(Lasso) VAR for Expectile
34 Pages Posted: 30 Nov 2020
Date Written: October 1, 2022
Expectile has recently received considerable attention in risk management. This paper extends the vector autoregressive (VAR) for conditional means to VAR for conditional expectiles (MCARE) to capture the interdependencies among the expectiles of multiple units. We further generalize MCARE to high-dimensional cases by imposing an L1-penalization (L-MCARE). We demonstrate how to estimate the model parameters and derive the estimators' asymptotic properties. As an empirical application, we apply MCARE and L-MCARE to the 2020 list of global systemically important banks. Formal out-of-sample model evaluation tests document our models significantly outperform single equation models. We also find clear asymmetric effects of past positive and negative shocks of different units as well as their volatilities on the expectiles. Moreover, network analyses based on L-MCARE report a clear decrease in the connectedness of most banks during the recent pandemic and an increase for several banks with exceptional performance during the pandemic.
Keywords: risk spillover, expectile, Lasso, network analysis, model evaluation
JEL Classification: G21, C32, C51
Suggested Citation: Suggested Citation