Statistical differences between models

[Movid]

Hi everybody:

I have estimated equation y c b1_X1 b2_X2 b3_X3 b4_X4 b5_X5 for my entire dataset but also (separately) for three subsets, based on region (North America, UK, and Continental Europe). Obviously the coefficients of the explanatory variables differ for each region but I want to find out if they differ significantly. How do I do this in Eviews?

I’m an Eviews rookie so the solution may be pretty simple.
Thanks in advance!

[startz]

Hi everybody:

I have estimated equation y c b1_X1 b2_X2 b3_X3 b4_X4 b5_X5 for my entire dataset but also (separately) for three subsets, based on region (North America, UK, and Continental Europe). Obviously the coefficients of the explanatory variables differ for each region but I want to find out if they differ significantly. How do I do this in Eviews?

I’m an Eviews rookie so the solution may be pretty simple.
Thanks in advance!

You can use the Chow breakpoint test under View/Stability tests.

[EViews Gareth]

A simple method would be to estimate over the entire sample, then use the Factor Breakpoint Test (under View->Coefficient Tests) to test whether there are differences between the sub-samples. Simply enter the series that contains the region data as the factor

[Movid]

You can use the Chow breakpoint test under View/Stability tests.

I was under the impression that the Chow test compares the coefficients of the subsets with the entire set, whereas I want to test for differences with one another. This is also possible using Chow?

[Movid]

A simple method would be to estimate over the entire sample, then use the Factor Breakpoint Test (under View->Coefficient Tests) to test whether there are differences between the sub-samples. Simply enter the series that contains the region data as the factor

I don't see that test, I have Eviews 5.0

[EViews Gareth]

Then you can use dummy variables to perform the test manually - simply create a dummy variable for each region and include it in the estimation.


Also, please post in the EViews 5 forum in the future.

[startz]

You can use the Chow breakpoint test under View/Stability tests.

I was under the impression that the Chow test compares the coefficients of the subsets with the entire set, whereas I want to test for differences with one another. This is also possible using Chow?

It comes to the same thing. Suppose the coefficients in the subsets are c1 c2 and c3. You want to test c1=c2=c3. If they are all the same, then they're the same as in the entire set.

[Movid]

Then you can use dummy variables to perform the test manually - simply create a dummy variable for each region and include it in the estimation.


Also, please post in the EViews 5 forum in the future.

I'm sorry, my bad.

Dummies can take values 0 and 1, how then will I be able to test for significant differences between the coefficients of three regions?

[EViews Gareth]

Use more than one dummy.

[Movid]

Use more than one dummy.

Hi I've been off for a week.
Let me explain my problem in more detail.
I already have estimation results from my three subsets. Each subset's coefficients differ from the other subsets, but now I need to find out whether these differences are in fact significant, i.e. the coefficient of explanatory variable A in the UK is significantly different from variable A in the USA and Continental Europe etc. I have 5 explanatory variables.

If I were to use dummies, does this mean I have to create three dummies (i.e. UK/USA/EU) for each variable? And if so, how should I interpret the estimation results? Is this really necessary when I already have all my results, but just not whether the differences are significant?

I now this might be slightly beyond the purpose of this forum but I'm not able to solve this problem by consulting my book on statistics.
Really appreciate your replies!

David

[startz]

Use more than one dummy.

Hi I've been off for a week.
Let me explain my problem in more detail.
I already have estimation results from my three subsets. Each subset's coefficients differ from the other subsets, but now I need to find out whether these differences are in fact significant, i.e. the coefficient of explanatory variable A in the UK is significantly different from variable A in the USA and Continental Europe etc. I have 5 explanatory variables.

If I were to use dummies, does this mean I have to create three dummies (i.e. UK/USA/EU) for each variable? And if so, how should I interpret the estimation results? Is this really necessary when I already have all my results, but just not whether the differences are significant?

I now this might be slightly beyond the purpose of this forum but I'm not able to solve this problem by consulting my book on statistics.
Really appreciate your replies!

David

You should be able to find out how to do this by looking up "Chow test" in a good econometrics text. Briefly.

Add together the sum of square residuals from all three separate regressions. Let's call this SSR. Then estimate all the data together in a single regression and call the sum of square residuals SSR*. Since you have six estimated parameters, if you have n observations then

F= [(SSR*-SSR)/((3-1)*6)]/[SSR/(n-3*6)] will be distributed F(12,n-18)

[Movid]

You should be able to find out how to do this by looking up "Chow test" in a good econometrics text. Briefly.

Add together the sum of square residuals from all three separate regressions. Let's call this SSR. Then estimate all the data together in a single regression and call the sum of square residuals SSR*. Since you have six estimated parameters, if you have n observations then

F= [(SSR*-SSR)/((3-1)*6)]/[SSR/(n-3*6)] will be distributed F(12,n-18)

Thanks for your help. I figured this out as well, but I don't think it entirely solves my problem. I want to be able to make statements like variable A does not differ significantly between the UK and USA, but the difference between Continental Europe and the UK is significant. Variable B in the UK differs significantly from its Continental European counterpart, but not from the USA etc. I don't think I will be able to do this using Chow, because Chow only indicates whether the subsets differ from the entire set, but not from one another, right? At least not to the extent that A1 and A2 differ significantly, but A1 and A3 not. And that is what I'm trying to find out...

Again thanks for your input.

[startz]

You should be able to find out how to do this by looking up "Chow test" in a good econometrics text. Briefly.

Add together the sum of square residuals from all three separate regressions. Let's call this SSR. Then estimate all the data together in a single regression and call the sum of square residuals SSR*. Since you have six estimated parameters, if you have n observations then

F= [(SSR*-SSR)/((3-1)*6)]/[SSR/(n-3*6)] will be distributed F(12,n-18)

Thanks for your help. I figured this out as well, but I don't think it entirely solves my problem. I want to be able to make statements like variable A does not differ significantly between the UK and USA, but the difference between Continental Europe and the UK is significant. Variable B in the UK differs significantly from its Continental European counterpart, but not from the USA etc. I don't think I will be able to do this using Chow, because Chow only indicates whether the subsets differ from the entire set, but not from one another, right? At least not to the extent that A1 and A2 differ significantly, but A1 and A3 not. And that is what I'm trying to find out...

Again thanks for your input.

Following along with Gareth's advice then, create three dummy variables. Then multiply these dummies by the real variables. As an example,

Code: Select all

ls y DUK DUS DEU DUK*X DUS*X DEU*X

[Movid]

Code: Select all

ls y DUK DUS DEU DUK*X DUS*X DEU*X

Thanks for your suggestion. When I enter equation: y c DUK DUS DEU DUK*X1 DUS*X1 DEU*X1 and so on for al five variables I get the error near singular matrix, also without entering the constant. This is probably because all dummies are mutually exclusive.
Starting to get a bit desperate here as I do not seem to be able to solve this problem that shouldn't be so difficult.

[EViews Gareth]

You're probably falling into the dummy variable trap. I'm sure if you post your workfile, including the equation, someone can take a look at tell you where you've gone wrong.