A0317
Title: Simultaneous inference for generalized impulse responses in VAR Models
Authors: Endong Wang - McGill University (Canada) [presenting]
Jean-Marie Dufour - McGill University (Canada)
Abstract: In macroeconomics, testing the non-causality hypothesis at multiple horizons is often crucial. It has been demonstrated that the test can be simply undertaken using linear regression through the coefficients in Vector Autoregression (VAR) at various horizon, named as generalized impulse responses. It is shown that lag-augmented Vector Autoregression (VAR) at multiple horizons can provide unit-roots robust inference and a simple formula for covariance matrix. The inference is derived by reparameterizing the model with real VAR parameters, which eliminates the concerns of unit roots and serially correlated residuals. The research results provide a theoretical foundation that conservative lag length selection is preferable in terms of preserving asymptotic efficiency and obtaining more accurate asymptotic variance estimates. In practice, the reparameterized model with VAR parameter estimates, analogous to using estimated residuals as regressors, can estimate the usual impulse responses jointly with a non-degenerated distribution. The simulation study illustrates that our estimates could save efficiency. The inference result for structural impulse responses with recursive identification is also presented. Lastly, an empirical application of generalized impulse responses is implemented to analyze the response of the macroeconomic variables to a positive shock on economic policy uncertainty, as measured by the Economic Policy Uncertainty (EPU) index.
