Networks as a mehod for portfolio selection=Τα δίκτυα ως μέθοδος επιλογής χαρτοφυλακίου

Περίληψη

This thesis is an extension to the existing literature on applications of network theory for portfolio selection. Until now, networks for portfolio selection are identified via Pearson correlation -a linear correlation measure- of stock returns while the stocks from which the portfolios are constructed are chosen based on common centrality measures. In the current thesis not only Pearson but also non-linear measures are applied which lead to both symmetric and asymmetric adjacency matrices. In total, 6 different types of (overlapping) networks are identified: Net Ι: Pearson correlation with replacement of negative values by zero, ii) Net ΙΙ: Absolute values of Pearson correlation, iii) Net ΙΙΙ: Normalized Mutual Information, iv) Net IV: Directed Normalized Mutual Information, v): Net V: Information Interdependence, vi): Net VI: Information Dependence (Asymmetric). For each node of those overlapping networks, the following are computed: strength, closeness centrality, betweenness centrality, eigenvector centrality, eccentricity; portfolios are constructed from the stocks with the highest and the lowest score on those measures. Apart from the application of non-linear measures and the identification of directed networks, a separate study for the 2008 financial crisis era is performed in order to also come up with the best performing networks during periods of extreme volatility. Portfolios are evaluated based on returns, total risk, systemic risk, adjusted to total risk return and adjusted to systemic risk return. Regarding the most crucial conclusions to be made, networks identified through Pearson Correlation achieve higher returns. However, non-linear measures are superior when it comes to building portfolios of less risk (both total and systematic). Concerning adjusted to risk return, top performance is shared between linear measures and the best performing non-linear ones. However, during the crisis the superiority of non-linear measures is evident, with the importance of directed networks during high volatility eras becoming lucid as well.

KEY WORDS Networks, Portfolio Selection