Archives for November 2019

A reader writes in, asking:

“What are the pros and cons of using the Monte Carlo tool for retirement planning?”

I wouldn’t focus so much on the pros and cons of Monte Carlo simulations, because there’s so much variation among how the Monte Carlo simulation concept is applied. Instead, I would focus on knowing what is going into (and coming out of) a specific Monte Carlo tool — or a study based on such.

For instance, what assumptions are being made about returns? Is the analysis using historical mean return as the mean return for each asset class? Or is it making a downward adjustment to account for today’s relatively low interest rates and high valuations? And what type of distribution is being assumed about returns? (For instance, many Monte Carlo tools assume a normal distribution for stock returns, which significantly understates their risk.) And what assumption is being made about reversion to the mean? That is, are the simulations assuming that several bad years in a row increases the likelihood that the next year is a good year? A set of simulations that does include such an assumption will have fewer very bad scenarios than a set of simulations that does not include such an assumption.

And what assumptions are being made about mortality? Are you (and your spouse, if married) assumed to live precisely to your life expectancy but no longer? Or is age at death one of the variables that the simulation is allowing to fluctuate? Or is mortality completely ignored, and the analysis is simply looking at a fixed length of time, such as 30 years?

And what assumptions are being made about spending? Most analyses don’t account for the possibility of spending shocks (i.e., an unanticipated and unavoidable large amount of spending in a given year). That’s fine, but it is important to recognize then that the analysis is leaving out one significant type of risk that exists in real life.

Overall point being, if you don’t know what assumptions are being made by the tool (or if you’re reading a study/article based on a set of simulations the writer performed and you don’t know what assumptions were made), it’s hard to get a lot of value out of the conclusions.

And what metrics are coming out of the simulations? For instance, if the simulations are regarding retirement strategies (i.e., a combination of spending decisions and asset allocation decisions — and possibly Social Security/tax decisions), does the tool give us something other than simply “likelihood of running out of money”? For instance, a few other metrics that are useful to know are:

• In scenarios in which the portfolio is depleted, when is it depleted (e.g., does the average/median depletion occur 15 years into retirement or 25 years into retirement)?
• In scenarios in which the portfolio is not depleted, what is the average/median bequest?
• If the strategy allows spending to fluctuate based on market performance, how much does it end up fluctuating?

(See this excellent paper paper from Wade Pfau, Joe Tomlinson, and Steve Vernon for a more thorough discussion of useful metrics for evaluating retirement plans.)

In short, one Monte Carlo analysis can vary significantly from another. So I wouldn’t worry so much about the pros and cons of Monte Carlo analysis in general, but rather make sure you understand what you’re dealing with when you use a given Monte Carlo tool or read about a Monte Carlo-based study. What types of risk are being excluded from the analysis? And what information is being left out of the output?