Sparse-structured time-varying parameter vector autoregression for high-dimensional network connectedness measurement

Abstract

This paper proposes a new sparse-structured time-varying parameter vector autoregression (SS-TVP-VAR) model to effectively measure high-dimensional network connectedness of financial enterprises. Previous VAR based network connectedness methods require high computational cost and are not feasible for high-dimensional networks of a large number of enterprises. The proposed SS-TVP-VAR not only reduces computational cost but also selects key coefficients when building the VAR model. As a result, it produces a more adaptive and effective connectedness to measure systemic risk caused by extreme events. Studies on 57 listed major financial enterprises in China mainland during a 15-year period show that the proposed SS-TVP-VAR achieves reliable and useful results, and it successfully identifies all the major market events during this time.