Impact of Covariance Matrix Estimator Choice on Yield Curve Decomposition With PCA

Abstract

This investigation evaluates the impact of different covariance matrix estimation methods when used as an input for principle component analysis in the context of yield curve decomposition, focusing on the economic interpretations of Level, Slope, and Curvature components. It was found that the constant correlation shrinkage estimator consistently outperforms others, including the sample covariance matrix. The investigation also reveals that short time periods yield less reliable results, with the issue attributed to the length of time sampled rather than sample sizes.

Keywords: PCA, principle component analysis, yield curve, covariance, constant correlation, yield curve decomposition