Planning to life expectancy is planning to run out half the time
The claim
Life expectancy is a median, so a plan funded to that horizon fails for roughly half of the people who use it, and individuals systematically misjudge their own survival odds at older ages.
Why we rate it strong
The statistical point is definitional. The behavioural point — that subjective survival expectations are biased in a consistent direction at older ages — is measured against realised mortality in large panel data.
Life expectancy at 65 is a median for a population, not a forecast for a person. Half of the cohort outlives it, by construction. A retirement plan funded exactly to that horizon therefore fails about half the time, which is not a standard anyone would accept if it were stated that way.
The couple case stretches it further. Planning has to cover the survivor, so the relevant number is the last death rather than the average death. For a 65-year-old couple, the probability that at least one is alive at 90 is far higher than the probability that any particular one of them is — which is why joint planning horizons of thirty years or more are conservative rather than paranoid.
Hurd and McGarry tested whether people's own survival estimates predict their actual mortality. They do carry real information — people know things about themselves that tables do not — but the estimates are biased, and older respondents in particular understate their chances of reaching advanced ages. The error runs in the direction that shortens plans.
Two more effects push the true horizon out. Period life tables ignore continuing mortality improvement, so a cohort table is the appropriate one and it is longer. And mortality has a steep socioeconomic gradient: the people most likely to be reading a retirement planning site are drawn from the longer-lived part of the distribution, so the population average understates their horizon specifically.
The practical consequence is not that everyone should plan to 100. It is that the horizon is a distribution, which is why this site runs withdrawal plans across every historical sequence rather than reporting one average, and why life-contingent income deserves a place in the answer.
Where this breaks down
- Health status dominates. Someone with a serious diagnosis is right to plan a shorter horizon, and averages are not evidence about them.
- Longer horizons are not free: planning to 100 means underspending for decades, and the cost of that is real even though it never shows up as a failure.
- Spending falls in later retirement for most households, so a longer horizon does not require the same real income throughout — the flat-spending assumption overstates what a long life costs.
- Mortality improvement has slowed in the US recently, and projecting past gains forward is itself an assumption.
Sources
Follow these rather than taking our word for the summary.
Michael D. Hurd and Kathleen McGarry (2002). The Predictive Validity of Subjective Probabilities of Survival
The Economic Journal, 112(482), 966-985
Finding: Subjective survival probabilities predict actual mortality but are systematically biased, with older respondents understating their likelihood of reaching advanced ages.
Social Security Administration, Office of the Chief Actuary (2026). Actuarial life table
ssa.gov
Finding: Period life tables give the median remaining lifespan at each age, from which the probability of survival to any later age can be read directly.