HYPOTHESIS TESTING ON LINEAR STRUCTURES OF HIGH DIMENSIONAL COVARIANCE MATRIX.

Area of Science:

• Statistics
• High-Dimensional Data Analysis
• Covariance Structure Analysis

Background:

• Traditional statistical methods struggle with high-dimensional data.
• Testing covariance structures is crucial in many statistical applications.
• Existing methods for high-dimensional covariance testing have limitations.

Purpose of the Study:

• To develop a unified framework for testing linear covariance structures in high dimensions.
• To propose two novel tests based on entropy and quadratic loss.
• To analyze the asymptotic properties and finite sample performance of these tests.

Main Methods:

• Construction of a consistent estimator for linear covariance structure parameters.
• Development of two significance tests utilizing entropy and quadratic loss.
• Application of high-dimensional random matrix theory to derive asymptotic distributions.
• Monte Carlo simulations to evaluate finite sample performance.

Main Results:

• The proposed tests effectively control Type I error rates.
• The quadratic loss-based test is asymptotically unbiased.
• Both tests show good approximation of null distributions by their limiting distributions.
• The quadratic loss test exhibits better power than the entropy loss test.

Conclusions:

• The developed framework provides a robust method for testing high-dimensional covariance structures.
• The proposed tests offer improved performance over existing methods.
• The quadratic loss-based test is recommended for its superior power and unbiasedness.