The main contribution of the paper is to employ the financial market network as a useful tool to improve the portfolio selection process, where nodes indicate securities and edges capture the dependence structure of the system. Three different methods are proposed in order to extract the dependence structure between assets in a network context. Starting from this modified structure, we formulate and then we solve the asset allocation problem. We find that the portfolios obtained through a network-based approach are composed mainly of peripheral assets, which are poorly connected with the others. These portfolios, in the majority of cases, are characterized by an higher trade-off between performance and risk with respect to the traditional Global Minimum Variance (GMV) portfolio. Additionally, this methodology benefits of a graphical visualization of the selected portfolio directly over the graphic layout of the network, which helps in improving our understanding of the optimal strategy.
IntroductionModern portfolio theory, which originated with Harry Markowitz's seminal paper in 1952 (see [23]) has stood the test of time and continues to be the intellectual foundation for real-world portfolio management. In this static framework, investors optimally allocate their wealth across a set of assets considering only the first and second moment of the returns' distribution. Despite the profound changes derived from this publication, the out-of-sample performance of Markowitz's prescriptions is often not as promising as expected. The poor performance of Markowitz's rule stems from the large estimation errors on the vector of expected returns (see [26]) and on the covariance matrix (see [17]) leading to the well-documented error-maximizing property discussed in [27]. The magnitude of this problem is evident when we acknowledge the modest improvements achieved by those models specifically designed to tackle the estimation risk (see [10]). Moreover, the evidence indicates that the simple yet effective equally-weighted portfolio rule has not been consistently out-performed by more sophisticated alternatives, as reported in [2,10]. The literature on portfolio selection has been extended on several directions. On one hand, some extensions modify the optimal problem, considering higher moments of returns' distribution (see [24] and the reference therein), exploiting alternative risk measures (see [6,19]) and as well as utility functions (see [37,13]), or proposing dynamic approaches (see among others [39,4]). On the other hand, there is now a vast literature on how to deal with the problem, that focuses on improved estimation procedures, considering that the optimal allocation is very sensitive to the estimation of moments and co-moments 1 .1 This sensitivity has generally been attributed to the tendency of the optimization to magnify the effects of estimation error. Michaud in [27] referred to "portfolio optimization "as "error maximization ". Efforts to improve parameters estimation procedure include among others the arXiv:18...