Measuring Time-Varying Connectedness: A Bayesian Approach = Időben változó összekapcsoltság mérése Bayes-i megközelítésben

Absztrakt (kivonat)

Interconnectedness and structural changes. Among the many lessons that the Global Financial Crisis has taught us, these two might be the most crucial for financial modelling and their importance has been enhanced once again during the European Sovereign Debt Crisis and lately, in the Covid-19 Pandemic. Especially during turbulent, uncertain periods on the financial market, the risk spills over across markets and different asset classes. In this study, I propose a framework to simultaneously incorporate the two concepts to estimate the change of connectedness due to structural breaks in a network. I introduce a time varying parameter vector autoregression with stochastic volatility (TVP-VAR-SV) estimation method, to capture the dynamic nature of financial relationships, into the Diebold-Yilmaz spillover framework, that is widely used to measure connectedness within and across different levels of markets. I also examine the estimating efficiency of two different prior specifications (horseshoe and inverse gamma priors) across three Monte Carlo simulations. First, to analyze the properties of the priors, a univariate model is simulated with breaks in the parameters and the stochastic volatility process too. Then I generate a bivariate model with three different structural break regimes and analyze the model fitting properties of the two prior specifications. Finally, I apply the Diebold-Yilmaz framework on the generatedand the fitted bivariate time series too, which would allow us to assess the accuracy of the estimation of networks. I find that the accuracy of the different priors ceteris paribus depends on the variance level of parameters, on the ratio between the variance of parameters and the size of the structural breaks, and on the existence of structural breaks in the system. Furthermore, accuracy-patterns in the network estimation are similar to the case where we estimate the underlying TVP-VAR-SV model.