Home » Blog » You Can’t Predict the “Best” Portfolio

You Can’t Predict the “Best” Portfolio

Glancing through my posts so far this year, I noticed how often I’ve been writing, “No one can predict the future”, or some variation on that theme.  So, I thought it would be interesting to drill down a bit more on this topic.

Uncertainty about the future is central to many of our investing decisions (and decisions in general).  It comes up most prominently and frequently in the area of asset allocation, or portfolio construction if you prefer.  Given all the different assets and their varied historical performances, how do we construct a portfolio that has a decent chance of performing well in a variety of uncertain futures?  That’s a relatively mindful way to frame this question.

A less mindful way to frame the question is something like, “How do I pick the best portfolio?”, or “What’s the optimal portfolio for the next 10 years?”  Unfortunately, I see these types of questions all too often in the media and blogs about investing.  This sort of question problematically implies that there’s one optimal portfolio out there that will perform well in all potential futures.  It’s like trying to find one set of clothes that will work best from now until the end of the year.  The future can produce more variations and surprises than we can ever dream up.

Unfortunately, many people and even “experts” tend to approach portfolio construction as an optimization exercise where historical data are used to select the portfolio that’s expected to perform best in the future.  This is often called “backtesting”.  Websites like Portfolio Charts and authors like Paul Merriman are dedicated to this premise.  Don’t get me wrong.  Portfolio Charts is a wonderful site that’s full of many useful backtesting tools.  And Paul Merriman presents back-tested portfolios that have plenty of merits.  The main issue is more about maintaining realistic expectations when using such resources.

The standard assumption is that the past will tell us about the future.  But of course, the problem is that the future will always turn out differently than the past, at least to some degree or in some unexpected ways.

A Thought Experiment

I decided that a thought experiment on portfolio construction is a good way to illustrate the problems with predicting the future using back-tests.  Let’s say it’s the year 2002 and you’re trying to pick a buy-and-hold portfolio for the next 16 years (through 2018).  Consistent with the idea of back-testing, you could look at the performance of various recommended portfolios in the prior 16 years (starting back in 1986).  (I picked these timeframes and dates based on the availability of data on various asset class returns.  Ideally, one would backtest for much longer periods, but longer spans of reliable and consistent data are often not available for one or more asset classes of interest.)

So, in our thought experiment, the timespan from 1986 to 2002 is the back-test period.  And from 2002 to 2018 is the forecast period, the period over which we hope our selected portfolio will perform well.

Methods – For this experiment I used some portfolios described at Portfolio Charts that have been popularized in the media.  I also threw in an example of a “mindful” diversified all-stock portfolio as well as two super simple portfolios that are popular in some circles: an all-US stock portfolio and the S&P 500 portfolio.  The names of each portfolio and the asset classes held in each are summarized in this table.

Asset All Season 60/40 No Brainer Perma-nent Three Fund Golden Butterfly Mindful Stock All US Stock All S&P 500
US Stock Market 30% 60% 25% 40% 20% 60% 100%
Long-term Treasury 40% 25% 20%
Intermediate Treasury 15% 40% 40%
US Large Cap 25% 100%
Gold 15% 25% 20%
US Small Cap 25%
Intl. Developed Stock ex-US 25% 25%
Short-term Treasury 25% 20%
Cash 25%
Intl. Stock Ex-US 20%
Emerging Market Stocks 15%
US Small Cap Value 20%

I plugged each of these portfolios into Portfolio Visualizer for both the back-test and forecast periods and summarized some of the resulting return and volatility statistics.  In most cases, tests like this will assume that the portfolios are rebalanced periodically to maintain the starting proportions of the assets.  But because portfolio rebalancing is often not the slam dunk that many people claim, I instead assumed no rebalancing.  In that sense, this experiment tests a true buy-and-hold strategy, where the portfolio is left to grow untouched after a one-time purchase of each asset.

Results – Here’s a table of the results for the back-test period from 1986 to 2002.

