unit root test & Johansen Coint
Posted: Thu Sep 02, 2010 12:57 pm
hi everyone,
my analysis involves finding the effect of a policy on prices. The policy change or event is designated by a dummy variable. My first problem is setting the maximum lag length for the unit root tests. I have monthly data for 10 years and it seems that the automatic selection procedure using SIC sets the actual lags used to the maximum, making me arrive at 40+ lags. What other rule is plausible in this case or should I go for so many lags? Using 12 gives me one I(2), one I(1) and 4 I(0) variables according to ADF.
Carrying on with Johansen, I tried to include the dummy variable as an exogenous variable, but then I get the error message: "near singular matrix". Neglecting the dummy, it works but again the lags cause big differences. 4 lags give at most 1 CE whereas 12 lags lead to 6CE. I have 6 variables in total +dummy. So again, why is dummy causing trouble and what can I do about the lags. Greatly appreciate any help!
Thx
my analysis involves finding the effect of a policy on prices. The policy change or event is designated by a dummy variable. My first problem is setting the maximum lag length for the unit root tests. I have monthly data for 10 years and it seems that the automatic selection procedure using SIC sets the actual lags used to the maximum, making me arrive at 40+ lags. What other rule is plausible in this case or should I go for so many lags? Using 12 gives me one I(2), one I(1) and 4 I(0) variables according to ADF.
Carrying on with Johansen, I tried to include the dummy variable as an exogenous variable, but then I get the error message: "near singular matrix". Neglecting the dummy, it works but again the lags cause big differences. 4 lags give at most 1 CE whereas 12 lags lead to 6CE. I have 6 variables in total +dummy. So again, why is dummy causing trouble and what can I do about the lags. Greatly appreciate any help!
Thx