F-test / Wald Test discrepancy

For technical questions regarding estimation of single equations, systems, VARs, Factor analysis and State Space Models in EViews. General econometric questions and advice should go in the Econometric Discussions forum.

Moderators: EViews Gareth, EViews Moderator

phoenix1914
Posts: 7
Joined: Sat Nov 17, 2018 1:48 pm

F-test / Wald Test discrepancy

Postby phoenix1914 » Thu Dec 13, 2018 2:27 am

Hello all

I have one question regarding the Wald Test in Eviews. I have the following regression: Y = β1 + β2Χ2 + β3Χ3 + β4Χ4 + β5Χ5 + β6Χ6 +u , and 145 observations .

I want to test the hypothesis that β3=β4=β5=β6=0 .

When I do manual calculation of F statistic (for 4,139 df) my result is ~2.77 . I calculating it using the R-square from the unrestricted equation and the R-square from the restricted.

But when I use the built-in Wald Test in Eviews my result is 2.11 . Do you know why there is this discrepancy?
Image

Interestingly enough when I run a different regression (different sample and variables) and I test the hypothesis β2=β3=0 I get same results using the manual calculation and Wald Test.

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: F-test / Wald Test discrepancy

Postby startz » Thu Dec 13, 2018 6:39 am

You'd have to post both sets of information for anyone to compare them.

As a wild guess, the sample changed when you did it manually.

phoenix1914
Posts: 7
Joined: Sat Nov 17, 2018 1:48 pm

Re: F-test / Wald Test discrepancy

Postby phoenix1914 » Sat Dec 15, 2018 11:28 am

Thank you Startz. Let's ignore the second example I mentioned, since it might complicate the matters. I will provide more details on the issue I have.

In the first screenshot below I run a multiple regression, let's consider it as the unrestricted one. The r-square is 0.432632 . What I want then is to test the hypothesisβ3=β4=β5=β6=0 . One way is to run the Wald-test and the result of the F statistic is what I posted above.

Image

The second way is to calculate manually the following statistic F= [(Rsquare-unrestricted - Rsquare-restricted)/ (k-1)] / [Rsquare-unrestricted/(n-k)]

where Rsquare unrestricted can be found above. The r-square restricted can be obtained by running another regression keeping only β1 and β2 and can be found here, Rsquare = 0.398089.

Image


If I use the R-squared from the above outputs and use the manual formula, for 4,139 d.f., the result I obtain is 2.774573. Which is very different from the result of the Wald-test.

Let me know if you need further clarifications and many thanks in advance for any help!

phoenix1914
Posts: 7
Joined: Sat Nov 17, 2018 1:48 pm

Re: F-test / Wald Test discrepancy

Postby phoenix1914 » Sat Dec 15, 2018 12:17 pm

I found it..my manual formula had an error. The denominator is (1-R-square) and not purely R-square. :oops:

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: F-test / Wald Test discrepancy

Postby startz » Sat Dec 15, 2018 12:35 pm

I think your R^2 formula is wrong. Maybe the k-1 should be the number of restrictions, r


Return to “Estimation”

Who is online

Users browsing this forum: No registered users and 16 guests