It has been said that if you want to start an argument on the Internet, all you need to do is state an opinion. From what I can tell, sharing a fact can do it too.
Many times, comments on articles or social media contain something of substance but often that is not the case. Some rants are funny or entertaining in a good way. Some are just plain mean and make you wonder how things got so nasty for the commenter that they became so unpleasant.
One of the few refuges from the ugliness has been the message boards of FPA and NAPFA. Sure, now and then someone gets, shall we say “passionate” about a topic, but most conduct themselves professionally. I have learned a thing or two just by reading through the digests, and they have inspired a few of these columns over the years.
Recently, a few FPA members had an interesting discussion about Monte Carlo simulations (MCS). Most commentary was negative and generally followed prior criticisms about the technique. A few suggested that MCS was misleading, even dangerous. With MCS so prevalent in financial and retirement planning software, I thought it might be worthwhile to bring some of the issues out.
The two criticisms I see most of the typical MCS function in planning software are that market returns have not conformed to a normal distribution curve and that with returns forecasted to be lower than historical averages, using historical data overstates a client’s chances of success.
I believe these two criticisms are valid, but I don’t agree with the assertion that MCS is, therefore, something planners should eschew for being misleading or dangerous. It can be a tool of great value. Like any tool, how you use it makes a difference.
Let’s look at these criticisms.
There should not be any debate about whether returns conform to a normal distribution curve. The data speaks for itself. A log-normal distribution is a better fit for returns. Some critics go a step further and point out that even if you use a log-normal distribution in MCS, actual returns demonstrate fat tails. Extreme events, good and bad, have occurred more frequently than the curve would predict.
I have even seen several disparaging things said about MCS because the simulations “miss” so-called black swans. A favorite seems to be that based on historical data for daily behavior of the Dow, MCS software should predict a one-day decline of 22 percent once every gazillion years. Such an event is basically impossible, statistically speaking. Yet, it happened on Black Monday in 1987, so MCS is a farce.
I think this criticism is off the mark. I don’t dispute the statistics are not wrong, but the conclusion that MCS is, therefore, useless makes no sense to me. That is like saying you shouldn’t keep a screwdriver in your toolbox because screwdrivers won’t hammer nails very well. Are there really people that use MCS to predict daily returns or otherwise time the markets? I hope not. That is an improper use of the tool.