ABSTRACT
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
Appendix 15.A: Properties of the Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
T HE CONVENTIONAL MEAN-VARIANCE APPROACH, which constitutes the primary basis forportfolio selection, assumes that asset returns follow normal distributions and/or that the investor has a quadratic utility function. Despite the long and widespread use of
the mean-variance method in portfolio management, its fundamental assumptions often
do not hold in practice. The returns of many financial securities exhibit skewed and
leptokurtic distributions. Derivatives, or securities with embedded options, have, by
construction, highly skewed return distributions. Many other investments are exposed to
multiple risk factors whose joint effect on portfolio returns often cannot be modelled by a
normal distribution.