Johansen approach for testing cointegration
Posted: Wed Dec 07, 2011 4:22 pm
Hi everyone,
I'm measuring the effect of crime on investments in Rio de Janeiro. To do so, I'm using the following investment function:
I = f(Y, K, r, E, O) ; where I = investments, Y = GDP, K = capital stock, r = interest rate, E = investor's confidence, O = openess of the economy (X+M).
In order to measure the effect of crime, I added the "c" variable (= homicide rate) into the function, yielding to: I = f(Y, K, r, E, O,c).
Technically, I proceeded like that:
1) I tested the stationarity level of each varaibel and found that they are all integrated of order 1, thus I(1).
2) I tested the lag lenght optimality (of the unrestricted VAR model using all variables as endogenous) by the following command:
var var_ndi.ls 1 12 ndi rgdp ks lrir inv_conf openess
var_ndi.laglen(12, vname)
... and conclude for 1 lag (as optimal number of lag (following the SBC and the AIC).
3) I tried to select the optimal model (with/without trend, with/without intercept) by applying the Pantula's principle...
4) Then, I will run the cointegration test with this code :
coint.(s,1) lndi lrgdp lks lrir linv_conf lopeness hom
... determining the cointegration time series
5) correcting cointegration by a VECM.
MY QUESTIONS:
1) How to apply the Pantula's Principle. I mean, what is the code for programing?!?
2) How to insert (by coding) the VECM in my initial regression code?!?
Thank you in advance guys!
D.
I'm measuring the effect of crime on investments in Rio de Janeiro. To do so, I'm using the following investment function:
I = f(Y, K, r, E, O) ; where I = investments, Y = GDP, K = capital stock, r = interest rate, E = investor's confidence, O = openess of the economy (X+M).
In order to measure the effect of crime, I added the "c" variable (= homicide rate) into the function, yielding to: I = f(Y, K, r, E, O,c).
Technically, I proceeded like that:
1) I tested the stationarity level of each varaibel and found that they are all integrated of order 1, thus I(1).
2) I tested the lag lenght optimality (of the unrestricted VAR model using all variables as endogenous) by the following command:
var var_ndi.ls 1 12 ndi rgdp ks lrir inv_conf openess
var_ndi.laglen(12, vname)
... and conclude for 1 lag (as optimal number of lag (following the SBC and the AIC).
3) I tried to select the optimal model (with/without trend, with/without intercept) by applying the Pantula's principle...
4) Then, I will run the cointegration test with this code :
coint.(s,1) lndi lrgdp lks lrir linv_conf lopeness hom
... determining the cointegration time series
5) correcting cointegration by a VECM.
MY QUESTIONS:
1) How to apply the Pantula's Principle. I mean, what is the code for programing?!?
2) How to insert (by coding) the VECM in my initial regression code?!?
Thank you in advance guys!
D.