[2H6-OS-2c] OS-2

5:30 PM - 5:45 PM JST(8:30 AM - 8:45 AM UTC)

[2H6-OS-2c-01] Extraction of Orthogonal Factors from Financial Data Using Principal Component Analysis and Gaussian Graphical Models

〇Koshi Watanabe¹, Ryota Ozaki², Kentaro Imajo², Masanori Hirano² (1. Hokkaido University, 2. Preferred Networks, Inc.)

Keywords:

Finance

Decomposing financial asset prices into market factors and residual factors is essential for both market analysis and portfolio selection. Residual factors represent asset-specific components that can be extracted using multivariate analysis techniques. However, unique characteristics of financial time series, such as rank deficiency, pose challenges to the stable extraction of residual returns. In this paper, we propose a hierarchical residual return extraction method that combines principal component analysis (PCA) with a Gaussian graphical model (GGM). Our approach leverages an MTP2-constrained GGM specifically tailored for financial time series, enabling more effective decomposition compared with conventional methods. Theoretical analysis shows that the proposed method reduces correlations among residual returns after removing principal components via PCA. Experimental results using time-series data from the S\&P 500 and TOPIX 500 constituents demonstrate that our method extracts residual returns with superior orthogonality.