (0.183) 0.351 (2.895)
Age 65−69 * Hm40 * H index –0.351 (0.250) 0.868 (7.111)
Age 70−74 * Hm40 * H index 3.364 (4.901) 0.162 (1.421)
Age 60−64 * F index –0.0486 (0.0874) –0.00118 (0.000790) 0.107 (0.0809) –1.967 (2.063)
Age 65−69 * F index –0.0666 (0.106) –0.000843 (0.000607) –0.0350 (0.0675) 0.492 (1.756)
Age 70−74 * F index 0.412 (0.201)* 0.000677 (0.00156) –0.118 (0.0591)* 2.352 (1.541)
Age 60−64 * H3040 * F index 0.155 (0.168) –0.000351 (0.00104)
Age 65−69 * H3040 * F index 0.297 (0.211) 0.00322 (0.00107)**
Age 70−74 * H3040 * F index 0.101 (0.139) 0.0154 (0.00980)
Age 60−64 * Hm40 * F index 0.00244 (0.107) 0.000448 (0.00111)
Age 65−69 * Hm40 * F index 0.216 (0.175) 0.00557 (0.00117)***
Age 70−74 * Hm40 * F index 0.109 (0.229) 0.0000363 (0.00284)
Age 60−64 * E index –0.402 (0.143)** –0.535 (4.415) 0.0130 (0.0393) –3.557 (2.214)
Age 65−69 * E index –0.486 (0.182)** –1.098 (3.296) 0.0907 (0.0601) –4.406 (2.335)
Age 70−74* E index 0.0933 (0.225) –2.259 (6.260) 0.0138 (0.0321) –3.152 (1.905)
Age 60−64 * H3040 * E index 0.150 (0.215) 5.103 (6.339)
Age 65−69 * H3040 * E index –0.0508 (0.451) –0.108 (6.097)
Age 70−74 * H3040 * E index –1.016 (1.169) 98.69 ( . )
Age 60−64 * Hm40 * E index 0.318 (0.168) 4.908 (6.874)
Age 65−69 * Hm40 * E index 0.507 (0.204)* 5.825 (8.722)
Age 70−74 * Hm40 * E index –1.645 (1.407) –15.97 (11.82)
OCR for page 199
Memory (word recall) –0.0411 (0.0410) –4.481 (36.74) –0.0773 (0.0742) –0.0970 (0.0797)
ADL limitations (any) 0.372 (0.347) 16.65 (142.0) –1.339 (0.412)** –1.505 (0.584)*
Grip strength –0.0111 (0.0106) –4.062 (33.36) 0.0454 (0.0327) 0.0436 (0.0340)
Working spouse –1.266 (0.398)** 7283.6 ( . ) –0.409 (0.228) –0.196 (0.251)
Providing care 1.131 (0.546)* 1021.3 (848.5) 0.280 (0.174) 0.233 (0.165)
No child –1.706 (1.254) –6020.0 (4603.5) 1.831 (0.929)* 1.361 (1.055)
Minimum child age –0.0749 (0.0402) –308.3 (126.2)* 0.0411 (0.0251) 0.0332 (0.0273)
Education_middle –0.143 (0.134) 0.775 (0.370)* –0.665 (0.946) 0.0551 (0.376)
EPI/MAI beneficiaries –0.220 (0.194) –0.145 (0.368) 0.133 (0.608) 0.552 (0.463)
Asset_m3500 –0.194 (0.170) –0.829 (0.588) 1.487 (1.505) 0.176 (0.517)
(Asset ≥ 35 million yen)
Asset_15003500 0.0916 (0.243) –1.452 (0.939) 1.327 (1.251) 0.109 (0.404)
(15 ≤ Asset < 35 million yen)
Asset_1001500 0.196 (0.212) 0.543 (0.604) 1.635 (1.346) 0.370 (0.526)
(1 ≤ Asset < 15 million yen)
Number of observations 526 500 690 680
R-squared 0.190 0.428 0.113 0.119
NOTE: Robust standard errors in parentheses. * denotes p < 0.1; ** p < 0.05; *** p < 0.01.
SOURCE: Data from JSTAR (2007).
