Analysing Asset Allocation Strategies


Recently there have been academic studies that claim that traditional practices regarding changing asset allocation based on age (e.g. allocate 100 - your age to stocks, Target Date Funds) are flat out wrong and even counterproductive. A good paper is available at, "The Glidepath Illusion - An International Perspective ssrn-id2217406.pdf". These papers identify a few simple allocation strategies that are the opposite of the traditional ones. They then run a Monte Carlo simulation using each strategy and examine the terminal wealths to compare them. One result of these tests is that the opposite strategies always turn out to be superior to the traditional ones.

That's useful information, but the papers don't even try to identify an optimal approach that an investor can employ. My question is can I use the Monte Carlo simulation of ESPlanner to simulate various retirement strategies to discover one that is effective for me? To do this I need to run trials that vary asset allocation by age and view the distribution of the terminal wealth of each trial. Expected return and standard deviation isn't enough: I need to look at the worst and best trials and at some percentiles (or at least the median) in between. Also I need to disable the consumption smoothing feature so the terminal wealths of the trials are comparable. So, does it make sense to use ESPlanner this way to identify a good allocation strategy?



dan royer's picture

Yes, I think I read that same report a week or two ago. This makes perfect sense to me and almost intuitive after my experience with MC and ESPlanner. After all, when very little of my living standard is contingent on market exposure, then risk relative to market exposure is low. Or as Kotlikoff says in his book, the poor can afford to take more risk with their retirement assets than the rich.

But to your question: I'm not sure. You can construct portfolios comprising different kinds of assets--10/90, 20/80, 40/60 etc. And then you can assign those portfolios to come into play at different years. I assume you knew that--if not, perhaps that goes some to answering your question.

Since the monte carlo shows five stops on the distribution from 5th percentile to 95, then you can see the distribution in any given year across those 500 trials.

You lose me when you mention turning off consumption smoothing and with your reference to "terminal wealths." (not sure what that is).

Regardless of whether or not you embrace the suggested strategies, the paper changed how I evaluate risk. It's not the variability of a given asset allocation, it's the worst case scenario as estimated by the Monte Carlo reports.

I ran some Monte Carlo analysis using ESPlanner Plus. All else being equal except that I inverted my allocation progression, the results pretty much replicated the author's findings. The lowest 5th percentile was better in the traditional allocation, but not by much. The lower 20th percentile was about even, but the mean was higher for the contrarian approach and the upper half of the chart was much higher for contrarian than for traditional.

For the benefit of other readers I'm about 4 years out from retirement at age 63. My current allocation is 60/40 debt/equity. My traditional plan is to switch to 80/20 TIPS/S&P500 a couple of years after retirement. In the contrarian approach I switch to 30/70 debt/equity at age 65 (yes, that a dramatic shift), then 20/80 at age 75.

I considered modeling transitions through 50/50 and 40/60 but I'm already past those breakpoints. I simplified and went straight to 30/70, which is approximately where I should be at that point, anyway.

As a quick test I switched out my current 60/40 debt/equity portfolio for 40/60 TIPS/S&P. That was worse that the result above. I assume that's because my current 60/40 portfolio appears to outperform 40/60 TIPS/S&P. That may be exaggerated by having only a short performance history on some of the custom investments I created for it.

Mean Median Ratios Beta
60/40 debt/equity 9.119185 7.962839 0.222499 -0.002244
40/60 TIPS/S&P 6.875641 5.656599 0.364652 0.006457

From personal correspondence with Mike I gather that by "terminal wealth" he refers to assets available for inheritance.

You may find several books by William Bernstein useful on this topic. For example:

- Rational Expectations: Asset Allocation for Investing Adults
- The Ages of the Investor: A Critical Look at Life-cycle Investing (Investing for Adults; [Book 1])
- The Intelligent Asset Allocator: How to Build Your Portfolio to Maximize Returns and Minimize Risk

The first two are recent and all are insightful.

To your question, more and more I question the ultimate value of Monte Carlo. Sure there is some useful information and testing different models can provide important insights along with testing spending behaviors, etc. However, correlations that used to hold true (for long established funds and asset classes) are breaking down/have broken in many instances and newer funds lack a track record along with odd correlations since at least 2007/8, if not before.

So...I seriously question if these MC assumptions/expectations are valid for the next 5, 10, 15 years or longer. One of these days, I expect we'll see a break. Perhaps high inflation, market crash, deflation, stopping major central bank printing/asset purchases/QE, who knows what or when except that what can't continue forever, ultimately won't.

I'm interested if other ESPlanner users have thoughts on this as well.


dan royer's picture

I share those concerns Brian. I feel skeptical about the history assumed by an asset class. I am not a hyperinflationista, and I am optimistic about US growth. But I'm just more comfortable with conservative assumptions I guess. I find the MC distribution tables interesting to some extent, but I prefer an approach that sees how low I can set a nominal return and still be happy with living standard and then allocate to try to match that return target.

Thanks Dan. I've come to think of asset allocation as secondary to ESPlanner profile optimization along with "what if" testing. What I mean is that through sensitivity analysis, you can find which variables (or combinations) lead to increased consumption. A number of these, such as SS start dates, have low risk (assuming you can afford to wait) and a nice increase. There are a ton of options here. With effort, these can be identified and people may find that they can increase their standard of living substantially while potentially reducing their overall risk. This approach combined with a reasonable asset allocation strategy makes sense to me.


dan royer's picture

Oh, yes, absolutely. Investigating those optimization strategies is ESPlanner's forte. That's always the place to begin I agree. And I'm often interested to learn that each new "case" has buried in it somewhere some way to optimize that was not available in another case. So the use of the imagination is also an important component when constructing a "what if."

We use cookies to deliver the best user experience and improve our site.