It seems that not a day passes without my seeing an article claiming modern portfolio theory is dead. That also seems to go for terms used synonymously with MPT: "asset allocation," "strategic allocation" and "buy and hold" investing.
What surprises me is that today's MPT critics seem to believe they have just discovered the truth, when in reality a new group of gurus consistently discover the same truth after every bear market.

If anyone deserves credit for thoughtful and legitimate criticism of MPT when misapplied, it is William Jahnke. Although we disagree on many issues and I've debated with Bill over many years, his February 1997 article in the Journal of Financial Planning, "The Asset Allocation Hoax," remains the most notable piece on the subject. Long before today's critics appeared, Bill was writing articles about it for the Journal, including his February 1999 piece, "Portfolio Management: Why Setting an Asset Allocation Policy Is a Bad Idea," and his June 2004 article, "It's Time to Dump Static Asset Allocation." In these papers, he lays out several criticisms of MPT:

The allocations are solely and simplistically based on projected historical data. "In essence," he writes, "we forecast that the long-term average historical correlation is an accurate estimate of the future correlation between investments."
Contemporary market valuations are ignored. Traditional methodology assumes valuation is irrelevant.
Allocations are determined at the beginning of the investment process and are never changed, except when they are rebalanced. "The idea that the most important investment decision should be fixed at some arbitrary point in time is strange advice."

A more recent criticism of MPT is Kenneth Solow's recently published Buy and Hold is Dead (Again). In this excellent book, Solow echoes and expands on the criticisms raised by Jahnke.

Although I agree that many practitioners who follow the tenets of MPT develop allocation models based simply on historical data, the criticism should be leveled at the practitioner, not the process. I can only assume that the critics have failed to read Harry Markowitz's seminal paper, "Portfolio Selection," from the Journal of Finance, 1952.

Markowitz writes this about portfolio selection: "The first stage starts with observations and experience and ends with beliefs about the future performances of available securities." (Emphasis mine.)

In detailing what he refers to as the "expected return-variance of return" (the "E-V rule") he writes, "To use the E-V rule in the selection of securities, we must have procedures for finding reasonable µi [the expected return] and sij [the covariance]."  The reader should pay close attention to the terms "future performances," "reasonable" and "expected," as I believe many critics missed these caveats.

Lest a reader not get the point, Markowitz continues, "These procedures, I believe, should combine statistical techniques and the judgment of practical men. My feeling is that the statistical computations should be used to arrive at a tentative set of µi and sij. Judgment should then be used ... on the basis of factors or nuances not taken into account by the formal computations." (Again, emphasis mine.)

He is quite clear in rejecting the approach of using historical projections. "One suggestion as to tentative µi, sij is to use observed µi, sij for some period of the past," he writes. "I believe that better methods, which take into account more information, can be found."

Financial planners who dipped into my own book Wealth Management over a decade ago would have also read my warning: "Since the optimization process [for asset allocation] is purely quantitative, we then apply a qualitative overlay based on investment judgment to arrive at our recommended allocations. Extrapolation of historical returns assumes that return series are stable. Unfortunately, they are not stable. ... Research to date suggests that simple historical projections do not seem to offer an acceptable solution for return projection. I do not believe that a wealth manager should use them as a basis for optimization input."