Gyroscopic Investing

Larry Swedroe just posed this excellent article on mistakes most investors (including a lot of financial planners) make in portfolio choices:

http://www.advisorperspectives.com/arti ... isors-make

PP adherents understandably take a lot of pride in holding an allocation based on having non-correlated assets whose whole is greater than the sum of its parts, but when looking at tweaks or variable portfolios I see that I - and many others - still rely all-too-often on backtesting.

IMHO Mr. Swedroe makes a compelling case that P/E and CAPE ratios are far more accurate predictors of equity returns going forward. Among the near-term practical implications are (a) lowered expectations for stock and bond returns vs. historical averages given comparatively rich equity valuations and unprecedentedly paltry bond interest rates; (b) TSM just isn't going to cut in on the equity side:

"...expected real returns to U.S. stocks of 4.5%, to developed market stocks of 7.3% and to emerging market stocks of 9.0%. To get an estimate of nominal returns, you can add the current spread between the 10-year nominal bond and the 10-year TIPS, which is currently about 1.5%."

Entranced as I am with Tyler's Portfolio Charts site, it's all about a detailed look in the rearview mirror. How to combine that in a systematic way with the far more useful predictive power of current valuations is an interesting challenge.

As much as I respect Swedroe, he lost me on this one.

The correlation of CAPE 10 to forward-looking returns promoted in the article is also based on studying historical returns. So he just as easily could have said that historical returns can be deceiving when taken out of context but useful depending on the application. That's not particularly controversial. But stating simply that the use of historical data is a "mistake" is sorta misleading.

I think the larger issue is the implied assumption that accurately predicting future returns of individual stock classes is required for investing success. He seems to be making an argument for active management using historical data in a very specific way as a predictive tool. Harry Browne's great insight was to take the approach of managing the risk of an unpredictable future using true diversification with no predictive gymnastics required. I personally prefer Harry's approach.

BTW, note that his entire opening argument about the problem with using past data to predict future returns is basically a rehash of the Start Date Sensitivity calculator. Luckily, not all portfolios are equally as start date dependent as the stock market. He sorta hinted at this at the end when talking about Monte Carlo simulations:

MC simulations also provide another important benefit: They allow investors to view the outcomes of various strategies and how marginal changes in asset allocations, savings rates and withdrawal rates change the odds of these outcomes. Looking at various alternatives will help you determine the right asset allocation for your unique situation.

I'd argue PortfolioCharts uses actual historical data to pursue the same goals. History ain't so useless after all!

This is just brilliant Tyler, and so helpful! I thought that I was missing something obvious and it turns out I was missing multiple somethings!

I really appreciate you taking the time to respond to the article. BTW Larry first shared the article over on Bogleheads and I see the from the responses there that no one has caught on to what you're saying, so I can console myself with not being the only semi-educated investor to have missed the boat here.

CAPE is tricky...it sort of works but by no means does it work as an accurate timing measure. Over priced stocks can last 1-2 decades in which time you can miss out on a lot of gains. If you are going to try timing do something like a 1.5-2 year moving average measured monthly (I.e. 18-24 month MA). It won't add anything to the bottom line but it will cut out the major bear markets. Be aware however that the best market performance comes out of the early parts of new bull markets.

Bottom line: there are no free lunches.

I agree with Swedroe on this one. I've long been suspicious of backtested results, except that they do give useful information on how specific portfolios will respond under different economic conditions. Differences of 1-2% of CAGR, however, are likely not to be replicated going forward.

Harry Browne had a very eloquent critique of backtests that he described in his books & radio show. It's worth reading/listening to get this viewpoint, but the Cliff notes version is that the world's state cannot be captured adequately in any data synthesis, it's impossible to predict what will happen in the near term, and previous results may not be valid because the state of the world will have been different. Thus, markets will react differently to the same stimulus in different time periods.