Gaussian Process Vector Autoregressions and Macroeconomic Uncertainty (avec N. Hauzenberger, F. Huber et N. Petz)
Séminaire Marcel-Dagenais en économétrie 2021-2022
conjoint avec le Département de sciences économiques, Université de Montréal
Responsable : Karim Chalak (U. de Montréal)
* Sur invitation seulement. Veuillez contacter le responsable si vous souhaitez y accéder.
Résumé: Linear models such as vector autoregressions (VARs) imply symmetry in the shocks and constancy in the parameters. The recent literature has relaxed these restrictions by introducing specific assumptions on how parameters change or whether shocks impact the economy differently over time. In this paper, we develop a non-parametric multivariate time series model that remains agnostic on the precise relationship between a (possibly) large panel of macroeconomic time series and their lagged values. The main building block of our model is a Gaussian process prior on the functional relationship that determines the conditional mean of the model. We control for changes in the error variances by introducing a stochastic volatility specification. To facilitate computation in high dimensions and introduce convenient statistical properties tailored to match stylized facts commonly observed in macro time series, we assume that the covariance of the Gaussian process is scaled by the latent volatility factors. We illustrate our model by analyzing the effects of macroeconomic uncertainty on US data with a particular emphasis on time variation and asymmetries in the transmission mechanisms of economic uncertainty.