Classifying Hedge Funds with Kohonen Maps: A First Attempt

Abstract

The purpose of this paper is to present an empirical study of a set of hedge funds on recent periods. Alternative investments are now widely used by institutional investors and numerous studies highlight the main features of such investments. As they are in general poorly correlated with the main world indexes, traditional asset pricing models yield poor adjustments, partially because of potential non-linearities in pay-off functions. Some funds, however, exhibit high reward to variability ratios and can be advantageously incorporated in a portfolio in a diversification perspective. After describing the dataset, we classify the funds employing the Kohonen algorithm. We then cross the classification with the one based on the style of strategies involved, wondering if such categories are enough homogenous to be relevant. The map of funds allows to characterize families of funds - whose conditional densities are different one to another -and to define a representative fund for each class. The structure of the network of funds is then described. In particular, we measure inter-class similarities and visualize them both on the network of funds and via a map of one-to-one distances between representative funds. Finally, we underline some of characteristics of classified fund families that may interest investors such as performance measurements.

Keywords: Kohonen maps, Classfication, Multidimensional Data Analysis, General Non-linear Models, Hedge Funds, Fund-Picking, Performance Measurements

JEL Classification: G10, G12, G13