Large Covariance Estimation by Thresholding Principal Orthogonal Complements

Area of Science:

• Statistics
• Econometrics
• Machine Learning

Background:

• High-dimensional covariance estimation is crucial in finance and statistics.
• Existing methods struggle with conditional sparsity and fast-diverging eigenvalues.
• Approximate factor models are often used but require careful handling of residual correlations.

Purpose of the Study:

• To develop a novel method for estimating high-dimensional covariance matrices with conditional sparsity.
• To introduce the Principal Orthogonal complEment Thresholding (POET) method.
• To analyze the theoretical properties and practical performance of the proposed estimator.

Main Methods:

• The study proposes the Principal Orthogonal complEment Thresholding (POET) method.
• POET assumes a sparse error covariance matrix within an approximate factor model.
• Mathematical insights are provided, linking factor analysis to principal component analysis.

Main Results:

• The POET estimator encompasses several existing methods as special cases.
• Rates of convergence for sparse residual and conditional sparse covariance matrices are derived.
• The impact of estimating unknown factors diminishes with increasing dimensionality.

Conclusions:

• The POET method provides a robust framework for high-dimensional covariance estimation with conditional sparsity.
• Theoretical results demonstrate vanishing impact of factor estimation and derived convergence rates.
• The method shows practical utility in applications like portfolio allocation.