Cholesky decomposition in SVAR

[Jose_j]

Hi all,

Could someone help me program Cholesky decomposition in structural VAR? I first run SVAR model in Eview (A*e_t = B*u_t) with 3 endogenous variables (the change in oil price as the top variable ordering followed by two sectoral output growth variables), and try to get impulse response of two output variables to 10% change in oil price. All variables are in log differenced. My A and B matrices look as follows which assume no contemporaneous effect on oil price equation:

A = 1 0 0 B= v1 0 0
b1 0.4 0.6 0 v2 0
b2 0.2 0.8 0 v3 0

I impose restrictions on parameters in A matrix and estimate only 5 parameters in SVAR (b1,b2 v1-v3). Eviews have an option to do structural decomposition to estimate impulse response but it assumes 1 standard deviation shock. One standard deviation of my price variable is 18%, so I want to rescale it to 10%. I also want to account for correlation across error terms by Cholesky decomposition. For this, I guess I need to specify my own impulse vector/matrix to estimate a user-specified impulse responses.

I've defined the impulse vector (shock) as follows and run the model. However, it does not seem to do Cholesky decomposition:

shock = 10
0
0

In my SVAR specification, could someone tell me how to define my impulse vector/matrix to rescale the price shock to 10% and to do Cholesky decomposition at the same time?

Thank you very much for your helps.

[trubador]

Use the following short-run pattern matrices:

Code: Select all

matrix mata = @identity(3,3) matrix(3,3) matb matb.fill NA,0,0,NA,NA,0,NA,NA,NA