Model-free connectedness measures

This paper introduces a new model-free version of the connectedness approach of Diebold and Yılmaz (2012). The main advantage of the model-free connectedness approach is that, contrary to the state-of-the-art models, it is not restricted by the number of variables in the network. Considering eight different datasets, we show that model-free dynamics closely reflect the VAR, QVAR, and TVP-VAR-based connectedness measures. Given that the model-free connectedness approach can be seen as a special case of the Diebold and Yılmaz (2012) connectedness approach, we argue that it should be used as a benchmark model to demonstrate the differences in dynamics when employing the VAR, QVAR, or TVP-VAR connectedness approach. Finally, we show how the model-free connectedness approach relates to the R2 goodness-of-fit measure and how the total and pairwise connectedness indices are affected.