Abstract:In this paper we study an ensemble of random matrices called Elliptic Volatility Model, which arises in finance as models of stock returns. This model consists of a product of independent matrices $X = \Sigma Z $ where $Z$ is a $T$ by $S$ matrix of i.i.d. light-tailed variables with mean 0 and variance 1 and $\Sigma$ is a diagonal matrix. In this paper, we take the randomness of $\Sigma$ to be i.i.d. heavy tailed. We obtain an explicit formula for the empirical spectral distribution of $X^*X$ in the particular case when the elements of $\Sigma$ are distributed as Student's t with parameter 3. We furthermore obtain the distribution of the largest eigenvalue in more general case, and we compare our results to financial data.

What problem does this paper attempt to address?

### What problems does this paper attempt to solve? This paper aims to study a class of random matrix ensembles called Elliptic Volatility Model (EVM), especially for modeling stock returns in the financial field. Specifically, the paper explores the following problems: 1. **Eigenvalue statistics of the Elliptic Volatility Model**: - It has studied the eigenvalue distribution of EVM, especially when the volatility parameter Σ has a heavy - tailed distribution. - In specific cases (for example, when the elements of Σ follow Student's t - distribution with a parameter of 3), an explicit formula for the empirical spectral distribution of the sample covariance matrix $X^*X$ has been derived. 2. **Distribution of the largest eigenvalue**: - It has explored the distribution of the largest eigenvalue of EVM in more general cases. - It has compared the differences between theoretical results and actual financial data to verify the validity and applicability of the model. ### Main contributions - **Eigenvalue density**: The paper reveals that when the volatility parameter has a heavy - tailed distribution, the eigenvalue density also exhibits heavy - tailed characteristics, which is different from the classical Marchenko - Pastur distribution. This phenomenon may imply a new universality phenomenon, which has been observed in multiple application fields such as calcium imaging, machine learning, and finance. - **Asymptotic behavior of the largest eigenvalue**: It has studied the asymptotic behavior of the largest eigenvalue and proved that under appropriate scaling, the largest eigenvalue can be approximated by the square of the maximum value of σ. This result is of great significance for understanding applications such as principal component analysis (PCA) in heavy - tailed data. ### Application background - **Financial mathematics**: In financial mathematics, the volatility process is usually defined as $X_t=\sigma_t Z_t$, where $Z_t$ is an independent standard normal random variable, and $\sigma_t$ is the volatility parameter. By introducing the volatility parameter with a heavy - tailed distribution, this paper better simulates the distribution characteristics of stock returns in the real world, especially their "cubic law" (that is, a power - law distribution with a tail index of about 3). - **Risk optimization**: The existence of heavy - tailed distributions has an important impact on portfolio optimization. The traditional mean - variance framework may no longer be applicable when dealing with heavy - tailed distributions, so new methods are needed to understand and optimize the risk of portfolios. ### Conclusion By in - depth study of the eigenvalue statistics of the Elliptic Volatility Model and the distribution of the largest eigenvalue, this paper provides new theoretical tools and methods for understanding and dealing with heavy - tailed phenomena in financial data. These results are not only helpful for improving the modeling and prediction of financial markets, but also have potentially important significance for other application fields involving heavy - tailed distributions.

Eigenvalue statistics of Elliptic Volatility Model with power-law tailed volatility

Prediction of Volatility Based on Realized-Garch-kernel-type Models: Evidence from China and the U.S.

On asymptotics of eigenvectors of large sample covariance matrix

Power law in Sandwiched Volterra Volatility model

Statistical properties of the volatility of price fluctuations

A stochastic volatility model with flexible extremal dependence structure

Entropy-Based Volatility Analysis of Financial Log-Returns Using Gaussian Mixture Models

Volatility: a hidden Markov process in financial time series

Student-t Stochastic Volatility Model With Composite Likelihood EM-Algorithm

Spectral properties of empirical covariance matrices for data with power-law tails

Modeling volatility using state space models

Extreme eigenvalues of sample covariance matrices under generalized elliptical models with applications

Eigenvalue spectra of finely structured random matrices

Risk measures with non-Gaussian fluctuations

An Empirical Model for Volatility of Returns and Option Pricing

Free Random Levy Variables and Financial Probabilities

A contribution to the systematics of stochastic volatility models

High-Dimensional Volatility Matrix Estimation with Cross-Sectional Dependent and Heavy-Tailed Microstructural Noise

Extreme eigenvalues of Laplacian random matrices with Gaussian entries

Stock market volatility: An approach based on Tsallis entropy