Comparing observed and perceived joint survival probabilities of married couples

We develop and estimate models for actual and perceived survival probabilities of married
couples. In the model, the potential lifetime dependence between spouses in a couple is generated
by correlated (un)observables that aect both spouses' survival probabilities, and by bereavement
eects - the impact of one spouse's bereavement on the surviving spouse's survival rate. We
use married couples' actual mortality experience and their survey reports of the probability to
survive until given target ages from the Health and Retirement Study. We account for possible
reporting biases in the elicited probabilities, jointly modelling couples' genuine subjective survival
probabilities and their reporting behaviour. We nd that actual and perceived lifetimes of the two
spouses in a couple are strongly positively correlated. The bereavement eect explains around 60-
98% of this correlation, with correlated observables and unobservables explaining the remainder. On
average, non-widowed people younger than 75 (older than 75) tend to underestimate (overestimate)
their actual expected remaining life time. Widowers substantially underestimate their expected life
time, while widows tend to underestimate their expected life-years if they are younger than 72.