Testing multiple restrictions in system window

[Bigbrotherjx]

I would like to test the hypothesis that all coefficients are the same across equations in a system e.g. if I have:

y1=C(1) + C(2)x1 + C(3)z1

y2=C(4) + C(5)x2 + C(6)z2

I want to test C(1) = C(4), C(2) = C(5), C(3)=C(6) i.e. for panel structure (which has efficiency gains if the null is true I believe, assuming no co-dependence in the error structure).

Can I do this using the Wald Coefficient Restrictions window or using other parts of the output?

[EViews Gareth]

Yep, through the Wald Coefficient Restrictions window.

[Bigbrotherjx]

So I can just leave a space between each restriction then...I guess what I was doing wrong was hitting the enter key to separate out restrictions so I thought you couldn't enter more than one at a time.

By the way...can I enter the coefficient restrictions into word and then paste them into the Wald restrictions window? Every time I click something else, I lose the results and it is tedious to type everything in again.

[EViews Gareth]

Yes.

(By the way, wouldn't it just be easier to try these things rather than post here?! )

[startz]

Yes.

(By the way, wouldn't it just be easier to try these things rather than post here?! )

Except you need to separate by commas rather than spaces.

[Bigbrotherjx]

Sorry, very lazy of me

Perhaps a more reasonable question is:

When I get:

"Restriction variance cannot be computed. Restrictions may not be unique."

I'm not entirely sure what this means. My best guess is that I don't have enough degrees of freedom to test all the restrictions at the same time?

Ideally, an LR test of some sort would be better right? I appear to be able to estimate the system with the restrictions in place.

[Bigbrotherjx]

The other diagnostic I get is:

"Positive or non-negative argument to function expected"

[startz]

Remember that we can't see your output. You might want to post pictures of your estimates or equations or the restrictions window to see if it jogs loose any suggestions from forum readers.

[Bigbrotherjx]

I have something like:

System: LEVPIIGS
Estimation Method: Generalized Method of Moments
Date: 03/18/10 Time: 20:34
Sample: 2001M09 2009M03
Included observations: 93
Total system (balanced) observations 455
Identity matrix estimation weights - 2SLS coefs with GMM standard
errors
Kernel: Bartlett, Bandwidth: Fixed (3), No prewhitening
Convergence achieved after 60 iterations

Coefficient Std. Error t-Statistic Prob.

