A0964
Title: On mixed causal-noncausal structural VAR models in macro-finance
Authors: Lison Christiaens - Maastricht University, Liege University (Belgium) [presenting]
Alain Hecq - Maastricht University (Netherlands)
Julien Hambuckers - University of Liege (Belgium)
Abstract: The aim is to propose a methodology to identify and interpret structural shocks, both causal and noncausal, in mixed causal-noncausal (structural) vector autoregressive (VAR) models. These models are relevant in macrofinance, where noncausal components capture forward-looking behavior, expectation-driven dynamics, and bubble-like episodes in asset prices or policy responses. In the first step, model parameters are estimated using the generalized covariance (GCov) estimator, which relies on nonlinear autocovariances in a non-Gaussian framework. However, it is observed that in multivariate settings, GCov may suffer from estimation instabilities, which is addressed by introducing refinements that enhance robustness. In the second step, identified restrictions are imposed to recover structural shocks and construct impulse response functions (IRFs) that reflect the asymmetric, time-irreversible propagation of causal and noncausal shocks. The methodology is evaluated through Monte Carlo simulations and applied to macroeconomic and financial time series.
