svars {svars}

svars: Data-driven identification of structural VAR models

Description

This package implements data-driven identification methods for structural vector autoregressive (SVAR) models as described in Lange et al. (2021) doi:10.18637/jss.v097.i05. Based on an existing VAR model object, the structural impact matrix B may be obtained via different forms of heteroskedasticity or independent components.

Details

The main functions to retrieve structural impact matrices are:

• id.cv

  Identification via changes in volatility,

• id.cvm

  Independence-based identification of SVAR models based on Cramer-von Mises distance,

• id.dc

  Independence-based identification of SVAR models based on distance covariances,

• id.garch

  Identification through patterns of conditional heteroskedasticity,

• id.ngml

  Identification via Non-Gaussian maximum likelihood,

• id.st

  Identification by means of smooth transition in covariance.

All of these functions require an estimated var object. Currently the classes 'vars' and 'vec2var' from the vars package, 'nlVar', which includes both VAR and VECM, from the tsDyn package as well as the list from MTS package are supported. Besides these core functions, additional tools to calculate confidence bands for impulse response functions using bootstrap techniques as well as the Chow-Test for structural changes are implemented. The USA dataset is used to showcase the functionalities in examples throughout the package.

Author(s)

• Alexander Lange [e-mail]

• Bernhard Dalheimer [e-mail]

• Helmut Herwartz [e-mail]

• Simone Maxand [e-mail]