Part I discusses the book's section on asset allocation. Part II discusses the book's other insights, my critique, and the author's responses.
Historical statistics should not be blindly fed into an optimizer.
Paul D. Kaplan, Ph.D.
The quote above belongs on a T-shirt. I'd like to give one to every quant headed for Wall Street with visions of spinning mathematical models into gold.
For now, though, I'll just continue with my review of Kaplan's Frontiers of Modern Asset Allocation, beginning with one of the book's key insights.
Sample Bias
As noted in the quote above, Paul Kaplan has tremendous respect for the limits of what optimization can achieve. Likewise, Laurence Siegel suggests limits on what extrapolation can achieve, since history offers us limited samples of the past.
In the book's foreword, Siegel noted that we treat the last century of equity returns in the U.S. and the U.K. as if this were a normal baseline period. But Siegel points out that it was impossible for an investor in 1900 to know in advance that the U.S. and the U.K. would succeed. After all, investors in 1900 who chose Japan, Russia, and Germany were completely wiped out. Siegel calls this an example of both survival bias and sample bias, since we tend to extrapolate from prior success, even when the period was not typical.
This is a critical insight for anyone who relies solely on long-term historical averages. What is the appropriate sample period?
Impact of Resampling on Monte Carlo Simulations
Siegel's comments also have implications for Monte Carlo simulations. Most Monte Carlo simulations take investment returns since 1926 and put them in a blender. Next, the returns are pulled out one at a time, just like the ping-pong balls in the NJ Lottery. This is called resampling, and it can create many scenarios from a limited sample of data.
When we simulate retirement, for example, we need to estimate 30 years of future returns. Monte Carlo simulations can do this by recreating thousands of 30-year periods based on returns since 1926. Unfortunately, resampling removes data from its historic context.
Kaplan does not specifically highlight the problem of resampling, though his methods certainly imply a healthy skepticism of blind extrapolation. Here is how I would illustrate the limitation of resampling: Bonds beat stocks from 1982-2011 due to a long-term decline in interest rates. Because of this trend, the returns in one year were dependent on the prior year (serial correlation), so the data were not randomly distributed.
Kaplan is certainly aware of economic trends, and he allows for serial correlation in his models for inflation. In my humble opinion, the reader would have benefited from a warning about the limits of resampling: Investors should understand that Monte Carlo analysis simulates long-term returns by connecting resampled data out of its original historic context.
Kaplan is more concerned, however, that investors are lulled into complacency by averages. As he notes in "The Flaw of Averages" on page 326, a mathematical average condenses a wide range of uncertain possibilities into a single number. Unfortunately, some investors focus too much on the average and not enough on volatility.
Extrapolation Incorporated
When I worked at Value Line in the 1980s, we used to poke fun at analysts who merely extrapolated prior trends: They should go work at "Extrapolation Incorporated."
The whole point of making a forecast is to figure out how the future will be both similar to the past and different from the past. This is why Kaplan cautions against overreliance on optimization and Monte Carlo simulations. His quote is worth repeating: "Historical statistics should not be blindly fed into an optimizer" (273).
Other Noteworthy Insights
- The Real Markowitz Portfolio: How did Harry Markowitz manage his own money? Did he use a complex algorithm to determine his asset allocation? On page 306 we find that Markowitz told Jason Zweig the answer in 1988: Markowitz had a 50/50 mix of stocks and bonds. Somehow I find it heartening to know that Harry Markowitz used heuristics for his own portfolio.
- Reported Volatility Is Too Low: Kaplan showed how the reported volatility is too low for small-cap stocks, real estate, and hedge funds. He offers detailed analysis of each, and proves his case.
- A Better Commodity Index: Commodity returns are a function of both the roll-yield and the price change in the asset. Kaplan shows how long-only commodity indices do not capture the roll-yield, and are poor investments when futures prices are in contango (when prices are rising, and the futures curve is positively sloped). Kaplan suggests that a momentum-based long/short index is a better solution, since it captures both price returns and the roll-yield.
- An Objective Annuity Framework: Kaplan provides an analytical framework for investors who are considering adding annuities to their portfolio. Kaplan addresses both withdrawals and longevity risk, and his analysis is not biased by the huge commissions that have tainted annuities as retirement product.
- Four Debates: The book includes candid interviews with luminaries such as Robert Arnott, BenoĆ®t Mandlebrot, and Roger Ibbotson. The topics range from MPT to fundamental indexation.
My Critiques and the Author's Responses
Although I thoroughly enjoyed the book, I did have some critiques of the content.
First, I'll note that all 27 of the chapters have been previously published as individual articles (15 by Morningstar and 12 by other professional journals). What's more, some of this material is a bit dated: 5 articles were from the 1990s, and 7 were from 2000-2008. Perhaps I expected more from a 2012 book called "Frontiers of Modern Asset Allocation." Since the book is a collection of articles, I found that it lacked a certain "flow." As such, I believe it is most useful as a reference book.
Paul Kaplan said that he intentionally arranged the book as a collection of articles, which is common for authors in his field. Kaplan did add a variety of new material:
- An introduction to the book
- An introduction to each section
- Three technical appendices that were not in the original articles (these are meant for the technically oriented institutional reader).
As for the "flow" of the book, Paul Kaplan pointed out that the afterword by Thomas Idzorek provides an excellent summary that puts everything in context.
This book is clearly aimed at audience of institutional investors. Nevertheless, I found that it had quite a few references to Morningstar's products and services. Paul Kaplan responded by noting that Morningstar owns the copyrights to this work, which he wrote as an employee with corporate support. Paul also pointed out that certain of the ideas and methodologies discussed in his work are now embedded in Morningstar products, suggesting relevancy to today's professional investor.
Finally, I'll note that the book isn't cheap at $95, though it is available now for $60 on Amazon. Paul Kaplan was kind enough to send me a free copy for this review, and I thank him for the opportunity.
Disclosure: I have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.
Additional disclosure: See here.