Probabilistic Thinking and Early Social Security Claiming Adeline Delavande Mike Perry Robert J. Willis RAND Corp. Univ. Nova de Lisboa CEPR University of Michigan University of Michigan 8th Annual Joint Conference of the Retirement Research Consortium “Pathways to a Secure Retirement” August 10-11, 2006 Washington, D.C. Motivation • Do people claim SS based on their survival expectations? • Hurd, Smith and Zissimopoulos - HSZ (2004) – Use direct measures of survival expectations – Findings: subjective survival of 0 associated with early claiming; otherwise, no effect • Is the effect found by HSZ too small? – Survey measures of survival expectations capture much individual heterogeneity in risk – But also have a lot of measurement error This Paper • We reexamine whether people claim SS based on their survival expectations • Correct measurement error in elicited subjective survival probabilities using rich set of risk factors as instruments • Findings: People act on their subjective survival beliefs – Statistically and economically significant effect of subjective survival on SS claiming for people working at 62 -elasticity of claiming probability with respect to survival probability = -1.24 This Paper (cont.) • Compare with predictions of objective survival probability based on same risk factors – Similar effect on SS claiming – Do not contain more information than subjective survival to explain SS claiming • Our findings suggest that people – have highly hetereogenous mortality expectations – their expectations are largely rational – they act on these beliefs in deciding when to claim Social Security benefits The Analytical Samples • Use data from the Health and Retirement Study (HRS) • Follow HSZ and study 2 groups 1. People who are retired by age 62 – Analyze SS claiming by age 64 2. People who are NOT retired by age 62 – Analyze joint decision to retire and claim by age 64 (all specifications) 30 20 10 • No effect of survival expectations 0 P e rc e n t • 79.2% claim in first year of eligibility • 89.6% claim by third year 40 Early Retiree Sample: Claiming by those retired by age 62 0 12 24 36 48 60 Months since 62nd birthday Months since 62nd birthday 72 84 Age 65 spike 10 N=1801 0 5 P e rc e n t • 21.2% claim in first year of eligibility • 62.2% claim by third year • Significant effects of survival expectations when corrected for measurement error 15 Late Retiree Sample: Claiming by those not retired by age 62 0 12 24 36 48 60 Months since 62nd birthday 72 Months since 62nd birthday 84 96 Correcting for Measurement Errors • Probabilistic beliefs about survival in HRS (On a scale from 0 to 100) What are the chances that you will live to be age 75 or more? • • Measurement error: rounding and heaping at ‘50’ and ‘100’ Use Instrumental Variable methods to correct for measurement error Four sets of instruments: (I) Basic demographic characteristics (II) Health variables (self-reported health and conditions) (III) Dummy variables on parental mortality (own and spouse) (IV) Optimism index 20 15 10 5 – many focal answers at “0”, “50” and “100” 0 P e rc e n t • Survey measure of survival beliefs are quite noisy 25 Heterogeneity of Survival Beliefs and Measurement Error 0 25 50 subjective survival to age 75 75 Subjective Probability of Survival to Age 75 100 Heterogeneity of Survival Beliefs and Measurement Error (cont.) .02 .03 .04 .05 Predicted Subjective Survival Probability 0 .01 Density • But there is a lot of individual variability in subjective mortality risk based on risk factors (see Table 5) 40 50 60 70 Linear prediction 80 90 The effects of subjective survival expectations on claiming behavior • Bivariate probit model with demographics, health and wealth variables Claim by 64 specification Without correction Subjective Prob. IV Subjective Prob. Coef. P value -0.002 0.132 With Correction Coef. -0.016 P value 0.004 Effect of subjective survivals on claiming •correction for measurement error increases magnitude of Predicted probability effectIV bysubjective eight-fold prob. of surviving until age 75 of claiming by age 64 •instrumented59 coefficient is highly significant based on 26 boot-strapped standard errors 69 30 79 34 The effects of subjective survival expectations on claiming behavior • Bivariate probit model with demographics, health and wealth variables Claim by 64 specification Without correction Subjective Prob. With Correction Coef. P value -0.002 0.132 IV Subjective Prob. Coef. P value -0.016 0.004 Effect of subjective survival probability on claiming IV subjective prob. of surviving until age 75 Predicted probability of claiming by age 64 59 26 69 30 79 34 The effects of objective survival expectations on claiming behavior • Use data on 8 to 12 years actual mortality to estimate and “objective” probability of survival to age 75 using same variables as for IV • Bivariate probit model Claim by 64 Specification IV Subjective Prob Objective probability Subj – Obj probability • Coef. P value -0.016 0.004 Coef. -0.013 P value Coef. P value -0.019 0.003 0.006 0.234 0.007 Similar effects of subjective and objective expectations on SS claiming • Objective expectations do not contain more information than subjective survivals to explain SS claiming Conclusion • Measurement errors in subjective probability are important • Mortality expectations have significant effect on SS claiming • People who expect to be long-lived delay claiming and enjoy larger benefits – Positive effect for the well-being of the elderly – Higher cost for tax payers – Ambiguous welfare effects on the whole population