Abstract
We develop threshold models that allow volatilities and copula functions or their association parameters to change across time. The number and location of the thresholds is assumed unknown. We use a Markov chain Monte Carlo strategy combined with Laplace estimates that evaluate the required marginal densities for a given model. We apply our methodology to financial time series, emphasizing the ability to improve estimates of risk characteristics, as well as measuring financial contagion by inspecting simultaneous changes of dependence and volatility structures.
Acknowledgements
We would like to thank the three anonymous referees for their helpful comments. In particular, the joint estimation of volatility and copula changes that greatly improved our proposed model is based on a recommendation of one referee to whom we are grateful.