Last week we started a discussion on investment management, specifically on the topic of diversification’s place in portfolio construction. This week, I’d like to continue this broad theme and discuss a potentially large flaw in the usage of Monte Carlo simulation. I’ll forgo the explanation and assume we all have a working knowledge of the topic.
Monte Carlo Simulation
Monte Carlo simulation has become extremely common in today’s advisory community. Common, yes. But is it being used properly? Often, the answer is no. I was introduced to MCS about 14 years ago. One of the first things I noticed is how most software programs used asset class data instead of data pertaining to the specific investments being analyzed. The problem is that the performance of any given investment can, and usually will, vary to some extent from that of its asset class. Hence, under these conditions, the simulation would yield erroneous results. When an investment strays far from its category, it is said to have a high “tracking error,” which we’ll discuss next week.
An asset class consists of a universe of similar types of investments, but the variance of the universe will range from small to large, depending on the particular asset class you are using. For example, the variance between short-term bonds will be much lower than that of small-cap stocks. Hence, using asset class data will greatly skew the results.
Remember the term “GIGO?” We all understand that better input equals better output. Therefore, if we are using asset class data for a portfolio and the specific investments tend not to track well with their asset class, the result will be unreliable. Therefore, rule number one should be to use actual investment data and forgo asset class data when utilizing MCS.
Even if the program allows for the use of specific investment data, there remains the question of what data to use. For example, should you use the entire period from 1926 to 2012? I recall that during the late 1990s, the 10- and 20-year trailing returns on the Dow were somewhere in the neighborhood of 14% to 18%. I may not have those figures exactly correct, but the point is that the returns over this period were far above the norm and using numbers based on it would have led to many unhappy investors.
If past returns are so unreliable, what can be done? There are several organizations that prepare capital market assumptions. Often, these assumptions are designed to be forward-looking based on expected economic data, earnings, etc. Of course, if they are wrong, then again, the results will be flawed.
Ironically, this is why MCS has become so popular. MCS forecasts the probability of future occurrences, and although it is a great improvement over the single-point, deterministic forecast, we need to be careful with our input. MCS does have a lot of sizzle, but let’s make sure we don’t burn ourselves by placing too much emphasis on it. I do use it and believe it offers a significant benefit. Unfortunately, it’s not a crystal ball, even though that’s the name of the Monte Carlo product I use. Ironic isn’t it?
Thanks for reading and have a great week!
See Mike Patton’s latest feature story from Investment Advisor on his process for choosing mutual funds.
View all of Mike Patton’s blogs and feature articles on his AdvisorOne home page.