High-dimensional covariance matrices in elliptical distributions with application to spherical test

Jiang Hu, Weiming Li, Zhi Liu, Zhou Wang

Abstract

This paper discusses fluctuations of linear spectral statistics of high-dimensional sample covariance matrices when the underlying population follows an elliptical distribution. Such population often possesses high order correlations among their coordinates, which have great impact on the asymptotic behaviors of linear spectral statistics. Taking such kind of dependency into consideration, we establish a new central limit theorem for the linear spectral statistics in this paper for a class of elliptical populations. This general theoretical result has wide applications and, as an example, it is then applied to test the sphericity of elliptical populations.

Introduction

In high-dimensional statistics, the study of the "spectrum" (eigenvalues) of sample covariance matrices is fundamental to understanding data structure. For decades, the "linear transform" model has been the standard, assuming data points are linear combinations of independent random variables. However, real-world data—particularly in finance and biology—often exhibits heavy tails and complex, non-linear correlations that classical models cannot capture.

Elliptical distributions, such as the multivariate $t$-distribution, provide a more robust framework for these scenarios. However, they break the independence assumptions required by classical Random Matrix Theory (RMT). This paper bridges this gap by establishing how eigenvalues behave when data originates from an elliptical population, specifically focusing on the fluctuations of linear spectral statistics (LSS).

Methods

The authors represent the data using a stochastic form: $\mathbf{x} = \xi A \mathbf{u}$, where $\xi$ is a random scalar (the radius), $A$ is a transformation matrix, and $\mathbf{u}$ is a random direction on a unit sphere. The core of the work is a new Central Limit Theorem (CLT) for LSS.

To achieve this, the authors:

  1. Define a specific class of elliptical distributions where the radius $\xi$ satisfies a fourth-moment condition characterized by a parameter $\tau$.
  2. Use the Stieltjes transform to translate the distribution of eigenvalues into a complex-valued function.
  3. Employ Martingale Decomposition to handle the non-linear dependencies introduced by the radius and direction components.
Table 1: Examples of elliptical distributions and the validity of Condition (2.1).

As shown in Table 1, the validity of the proposed CLT depends on Condition (2.1). While distributions like Normal, Double Exponential, and Exponential Power satisfy this condition with specific $\tau$ values, others like the Student-$t$ (with $v > 4$) or Normal scale mixtures do not necessarily hold, highlighting the boundaries of the proposed theory.

Key Insights and Results

The researchers discovered a "tug-of-war" between the radius $\xi$ and the direction $\mathbf{u}$. In standard normal distributions, these effects cancel out, but in other elliptical distributions, they do not. The parameter $\tau$ effectively captures this impact, allowing for a simplified transition from standard RMT to "elliptical" RMT.

Sphericity Testing and Financial Application

The authors developed a "John-type" test ($T_m$) using a spatial-sign transformation, which normalizes each data vector to unit length. This effectively "strips away" the influence of the radius $\xi$, making the test universal across different elliptical distributions.

Applying this to S&P 500 stock returns from 2013, the authors found:

  • Intra-sector correlation: Stocks within the same sector are strongly correlated and fail the sphericity test ($p$-values $< 10^{-9}$).
  • Inter-sector structure: Stocks chosen across different sectors are closer to being "spherical" (uncorrelated), though they still exhibit some high-dimensional structure.

Limitations and Future Work

The current theory requires the radius to have a finite fourth moment ($E(\xi^4)$), which excludes extremely heavy-tailed distributions like the multivariate Student-$t$ with $v \leq 4$. Future research may explore "spiked" covariance models where a few eigenvalues are significantly larger than others, and extend these findings to complex-valued data common in signal processing.

Summary updated

Abstract

This paper discusses fluctuations of linear spectral statistics of high-dimensional sample covariance matrices when the underlying population follows an elliptical distribution. Such population often possesses high order correlations among their coordinates, which have great impact on the asymptotic behaviors of linear spectral statistics. Taking such kind of dependency into consideration, we establish a new central limit theorem for the linear spectral statistics in this paper for a class of elliptical populations. This general theoretical result has wide applications and, as an example, it is then applied to test the sphericity of elliptical populations.
BibTeX
@article{hu2018high,
  author = {Jiang Hu and Weiming Li and Zhi Liu and Zhou Wang},
  title = {{High-dimensional covariance matrices in elliptical distributions with application to spherical test}},
  journal = {The Annals of Statistics},
  year = {2018},
  doi = {10.1214/18-aos1699},
}
Figure