INAHSIM: A Japanese Microsimulation Model

1
INAHSIM A Japanese Microsimulation Model
2nd General Conference of the International
Microsimulation Association Microsimulation
Bridging Data and Policy Ottawa, Canada, June
8th to 10th, 2009

• Seiichi Inagaki
• Research Institute for Policies on Pension and
  Aging
• 1-3-13 Takanawa, Minato-ku, Tokyo 108-0074, Japan
• sinagaki_at_tiu.ac.jp Tel [phone], Fax
  81-3-5793-9414

2
History of INAHSIM (Integrated Analytical Model
for Household Simulation)

• Revision 1 (1986)
• 32,000 persons / 10,000 households 1975 2025
• 1974 CSLC (private household only)
• A tool for household simulation
• Revision 2 (2005)
• 126,000 persons / 46,000 households 2001 2100
• 2001 CSLC (private household only)
• Employment status, health status, earnings
• Revision 3 (2008)
• 128,000 persons / 49,000 households 2004 2100
• 2004 CSLC (aligned with the population census)
• Employment status, health status, earnings,
  public pension

(Note) Fukawa used his modified INAHSIM for his
study, Household projection and its application
to health/long-term care expenditures in Japan
using INAHSIM-II.
3
Basic structure of the model population

• The model population comprises three tables.
• There are links between the family and individual
  segments and between the household and individual
  segments.

Individual segment
Household membership
Parents-children Husband-wife
Family segment
Household segment
4
Characteristics in each segment

• Individual segment
• the year of birth, sex, marital status, health
  status, employment status, earnings, and the
  pension amount
• family segment ID number (as husband or wife)
• family segment ID number (as child)
• household segment ID number
• Family segment
• the year of marriage, number of children, the
  year the marriage dissolved, and cause of
  separation
• Household segment
• the year of formation of the household, number of
  household members, private/institutional
  household, total household earnings, and
  household structure

5
Key feature kinship relationships

• This model is not limited to parents, children,
  husbands, and wives, and can simulate all kinds
  of kinship relationships including those
  involving uncles, nieces, cousins, sons of
  separated parents, grandnephews, and
  great-grandnieces.
• This information is very important to simulate
  household changes in Japan since household
  mergers among family membersfor example, adult
  children resettling to care for their aged
  parents or returning to their parents households
  following divorceare common.

6
Family comprising a couple and children

• Family in this model comprises a couple and
  their children, and the group of children is
  defined by a list structure.
• A family below comprising a coupleJim and
  Maryand their three childrenKen, Karen, and
  Tom.

(Family segment)
Youngest childs ID
Wifes ID (Mary)
Husbands ID (Jim)
Other characteristics of the family
Child_1 (Ken)
Child_2 (Karen)
Child_3 (Tom)
(Individual segment)
7
Jims family when he was a child

• Jim, as a child, is also a member of another
  family.
• From these two family segments, we can know the
  relationships among these persons.
• For example, Ann is Kens aunt, Ken is Anns
  nephew, John is Toms grandfather, and so on.

(Family segment)
Youngest childs ID
Husbands ID (John)
Wifes ID (Liz)
Other characteristics of the family
Child_1 (Ann)
Child_2 (Jim)
(Individual segment)
8
Household comprising three persons

• A group of household members is defined by a list
  structure.

(Household segment)
Household Members ID
Other characteristics of the household
person_1
person_2
person_3
(Individual segment)
9
Source of the initial population

• The Comprehensive Survey of the Living Conditions
  of People on Health and Welfare (CSLC)
• The survey is conducted every three years using
  large sample sizes by the Ministry of Health,
  Labor and Welfare.
• In the 2004 survey, the sample size was 25,091
  households and 72,487 household members.
• This survey covers kinship relationships within
  household members, marital status, employment
  status, health status, earnings, pension amounts,
  and other socioeconomic characteristics

10
Initial population

• Private households (49,089) and household members
  (126,487) is prepared by resampling with
  replacement from the micro data.
• The elderly population of 1,200 persons in
  institutional households is prepared separately
  and is added to the initial population.
• The initial population is 127,687 persons, and
  reflects Japans society on a 1/1000 scale.

11
Adjustment of collection rates

• Another problem with the CSLC is its collection
  rate. It was 54.7 in the 2004 survey note that
  this rate varies according to sex, age, and
  household structure.
• The collection rate of single-person households
  was very low, and that of young people was also
  very low.
• These differences are adjusted by weighing the
  resampling rates when the initial population was
  prepared.

12
Imputation of kinship relationships

• The kinship relationships between the persons
  living in different households cannot be obtained
  from the CSLC.
• Imputation method
• List the persons or couples who have children but
  live separately using the CSLC results. The CSLC
  surveyed the number of children who live
  separately for each person.
• Randomly draw children whose parent(s) would be
  alive using the probabilities by childs age that
  his/her mother or father is alive.
• Make a match between the couples on the list (a)
  and the children on the list (b) in order of age.

13
Simulation cycle
New year
Birth
Entering an institution
Determining pensions
Change in health status
Death
Living with elderly parents
Marriage
Young people leaving home
Divorce
Estimating earnings
Change in employment status
14
Birth

• Marital fertility rates by parity and mothers
  age, and sex ratio (boys to girls) are used for
  birth.
• This model does not take into account
  illegitimate children since their proportion is
  very low in Japan. Therefore, total fertility
  rate is strongly affected by marriage rates. In
  fact, one of the major reasons why fertility
  rates in Japan are declining is the low marriage
  rates among young women.
• Newborn babies will belong in their mothers
  households.

