Revisiting Modern Portfolio Theory
Modern portfolio theory is the dominant force in investing. It extends from simple statistical relationships to statements about the pricing of assets in the form of the Capital Asset Pricing Model (CAPM) to methods for building portfolios. For institutions seeking to maximize gains for a given level of risk, mean-variance optimized (MVO) portfolios are the standard. In retail products, the same principles have filtered down into balanced mutual funds, life cycle and target date plans. It is hard to overstate the influence of MPT or its connection to deeply held beliefs about market behavior and prudent ways to invest.
But time after time, it fails to provide any real protection. After each new market crisis, no matter how disappointed we get, we always come back to it. Maybe because it is beautiful, it is everywhere and there is no obvious better choice.
In his book Capital Ideas Evolving (2007), Peter Bernstein talks about reliance on the CAPM as a paradox. He thinks the CAPM has turned into the most fascinating and influential of all the theoretical developments in investing today: “Yet repeated empirical tests of the CAPM, dating all the way back to the 1960s, have failed to demonstrate that the theoretical model works in practice.” In researching the book, Bernstein interviewed Markowitz to get an update on what he was working on. Markowitz told him, “You will be completely surprised if I tell you about my latest research.” Bernstein said, “He is no longer the same Harry Markowitz whose view [of securities] put Bill Sharpe to work on the [CAPM]. Markowitz has lost faith in what he terms the traditional neoclassical ‘equilibrium models.’”6
A lot of people have lost faith. Richard Ennis, in his article “Parsimonious Asset Allocation,” wrote:
- Over the past 25 years, institutional investors have become increasingly reliant on asset allocation models that use a complex set of assumptions about the future. ... As a result, institutional investors of all types experienced losses far greater than the “worst-case” outcomes predicted by their asset allocation models. It is important to realize that, over time, asset-class return correlations are unstable—really unstable. ... What good is a system of risk control that fails when you need it most?7
Psychologically, it is hard to accept that a system that works so well 90% of the time is not going to help the other 10% of the time. Even if you accept that markets crash, that the declines are severe, and risk control fails, there is still the possibility that something was missed in execution or that next time will be different. To make progress, it is helpful to understand why the system breaks down. Otherwise, it is hard to know if and how to work with it. At this point, there is a great deal of research that fills in the details. It is widely known that severe markets events can cause all asset classes to decline at the same time, a form of contagion that eliminates any positive effect of diversification. Looking closer at this behavior, there are two related issues, implicit beta exposure and optimistic correlation matrix construction.
Martin Leibowitz, in his work with institutional investors, identified what he calls implicit beta exposure. He noticed that as endowments and others began to add alternative investments, the portfolios looked dramatically different from each other, but performed about the same. In trying to understand why these portfolios act like each other, and much like a traditional 60% equity/40% bond portfolio, he realized it is because so many assets are linked, either directly or indirectly, to the U.S. equity markets. Because of the linkage, many of the changes were having no real effect on the overall returns or risk measures.
Optimistic correlation matrix construction refers to the use of “average” correlations between asset classes to estimate future losses rather than using the “stress” correlations that existed during prior market crashes. Average correlations may work well across market cycles, but it doesn’t make sense to use these same correlations to estimate the magnitude of losses in market crashes. Continuing to set risk policy using average correlations is something like building a house in an earthquake zone and assuming there will be no earthquakes.
But, regardless of the mechanics of the failure, the ability to accept that failure occurs is important to making a commitment to change. Sometimes, it is best just to see a flat statement. In the monograph from The Research Foundation of the CFA Institute, Investment Management after the Global Financial Crisis, the limitations of MPT are stated bluntly. “MPT does not offer the promise of eliminating losses—even large losses—even under the most favorable assumptions.”8