The Effect of Subjective Survival Probabilities on Retirement and Saving

1
The Effect of Subjective Survival Probabilities
on Retirement and Saving

• David E. Bloom
• Harvard School of Public Health
• David Canning
• Harvard School of Public Health
• Michael Moore
• Queens University, Belfast
• Younghwan Song
• Union College

2
The question

• What are the effects of increased life
  expectancy, coupled with compression of
  morbidity, on
• labor supply/retirement?
• savings/wealth accumulation over life cycle?

3
Literature

• Life cycle model
• save during working years to finance
  consumption in retirement
• Institutional constraints on working (Blöndal and
  Scarpetta 1997 Gruber and Wise 1998)
• Social security provisions in OECD countries
  create powerful incentives to retire at
  particular ages

4
Empirical macro literature

• Lee, Mason, and Miller (1998, 2000)
• A rise in savings in Taiwan can be accounted
  for by increased longevity in the presence of a
  fixed retirement age
• Bloom, Canning, Mansfield, and Moore (2007)
• Cross-country panel analysis shows that
  increased longevity raises savings rates in
  countries with mandatory retirement, but has
  little effect when there are age-neutral
  incentives

5
Empirical micro literature

• Hurd, Smith, and Zissimopoulos (2004)
• People with very low subjective survival
  probabilities (zero) retire earlier (HRS)
• Hurd, McFadden, and Gan (1998)
• Saving increases with subjective survival
  probabilities among couples (AHEAD)

6
Contributions of this paper

• Examines the effect of subjective survival
  probabilities on both retirement and wealth,
  using micro data from various countries
• Instruments subjective survival probabilities
  using parental mortality experience to address
  measurement error and reverse causality
• Validation test using panel data from the HRS

7
Outline of talk

• HRS data Subjective survival probabilities
• Validation using panel data from HRS
• Retirement and wealth analyses
• - HRS (Health and Retirement Study) US
• - SHARE (Survey of Health, Ageing, and
  Retirement in Europe)
• - ELSA (English Longitudinal Study of Ageing)
• Conclusion

8
Health and Retirement Study (HRS)

• Panel survey of individuals aged 51-61 in 1992
  and their spouses or partners in U.S.
• 12,562 individuals in the first wave
• Collects extensive information about retirement,
  health, and economic well-being of the
  respondents
• Subsequent waves of interviews were fielded
  biennially

9
Subjective survival probabilities

• In the first wave of the HRS
• Using any number from zero to ten where 0 equals
  absolutely no chance and 10 equals absolutely
  certain, what do you think are the chances you
  will live to be 75 or more?
• 85 or more?

10
Subjective survival probabilities

• Hurd and McGarry (1995)
• Aggregate well to averages close to survival
  probabilities from life tables
• Vary plausibly with income, wealth, schooling,
  and risk factors such as smoking
• Hurd and McGarry (2002)
• Predict actual mortality by wave 2, independent
  of health status

11
Measurement errors and focal-point response

• Hurd and McGarry (1995)
• About 2.5 percent reported larger values for the
  survival probability to 85 than for the survival
  probability to 75
• Hurd, McFadden, and Gan (1998)
• AHEAD Many respondents systematically provided
  focal-point answers (0, 0.5, or 1) Averages
  higher than the averages from life tables
• Bassett and Lumsdaine (2001)
• Respondents with probabilistically inconsistent
  answers in Social Security questions reported
  higher values for subjective survival
  probabilities

12
Efforts to improve subjective survival
probabilities

• Hurd, McFadden, and Gan (1998)
• AHEAD Transformed the error-ridden survival
  probabilities with focal points to continuous
  probabilities
• Saving increases with their new survival
  probabilities among couples, but not among
  singles
• No validation of their new survival probabilities

13
Our instruments

• Parents mortality experience
• Parents ages or ages at death are substantially
  correlated with the subjective survival
  probabilities of respondents
• The relationship between the parents ages at
  death and the subjective survival probabilities
  is nonmonotonic (Hurd and McGarry 1995)
• Respondents modify their subjective survival
  probabilities in response to the death of a
  parent (Hurd and McGarry 2002)

14
Validation

• Sample
• 10,070 respondents whose actual mortality is
  known by wave 6 of the HRS
• Age 45-65 in 1992
• No proxy response
• Use subjective survival probabilities to 75 at
  wave 1

15
Cumulative number and proportion of respondents
deceased by wave 2 through 6
16
Histogram of the reported survival probability to
Age 75
17
Marginal effects of subjective survival
probability from probit estimation of the
observed mortality, waves 2 through 6
18
Marginal effects of subjective survival
probability from probit estimation of the
observed mortality, waves 2 through 6
19
Marginal effects of subjective survival
probability from probit estimation of the
observed mortality, by selected variables, waves
2 through 6
20
Marginal effects of subjective survival
probability from probit estimation of the
observed mortality, by selected variables, waves
2 through 6
21
Marginal effects of subjective survival
probability from probit estimation of the
observed mortality, by selected variables, waves
2 through 6
22
Marginal effects of subjective survival
probability from probit estimation of the
observed mortality, by selected variables, waves
2 through 6
23
First-stage regression results in IV probit
estimation of observed mortality linear
probability model of survival to age 75
24
First-stage regression results in IV probit
estimation of observed mortality linear
probability model of survival to age 75
25
First-stage regression results in IV probit
estimation of observed mortality linear
probability model of survival to age 75
26
Distribution of the instrumented survival
probability to age 75
27
Marginal effects of instrumented subjective
survival probability from IV probit estimation of
the observed mortality, waves 2 through 6
28
Marginal effects of instrumented subjective
survival probability from IV probit estimation of
the observed mortality, by selected variables,
waves 2 through 6
29
Marginal effects of instrumented subjective
survival probability from IV probit estimation of
the observed mortality, by selected variables,
waves 2 through 6
30
Retirement analysis (HRS)

