Title: Statistical inference for financial connectedness
Authors: Ruben Hipp - University of Mannheim (Germany) [presenting]
Carsten Jentsch - TU Dortmund University (Germany)
Abstract: By using forecast error variance decompositions to identify networks, some authors established a new standard in estimating networks by introducing a concept of financial connectedness. Although many papers devoted their application to this concept of financial connectedness, the corresponding literature is lacking rigorous asymptotic theory that allows for statistical inference. To fill this gap, we make use of the framework of locally stationary processes. For this purpose we propose a general class of local-linear estimators for the time-varying VAR coefficient matrices and the innovations variance matrix that covers e.g. the previous approach. We derive explicit expressions for the limiting bias and variance and prove a CLT for these estimators to deduce corresponding limiting results for various connectedness measures. As the limiting distributions turn out to be complicated, we propose a model-based bootstrap procedure that builds on our estimates. In an application on the U.S. financial market we adopt this bootstrap method to construct confidence intervals. Following these results, we point out practical issues in the setting of financial connectedness and advice on how to handle bandwidth selection and estimation imprecision in periods of financial stress.