Comparing coefficients between two samples

[Weaver102406]

I have partitioned my data according to the financial accounting standards used. The first regression will contain the years that the company used domestic standards. The second regression will contain the years the company used IFRS. Same company examined in both samples, but different years' data corresponding to stds used. Is there a way in Eviews to test to see if the difference in coefficient values on one of the variables between the two regressions is significant?

Thanks!

[startz]

Put all the data in one regression. Use a dummy variable for the sample interacted with the variables to distinguish the samples. For example, if D is true in the first sample

Code: Select all

ls y c d x1 x1*d x2 x2*d

Then test whether the coefficient on x2*d is significant.

[EViews Glenn]

On a related note, depending on the exact nature of the test that you want to perform, you could use the facbreak view of the equation. Note that unlike Startz's example where the coefficients on the constant and X are also regime specific but not tested, this form of the test will only allow the interactions on the variables to be evaluated. You could, of course, set up an equation which had those interactions and then test on the subset variables.

[Weaver102406]

I'm sorry, but I don't think that I explained what I needed very well in my first post. The paper that I am following (Hung and Subramanyam 2007) ran separate regressions - same model (price = a + book value + net income + e), but different data (one set when the firm used German accounting principles and one set when the same firm used international accounting principles). Then they used two-tailed t-tests to see if the difference between the coefficients on the intercept, book value and net income under each accounting stds was significant. The formulas for t-tests that I have found use the means of the variables and I want to test the difference in the regression coefficients. I'm confused as to how combining the two subsets and running the regression will give me what I am looking for. [As I'm sure you can see, statistics is not one of my areas of expertise. :-) ]