Ask Your Question

# Revision history [back]

Hi, Here's a procedure that can be used to compute impulse response functions and to decompose the forecast-error variance. Hope you find it useful.

Firstly, estimate a VECM or VAR using OLS, SYSTEM, or the procedure file JOHANSEN.PRO in conjunction with the input file JOHANSEN.SHA. (Both files are available from Resources > Procedure Library in Shazam, version 11.1.2 or later.) Then, identify the coefficient matrix for y[t], or its inverse, in the structural representation of your model based on the identifying restrictions. (For example, if you assume triangular structure for identification, the Cholesky-decomposition matrix of the variance-covariance matrix of the residuals is the inverse of the coefficient matrix for y[t]. Or, in the simplest case of orthogonal errors, the matrix is an identity matrix.)

Once the coefficients of the structural VAR, reduced-form VAR, or VECM representation of your model is identified, use the following procedure (IMPULSE&VARDEC.PRO), after providing relevant information in the input file, IMPULSE&VARDEC.SHA, to compute the impulse response functions and the proportions of the forecast-error variance due to shocks to the endogenous variables. In a typical case of SVAR analysis, the user would only need the impulse response functions to structural shocks. However, impulse response functions to shocks to the reduced-form residuals are also computed for those who might be interested in them. More explanations are provided in both files. An illustration is provided in IMPULSE&VARDEC-EX.SHA using the data in IMPULSE&VARDEC-EX.DAT.