Mean–variance approximations to expected utility☆
Highlights
► Critically evaluates attacks on assumptions of modern portfolio theory. ► Argues attacks confuse sufficient versus necessary conditions for applying theory. ► Reviews a half-century of research on mean–variance approximations to expected utility.
Section snippets
Why not just maximize expected utility?
If one believes that action should be in accord with the max EU rule, why seek to approximately maximize it via a mean–variance analysis? Why not just maximize expected utility? In addressing this question I distinguish three types of expected utility maximization:
- •
explicit,
- •
MV-approximate,
- •
implicit.
I refer to it as “explicit” EU maximization when a utility function is given and analytical or numerical methods are used to find the portfolio that maximizes the expected value of this function. In
Utility of return vs. utility of wealth
Eqs. (3), (4) assume that an investor seeks to maximize the expected value of Ln(1 + R), where R is the return on the portfolio during the forthcoming period.
Loistl’s erroneous analysis
In a paper titled “The Erroneous Approximation of Expected Utility by Means of a Taylor’s Series Expansion: Analytical and Computation Results,” Loistl (1976) concludes that “the mean–variance approximation is not a good approximation of the expected value of utility at all.” [p. 909]. The reason Loistl reaches such a negative conclusion is as follows: note that in Table 1 a 6% return is represented as R = 0.06, not as R = 6.0. In particular, in Eq. (9), if starting wealth is Wt = $1,000,000 and the
Levy and Markowitz (1979)
The Levy–Markowitz study had two principal objectives:
- (1)
to see how good mean–variance approximations are for various utility functions and portfolio return distributions; and
- (2)
to test an alternate way of estimating expected utility from a distribution’s mean and variance.
The Levy–Markowitz “alternate way” was to fit a quadratic approximation to U at three values of R:They tried their approach forOf these, k = 0.01 did best in almost every case. This is
Highly risk-averse investors
The Levy–Markowitz results for the exponential with b = 10 differ markedly from those of the other utility functions reported in Table 2. In this section we explore the reason for this.
For E = 0.1 and σ = 0.15, Table 3 compares the exponential utility function with the quadratic approximation, QE in Eq. (2). The utility function is rescaled as followsWith this scaling, the difference between U (0.5) and U (−0.3) is of the same order of magnitude as that for Ln(1 + R) in Table 1, namely,
Highly risk-averse investors and a risk-free asset.
Simaan (1993) explores the efficacy of MV-approximate maximization for investors with the exponential utility function when a risk-free asset is available versus when such a risk-free asset is not available. He finds that, for investors with exponential utility functions with large values of b, MV-approximate EU maximization is highly efficacious when a risk-free asset is available, and much less so when it is not.
In deriving these results, Simaan assumes that security returns follow a factor
Portfolios of call options
It is often said that mean–variance analysis is not valid if applied to portfolios which contain securities with asymmetric distributions, or whose returns vary in a nonlinear manner as a function of some security or index. Hlawitschka (1994) examines the efficacy of mean–variance and higher order approximations to expected utility for portfolios of calls, and concludes that the afore mentioned view is wrong. For his individual stock return series, Hlawitschka uses the monthly returns for 10
Ederington’s quadratic and Gaussian approximations to expected utility.
Ederington (1995) principal results are summarized in our Table 7, extracted from Ederington’s Table 5. Ederington refers to the series represented in the table as “semi-ex ante simulated returns.” He argues that historical return series such as those used by Levy and Markowitz are an “ex post” sample drawn from some unknown population. Other possible samples from this same population might have randomly drawn more extreme returns than the one actual historical series, and therefore would have
Mean–variance approximations to the geometric mean
Markowitz (2012a) presents historical comparisons between the geometric mean G and six mean–variance approximations to it, for two data sets: one consisting of the historical returns on asset classes widely used in asset allocation decisions, the other containing the real returns during the 20th Century of the equity markets of sixteen countries. I will refer to the first of these as the AC (Asset Class) database, and the second as the DMS (Dimson et al., 2002) database. Both were obtained
Other measures of risk
As described in the preceding section, Markowitz (2012a) compared six different MV methods for approximation ELn(1 + R). Among these the Markowitz (1959) approximation in Eq. (6) here did as well as any. Markowitz (forthcoming) asks could another of the commonly proposed risk-measures have done better. The measures compared in Markowitz (forthcoming) are:
Variance (V),
Mean Absolute Deviation (MAD),
Semivariance (SV),
Value at Risk (VaR),
Conditional Value at Risk (CVaR).
From Eq. (6) we see that
Other pioneers
In previous sections I have tried to review what I consider to be the most critical steps towards our present understanding of MV approximations of EU, these include: the Levy, Markowitz observation that MV approximations to EU are quite good for many U(R)—except those with extremely high risk-aversion, i.e., implausibly low VBC; Simaan’s observation that such investors will nevertheless be well served if the MV analysis includes a risk-free asset; Hlawitschka’s observation that MV
Conclusion
It is now over a half-century since Markowitz (1959) first defended MV analysis as a practical way to approximately maximize EU. In light of repeated confirmation since then of the efficacy of MV approximations to EU, the persistence of the Great Confusion—that MV analysis is applicable in practice only when return distributions are Gaussian or utility functions quadratic—is as if geography textbooks of 1550 still described the Earth as flat.
References (33)
A characterization of the distributions that imply mean–variance utility functions
Journal of Economic Theory
(1983)- et al.
Portfolio selection in a lognormal market when the investor has a power utility function: computational results
- Dimson, E., Marsh P., Staunton M., 2002. Triumph of the Optimists. Princeton Univercity...
- Ederington, L.H., 1995. Mean–variance as an approximation to expected utility maximization: semi ex-ante results. In:...
Normality, solvency, and portfolio choice
Journal of Financial and Quantitative Analysis
(1986)Capital growth and the mean–variance approach to portfolio selection
Journal of Financial and Quantitative Analysis
(1971)The empirical nature of Taylor-series approximations to expected utility
The American Economic Review
(1994)- et al.
Geometric mean approximations
Journal of Financial and Quantitative Analysis
(1983) - et al.
Mean variance versus direct utility maximization
Journal of Finance
(1984) Criteria for Choice Among Risky Ventures
Journal of Political Economy
(1959)
Criteria for Portfolio Building
The Journal of Finance
Approximating expected utility by a function of mean and variance
American Economic Review
The erroneous approximation of expected utility by means of a Taylor’s series expansion: analytic and computational results
The American Economic Review
Portfolio selection
The Journal of Finance
Portfolio Selection: Efficient Diversification of Investments
Foundations of portfolio theory
Journal of Finance
Cited by (220)
Does Islamic investing modify portfolio performance? Time-varying optimization strategies for conventional and Shariah energy-ESG-utilities portfolio
2024, Quarterly Review of Economics and FinanceFirst passage times in portfolio optimization: A novel nonparametric approach
2024, European Journal of Operational ResearchFifty years of portfolio optimization
2024, European Journal of Operational ResearchA general framework for portfolio construction based on generative models of asset returns
2023, Journal of Finance and Data ScienceEstimating ensemble weights for bagging regressors based on the mean–variance portfolio framework
2023, Expert Systems with Applications
- ☆
This paper is based on Chapter 2 of a book which the author is writing under the sponsorship of 1st Global of Dallas, TX. The author is delighted to thank 1st Global in general and, more specifically, its CEO Stephen A (Tony) Batman, its President David Knoch, and Kenneth Blay, my principal day-to-day contact at 1st Global.