The RATS Software Forum

Hi,

I hope this finds you well,

I compute the IRF of a SVAR model identified with B&Q methodology. the IRF that I got fluctuate around zero before completly decay to zero.

I found applications of the same model, however thier IRF are like smooth lines around zero, authors note "Percent deviations from base line".

I think that the difference between my IRF that I compute with rats and thiers is not the difference of data, but the way to present, do you know how should I do with rats to have similar IRF.

I hope you got my point.

Thank you in advance,

Ahmed Sahloul

We have a replication for the original Blanchard-Quah paper. You have to accumulate the responses to the differenced variable as is done in that (the ACCUM option):

*
* Figures 1 and 2. Responses of output (1st variable) are accumulated.
*
@varirf(model=bqmodel,decomp=factor,steps=40,page=byshocks,$
variables=||"Output","Unemployment"||,shocks=||"Supply","Demand"||,accum=||1||)

If you don't do that, then you will get responses which aren't smooth since they're the responses of the change and not of the level.

Ok, thank you.

But I don't find the option "decomp" among the options of the VARIRF procedure. However, it works!! and it gives a different variance decomposition than the option "factor" could you kindly tell me why?

Another question, how could I trace all the accumulated IRF for one variable on the same garph ?

thank you very much.

Hi,

Excuse me to buther you with another question,

I found that the GDP on which I work has a determinstic trend instead of a unit root, could I in this case refer to the accumulated shocks, or essentially apply the identification with B&Q?

Thank you.

Ahmed

AhmedSahlool wrote:Hi,

Excuse me to buther you with another question,

I found that the GDP on which I work has a determinstic trend instead of a unit root, could I in this case refer to the accumulated shocks, or essentially apply the identification with B&Q?

Thank you.

Ahmed

The B-Q restriction doesn't really make sense in that case since the long-run response of a stationary series is already zero.

Could you kindly recommend me a solution.

I try to determine the sources of macreconomic fluctuations for a small open economy.

The model contains 5 variables: 1st diff of imports price index, 1st difference of global interrest rate, domestic detrended GDP, domestic exchange rate, domestic inflation.

I don't want to restrict the data in the short run. So did you meet a simlar case before?

Thank you

Again, there is no point in restricting the long run with stationary data, since they all have zero long-run responses. There's a separate literature on the use of short-but-not-impact restrictions for identifying shocks. Among the examples are

Uhlig(2005)
Faust(1998)

both of which focus on identifying a single shock and

Mountford and Uhlig(2009)

which is an extension of the original Uhlig ideas to multiple shocks.