Portfolio Annualized Return (CAGR) St. Dev. (Volatility) Return/Risk Ratio Max Drawdown Return Rank St. Dev. Rank
All Season 9.25% 7.87% 1.18% -12.36% 5 2
60/40 9.83% 11.49% 0.86% -30.21% 2 5
No Brainer 8.88% 12.75% 0.70% -33.45% 7 6
Permanent 7.68% 6.86% 1.12% -15.67% 8 1
Three Fund 9.10% 10.80% 0.84% -27.72% 6 4
Golden Butterfly 9.33% 8.84% 1.06% -14.45% 4 3
Mindful 6.12% 16.49% 0.37% -44.01% 9 7
All US Stock 9.50% 17.03% 0.56% -44.11% 3 9
All US Large Cap Stock (S&P 500) 10.24% 16.83% 0.61% -44.82% 1 8

Eee gads!  The Mindful portfolio had the worst annualized return, mainly due to the very poor performance of international and emerging market stocks in this period.  The best annualized return was for the simplest portfolio consisting of only the S&P 500.  While mindful investors care less about routine volatility, it’s worth noting that the Permanent portfolio had the least volatility and the All Season portfolio had the best risk-adjusted return (using the return/risk ratio).  This makes sense given that these two portfolios hold a lot of bonds and/or cash, which typically are more stable than stocks but generate lower returns than stocks.

So, if in 2002 you were interested solely in returns, you might conclude that the best portfolio for the next 16 years would be the super-simple S&P 500 index fund.  And if you were interested solely in minimizing volatility, you might pick the Permanent portfolio instead.

What actually happened in the forecast period from 2002 to 2018?  Here’s a table of those results.

Portfolio CAGR St. Dev Return/Risk Max Drawdown Return Rank St. Dev. Rank
All Season 6.76% 7.51% 0.90% -14.63% 4 1
60/40 6.16% 7.83% 0.79% -27.38% 7 2
No Brainer 6.07% 11.95% 0.51% -43.02% 8 6
Permanent 6.47% 8.64% 0.75% -16.88% 6 5
Three Fund 5.81% 8.50% 0.68% -32.55% 9 4
Golden Butterfly 6.94% 8.40% 0.83% -17.64% 3 3
Mindful 7.05% 15.92% 0.44% -56.24% 2 9
All US Stock 7.20% 14.36% 0.50% -50.89% 1 8
All US Large Cap Stock (S&P 500) 6.71% 14.01% 0.48% -50.97% 5 7

The Mindful portfolio had the worst returns in the back-testing period, but in the forecast period, it was the second-best performer.  Yay mindfulness!  Similarly, the S&P 500 portfolio dropped from the first rank for returns to the fifth rank.  The All US Stock portfolio managed to rank pretty well for returns in both periods.  From a volatility standpoint, the least volatile portfolio in the back-test period was the Permanent portfolio, but it fell to the fifth rank for low volatility in the forecast period.  The All Season portfolio had very low relative volatility in both periods, again because it holds 55% in bonds.

The other interesting thing about the results in the forecast period is that the rates of annualized return are all tightly grouped between a little less than 6% and a little more than 7%.  It turned out that in 2002 it hardly mattered which portfolio you picked.  But that outcome wasn’t obvious at all based on the back-test results.  In the end, all that back-testing in 2002 provide very little help in finding the best performers for the future.

Conclusions

My little experiment shows that back-testing can provide some useful but limited information.  In this case, the back-tests did a pretty good job of differentiating future low volatility portfolios from high volatility portfolios.  But the back-test ranks weren’t really predictive of future ranks for either returns or volatility.  In my view, just looking at the ratios of stocks to ballast (bonds/cash) within each portfolio would have provided nearly as good a prediction without the need for any calculations.  Back-tests are like looking at the future through binoculars; they’re certainly not a microscope.

Given that the supersimple portfolios produced pretty decent returns in both periods, why diversify at all?  There’s certainly a case to be made for a simple portfolio composed entirely of one S&P 500 index fund.  But the most mindfully simple reason to diversify your portfolio is the same inherent uncertainty about the future that we started with.  No one can say for sure which single asset will perform best in the future.  So, spreading your bets across several assets will minimize the chances that you’re stuck with one clear loser.

By the same token, diversification also means that you’re almost guaranteed to underperform some of the less diversified (and sometimes more diversified) portfolios in any given period.  There will always be something relatively sucky in your portfolio because there will always be a top performer in your portfolio.

If back-tests could tell us the best combination of assets for all possible futures, then it could just as easily pick out the single best performer, which would alleviate the need for diversification entirely.  Mindfulness can help us accept the fact that it’s impossible for back-tests, other calculations, and “experts” to consistently predict the future.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.