ABSTRACT
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
T HE PORTFOLIO SELECTION PROBLEM is one of the basic problems within the researcharea of computational finance. It has been studied intensively throughout the last 50 years, producing several relevant contributions described in the specialized
literature. Portfolio selection originates from the seminal paper of Markowitz (1952),
who introduced and motivated the mean-variance investment framework. This conven-
tional approach to portfolio selection, which has received increasing attention, consists of
two separate steps. The first step concerns distributional assumptions about the behaviour
of stock prices, while the second step is related to the selection of the optimal portfolio
depending on some objective function and/or utility function defined according to the
investor’s goal. This conceptual model has proved in the past to be useful in spite of the
many drawbacks that have been pointed out by finance practitioners, private investors and
researchers. Indeed, the first step, related to distributional assumptions concerning the
behaviour of stock prices, encounters many difficulties because the future evolution of
stock prices is notoriously difficult to predict, while the selection of a distribution class
inevitably brings a measure of arbitrariness. These problems become even more evident
and dramatic in the case where there are reasons to believe that the process that governs
stock price behaviour changes over time.