A0740
Title: Co-extremal shocks and VAR analysis
Authors: Stefan Mittnik - Ludwig Maximilians University Munich (Germany) [presenting]
Abstract: There is substantial empirical evidence that macroeconomic and financial shocks deviate from Gaussianity. A past study offered a theoretical explanation for this phenomenon by demonstrating that aggregation through production networks can transform micro disturbances into non-Gaussian macro shocks. Other studies cautioned that neglecting tails' properties may bias inference. In light of this, vector autoregressive (VAR) models have been refined to reflect this property by allowing errors to follow a multivariate Student's t-distribution, a Laplace distribution, or a generalized hyperbolic distribution. Based on the assumption of local ellipticity, the use of quantile-implied tail correlations is proposed as a natural approach to capturing non-Gaussianity and, in particular, to assessing the implications of extreme shocks in VAR modeling. When applied within a Global VAR (GVAR) framework, this approach can sharpen cross-country spillover analysis by better capturing the occurrence and implications of joint extremes in international shocks. This could lead to more realistic risk assessments and deeper, crisis-relevant policy insights at the global level.
