LAG SELECTION IN ARDL
Posted: Sun Mar 31, 2013 5:39 am
I am woking to find out long run relationship between dependent variable i.e. "Y" and independent variables i.e. "X1","X2", "X3", "X4", "X5", "X6" through ARDL method. I have learnt the following procedure:
Processing Steps:
(1) Check Unit Root. If Series are of different integrating orders, proceed with ARDL.
(2) Select Lag through VAR process - Lag Structure - Lag Length Criteria. Choose whichever is less between SBC & AIC (prefer SBC)
(3) Suppose, proposed lag length for overall model is 1, below equation is developed for 7 variables.
∆(Y) C ∆(Y(-1)) ∆(X1) ∆(X1(-1)) ∆(X2) ∆(X2(-1)) ∆(X3) ∆(X3(-1)) ∆(X4) ∆(X4(-1)) ∆(X5) ∆(X5(-1)) ∆(X6) ∆(X6(-1)) Y(-1) X1(-1) X2(-1) X3(-1) X4(-1) X5(-1) X6(-1)
(4) Check probabilities in the output against all of the above explanatory variables and explain results.
(5) Execute WALD Test with restrictions: C(14)=C(15)=C(16)=C(17)=C(18)=C(19)=C(20)=0
(6) If probability of resulting F Test is less than 0.05, it implies rejection of null hypothesis and confirm presence of long run relationship between significant variables and dependent variables.
Queries:
(a) If above procedure is correct?
(b) Is above equation valid for up to Lag 1?
(c) What are steps to introduce further lags without following any auto criteria like AIC, SBC?
(d) If lags are required to be increased (for satisfaction only!), do we need to introduce the same in explanatory variables without difference in above equation i.e. Y(-1) X1(-1) X2(-1) X3(-1) X4(-1) X5(-1) X6(-1) OR lag is required to be introduced in full equation simultaneously.
(e) I have tried to introduce Lag 2 in each of "Y(-1) X1(-1) X2(-1) X3(-1) X4(-1) X5(-1) X6(-1)" but error appeared as "near singular matrix". However, when I introduced Lag three in the same instead of Lag 2, results were calculated by EVIEWS. Why?
(f) With every introduction in LAG for any explanatory variable, probabilities of all variables are changed. Why? Is that correct?
I request for an expert advise on above queries.
Processing Steps:
(1) Check Unit Root. If Series are of different integrating orders, proceed with ARDL.
(2) Select Lag through VAR process - Lag Structure - Lag Length Criteria. Choose whichever is less between SBC & AIC (prefer SBC)
(3) Suppose, proposed lag length for overall model is 1, below equation is developed for 7 variables.
∆(Y) C ∆(Y(-1)) ∆(X1) ∆(X1(-1)) ∆(X2) ∆(X2(-1)) ∆(X3) ∆(X3(-1)) ∆(X4) ∆(X4(-1)) ∆(X5) ∆(X5(-1)) ∆(X6) ∆(X6(-1)) Y(-1) X1(-1) X2(-1) X3(-1) X4(-1) X5(-1) X6(-1)
(4) Check probabilities in the output against all of the above explanatory variables and explain results.
(5) Execute WALD Test with restrictions: C(14)=C(15)=C(16)=C(17)=C(18)=C(19)=C(20)=0
(6) If probability of resulting F Test is less than 0.05, it implies rejection of null hypothesis and confirm presence of long run relationship between significant variables and dependent variables.
Queries:
(a) If above procedure is correct?
(b) Is above equation valid for up to Lag 1?
(c) What are steps to introduce further lags without following any auto criteria like AIC, SBC?
(d) If lags are required to be increased (for satisfaction only!), do we need to introduce the same in explanatory variables without difference in above equation i.e. Y(-1) X1(-1) X2(-1) X3(-1) X4(-1) X5(-1) X6(-1) OR lag is required to be introduced in full equation simultaneously.
(e) I have tried to introduce Lag 2 in each of "Y(-1) X1(-1) X2(-1) X3(-1) X4(-1) X5(-1) X6(-1)" but error appeared as "near singular matrix". However, when I introduced Lag three in the same instead of Lag 2, results were calculated by EVIEWS. Why?
(f) With every introduction in LAG for any explanatory variable, probabilities of all variables are changed. Why? Is that correct?
I request for an expert advise on above queries.