1
Estimating a Structural VAR
To estimate a structural VAR you begin as you would any VAR estimation but selecting
the series in your VAR and estimating a conventional VAR. The following example is
from a paper co-written with a former graduate student and deals with inflation/deflation
in China.
Step 1: Choose the series you are interested in. Here, CPI inflation and real GDP
growth (a constant and a dummy for the deflation are assumed to be exogenous
variables). The VAR is assumed to have 4 lags.
2
Step 2: Estimate the unrestricted VAR which results in the following window appearing
3
Step 3: Click on “Procs” which produces the following menu
4
Step 4: Choose “Estimate Structural Factorization” (the last item on the menu) and the
following will appear
5
Step 5: Now you need to choose the structural factorization. Several choices are
available. First, you can enter the long or short-run restriction via text and Eviews gives
an example. We shall also go over this approach in class. Alternatively (and preferably),
you can create a matrix, here called “patc” which contains the unrestricted (denoted by
NA) and the restricted (here a numerical value is used but it could be some other value).
The matrix is created as follows (this can be done in the top window where commands
are entered):
matrix (2,2) patd
matrix is the command, (2,2) tells Eviews the size of the matrix (here 2 rows and 2
columns), while patd is the name given to the matrix. As shown below, when the matrix
is created the rows and columns are zeroes and they can be edited just as any series value
can be edited in the worksheet.
6
Step 6: Once you have decided whether the restrictions are of the long-run of short-run
variety you enter the matrix name as shown below and estimate the SVAR. In class we
will discuss the choices that can be made as well as the differences between short and
long-run restrictions
a. patc is the matrix used and a long-run restriction is assume