[Paper Review] Cross-sectional Learning of Extremal Dependence among Financial Assets

Yan Xing, Qi Wu+1 more|arXiv (Cornell University)|May 31, 2019

Financial Risk and Volatility ModelingEconomics, Econometrics and Finance18 references3 citations

TL;DR

This paper proposes a novel transformation-based probabilistic model that separately models tail dependence from correlation in multivariate financial returns, enabling flexible, distinct pairwise tail dependencies and distinct marginal tail heaviness. When combined with GARCH filtering, the model significantly improves multivariate coverage tests, reducing rejections from 52 (Gaussian) and 43 (t) to 32, and reveals asymmetric tail behavior in market and idiosyncratic components.

ABSTRACT

We propose a novel probabilistic model to facilitate the learning of multivariate tail dependence of multiple financial assets. Our method allows one to construct from known random vectors, e.g., standard normal, sophisticated joint heavy-tailed random vectors featuring not only distinct marginal tail heaviness, but also flexible tail dependence structure. The novelty lies in that pairwise tail dependence between any two dimensions is modeled separately from their correlation, and can vary respectively according to its own parameter rather than the correlation parameter, which is an essential advantage over many commonly used methods such as multivariate $t$ or elliptical distribution. It is also intuitive to interpret, easy to track, and simple to sample comparing to the copula approach. We show its flexible tail dependence structure through simulation. Coupled with a GARCH model to eliminate serial dependence of each individual asset return series, we use this novel method to model and forecast multivariate conditional distribution of stock returns, and obtain notable performance improvements in multi-dimensional coverage tests. Besides, our empirical finding about the asymmetry of tails of the idiosyncratic component as well as the market component is interesting and worth to be well studied in the future.

Motivation & Objective

• To address the limitation of existing models in separately modeling tail dependence from correlation in multivariate financial returns.
• To develop a flexible, interpretable, and easy-to-sample model that captures distinct pairwise tail dependencies and marginal tail heaviness.
• To improve multivariate conditional distribution forecasting of asset returns using a GARCH-filtered framework.
• To empirically investigate asymmetries in tail behavior of market-wide and idiosyncratic components.
• To provide a scalable alternative to copula and elliptical models for high-dimensional tail dependence.

Proposed method

• Propose a transformation of known random vectors (e.g., standard normal) to generate multivariate heavy-tailed distributions with flexible tail dependence.
• Design a one-factor model where asset returns depend on a common market factor and idiosyncratic components, each with separate tail parameters.
• Use a lower-triangular structure and a one-factor structure to model pairwise tail dependencies independently of correlation.
• Apply GARCH models to filter serial dependence in individual return series before applying the tail dependence model.
• Develop an algorithm to estimate model parameters using data, with modifications to standard methods for fitting.
• Use multi-dimensional coverage tests to evaluate model performance against competing one-factor Gaussian and t-distribution models.

Experimental results

Research questions

• RQ1 Can tail dependence between financial assets be modeled independently of their linear correlation, enabling more realistic risk assessment?
• RQ2 How does separating tail dependence from correlation improve multivariate distribution forecasting in financial returns?
• RQ3 What are the empirical patterns of asymmetry in tail behavior for market-wide and idiosyncratic components?
• RQ4 Can a transformation-based model achieve better performance than standard multivariate t or Gaussian models in capturing extremal co-movements?
• RQ5 What are the implications of observed asymmetries—such as right-skewed idiosyncratic components and left-skewed market factors—for asset pricing and systemic risk?

Key findings

• The proposed model reduces the number of rejections in multi-dimensional coverage tests from 52 (one-factor Gaussian) and 43 (one-factor t) to 32, indicating significant performance improvement.
• Empirical results show that all 15 individual stocks have right-skewed idiosyncratic components (u_i > v_i), indicating higher sensitivity to negative shocks.
• The market factor exhibits left-skewed tail behavior (u_i^M < v_i^M) for most stocks, suggesting stronger downward tail dependence during market-wide downturns.
• For many stocks, the market factor has u_i^M = 1, indicating no tail sensitivity on the upside, which is consistent with asymmetric market impact.
• The model successfully generates distinct pairwise tail dependencies, independent of correlation, validating its core design advantage over multivariate t and elliptical distributions.
• The empirical asymmetry in tail behavior—right-skewed idiosyncratic components and left-skewed market factors—deserves further theoretical and empirical investigation.