# How do you remove expected returns from asset allocation strategies?

The classic mean-variance optimization problem tries to minimize variance of a portfolio for a given expected return:

$$\underset{w}{\arg \min} \quad w^T \Sigma w \quad \text{s.t} \quad \mu^Tw \geq \bar{\mu}$$

However the expected returns $\mu$ are can be computed in many different ways, but many use very subjective estimations. The problem is that the result of the optimization is very sensitive to these expected returns.

What asset allocation strategies exist where the expected returns are removed from the framework or where the sensitivity of the result (to changes in expected returns) is limited?

For example, the "min-variance" optimization removes the expected returns by return only the portfolio with minimum variance as follows:

$$\underset{w}{\arg \min} \quad w^T \Sigma w$$

-
I'm not sure if there are any more besides those we've mentioned in previous questions, such as Choice of prior as a shrinkage target in portfolio construction? and What methods do you use to improve expected return estimates when constructing a portfolio in a mean-variance framework? Still, I look forward to reading others' thoughts. –  Tal Fishman Jan 3 '12 at 18:41
Agreed with comment. Bear in mind that are implied expected returns tilts by so-called "Risk only" portfolio utility functions. See the Wai Lee paper cited in "Choice of prior as shrinkage target..." to see more. –  Quant Guy Jan 5 '12 at 4:30

There is a vast, growing body of literature on risk parity, much of which is predicated on this idea (i.e. of optimizing a portfolio allocation without including expected return). As an example, The Journal of Investing put out an entire issue dedicated to the subject last year: see "Latest Approaches to Risk Parity and Diversification".

Risk Parity (RP) is a relatively simple idea. Very loosely defined, RP attempts to create a portfolio in which the various asset classes contribute equally to the overall risk of the portfolio.

This is typically viewed against a 60/40 portfolio, and the argument is made that a 60/40 allocation actually allocates 90% of the risk to stocks. There is a some amount of disagreement over whether risk parity is the best approach, as will be clear from reading a selection of these papers. Some argue that there is no theoretical basis for why an RP portfolio should outperform an MVO-portfolio, and that it is inherently inefficient; "Leverage Aversion and Risk Parity" (2011) lays out a positive case based on leverage aversion. There are also many different implementations of this idea (see "Risk Parity for the Masses" for an example).

Some blog posts which have further references:

A related idea is the Most Diversified Portfolio (see, for example, "Properties of the Most Diversified Portfolio", 2011). This optimizes the portfolio based on the diversification ratio, which is simply the sum of the individual asset volatilities over the portfolio volatility.

-

Do you know the Black-Litterman Model?

In principle Modern Portfolio Theory (the mean-variance approach of Markowitz) offers a solution to this problem once the expected returns and covariances of the assets are known. While Modern Portfolio Theory is an important theoretical advance, its application has universally encountered a problem: although the covariances of a few assets can be adequately estimated, it is difficult to come up with reasonable estimates of expected returns. In other words, composing a portfolio based only upon statistical measures of risk and returns yields simplistic results; these are known as unconstrained optimizations. Black–Litterman overcame this problem by not requiring the user to input estimates of expected return; instead it assumes that the initial expected returns are whatever is required so that the equilibrium asset allocation is equal to what we observe in the markets. The user is only required to state how his assumptions about expected returns differ from the market's and to state his degree of confidence in the alternative assumptions. From this, the Black–Litterman method computes the desired (mean-variance efficient) asset allocation.

More references on this Black-Litterman model can be found easily on Internet

-