I have run 10 regressions using same model for a sample by diving the the sample on the basis of some criteria.
The problem I am facing is that I can test the intercepts for individual sample regression models but unable to conclude whether the intercepts of the model are jointly equal to zero.
Please help.
I would take the following path using a single regression with dummies. Let's say you have "two regressions" (you can then generalise). You pool all the observations.
y = a1 + d2 + bX + e
d2 is a dummy variable that takes the value 0 for group 1 and the value 1 for group 2. The intercept for group 1 is a1. The intercept for ggroup two a1 + d2. Therefore, you can test if both intercepts are equal if d2 = 0 (a simple t-test).
With 3 groups
y = a1 + d2 + d3 + bX +e
You would test d2=d3=0, which is also quite easy to do in any software (e.g., Stata).
Furthermore, you can allow "b" to vary within samples (instead of impossing a common structure interacting the dummies with the X).
I hope it helps.
José-Ignacio
You can test whether the intercepts from several regressions are jointly equal to zero. This is a test that is quite common in empirical asset pricing where one runs several time series regressions (in each regression excess returns of a "test asset" are regressed on a set of factors) and then tests whether the intercepts are jointly equal to zero, usually using the Gibbons, Ross and Shanken (1989) "GRS" test. You can find a description of the methodology in chapter 5 of Campbell, Lo and MacKinlay's book "The Econometrics of Financial Markets".
If you're using Eviews, set up a system object and specify all your equations and then run the analysis using ols or any other option. After the regression analysis, you pick out the intercept coefficients for each equation and conduct the Wald test.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression