Working Paper: CEPR ID: DP10801
Authors: Andrea Carriero; George Kapetanios; Massimiliano Marcellino
Abstract: We address the issue of parameter dimensionality reduction in Vector Autoregressive models (VARs) for many variables by imposing specific reduced rank restrictions on the coefficient matrices that simplify the VARs into Multivariate Autoregressive Index (MAI) models. We derive the Wold representation implied by the MAIs and show that it is closely related to that associated with dynamic factor models. Next, we describe classical and Bayesian estimation of large MAIs, and discuss methods for the rank determination. Then, the theoretical analysis is extended to the case of general rank restrictions on the VAR coefficients. Finally, the performance of the MAIs is compared with that of large Bayesian VARs in the context of Monte Carlo simulations and two empirical applications, on on the transmission mechanism of monetary policy and the propagation of demand, supply and financial shocks.
Keywords: Bayesian VARs; Factor Models; Forecasting; Large Datasets; Multivariate Autoregressive Index Models; Reduced Rank Regressions; Structural Analysis
JEL Codes: C11; C13; C33; C53
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| reduced rank structure (R50) | simplification of the model's complexity (C52) |
| Bayesian estimation of MAI model (C51) | more reliable impulse response functions (C51) |
| monetary policy shocks (E39) | various economic indicators (industrial production and employment) (E24) |
| demand shocks (E39) | macroeconomic variables (E19) |
| supply shocks (E39) | macroeconomic variables (E19) |
| financial shocks (F65) | macroeconomic variables (E19) |