Retirement “Probabilities” Are Nonsense
The claim
Personal finance tools of all types—ranging from free internet calculators to professional-grade software packages—purport to calculate probabilities of success with respect to your retirement or investment objectives.
Don’t believe it.
Impossible Math
When applied to portfolio simulations, terms like “probability” and “likelihood” convey a false sense of scientific rigor to unsuspecting users. The effect is often amplified by the spurious precision of multiple decimal places (“…your probability of a successful retirement is 88.65%…”).
Here’s the truth: unlike rolling dice or pulling from a deck of cards, risky asset returns do not have a known probability distribution—a fact that prevents rigorous probability calculations. The historic variability of stock market performance is particularly noteworthy. There happens to be a very colorful and controversial academic history on this subject, which we summarize here. There is a clear consensus among researchers: the fundamental statistical properties of long-term asset returns cannot be modeled with any serious degree of precision.
But the problem runs deeper than this. Even if returns did obey the elegant lines of a bell curve, we would still have no way of knowing the correct forward-looking values (e.g., standard deviation and expected return) with which to parameterize our models. To use a gambling metaphor: you can’t determine the likelihood of rolling a six if you know neither the shape nor the markings of the die.
Many advisors and pop-culture finance gurus suggest using historical investment performance to garner an understanding of how investments should perform in the future. But reality is a bit more complicated than that. Running simulations based on past averages across various economic cycles, political regimes, and interest rate environments is fundamentally flawed. For instance, what do 15% U.S. Treasury rates from the early 1980s have to do with projected bond returns today? The same problem exists with the past performance of equities. Prior stock returns are the product of the environment in which they were produced, which depends on valuation levels, prevailing interest rates, economic conditions, and countless other factors.
In short: people simply don’t know enough about investments and markets to make intellectually honest portfolio probability calculations.
Flawed interpretations
Even if we could appropriately parametrize our models, the traditional manner in which simulation outcomes are characterized is terribly flawed. Most software splits the modeling output into two categories: (i) “successful” trials, in which the retiree doesn’t run out of money, and (ii) “failed” trials, in which the retiree goes broke before the end of their life. But this simple dichotomy is flawed and misleading, and—all things considered equal—actually overestimates the likelihood of going broke.
For example, take two “failure” trials that run out of money at the respective ages of 75 and 88, each falling shy of an estimated lifespan of 90 years. Simply classifying both of these scenarios as “failures” is misleading and denies the user some very useful information. The latter scenario, for instance, could have likely survived until age 90 with moderate lifestyle adjustments—and such adjustments are actually consistent with observed retiree behavior.
In fact, a sizable body of research indicates that retirees rarely run out of money, even when markets underperform. Instead, people simply adjust their spending based on portfolio performance. In other words, when times are bad, budgets are trimmed—and it’s remarkable how much seemingly minor adjustments in spending can improve the rate of simulation success. Any particular trial that’s doomed for failure on a fixed budget can often be rescued with a variable budget. This means that many “failure” trials could more accurately be described as trials requiring “spending adjustments”. With this perspective, even a simulated success rate of just 50% can be reasonable. Seriously.
A more thoughtful approach
Our approach to simulation analysis uses fatter tails, multiple time series models, and forward-looking capital market expectations. We also allow users to evaluate the impact of untimely “Black Swan” events, and we steer clear of the traditional “success” or “failure” presentation. We also avoid using terms like “probability” and “likelihood” when describing our analysis, and emphasize that future events are not limited to historical precedent. The market’s behavior in 2020 should have made this clear to everybody. Unfortunately, we tend to forget quickly.
