Modeling dependence dynamics through copulas with regime switching

• Published in: Insurance, mathematics & economics Vol. 50; no. 3; pp. 346 - 356
• Main Authors: Silva Filho, Osvaldo Candido da, Ziegelmann, Flavio Augusto, Dueker, Michael J.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-05-2012; Elsevier Sequoia S.A
• Subjects: Asymmetric dependence; Asymmetric information; Bootstrap mechanism; Bootstrap test; Copulas; Econometrics; Economic crisis; Financial crisis; International finance; Markov analysis; Markov switching; Markovian processes; Modelling; Monte Carlo simulation; Securities markets; Statistical analysis; Studies; United Kingdom > UK; Brazil; United States > US; Copulas; C46; Bootstrap test; Asymmetric dependence; Markov switching; C15; G15
• ISSN: 0167-6687; 1873-5959
• DOI: 10.1016/j.insmatheco.2012.01.001

Measuring dynamic dependence between international financial markets has recently attracted great interest in financial econometrics because the observed correlations rose dramatically during the 2008–09 global financial crisis. Here, we propose a novel approach for measuring dependence dynamics. We include a hidden Markov chain (MC) in the equation describing dependence dynamics, allowing the unobserved time-varying dependence parameter to vary according to both a restricted ARMA process and an unobserved two-state MC. Estimation is carried out via the inference for the margins in conjunction with filtering/smoothing algorithms. We use block bootstrapping to estimate the covariance matrix of our estimators. Monte Carlo simulations compare the performance of regime switching and no switching models, supporting the regime-switching specification. Finally the proposed approach is applied to empirical data, through the study of the S&P500 (USA), FTSE100 (UK) and BOVESPA (Brazil) stock market indexes.
► Copula functions are used to model dependence dynamics between random variables. ► We include hidden first order Markov Chain to describe the dependence dynamics. ► We implement a block bootstrap procedure to estimate the covariance matrix. ► The method performance is tested by Monte Carlo simulation and empirical applications. ► Our model offers relevant insights about the dependence dynamics.