그래프 필터링 및 중심성 측도에 기반한 포트폴리오 구성 전략

초록

This study investigates the impact of graph filtering methods and centrality measures on portfolio stock selection. Two graph filtering methods - Minimum Spanning Tree (MST) and Triangulated Maximally Filtered Graph (TMFG)-are combined with three centrality measures-Degree, Eigenvector, and Subgraph centrality-to identify peripheral assets in financial networks. The analysis is conducted on 29 stocks from the Dow Jones Industrial Average with no constituent changes. Portfolios are constructed by selecting the least central assets at varying portfolio sizes(5,10,15, and 20 assets) and applying equal weight allocation. The empirical results indicate that the combination of TMFG and Eigenvector centrality shows the highest Sharpe ratio of 2.65 with 10 selected assets, and maintains high performance with 5 and 15 assets. Degree centrality exhibits improving performance as portfolio size increases across both filtering methods ; for MST with Degree centrality, the Sharpe ratio rises from 1.24 with 5 assets to 2.55 with 20 assets. These findings clarify the impact of the choice of filtering method and centrality measure on portfolio performance, demonstrating that portfolios composed of peripheral assets achieve higher risk-adjusted returns.

키워드