Dear community,
I am estimating a VARX model with fixed coefficients for endogenous variables (as the dataset for the endogenous variables is longer than the one for exogenous variables). The problem I have is that when I constrain all the coefficients in the VARX according to the coefficients of a VAR estimation with longer sample, then the orthogonalized impulse response functions (OIRFs) are not the same for the VAR and the VARX model.
I have 40 years of data on the endogenous variables and 10 years of data for the exogenous variables. I first estimate a VAR with the long sample. Thereafter, I fix the coefficients for the endogenous variables and estimate a VARX with short sample.
The OIRFs are the same if I estimate the VAR with long sample with and without fixed coefficients. However, when I move from the long sample to the short sample (i.e. drop ¾ of my endogenous variables observations), the orthogonalized impulse response functions are not the same even though all the coefficients for the endogenous variables are fixed. What could be causing the problem?
A further curiosity is that the problem is only related to orthogonalized impulse response functions. When I use generalized impulse response functions, the IRFs are exactly the same for both VAR with long sample and VARX with short sample.
Thank you for your help
Jussi
I am estimating a VARX model with fixed coefficients for endogenous variables (as the dataset for the endogenous variables is longer than the one for exogenous variables). The problem I have is that when I constrain all the coefficients in the VARX according to the coefficients of a VAR estimation with longer sample, then the orthogonalized impulse response functions (OIRFs) are not the same for the VAR and the VARX model.
I have 40 years of data on the endogenous variables and 10 years of data for the exogenous variables. I first estimate a VAR with the long sample. Thereafter, I fix the coefficients for the endogenous variables and estimate a VARX with short sample.
The OIRFs are the same if I estimate the VAR with long sample with and without fixed coefficients. However, when I move from the long sample to the short sample (i.e. drop ¾ of my endogenous variables observations), the orthogonalized impulse response functions are not the same even though all the coefficients for the endogenous variables are fixed. What could be causing the problem?
A further curiosity is that the problem is only related to orthogonalized impulse response functions. When I use generalized impulse response functions, the IRFs are exactly the same for both VAR with long sample and VARX with short sample.
Thank you for your help
Jussi
