계층적 동일비중 포트폴리오의 연결법에 따른 성과 비교

초록

The mean-variance optimization is subject to the instability of covariance matrix inversion, leading to extreme asset weights and poor out-of-sample performance. Hierarchical clustering-based approaches that do not require matrix inversion have been proposed as alternatives. This study employs a hierarchical equal-weight portfolio, a variant of Hierarchical Risk Parity that replaces inverse-variance weighting with equal allocation, to isolate the effect of hierarchical cluster structure on portfolio diversification. Using 29 stocks from the Dow Jones Industrial Average, four linkage methods—single, complete, average, and Ward—are compared in terms of dendrogram structure and out-of-sample performance. The results show that single linkage produces a step-like dendrogram with the highest volatility (14.50%) and the lowest Sharpe ratio (1.67). Average linkage achieves the highest Sharpe ratio (2.38) and the smallest maximum drawdown (-4.94%), while Ward's method delivers a Sharpe ratio (2.34) nearly identical to the equal-weight benchmark (2.33) with improved volatility and downside risk. Complete linkage exhibits the lowest volatility (9.56%) but also the lowest return. These findings suggest that hierarchical clustering-based portfolios can mitigate downside risk relative to equal-weight allocation, and that the choice of linkage method is a critical determinant of portfolio outcomes.

키워드