By LARRY SWEDROE
When developing a financial plan, one of the most important assumptions is life expectancy. And while no one knows exactly how long they will live, the estimate should be carefully considered to make sure it is as accurate as possible, i.e., adjusting mortality tables for one’s unique health profile. That is because errors in forecasts can have a significant impact on estimates of how much one must save for — and spend in — retirement.
Errors in longevity estimates can occur because one might not be aware that while life expectancy (measured in years) declines with age, the average expected age of survival increases with age. For example, the life expectancy of a 65-year-old male is 18 years, which implies that the average 65-year-old will live to age 83. That is six years longer than that of a newborn.
In addition, mortality tables are based on estimates for the probability of survival for a certain population of individuals for a given age (or age range). However, mortality rates can vary significantly across different cohorts of individuals based on different attributes or behaviours (such as smokers). Therefore, it is important to understand the respective population in addition to how well it fits a given individual.
David Blanchett, author of the paper Minding the Gap in Subjective Mortality Estimates published in the Fall 2021 issue of The Journal of Retirement, examined the accuracy of subjective life expectancy estimates using data primarily from the Health and Retirement Study (HRS). Following is a summary of his findings:
Consistent with past research, while individuals appear to have some sense about their likelihood of survival (on average, their estimates are close to those provided by Social Security), there are notable gaps in their estimates. For example, respondents in the first wave (1992) of the HRS who said they had a 0% probability of surviving to age 75 actually had about a 50% chance, and those who said they had a 100% probability actually had about an 80% chance.
While individuals do a relatively good job incorporating their health status into projections, they do a relatively poor job incorporating things like smoking (negatively correlated with longevity) and income and wealth (both of which are positively correlated with longevity).
Smoking was the most understated variable. Smokers were overly optimistic about their survival chances — individuals who smoked reported a 4% lower probability of surviving to age 75, holding the other variables constant, while the actual probability impact of smoking on survival was closer to -15%.
Males had only a 4% lower subjective survival probability but a 10% lower actual probability, holding the other variables constant.
Blacks/African Americans had a higher subjective survival probability but a significant negative gap, overestimating survival by approximately 12%.
Health was the most important indicator associated with survival. For example, an individual in poor health had a 25% lower probability of survival, holding all other variables constant, and households did a good job considering it when creating subjective probabilities. However, a mortality model built entirely on self-reported health would have been more predictive than one based on subjective probabilities.
There was little improvement in subjective predictability since the first wave in 1992, with two exceptions—the newer survey respondents correctly recognised that couples tend to live longer, and they also incorporated the income effect (i.e., households with higher incomes have higher life expectancies).
While average longevity estimates are fairly accurate (a person can drown in a stream with an average depth of six inches), the average person who overestimates his/her life expectancy would be expected to save 25% more than required, and the average person who underestimates his/her life expectancy would be expected to save 25% less than required.
His findings led Blanchett to conclude: “A predictive survival model built from health status alone would have outperformed subjective mortality estimates historically and considering other objective factors that are readily available (e.g., age and gender) would further improve the model. Therefore, given the gaps that exist in subjective estimates, objective information should form most (or all) of the basis of any type of mortality estimate.
In other words, opposed to asking someone how long he/she thinks they will live, combining generally available information (e.g., age, gender, income, etc.) with questions about notable objective drivers of mortality rates (e.g., health, smoking, etc.) is likely to result in a more accurate estimate of mortality.”
Investor takeaway
As Blanchett noted, personalised mortality estimates can have a considerable impact on a financial plan, as the assumed retirement period not only affects things like required savings and optimal spending but also important decisions, such as when someone can retire and how much he or she needs to save.
Additionally, decisions about how to fund retirement—whether to delay claiming Social Security retirement benefits, to purchase an annuity, and others — will be affected. Thus, it is important to ensure that the estimate is as accurate as possible.
Adding an objective and knowledgeable voice to the discussion is one way a financial adviser can add considerable value. For them to do so, they must be aware of the biases of their clients and the impact of those biases.