15
Change in health status

• The health status is classified as good or poor,
  and is assumed to deteriorate with age.
• The deterioration rates are specified by age and
  sex.

16
Death

• Mortality rates by sex and age are used for
  death.
• In the case of a persons death, if there exists
  a nominee (the deceaseds spouse) who is eligible
  to a survivors pension, the pension amount is
  determined in this life event. The pension amount
  is principally three-fourths of the
  earnings-related part of the deceaseds pension
  amount.

17
Marriage

• First-marriage rates and re-marriage rates by
  sex, age, and employment status are used for
  marriage.
• When a marriage takes place, the couple decides
  to live with the grooms parents, live with the
  brides parents, or start a new household.
• Adjust the numbers of brides and grooms
• Select the candidate brides and grooms using the
  marriage rates twice.
• Calculate the average number of candidates. The
  number of couples would be the average figure
  divided by two.
• Take a sampling of the candidate brides and
  grooms.
• Form couples between the sampled brides and the
  grooms.

18
Divorce

• Divorce rates by wifes age are used for divorce.
  
• The custody of the children is determined by the
  given probabilities. The children will live with
  the parent who has their custody.
• When a divorce takes place, the divorced
  husband/wife decides to return to his/her
  parents household or form a new household. This
  event is important since around half of the
  divorcees return to their parents households in
  Japan.

19
Change in employment status

• With regard to the employment status, the
  individual is classified as regular employees,
  non-regular employees, self-employed, or
  unemployed.
• Transition probabilities between these four
  statuses by sex and age are used for the change
  in employment status.
• As for women, those transition probabilities are
  differentiated with respect to their marital
  status since the employment pattern among
  Japanese women differs with their marital status.

20
Estimating earnings

• Earnings are assumed to conform to a log-normal
  distribution by sex, age group, and employment
  status.
• In this model, z-score of the earnings-distributio
  n for each person is given in advance, and
  persons earnings are estimated on the basis of
  ones z-score assuming the earnings-distribution
  by sex, age group, and employment status each
  year.
• The z-score does not change over the lifetime of
  the individual.

21
Young people leaving home

• Young people leave their parents household for
  purposes of higher education, finding employment,
  or changing jobs.
• The probabilities of never-married young people
  leaving home by sex, age, and employment status
  are used as the transition probabilities of young
  people leaving home.

22
Living with elderly parents

• When elderly people, who do not live with their
  children, become very old age and need care, many
  children move in with their elderly parents to
  take care of them. This is still an important
  life event to secure the life of the elderly in
  Japan.
• This life event is referred to as living with
  elderly people in this model, and the
  probabilities by parents sex and age are used to
  simulate it.

23
Determining pensions

• This is a determination of the pension amount to
  a pension subscriber who has reached his/her
  pensionable age.
• Early and deferred payments are not considered.
• The amount is estimated on the basis of
  pensioners z-score and subscription category
  assuming the distribution of newly awarded
  pension amounts.

24
Entering an institution

• Probabilities by sex, age, and marital status of
  entering an institution are used for this life
  event.

25
Baseline scenario

• The baseline scenario is supposed to serve as a
  benchmark for assessing the impact of policy
  change or behavioral changes.
• The scenario is assumed that peoples behavioras
  it was in 2001would not, in principle, change in
  the future.
• However, it is assumed the declining trends in
  mortality and first marriage rates.

26
Population
(Source) Population Projections for Japan
(National Institute of Population and Social
Security Research)
27
Number and size of private households
(Source) Household Projections for Japan
(National Institute of Population and Social
Security Research)
28
Subscribers of pension scheme
(Source) Actuarial valuation 2004 on Employees
Pension Insurance and National Pension in Japan
(Actuarial Affairs Division, Pension Bureau,
Ministry of Health, Labor and Welfare)
29
Elderly population by family type
30
Income distribution
31
Gini coefficient
32
Lifetime income distribution
33
Future directions

• This model is a comprehensive microsimulation
  model for the Japanese population. As discussed,
  it provides us plenty of valuable simulation
  results, which are well aligned with the official
  projection results.
• However, it does not make projections for some
  characteristics such as wealth, education,
  housing, and social security contribution and tax
  burdens.
• International migration should be considered
  since it will become an important political issue
  in the super-aging Japanese society.

34
Future directions

• Microsimulation models are not common in Japan.
• However, the output of microsimulation models,
  especially that which pertains to the
  distributional aspect of any social change, is
  more important in Japan than in any other country
  as Japan will become the most rapidly aging
  society in the world.
• Some researchers and policy makers have
  acknowledged the importance of microsimulation
  models in social policy making.

35
Future directions

• Japan is sufficiently well-versed with the
  development of microsimulation models and has the
  necessary pre-requisites
• availability of suitable micro data
• the demands made by policy makers
• computer technologies
• A supercomputer may be used for the simulation.
• The next task is to not only improve the
  simulation but also introduce the microsimulation
  models to researchers and policy makers.

36
Computer language

• Computer language
• This model is written in FORTRAN90.
• Execution time
• If the initial population is 127 thousand
  persons, that is, 1/1000 of Japanese population,
• it takes about 1 minute to make a 100-year
  simulation using a PC with 2GB RAM and Intel
  Core(TM2) 6600, 2.4GHz processor.

Thank you