199
OCR for page 200
200 AGING IN ASIA
large, as almost everyone stays retired with high probability in any case as
seen in Figures 8-7a and 8-7b for working hours set at 0 (discussed below).
For females who worked in 2007, we examine the retirement decision
in the same way we did for males using Regression 1. The results are
reported in Table 8-4’s first column, which we use to predict retirement
rates in 2009 by age and hours-worked group below. When females work
at all, the probability of retiring in two years is significantly less; it is less
than 28% (slightly higher than males’ 26%), and there is not much differ-
ence across different age groups when they are working less than 30 hours
per week. The probability of retirement in 2009 is around 20 to 28%, the
same as males’ results. However, there is a significant difference across
index values. For females, socioeconomic variables affect the retirement
decision in a statistically significant way.
Comparing Figures 8-7a and 8-7b, females in their 50s on average are
not affected much by the index value. Regardless of the index value, they
retire with about 20% probability when they work less than 30 hours per
week but retire with about 10% probability when they work more. Those
who are above 70 with a higher index value retire with much higher prob-
ability, at 55% when they work less than 30 hours. In contrast, for those
with a lower index value, the probability of retirement declines to close
to zero for females who work at all.
On the other hand, females in their 60s who work less than 30 hours
per week retire with less probability when their indices values are high
(around 0%) compared to those who have lower indices values, at around
(40 to 50%). The difference is still large for the 60–64 age group when
females work between 30 to 40 hours per week (about 5% versus 40%).
Turning to the working hours decision in 2009, the health index
affected for males, but the family index affected for females. The marital
status variable and the minimum child’s age (higher age implies less
index value) are the statistically significant variables in the family index.
This can be seen in the second column of Table 8-4. It reports results from
Regression 3, which we use to predict working hours in 2009 by age and
hours-worked group in Figures 8-8a and 8-8b. First, unlike males, one
cannot see a clear difference between the age groups. Second, all groups
seem to be predicted to work less in 2009 than the hours worked per week
in 2007. Third, it is observed that females with higher indices values work
more hours if they are in their 50s or 60s.
CONCLUSION
We have examined the transition of work status and working hours
for Japanese males and females who were between ages 50–75 in 2007
using the JSTAR data. Here we summarize our findings.
OCR for page 201
201
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
A. Top 50th percentile, female
1.0
0.9
0.8
0.7
0.6
50–59
0.5
60–64
0.4
65–69
0.3
70–74
0.2
0.1
0
0 0–30 30–40 40–
Working hours
B. Bottom 50th percentile, female
1.0
0.9
0.8
0.7
0.6
50–59
0.5
60–64
0.4
65–69
0.3
70–74
0.2
0.1
0
0 0–30 30–40 40–
Working hours
FIGURE 8-7 Predicted retirement rate (female).
SOURCE: Data from JSTAR (2007).
R02177
Figure 8-7
parts a and b combined to fit on one page
two uneditable bitmaps, plus vector labels for A and B
all text redrawn
OCR for page 202
202 AGING IN ASIA
A. Top 50th percentile, female
50
45
40
35
30
50–59
25
60–64
20
65–69
15
70–74
10
5
0
0 0–30 30–40 40–
Working hours
B. Bottom 50th percentile, female
50
45
40
35
30
50–59
25
60–64
20
65–69
15
70–74
10
5
0
0 0–30 30–40 40–
Working hours
FIGURE 8-8 Predicted working hours (female).
SOURCE: Data from JSTAR (2007).
R02177
Figure 8-8
parts a and b combined to fit on one page
two uneditable bitmaps, plus vector labels for A and B
OCR for page 203
203
HIDEHIKO ICHIMURA and SATOSHI SHIMIZUTANI
For males and females, we find strong evidence that those who retire
stay retired two years later once they are aged 60 and older for males and
for females in general. This decision does not seem to be affected much by
the health, family, and socioeconomic indices, although there are statisti-
cally significant indices for females. Males in their 50s, on the other hand,
do seem to come back to work to some extent. Interestingly, among this
age group, it is the unhealthy who are predicted to work longer hours two
years later (16 hours per week versus 6 hours per week).
