« LM Tests for Heteroskedasticity | Main | Homework 3 »

Monday, January 17, 2011



1. White-test(nonparametric test)
a. Does not rely on a specific form of heterosk. If you know the exact specification of the LM test and GQtest, parametric test makes things better. But if you do not know the specific specification, nonparametric makes better. You always want to do white test to check the specification correctness. (if two types of test get the same answer, that is fine; if dif answer, you need to check)
b. Closely related to model a. It is a large sample LM test
c. Does not require normality of errors. So it is a more robust test.
2. Estimation
a. White’s correction: white’s hetero consistant covariance matrix (HCCM)
b. Generalized or weighted least squares(GLS estimator, WLS estimator)
i. Trick: divide the original model both sides ∂i, to make the original heterosk not blue model to be homo blue model. Bonus: the reason constant does not change the variance is that constant only shift the distri. Not twist the spread of the distri.
ii. So all next we need to do is to find ∂i.
1. Linear algebra, vector, matrix algebra, transpose(vertical and horizontal), variance-covariance matrix, diagonal element, positive definite matrix, eigen value, identity matrix
iii. Feasible generalized least square(FGLS)
1. Steps
2. Problems: no guarantee ∂i>0, since it is form estimation. It might be negative. We need to take the absolute value.

The comments to this entry are closed.