06.06.2013 Views

STOCHASTIC

STOCHASTIC

STOCHASTIC

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

proportions to maximize expected utility. A modification of the Frank-Wolfe<br />

method is developed in Exercise ME-25 to solve the problem. The objective<br />

has the required concavity and differentiability properties to guarantee<br />

convergence, and the algorithm should be quite efficient because the directionfinding<br />

problem may be solved in a trivial way in each iteration.<br />

It is important to know when investors will diversify their holdings. As is<br />

shown in Exercise CR-10, it is sometimes optimal in mean-variance models to<br />

invest entirely in one asset even if all assets have equal mean returns. Samuelson<br />

considers quite general diversification problems in his second paper. He<br />

utilizes an expected utility framework and assumes that the utility function is<br />

strictly increasing, strictly concave and differentiable. The random investments<br />

are assumed to possess equal means and positive but finite variances.<br />

A natural question to pose is when is it desirable to invest equally in all<br />

securities? Samuelson shows that such an allocation is optimal if the investments<br />

have identical independent distributions or, more generally, if they have<br />

nontrivial symmetric distribution functions. He also shows that equal diversification<br />

will generally not be optimal unless some type of symmetry is present.<br />

The finiteness assumption on the variances is crucial for equal allocation to<br />

be the unique optimal solution. Indeed, as is shown in Exercise CR-8, all<br />

possible allocations between independent identically distributed Cauchy<br />

investments are equally good when expected utility is infinite. In fact for<br />

symmetric distributions, an equal allocation policy is optimal but generally<br />

not unique regardless of the finiteness of the means and variances.<br />

Perhaps it is more basic to know when a particular investment will form a<br />

part of the optimal portfolio. Utilizing the same framework Samuelson shows<br />

that an investment whose return is distributed independently of the returns of<br />

the remaining investments will always be purchased if its mean is not exceeded<br />

by any of the remaining investments. Indeed if all investments have a common<br />

mean and have independent distributions with finite positive variances, they<br />

will enter positively in the optimal portfolio. When a mean-variance or meandispersion<br />

analysis is optimal, the investment proportions are inversely related<br />

to the variances or dispersions, respectively (Exercise CR-7). However, unless<br />

the strict independence assumption holds, it is not true that every investment<br />

in a group with equal means must enter the optimal portfolio.<br />

Although it is natural to suppose that negative interdependence improves<br />

the case for diversification, the appropriate measure of interdependence is not<br />

clear. For example, if all investments have a common mean and have negative<br />

correlations, then all enter into the optimal portfolio if the utility function is<br />

quadratic or if the investments have a normal distribution. However, negative<br />

correlation is neither necessary nor sufficient for complete or even any diversification<br />

for general strictly concave functions (Brumelle, 1974) (Exercise<br />

ME-27). In the two-asset case a concept of negative dependence that leads to<br />

210 PART III STATIC PORTFOLIO SELECTION MODELS

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!