GLS / "Help! My variables are insignificant!"

[SophieSmith]

Hello Everyone, I hope I am posting in the right place. I'm a MSc Economics student working on my dissertation and have got a bit stuck with trying to use GLS on eviews.

I am following the technique of an ECB paper by Heinz and Rusinova and they use GLS because of heteroskedasticity (i tested for this with my data, and it was present) so I want to follow them and use GLS also. The only problem is I have no idea how to do this on eviews, we never did it in class. Sorry for my ignorance, I have been searching though this forum and the help documents, can someone point me in the right direction? In addition, when estimating my model with OLS (so i could do BPG test etc) almost all of my variables are insignifcant which is really worrying as I need to test this equation for a structural break as the main part of my dissertation. Is this because of OLS instead of GLS or a bigger problem? Thanks!

n.b. using 3 different explantory variables (with two lags based on SIC test), lagged explanatory and error correction term (from engle-granger technique).

[SophieSmith]

The paper I am following uses FGLS. I'm trying to work out the difference.

[SophieSmith]

So i followed these steps for doing a FGLS that I found online:

"Or, we can transform the model and run OLS on this transformed model. How?
Feasible Generalized Least Squares procedure:
- generate the estimated squared residuals (the residuals from the model ;
- regress the log of the estimated squared residuals on the independent variables (why the log?), obtain the fitted values of this regression
- exponentiate the fitted values to get
- estimate the equation by WLS, using as weights

Since we have to estimate h, FGLS will not be unbiased but it is consistent and asymptotically more efficient than OLS.
If cigs_residsq stands for the estimated h, we have to divide the model by
1/square root(h). Why? See book (there is a univariate example there for savings, start from there)."

Does that sound right? My variabes have gone from many not being significant:

Dependent Variable: D_LNAVNOMWAGE
Method: Least Squares
Date: 08/21/11 Time: 14:06
Sample (adjusted): 1971Q4 2011Q1
Included observations: 158 after adjustments

Coefficient Std. Error t-Statistic Prob.

C 0.001359 0.001324 1.026975 0.3061
CYCLICAL25600 0.009682 0.004823 2.007496 0.0466
CYCLICAL25600(-1) -0.020928 0.008810 -2.375475 0.0188
CYCLICAL25600(-2) 0.009208 0.004822 1.909495 0.0582
D_LNPRODGROWTH 0.479939 0.069518 6.903775 0.0000
D_LNPRODGROWTH(-1) 0.159963 0.078141 2.047095 0.0425
D_LNPRODGROWTH(-2) 0.011363 0.076664 0.148216 0.8824
ERRORCORRECTION(-1) -0.006955 0.012012 -0.579046 0.5635
D_LNRPIINDEX -0.000622 0.000773 -0.805127 0.4221
D_LNRPIINDEX(-1) 0.000109 0.000750 0.145935 0.8842
D_LNRPIINDEX(-2) -0.001499 0.000712 -2.105376 0.0370
D_LNAVNOMWAGE(-1) 0.115093 0.085865 1.340398 0.1822
D_LNAVNOMWAGE(-2) 0.326917 0.086801 3.766279 0.0002

R-squared 0.716942 Mean dependent var 0.019605
Adjusted R-squared 0.693517 S.D. dependent var 0.016282
S.E. of regression 0.009014 Akaike info criterion -6.501394
Sum squared resid 0.011781 Schwarz criterion -6.249408
Log likelihood 526.6101 Hannan-Quinn criter. -6.399059
F-statistic 30.60522 Durbin-Watson stat 1.905223
Prob(F-statistic) 0.000000

To nearly all being significant:

Dependent Variable: D_LNAVNOMWAGE
Method: Least Squares
Date: 08/21/11 Time: 14:23
Sample (adjusted): 1971Q4 2011Q1
Included observations: 158 after adjustments
Weighting series: 1/SQR(WEIGHTSBEST)

Coefficient Std. Error t-Statistic Prob.

C -0.075206 0.008360 -8.996339 0.0000
CYCLICAL25600 0.063368 0.009651 6.565704 0.0000
CYCLICAL25600(-1) -0.143500 0.017232 -8.327359 0.0000
CYCLICAL25600(-2) 0.077920 0.008539 9.125722 0.0000
D_LNPRODGROWTH 1.412376 0.085823 16.45692 0.0000
D_LNPRODGROWTH(-1) 1.014297 0.130246 7.787525 0.0000
D_LNPRODGROWTH(-2) 0.847152 0.095818 8.841297 0.0000
ERRORCORRECTION(-1) -0.174817 0.036494 -4.790253 0.0000
D_LNRPIINDEX -0.005352 0.000843 -6.349094 0.0000
D_LNRPIINDEX(-1) -0.004186 0.000414 -10.11890 0.0000
D_LNRPIINDEX(-2) -0.001460 0.000673 -2.168971 0.0317
D_LNAVNOMWAGE(-1) -0.152448 0.089330 -1.706562 0.0900
D_LNAVNOMWAGE(-2) 0.474760 0.100039 4.745754 0.0000

Weighted Statistics

R-squared 0.999783 Mean dependent var 0.090531
Adjusted R-squared 0.999765 S.D. dependent var 0.846902
S.E. of regression 0.001693 Akaike info criterion -9.846125
Sum squared resid 0.000416 Schwarz criterion -9.594139
Log likelihood 790.8439 Hannan-Quinn criter. -9.743790
F-statistic 55611.29 Durbin-Watson stat 1.406656
Prob(F-statistic) 0.000000

Unweighted Statistics

R-squared -8.937302 Mean dependent var 0.019605
Adjusted R-squared -9.759699 S.D. dependent var 0.016282
S.E. of regression 0.053409 Sum squared resid 0.413609
Durbin-Watson stat 0.229395

But now I am worried about why my unweighted statistics R^2 come out as -8.937? Have I done something wrong here?

[SophieSmith]

In the help book it says: "Following estimation, the unweighted residuals are placed in the RESID series.
If the residual variance assumptions are correct, the weighted residuals should show no evidence
of heteroskedasticity. If the variance assumptions are correct, the unweighted residuals
should be heteroskedastic, with the reciprocal of the standard deviation of the residual at
each period being proportional to ." Where can i access the weitghed residuals so I can check this??