## Impulse response function and decomposition

Questions and discussions on Vector Autoregressions

### Impulse response function and decomposition

Hi Tom,

I have used the engle-granger 2 step to get the residuals, test residuals for stationarity, then use the residuals to create a VECM. How do I create the IRF and decomposition takign into consideration that the ECM (T-1) also influence the dependent var? i tried using the impulse function but thought otherwise. do you know what is the calculation involved in tabulating the IRF and decomposition if i am to do it manually if i can't use it using any RATS procedure? (using cholesky, residuals with one unit/standard deviation or general response)

for example, the 1st response in the IRF table was WTI{1} + coefficient from ecm{1}. however, i was not able to derive the 2nd and subsequent figures....

is it possible to use @stampdiags on linear regressions, bai-perron and other equations which you recomended to diagnose the state space model?

how do I create the DW and hetereoscedasiticty tests to compare to @stampdiags? or vice versa?

also, how do i create the Yhat for fitted values after a linear regression?

apologies for asking such minor silly questions.

Thanks once again.

Rgds,
Des

Posts: 30
Joined: Wed Sep 14, 2011 2:11 am

### Re: Impulse response function and decomposition

I have used the engle-granger 2 step to get the residuals, test residuals for stationarity, then use the residuals to create a VECM. How do I create the IRF and decomposition takign into consideration that the ECM (T-1) also influence the dependent var? i tried using the impulse function but thought otherwise. do you know what is the calculation involved in tabulating the IRF and decomposition if i am to do it manually if i can't use it using any RATS procedure? (using cholesky, residuals with one unit/standard deviation or general response)

for example, the 1st response in the IRF table was WTI{1} + coefficient from ecm{1}. however, i was not able to derive the 2nd and subsequent figures....

Just use the features built in to RATS for estimating error correction models. Instead of putting in the residuals, you create an ECT instruction which gives the cointegrating vector. When you estimate the model, it generates a properly expanded model. The point estimates will be identical to differencing the variables and regression them with the ECT residuals, but it will be easier to use.

chiade wrote:is it possible to use @stampdiags on linear regressions, bai-perron and other equations which you recomended to diagnose the state space model?

Diagnostics for one model aren't necessarily interesting for another type of model. They don't compare Model A with Model B; Model A's compare's A with some (often rather vague extension) A* and B with B*.

chiade wrote:how do I create the DW and hetereoscedasiticty tests to compare to @stampdiags? or vice versa?

also, how do i create the Yhat for fitted values after a linear regression?

"Fitted values" is in the index for the manuals.
TomDoan

Posts: 2727
Joined: Wed Nov 01, 2006 5:36 pm