On completing the connectedness analysis—A bootstrap-based DCC-GARCH approach

Abstract

By mapping high-dimensional systems to directed weighted networks, VAR-based Diebold-Yilmaz connectedness framework provides a novel and nuanced perspective on spillovers. Nonetheless, its reliance on rolling windows and the absence of formal statistical evidence for event-based analysis may limit its applicability. To overcome these two shortcomings, this study develops an alternative connectedness framework based on the dynamic conditional correlation—generalized autoregressive conditional heteroskedasticity (DCC-GARCH) model and a bootstrap technique, augmented by a probabilistic analysis of an increase in connectedness in response to major events. We apply our framework to four major currencies against the US dollar. In terms of event probability analysis, it is observed that 15 out of the 20 identified events correspond to a probability exceeding 90% for an increase in total connectedness, which predominantly pertain to geopolitical crises, financial market collapses, and global health emergencies. Therefore, although traditional event analysis frequently capture increases in connectedness, our methodology underscores that these associations may lack statistical rigor. Beside this, we find that the total connectedness also responds instantaneously to such events with rapid dissipation or manifests a delayed reaction.