t-statistics in FMOLS and CCR estimations

[wolly77]

Dear guys,
I have a simple question. I'll want to say if the t-statistics reported in Fully Modified OLS and Canonical Cointegrating Regression (Eviews 7) are valid.

[EViews Glenn]

They are, under the appropriate assumptions.

[ariuna97@hotmail.com]

I would like to ask some technical questions about estimating CCR on Eviews 7, I am new to cointegration methods, so my questions mught look silly.
This is my output: LY is my dependent var and LLY is exogen. variable:
Dependent Variable: LY
Method: Canonical Cointegrating Regression (CCR)
Date: 10/28/10 Time: 19:12
Sample (adjusted): 1972 2006
Included observations: 35 after adjustments
Cointegrating equation deterministics: C
Long-run covariance estimate (Prewhitening with lags = 2, Bartlett kernel,
Andrews bandwidth = 2.3903)

Variable Coefficient Std. Error t-Statistic Prob.

LLY 0.028093 0.002599 10.80882 0.0000
C 24.76078 0.139754 177.1738 0.0000

R-squared 0.940371 Mean dependent var 26.35429
Adjusted R-squared 0.938564 S.D. dependent var 0.323531
S.E. of regression 0.080191 Sum squared resid 0.212211
Durbin-Watson stat 0.272007 Long-run variance 0.024146
My queations are as followings:
1. can I have more than 1 exon. var?
2. is the coeff. here same as β hat in CCR equation?
3. what is band-with? is it same as lag lenght?
4. does CCR produce H(0,1) and H(0, q)? if does, how i can do that?
5. can i test Park's J and G tests? if yes, how?

Thank you very much.

[EViews Glenn]

1. You can have more than one cointegrating and more than one deterministic regressor. Deterministic regressors not in the trend specification combo cause some problems for inference.

2. Not sure what you mean.

3. No, it's the bandwidth for the kernel estimator of the long-run variance. See the Appendix in EViews UGII for details.

4. See the section on testing in the chapter for a description of how to do the Park test.

5. No.

[ariuna97@hotmail.com]

Thank you very much.