Asymmetric VAR

[PTillmann-436]

Dear Tom,

Are you aware of RATS codes of an asymmetric VAR, i.e. one in which positive shocks and negative shocks could have different effects in absolute terms?

Thanks for your help.

Peter

[TomDoan]

Do you have a reference on that? Asymmetric VAR usually means that different variables have different lags in different equations.

[PTillmann-436]

Dear Tom,

I work with the VAR model with shocks to exogenous variables and want to distinguish positive and negative innovations to the exogenous variable. Currently, I include the change in the interest rate as an exogenous variable and split it into positive and negative ones:

system(model=varmodel)
variables var1 var2 var3
lags 1 to lags
det constant rate_pos rate_neg
end(system)

equation(empty) rateeq_pos rate_pos
equation(empty) rateeq_neg rate_neg

The problem here is that rate_pos and rate_neg each include many zeros, which make the results insignificant. Do you see a way to include the total rate change and a dummy for positive changes instead of including positive/negtive changes separately?

Thanks and best regards

Peter

[TomDoan]

The fact that each of the + and - branches has many zeroes really isn't that important. After all, a single time period dummy, which has T-1 zeros and only one non-zero will come in significant if that data point has a residual from a non-dummied model which is more than roughly two standard errors. It seems that it's really telling you that there isn't that much difference between the + and - branches.

Maybe I'm missing what you're asking but the total change combined with a plus branch, or total change combined with a minus branch spans exactly the same space as plus branch and minus branch, so you would get the same results.

[PTillmann-436]

Yes, I fully agree. The advantage is that I can discuss the asymmetry in terms of a single vector of asymmetry-dummies, not two vectors.

Thanks,
Peter