Individual Defense and Joint Defense: A New Defensive Portfolio Selection Method Based on Stock Network Structure

Abstract

We establish a stock network using the correlation of return sequences and model the optimization problem of portfolio weights using two types of topological metrics within the network to construct a method for selecting investment portfolios based on stock networks. At the individual level, centrality features are utilized to describe the importance of a stock within the network, reflecting the systemic risk (beta value) of the individual stock; at the connection level, structural dissimilarity features are used to describe the positional differences between two stocks within the network, reflecting the nature differences of different stocks. Regarding network structure, we propose several approaches for constructing investment portfolios: firstly, selecting stocks located at the periphery of the network as much as possible, which implies minimizing the centrality features of the portfolio. Secondly, maximizing the dispersion of stocks within the network, which implies maximizing the structural dissimilarity of the portfolio. Thirdly, simultaneously optimizing centrality features and structural similarity features. Empirical analysis on the Chinese market from 2002 to 2022 demonstrates that our investment portfolios outperform Markowitz's classic method and the Naive method in terms of risk (value-at-risk) and return (cumulative return rate). Interestingly, our portfolios select pro-cyclical stocks, particularly those from the financial sector, during bull markets and counter-cyclical stocks, particularly from the IT sector, during bear markets, reflecting the defensive capability of the portfolios against market risk.

Keywords: Portfolio Selection, Stock Network, Topological Centrality