GJR-GARCH in mean very large estimates

[Muhammed Shaker]

Hi Statlisters
I am using GJR-GARCH in mean for portfolios. each has 240 observations.
I got very high parameters like the following

ARCH family regression

Sample: 7 - 245, but with a gap Number of obs = 238
Distribution: Gaussian Wald chi2(4) = 33.22
Log pseudolikelihood = 249.9526 Prob > chi2 = 0.0000


Semirobust
hwt61 Coef. Std. Err. z P>z [95% Conf. Interval]

hwt61
hmktrf1 .5130338 .1149479 4.46 0.000 .2877401 .7383276
hsmb1 .285152 .1821982 1.57 0.118 -.0719498 .6422538
hhml1 .3326754 .2219941 1.50 0.134 -.1024251 .7677758
_cons 2.57485 6.568659 0.39 0.695 -10.29948 15.44919

ARCHM
sigma2 -351.6706 915.2793 -0.38 0.701 -2145.585 1442.244

ARCH
arch
L1. .0191179 .0522311 0.37 0.714 -.0832531 .121489

tarch
L1. -.0348283 .0925937 -0.38 0.707 -.2163087 .1466521

garch
L1. .7722554 .1027877 7.51 0.000 .5707952 .9737156

_cons .0016147 .0007723 2.09 0.037 .000101 .0031283

please look at the ARCH in mean factor. it is -351.67. kindly note this time series does not have unit root



I also used the following code

arch wt6 ktrf smb hml , arch(1/1) tarch(1/1) garch(1/1) archm vce(robust)

I doubt this finding
any help would be greatly appreciated

[Cyrus Levy]

There may be something wrong with one of your variables. Can you post a summary statistics table here?

summ wt6 ktrf smb hml

[Muhammed Shaker]

. Many thanks for your response
the following is the summary
sum wt6 mktrf smb hml

Variable Obs Mean Std. Dev. Min Max

wt6 241 .0602204 .1503794 -.4401423 .4390088
mktrf 240 .603 4.31711 -10.1 8.22
smb 239 .0006689 .0322044 -.135625 .0894906
hml 239 .001862 .0375217 -.154349 .1464016