C(58) 52.77308 14.88774 3.544734 0.0004
C(59) 0.437210 0.169151 2.584726 0.0101
C(60) 7.768349 1.963168 3.957047 0.0001
C(61) -0.118378 0.175295 -0.675307 0.4999
C(62) -0.585736 0.220468 -2.656786 0.0082
C(63) 0.018482 0.138400 0.133542 0.8938
C(64) -0.281555 0.377189 -0.746456 0.4559
C(65) 145.4637 55.29076 2.630886 0.0089
C(66) -0.472808 0.246933 -1.914722 0.0563
C(67) 8.672818 3.433472 2.525961 0.0119
C(68) 8.259275 2.009164 4.110802 0.0000
C(69) -3.750367 1.588043 -2.361628 0.0187
C(70) -1.143084 0.465690 -2.454604 0.0145
C(71) -1.129445 3.436282 -0.328682 0.7426
C(72) 0.618872 0.182222 3.396245 0.0008
C(73) 23.94576 10.95180 2.186469 0.0294
C(74) 0.429730 0.085998 4.996994 0.0000
C(75) 6.070046 1.219489 4.977534 0.0000
C(76) 1.734919 0.332571 5.216685 0.0000
C(77) 0.233041 0.189330 1.230871 0.2191
C(78) 0.202925 0.080986 2.505673 0.0126
C(79) -1.620722 2.444077 -0.663123 0.5077
C(80) 58.23137 21.10695 2.758872 0.0061
C(81) -0.096662 0.122427 -0.789550 0.4303
C(82) 3.785627 1.371185 2.760844 0.0060
C(83) 2.583086 0.445734 5.795129 0.0000
C(84) 1.010292 0.331153 3.050831 0.0024
C(85) -1.400566 0.692290 -2.023093 0.0438
C(86) 1.178259 5.193363 0.226878 0.8206
C(87) 52.05620 9.951916 5.230772 0.0000
C(88) 0.252408 0.113176 2.230237 0.0263
C(89) 7.805538 0.921175 8.473462 0.0000
C(90) 0.094391 0.252193 0.374282 0.7084
C(91) -0.647692 0.187502 -3.454315 0.0006
C(92) 0.052455 0.070782 0.741078 0.4591
C(93) 1.760221 1.953911 0.900871 0.3682
C(94) 35.22975 68.92998 0.511095 0.6096
C(95) -0.634179 0.131175 -4.834611 0.0000
C(96) 5.129053 1.482788 3.459060 0.0006
C(97) 3.114738 1.269461 2.453591 0.0146
C(98) -0.408627 1.625074 -0.251451 0.8016
C(99) -0.390117 0.255830 -1.524910 0.1281
C(100) -0.370916 6.487373 -0.057175 0.9544
C(101) 0.657144 0.140135 4.689351 0.0000
C(117) 10.03534 2.904693 3.454871 0.0006
C(118) 0.630392 0.102592 6.144668 0.0000
C(119) 2.853615 0.971547 2.937188 0.0035
C(120) 0.274130 0.260057 1.054115 0.2925
C(121) 0.308724 0.153025 2.017481 0.0443
C(122) -0.067013 0.159157 -0.421052 0.6740
C(123) 0.111578 0.976660 0.114244 0.9091
C(124) 6.368405 14.11566 0.451159 0.6521
C(125) 0.018998 0.209797 0.090554 0.9279
C(126) 0.318870 1.602242 0.199015 0.8424
C(127) -2.209466 2.676024 -0.825653 0.4095
C(128) -0.038576 1.993564 -0.019350 0.9846
C(129) -0.415032 0.696885 -0.595554 0.5518
C(130) 5.257025 3.625288 1.450098 0.1479
C(131) 16.83561 4.393737 3.831729 0.0001
C(132) 0.443533 0.154509 2.870594 0.0043
C(133) 3.724165 1.402876 2.654664 0.0083
C(134) 1.318449 0.462429 2.851139 0.0046
C(135) 0.000587 0.031370 0.018700 0.9851
C(136) -0.422975 0.124523 -3.396753 0.0008
C(137) 0.392814 0.633177 0.620386 0.5354
C(138) 41.08149 26.38655 1.556910 0.1203
C(139) -0.643601 0.198281 -3.245901 0.0013
C(140) 3.628314 1.668935 2.174029 0.0303
C(141) 0.780240 1.393965 0.559727 0.5760
C(142) 1.163556 0.944952 1.231339 0.2190
C(143) 0.075894 0.592469 0.128097 0.8981
C(144) 7.851271 7.272398 1.079599 0.2810
C(145) -0.491028 0.195116 -2.516598 0.0123

Determinant residual covariance 2034.899
J-statistic 467.0650

And I'm imposing:

C(58)=C(73)=C(87)=C(117)=C(131),
C(59)=C(74)=C(88)=C(118)=C(132),
C(60)=C(75)=C(89)=C(119)=C(133),
C(61)=C(76)=C(90)=C(120)=C(134),
C(62)=C(77)=C(91)=C(121)=C(135),
C(63)=C(78)=C(92)=C(122)=C(136),
C(64)=C(79)=C(93)=C(123)=C(137),
C(65)=C(80)=C(94)=C(124)=C(138),
C(66)=C(81)=C(95)=C(125)=C(139),
C(67)=C(82)=C(96)=C(126)=C(140),
C(68)=C(83)=C(97)=C(127)=C(141),
C(69)=C(84)=C(98)=C(128)=C(142),
C(70)=C(85)=C(99)=C(129)=C(143),
C(71)=C(86)=C(100)=C(130)=C(144)

It works OK if I leave out some of the restrictions, but messes up if I include all.

A sample of one of the equations would be like:

gresprsa = C(58)+C(59)*greSPRSA(-1)+C(60)*PCSA+C(61)*greB+C(62)*greD+C(63)*greF+C(64)*greBA+C(65)*CRI+C(66)*CRI*greSPRSA(-1)+C(67)*CRI*PCSA+C(68)*CRI*greB+C(69)*CRI*greD+C(70)*CRI*greF+C(71)*CRI*greBA+[ar(1)=C(72)] @ greSPRSA(-1) PCSA greB2 greD greF greBA CRI CRI*greSPRSA(-1) CRI*PCSA CRI*greB CRI*greD CRI*greF CRI*greBA

[startz]

Now that I can see what you're doing, I see why you're puzzled. It looks okay to me.

[Bigbrotherjx]

Is a Wald test the only way to proceed here?

Are there any LR/LM tests that use other bits of the output? I do seem to be able to estimate the restricted specification even though I can't test for it outright using Wald.