"Thin-slicing" traditional asset classes into their component parts can also help. For example, long-term U.S. Treasury securities are often lumped into a more general domestic fixed-income category. But they are more negatively correlated with U.S. equities than the fixed-income asset class as a whole. Investors who had separated them in both 2008 and 2011 would have seen better results.
Markowitz also assumes that we will bring with us a set of expectations about the future returns, variances and covariances of the assets classes we use in our portfolios. He doesn't tell us how to develop these, although he says we might use historical data as a starting place. He goes on to say that "better methods can be found." He was right about that.
This is another area where the practices adopted by advisors have caused modern portfolio theory to fall short of its full potential. Most investors have never developed a solid process for reasonably anticipating future returns, risk and correlations. Instead, they have defaulted to the use of long-term historic averages. These averages often mask reality rather than reveal it.
If you have read Jeremy Siegel's book, Stocks for the Long Run, you know that he went back to the early 1800s and determined that the long-term real rate of returns for stocks was about 7%. You also know that, year to year, the return of the stock market is almost never 7% or anything close to that. Even if you look at 10-year returns for stocks, their return is rarely close to 7%.
Long-term historic averages also don't help investors anticipate volatility, like that seen in 2011. Volatility changes over time, just like returns do.
Correlations also change over time, and in a more treacherous way. Asset classes that appear relatively uncorrelated day to day become more highly correlated just when you don't want them to-when markets are in a free fall. Long-term historic averages do not reveal those relationships.
A better way to approach correlations is to look at them during the times that matter most-during periods when the markets are moving strongly in one direction or the other. By disregarding the periods when the markets are just bumping around, you eliminate much of the noise from long-term historical averages.
Once you develop a framework for establishing return, risk and correlation expectations that are truly forward-looking, your portfolio allocations will change over time to reflect your changing expectations about the future. If you have no framework, buying and holding will suffice, though it is not the best approach.
I am not advocating market timing here, which is worse than buying and holding. But there are sound ways to develop long-term expectations about the various asset classes. Most are valuation-based. Some involve macroeconomic overlays. Don't confuse these with market-timing schemes, which should be avoided at all costs. The hall of fame for market timers is an empty room.
Modern portfolio theory has also disappointed practitioners who pick the wrong managers. The standard practice is to develop an asset allocation strategy with indexes for each asset class, and then pick "best-of-breed" managers for each slice of the asset allocation pie.