The connectedness of the banking system has often been analysed through Diebold-Yilmaz framework, a VAR based approach using the forecast error variance decomposition to measure the strength and direction of the relationships between financial institutions. The framework has spread widely, still the application of VAR on multiple time series can be problematic. Most analysis include dynamic rolling-window estimation, but the stationarity of the processes or the VAR lag selection test is not executed on the distinct windows. Furthermore, most often the cointegration tests are also missing. Through DGP I build a simulated system of multiple time series to measure the error caused by ignoring cointegration. In addition, I test another multiple time series model, VECM, on the system that is able to capture cointegrating relationships. Based on the simulation, ignoring cointegration leads to higher standard error and inconsistent estimation that can be handled through applying VECM framework rather than VAR.
