[Paper Review] Bayesian inference for dynamic vine copulas in higher dimensions

Alexander Kreuzer, Claudia Czado|arXiv (Cornell University)|Nov 2, 2019

Financial Risk and Volatility Modeling38 references4 citations

TL;DR

This paper proposes a novel Bayesian inference approach for dynamic vine copulas in higher dimensions, enabling time-varying dependence structures through latent AR(1) processes for copula parameters. Unlike prior methods, it avoids point-estimation of lower-tree parameters, preserving uncertainty propagation and achieving superior one-day-ahead forecasting accuracy on 21 exchange rates compared to static and dynamic C/D-vine models.

ABSTRACT

We propose a class of dynamic vine copula models. This is an extension of static vine copulas and a generalization of dynamic C-vine and D-vine copulas studied by Almeida et al (2016) and Goel and Mehra (2019). Within this class, we allow for time-varying dependence by driving the vine copula parameters with latent AR(1) processes. This modeling approach is very flexible but estimation is not straightforward due to the high-dimensional parameter space. We propose a Bayesian estimation approach, which relies on a novel approximation of the posterior distribution. This approximation allows to use Markov Chain Monte Carlo methods, such as elliptical slice sampling, in a sequential way. In contrast to other Bayesian sequential estimation procedures for vine copula models as proposed by Gruber and Czado (2015), there is no need to collapse copula parameters to point estimates before proceeding to the next tree. Thus more information and uncertainty is propagated from lower to higher trees. A simulation study shows satisfactory performance of the Bayesian procedure. This dynamic modeling and inference approach can be applied in various fields, where static vine copulas have already proven to be successful, including environmental sciences, medicine and finance. Here we study the dependence among 21 exchange rates. For comparison we also estimate a static vine copula model and dynamic C-vine and D-vine copula models. This comparison shows superior performance of the proposed dynamic vine copula model with respect to one day ahead forecasting accuracy.

Motivation & Objective

• To develop a flexible class of dynamic vine copula models that generalize static and existing dynamic C/D-vine structures for higher-dimensional dependence modeling.
• To address the challenge of high-dimensional parameter estimation in dynamic vine copulas through a novel Bayesian framework with full uncertainty quantification.
• To enable sequential estimation without collapsing lower-tree parameters to point estimates, thereby preserving uncertainty across tree levels.
• To demonstrate the method's superiority in forecasting accuracy using real financial data on 21 exchange rates.

Proposed method

• The model extends static vine copulas by driving copula parameters with latent AR(1) processes to capture time-varying dependence.
• A novel approximation of the posterior distribution enables efficient use of Markov Chain Monte Carlo methods, such as elliptical slice sampling.
• The Bayesian estimation proceeds sequentially across tree levels without fixing lower-tree parameters to point estimates, thus propagating full uncertainty.
• Parameter sharing is implemented across copula families using the Fisher’s Z-transform of Kendall’s τ to improve estimation efficiency.
• The method supports general regular vine structures, including C-vine and D-vine as special cases, enhancing flexibility.
• The approach is validated through a simulation study and applied to 21 exchange rate returns for forecasting performance comparison.

Experimental results

Research questions

• RQ1 Can a general dynamic vine copula model be constructed that supports time-varying dependence in higher dimensions beyond C- and D-vine structures?
• RQ2 How can Bayesian inference be efficiently applied to high-dimensional dynamic vine copulas with complex latent parameter structures?
• RQ3 Does preserving full uncertainty from lower to higher trees improve forecasting accuracy compared to point-estimate-based sequential estimation?
• RQ4 How does the proposed model perform in forecasting one-day-ahead dependence compared to static and dynamic C/D-vine models in financial data?
• RQ5 Is parameter sharing across copula families justified based on similarity in estimated Kendall’s τ values?

Key findings

• The proposed dynamic vine copula model achieves superior one-day-ahead forecasting accuracy compared to static vine copulas and dynamic C- and D-vine models on 21 exchange rate returns.
• The simulation study confirms satisfactory performance of the Bayesian estimation procedure, with estimated parameters closely tracking true values.
• Parameter sharing across copula families is justified, as estimated Kendall’s τ values for Gaussian, Student t, Clayton, and Gumbel copulas are consistently close across different true τ values.
• The stationary distribution of the AR(1) process for copula parameters is analytically derived, enabling proper prior specification and posterior inference.
• The method successfully propagates uncertainty from lower to higher trees, unlike prior approaches that collapse lower-tree parameters to point estimates.
• The application to exchange rates demonstrates the model’s practical utility in capturing complex, time-varying dependence in financial time series.