Pension systems in developed countries are creaking under the pressure of population aging, and responsibility for planning a long life is increasingly being shifted onto individuals themselves. To allocate savings rationally, you need to know how long you'll live, but no one has a precise answer. Subjective life expectancy (SLE) is a person's self-assessment of the age they will live to. The literature has already shown that SLE is linked to actual mortality, but most studies used a probabilistic method (chances of surviving to a certain age) and Western samples. Japan, with its record longevity and rapidly growing old-age dependency ratio, remains a blank spot. This work fills that gap by analyzing point estimates of SLE among elderly Japanese over nearly three decades.
The study used data from the National Survey of the Japanese Elderly (NSJE), wave 4 (1996), in which respondents aged 60+ were asked: "Until what age do you expect to live?" Answers ranged from 63 to 120 years, and 393 out of 2,447 people answered "don't know." Over 28 years of follow-up (until wave 11 in 2024), 1,514 of the 2,054 who gave a numerical answer died. To assess the predictive power of SLE, the researchers used parametric survival analysis with a Gompertz distribution, selected by the Akaike information criterion. To measure inequality in forecast accuracy, they constructed concentration curves and calculated the concentration index (CI). Determinants of errors were estimated using linear regression and multinomial logistic regression. To reduce self-report bias, health was modeled as a combined index based on physical, mental, and behavioral indicators. Attrition was accounted for using inverse probability weights, and robustness was checked with multiple imputation of missing data.
Subjective life expectancy significantly predicted actual survival: each additional year of SLE reduced the risk of death by 1.6–1.7% (hazard ratio 0.984). However, the relationship was inelastic: the marginal effect was only 0.119, meaning that a one-year increase in subjective estimate raised actual life expectancy by an average of just 0.12 years. Most respondents lived longer than they expected, especially women and people with higher education. For example, the probability of "outliving your forecast" was 52.8 percentage points higher for women, and each additional year of education added 1.5 percentage points. The less educated more often gave "round" answers (multiples of 5 or 10 years) and more often said "don't know," which reduced accuracy. The distribution of answers clustered around age 80, close to the average life expectancy at birth for both sexes in 1996 (77.0 for men and 83.6 for women).
This work is the first, using a representative national sample with long-term follow-up, to show that a simple point question about desired age at death has real predictive power, although it is systematically biased. Unlike neighboring studies that focused on probabilistic estimates or Western populations, this one used a less cognitively demanding method and a unique Japanese context with extreme longevity. The identified systematic errors (underestimation by women and educated people, and the tendency of the less educated to give "round" answers) point to specific groups that need help with financial planning. This is not just a statistical quirk but a direct signal for pension consultants and developers of educational programs.
The topic could develop in several directions. First, repeated measurements of SLE over a lifetime would show how expectations change with age and new events (diagnoses, deaths of loved ones, medical breakthroughs). Second, studying young and middle-aged cohorts would reveal how early forecasting errors form, which is critical for long-term savings. Third, comparing point and probabilistic methods in the same sample would clarify which is more accurate and for whom. Finally, incorporating data on parental longevity (especially the same-sex parent) could improve models of SLE formation.
The results directly concern pension policy, financial education, and behavioral economics. They can help develop personalized information materials for retirement planning and choosing when to start receiving benefits.
The next logical step is to analyze data from younger NSJE cohorts or similar surveys to trace the evolution of SLE errors with age. It is also important to test whether providing people with information about age- and sex-specific life expectancy improves the accuracy of their forecasts.
For gerontology and demography, the key unresolved problem is why some people systematically underestimate and others overestimate their life expectancy, and how these errors relate to cognitive abilities, social environment, and access to medical information. The work shows that education and sex are important moderators, but the mechanisms (e.g., the role of optimism, reference points, bounded rationality) require further study.
Nearly 40% of respondents named a "round" age, a multiple of 5 or 10 years, and the most popular answer was 80 years, exactly how long Japanese people lived on average in 1996. Without knowing it, people were using birth statistics as an anchor for their own fate.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.