• Definition of retirement
• Not in the labor force
• In the Labor force working full-time/part-time
  or unemployed
• Not in the labor force retired, partially
  retired, disabled, or not in the labor force in
  the summary labor force status

31
Retirement analysis (HRS)

• Sample
• 9,155 individuals aged 50-65 in 1992
• No proxy response
• Separate analysis by household living
  arrangements (single or couple) and by sex

32
Proportion retired by age, whole sample, HRS
33
Proportion retired by age, whole sample, HRS
34
Retirement equation (HRS)

• Dependent variable retirement dummy
• Explanatory variables
• - Subjective survival probabilities to 75
• - Age
• - Race/ethnicity
• - Years of education
• - Number of children
• - Self-reported health status
• - Presence of a health problem
• For couples the same set of explanatory
  variables for the spouse

35
Retirement probit coefficients, singles, HRS
36
Retirement probit coefficients, couples, HRS
37
Wealth analysis (HRS)

• Wealth Sum of all household wealth components
  less all debt, exclusive of Social Security and
  pension wealth (in 1,000s)
• Unit of analysis household
• Separate analysis by household living
  arrangements (single or couple)
• The same set of explanatory variables as those
  used in retirement analysis

38
Kernel density graph of wealth, couples, HRS
39
Wealth regression results, single households,
HRS Dependent variable Household wealth (in
1,000s)
40
Wealth regression results, couple households,
HRS Dependent variable Household wealth (in
1,000s)
41
Survey of Health, Ageing and Retirement in Europe
(SHARE)

• Longitudinal survey of individuals aged over 50
  and their spouses in 11 European countries
  (Austria, Germany, Sweden, the Netherlands,
  Spain, Italy, France, Denmark, Greece,
  Switzerland, and Belgium) that began in 2004
• Designed after the HRS
• Release 1 of the SHARE 22,777 individuals from
  10 countries, exclusive of Belgium

42
Subjective survival probability in SHARE

• (On a scale from 0 to 100) What are the chances
  that you will live to be age T or more?
• Target age, T, varies by respondents age

43
Subjective survival probability in SHARE

• Hurd, Rohwedder, and Winter (2005)
• - Subjective survival probabilities in the SHARE
  also vary with income, wealth, and health
  conditions, as well as with the parental
  mortality experience.
• - Measurement error and focal point responses

44
Retirement analysis (SHARE)

• Definition of retirement
• Not in the labor force
• In the Labor force employed, including
  self-employed, or unemployed
• Not in the labor force retired, permanently sick
  or disabled, or homemaker

45
Retirement analysis (SHARE)

• Sample
• 10,706 individuals aged 50-65 in 2004
• Those reported subjected survival probability to
  age 75
• Separate analysis by household living
  arrangements (single or couple) and by sex
• Pooled observations from all ten countries with
  country dummies
• Explanatory variables for spouse not included for
  couples

46
Retirement probit coefficients, singles, SHARE
47
Retirement probit coefficients, couples, SHARE
48
Wealth analysis (SHARE)

• Wealth Household net wealth (in 1,000s, ppp
  adjusted)
• Hotdeck using 5 imputed asset data sets
• Unit of analysis individual
• Separate analysis by household living
  arrangements (single or couple)

49
Wealth regression results, single households,
SHARE Dependent variable Household wealth (in
1,000s)
50
Wealth regression results, couple households,
SHARE Dependent variable Household wealth (in
1,000s)
51
English Longitudinal Study of Ageing (ELSA)

• Longitudinal survey of individuals aged over 50
  and their spouses living in England that began in
  2002
• Designed after the HRS
• Wave 1 of the ELSA 12,100 individuals
• Question on the subjective survival probability
  in ELSA also has a target age that varied by the
  respondents age

52
Retirement analysis (ELSA)

• Definition of retirement
• Not in the labor force
• In the Labor force employed, including
  self-employed, or unemployed
• Not in the labor force retired, semi-retired,
  permanently sick or disabled, or looking after
  home or family

53
Retirement analysis (ELSA)

• Sample
• 5,683 individuals aged 50-65 in 2002
• Those reported subjected survival probability to
  age 75
• Separate analysis by household living
  arrangements (single or couple) and by sex
• Explanatory variables for spouse not included for
  couples

54
Retirement probit coefficients, singles, ELSA
55
Retirement probit coefficients, couples, ELSA
56
Wealth analysis (ELSA)

• Wealth Household net wealth (in 1,000s)
• Unit of analysis individual
• Separate analysis by household living
  arrangements (single or couple)

57
Wealth regression results, single households,
ELSA Dependent variable Household wealth (in
1,000s)
58
Wealth regression results, couple households,
ELSA Dependent variable Household wealth (in
1,000s)
59
Conclusions

• Instrumented subjective survival probabilities
  predict actual long-term mortality
• HRS ELSA There is no evidence that higher
  subjective survival probabilities decrease the
  probability of retirement, due to institutional
  constraint
• HRS ELSA Higher subjective survival
  probabilities significantly increase wealth
• SHARE not as informative due to small samples
  need to impose more structure

60
Future research

• Analyze full HRS panel
• Merge HRS with Social Security data on lifetime
  income to analyze rates of lifetime savings
• Look at interaction of longevity effects and
  institutional arrangements
• Sensitivity analyses (definition of retirement
  inclusion of social security and pension wealth
  instrument subjective probability of survival to
  age 75 with estimates from national life tables
  conditional on age, race, and sex)