For males and females who were not retired in 2007, retirement
probabilities are predicted to be between 20–28% when the three indices
are evaluated at the mean values. However, the retirement decisions of
males and females seem to be affected by different factors. The important
index affecting males’ retirement decisions seems to be the family index,
whereas the socioeconomic index affects in a statistically significant way
the retirement decision for females. Although the sources and the mag -
nitude of the effect of the indices are different, the direction of the effects
is the same across males and females. For both males and females, those
who are in their 60s retire with lower probability when they have a
higher index. The largest effects are observed among males who are
aged 65 and older when they work less than 30 hours, females who
are aged 60 and older when they work less than 30 hours, and females
aged 60–64 when they work between 30–40 hours per week.
In terms of hours worked, the Regression 3 results for males and
females show that males and females with a lower index tend to reduce
hours worked more quickly than those with a higher index. Overall,
higher-index males seem to keep working at current working hours lon -
ger than their lower index values counterparts. If their working hours are
reduced to 30 hours or less per week when they are in their 50s or above
70, higher index value persons retire with higher probability than those
with lower index values. If they reach 30 working hours or less per week
when they are in their 60s, they tend to stay in the labor market longer if
they have lower index values.
The pattern we have described above is of course tentative to the
extent we have assumed stationarity of behavior across different cohorts.
To what extent this assumption holds up needs to be examined using
longer panel data.
We also need to examine to what extent the pattern described depends
on current institutional arrangements. In order to examine this, we need
to find some variations in data that can be regarded equivalent to insti-
tutional changes.
OCR for page 204
204 AGING IN ASIA
REFERENCES
Arellano, M., and C. Meghir. (1992). Female labour supply and on-the-job search: An em -
pirical model estimated using complementary data sets. Review of Economic Studies
59(3):537-559.
Banks, J., and S. Smith. (2006). Retirement in the UK. Oxford Review of Economic Policy
22(1):40-56.
Gustman, A., and T. Steinmeier. (2009). Integrating Retirement Models. NBER Working Paper
#15607. Cambridge, MA: National Bureau of Economic Research.
Ichimura, H., H. Hashimoto, and S. Shimizutani. (2009). JSTAR First Results: 2009 Report.
RIETI Discussion Paper Series #09- E-047, Research Institute on Economy, Trade and
Industry. Project on Intergenerational Equity and Center for Intergenerational Studies,
Hitotsubashi University, Discussion Paper #443-447. Available: http://www.rieti.go.jp/
jp/publications/dp/09e047.pdf.
Japanese Study on Aging and Retirement. (2007). Research Institute of Economy, Trade and
Industry. Available: http://www.rieti.go.jp/en/projects/jstar/index.html.
Lazear, E. (1986). Retirement from the labor force. Pp. 305-355 (Chapter 5) in Handbook of
Labor Economics, Volume 1, O.C. Ashenfelter and R. Layard (Eds.). Amsterdam: Elsevier.
Lumsdaine, R., and O.S. Mitchell. (1999). New developments in the economic analysis of
retirement. Pp. 3,261-3,307 (Chapter 49) in Handbook of Labor Economics, Volume 3C,
O.C. Ashenfelter and D. Card (Eds.). Amsterdam: Elsevier.
Organisation for Economic Co-operation and Development). (2008). Ageing and Employment
Policies—Statistics on Average Effective Age of Retirement. Paris: Organisation for Eco-
nomic Co-operation and Development.
Oshio, T., and S. Shimizutani. (2011). Disability Pension Program and Labor Force Participa-
tion in Japan: A Historical Perspective. NBER Working Paper #17052. Cambridge, MA:
National Bureau of Economic Research.
Oshio, T., A. Oishi, and S. Shimizutani. (2011). Social security reforms and labor force par-
ticipation of the elderly in Japan. Japanese Economic Review 62(2):248-271.
Rust, J. (1989). Dynamic programming model of retirement behavior. Pp. 359-398 in The
Economics of Aging, D. Wise (Ed). Chicago: University of Chicago Press.
Shimizutani, S. (2011). A new anatomy of the retirement process in Japan. Japan and the World
Economy 23(3):141-152.
Shimizutani, S., and T. Oshio. (2011). Claiming Behavior of Public Pension Benefits: New
Evidence from Japanese Study on Aging and Retirement (JSTAR). Mimeograph.
