SVAR Blanchard and Perotti methodology
Posted: Sun Oct 04, 2015 4:33 pm
I'm trying to estimate an SVAR model following the Blanchard and Perotti strategy. I have read the papers of Blanchard and Perotti (2002), Perotti (2002) and Raffaela, Momigliano, Neri and Perotti (2007), which use the same methodology. I understood that Blanchard and Perotti estimated all equations using OLS(or IV in the case of GDP equation) and then use the A and B matrices to generate the IRFs. My questions are the following:
1) Standard books such as Lutkepohl (New introduction to Multiple Time Series Analysis) pag 372, suggest that an AB model should be estimated through maximum likelihood method. If I try to estimate the same model of Blanchard and Perotti using maximum likelihood method, then the restriction a2=0 (or b2=0) is not necesary. Am I right?
2) If the statement above is right, then c1 and c2 can be estimated via maximum likelihood (together with a2 and b2), too. Could I follow a mixed procedure? It means that for instance: I estimate c1 and c2 using Blanchard and Perotti procedure and then estimate a2 and/or b2 using maximum likelihood.
3) What is the best method to estimate an AB model? Why?
4) When the variables are in log-level form, is necessary to divide the IRFs to the average fiscal variable - output ratio. This ensure that the IRFs have the interpretation of fiscal multipliers. I understand the maths of this. However, one specification of Blanchard and Perotti set the variables in log-differences. Is necessary any transformation of the impulse response funtions to get these as fiscal multipliers? What transformation?
1) Standard books such as Lutkepohl (New introduction to Multiple Time Series Analysis) pag 372, suggest that an AB model should be estimated through maximum likelihood method. If I try to estimate the same model of Blanchard and Perotti using maximum likelihood method, then the restriction a2=0 (or b2=0) is not necesary. Am I right?
2) If the statement above is right, then c1 and c2 can be estimated via maximum likelihood (together with a2 and b2), too. Could I follow a mixed procedure? It means that for instance: I estimate c1 and c2 using Blanchard and Perotti procedure and then estimate a2 and/or b2 using maximum likelihood.
3) What is the best method to estimate an AB model? Why?
4) When the variables are in log-level form, is necessary to divide the IRFs to the average fiscal variable - output ratio. This ensure that the IRFs have the interpretation of fiscal multipliers. I understand the maths of this. However, one specification of Blanchard and Perotti set the variables in log-differences. Is necessary any transformation of the impulse response funtions to get these as fiscal multipliers? What transformation?