A1266
Title: Bayesian estimation of multiple change points in factor copula models
Authors: Marvin Borsch - University of Cologne (Germany) [presenting]
Roman Liesenfeld - University of Cologne (Germany)
Dominik Wied - University of Cologne (Germany)
Abstract: The aim is to introduce a methodology for detecting multiple change points in the loadings of a factor copula model, resulting in shifts in the underlying correlation structure. By leveraging a copula framework, the marginal distributions are modeled independently of their dependence structure. A Bayesian procedure is developed to jointly estimate the factor loadings and the locations of the change points, assuming a fixed number of structural changes. Posterior samples are obtained via an independent Metropolis-Hastings within Gibbs sampler. In this approach, the conditional posterior for the loadings depends solely on the likelihood within each segment and its prior, while each changepoints conditional posterior is determined only by the likelihood of the data between adjacent change points and its prior. The proposed approach is evaluated through Monte Carlo simulations under various change scenarios, and its practical relevance is demonstrated through an empirical application to the correlation structure of EURO STOXX 50 companies across different sectors during the recent Covid-19 pandemic.
