Cochrane–Orcutt procedure

[sven77]

Hi people. I'm trying to estimate potential GDP using a 3-input cobb douglas production function. A first estimation of the linear equation (after taking logs of both sides) shows autocorrelation amongst residuals (dw stat equals ~0.6). After implementing CO procedure (rho=0.6966685366345294), Dw stat hasn't been improved then the serial correlation still exist.
Could you please help me in removing autocorrelation? It's urgent. Any answer will be appreciated.

ls y c k h l
genr e=resid
ls e e(-1)
scalar rho=c(1)
smpl 2000q1 2000q1
genr y_star=((1-rho^2)^0.5)*y
genr h_star=((1-rho^2)^0.5)*h
genr k_star=((1-rho^2)^0.5)*k
genr l_star=((1-rho^2)^0.5)*l
genr e_star=((1-rho^2)^0.5)
smpl 2000q2 2013q4
genr y_star=y-rho*y(-1)
genr l_star=l-rho*l(-1)
genr k_star=k-rho*k(-1)
genr h_star=h*rho*l(-1)
genr e_star=1-rho
smpl @all
ls y_star e_star k_star h_star l_star

[startz]

Note that your data seems to have a unit root, although that wasn't what you asked.
The command

Code: Select all

ls y c k h l ar(1) ar(2)

will take care of the serial correlation.

[sven77]

so in this case CO method shouldnt have been implemented? I checked the stationarity matter before but the results are kind of misleading (adf test showed unit root, but kpps test cannot reject the null hypothesys). Anyway, looking at the graphs indicates nonstationarity. Is CO procedure dealing only with stationarity data? thank you for your reply

[startz]

Putting in AR(1) is almost exactly the same as iterated Cochrane-Orcutt. CO, as you have done it, adjusts for first-order serial correlation. The model seem to have second order serial correlation.

And I think you are misreading the KPSS test.

However, the real problem is probably that the right hand side variables are endogenous.