Abstract:We introduce a new class of multivariate heavy-tailed distributions that are convolutions of heterogeneous multivariate t-distributions. Unlike commonly used heavy-tailed distributions, the multivariate convolution-t distributions embody cluster structures with flexible nonlinear dependencies and heterogeneous marginal distributions. Importantly, convolution-t distributions have simple density functions that facilitate estimation and likelihood-based inference. The characteristic features of convolution-t distributions are found to be important in an empirical analysis of realized volatility measures and help identify their underlying factor structure.

What problem does this paper attempt to address?

This paper mainly discusses a new type of multivariate heavy-tailed distribution called convolution-t distribution, which is composed of the convolution of heterogeneous multivariate t distributions. Unlike commonly used heavy-tailed distributions, the convolution-t distribution can reflect clustering structures with flexible non-linear dependencies and heterogeneous marginal distributions. Importantly, the convolution-t distribution has a simple density function that facilitates estimation and likelihood-based inference. The paper empirically analyzes the implementation of volatility measurement to demonstrate the importance of the distribution's characteristics in identifying its underlying factor structure. The main contributions of the paper include: 1. Introducing a class of convolution-t distributions and obtaining their marginal densities and cumulative distribution functions through characteristic functions. 2. Addressing identification issues and showing that convolution-t distributions can help reveal latent factor structures that cannot be identified solely based on covariance structure. 3. Deriving scores, Hessian matrices, and information matrices from the log-likelihood function and proving the consistency and asymptotic normality of the maximum likelihood estimator. 4. Extending Patil's (1965) moment approximation method for arbitrary numbers of t distribution convolutions. 5. Applying the convolution-t framework to model ten financial volatility sequences, demonstrating significant empirical advantages and revealing clustering structures of identified common market factors and sector classification based on non-linear dependencies. The paper also discusses the differences between convolution-t distribution and other related distributions, such as multivariate t distribution, elliptical distribution, and random-scale mixtures of Gaussian distribution. It particularly emphasizes the role of clustering structure and novel rotation-scale matrix of convolution-t distribution in defining specific non-linear dependencies, which can be used to identify clusters and latent factor structures in the data.

Convolution-t Distributions

Characterization of Multivariate Heavy-Tailed Distribution Families Via Copula

A Probability Inequality for Convolutions of MTP2-Distribution Functions

A new approach in two-dimensional heavy-tailed distributions

On the Conditional Distribution of the Multivariate $t$ Distribution

A new multivariate t distribution with variant tail weights and its application in robust regression analysis

Cluster GARCH

Multivariate distributions from mixtures of max-infinitely divisible distributions

Convex combinations of random variables stochastically dominate the parent for a new class of heavy-tailed distributions

On Generalized t-Transformation of Free Convolution

A Wide Class of Heavy-Tailed Distributions and Its Applications

Generalized Convolution Properties Based on the Modified Mittag-Leffler Function

On a transformation between distributions obeying the principle of a single big jump

Convolutions and Mixtures of Gamma, Stable and Mittag-Leffler Distributions

A Realization of Time-Frequency Distributions with Complex-lag Argument for Multicomponent Signal

Heavy-Tailed Distributions and Their Applications

The Weinstein Transform Associated with a Family of Generalized Distributions

Probabilistic PCA for T Distributions.

About the convolution of distributions on groupoids

A new distribution for robust least squares

Closure properties and heavy tails: random vectors in the presence of dependence