Rolling a die and betting it comes up with a two on top creates an 83 percent risk that we’ll be wrong—the risk is knowable and can be predicted and quantified. What is the risk that we’ll run out of money early by withdrawing three percent or four percent or five percent from our investment portfolio? Although this is a risk, what we are really talking about is uncertainty, since too many of the variables—such as living too long or earning too little—cannot be accurately predicted.
Wall Street pretends there is no uncertainty, but instead quantifiable risk, by creating models and then proclaiming that there is a “78 percent probability that their model will perform.” However, the reality is there is neither a 78 percent probability that their model will be right nor a 22 percent risk that it will be wrong due to model uncertainty. This uncertainty takes three forms: The first is that the parts used in building the model are subjective and art rather than science (the composition of the investment portfolio, inflation rate, past period used, longevity); the second is assuming that, even if we got all the parts correct, that the performance and economic forces of the past will repeat; and the third is assuming that people will act like the model requires—a supposition that is routinely contradicted by real life behavior.
Wall Street’s models have gotten better—Monte Carlo models are better than simply using average performance because they show the range of returns—but every one of them suffers from model uncertainty. Of these, the primary flaw is using a static past to predict the future. In broad terms the past is useful…markets go up and down…but the past is largely useless in predicting specific periods. As a small example: The market of the ‘80s and ‘90s did not reflect what has happened so far in this millennium, and a retiree relying on a Monte Carlo model from that period could be in serious trouble today due to sequence of return risk and lower overall returns. These models could be improved if Bayesian statistics were more commonly used, since Bayes tries to incorporate current changes and new assumptions into old data, but I have not found Bayesian statistics commonly utilized by Wall Street advisors.
Of course, a way to avoid this type of retirement uncertainty is to simply purchase an annuity with a guaranteed lifetime income benefit. If a $30,000 a year inflation adjusted retirement income is desired, today’s Wall Streeter usually says you need $1 million. However, at age 66 you can also get an inflation-adjusting $30,000 initial income for $634,000 with an immediate annuity (go2income.com). Or, if you have a bit of time, a guaranteed lifetime increasing-income withdrawal benefit can get you that $30,000 for quite a bit less than $1 million and you retain access to the balance.
Buying a life-income annuity for retirement does not completely get rid of general uncertainty, because there is the possibility of carrier failure (notwithstanding the promises of state guaranty associations), but it does eliminate almost all of the model uncertainty that taints Wall Street investment-based models. Buying an annuity is rolling a pair of dice with a two on every side.
