Application of Threshold-based Filtered Networks in Stock Portfolio Selection and Performance Evaluation

Network analysis is one of the methods of attention of analysts to analyze complex relationships in data in an intuitive way. One of the applications of network analysis is illustrating the relationships between different classes of assets. Identifying stock market dynamics is essential for actors, investors, and financial policymakers. The stock market is considered a complex system that shows its complex dynamics. The complexity of the stock market can have several reasons that the interdependence of stocks can be one of the most prominent of these factors. One of the most important concerns of people in the capital market is finding a way to present and analyze stock data of different companies. There are different companies in the stock market and portfolio managers and investors, in choosing the right stock portfolio, need to consider the best way to form a stock portfolio. This article discusses the formation of diverse and non-diverse portfolios through network theory. To conduct this research, the adjusted final price of 138 listed companies for the period 2017-01-01 to 2021-07-06, equivalent to 1648 trading days, has been used. To describe the effect between stocks, the Adjacency Matrix is used and using the optimal threshold, diverse and non-diverse portfolios are obtained. We implement the results of selected stocks for the portfolio using the Hierarchical Risk Parity (HRP) approach based on clustering methods and the results with three methods of Minimum Variance (MVP), Uniform Distribution (UNIF), and Risk Parity (RP) for both in-sample and out-of-sample periods are compared for both diverse and non-diversified portfolios. Finally, the results have been compared using the four criteria of Sortino, Sharpe, Maximum DD, and Calmar. The results show the superiority of the non-diversified portfolio approach in market downturns and the superiority of the diversified portfolio approach in other periods.