Hi everyone
Dear friends
I have a panel data model with heteroskedasticity, I know one way for overcoming this problem is to use FGLS method. I've searched a lot, but i don't get how this method can eliminate heteroskedasticity.
Can someone please explain me in simple way, how this method works?
If you give me reference of what you explain, I'll be thankful.
regards.
This is not the easiest way at all. It is much more convenient to use robust standard errors, which any software can implement very easily (e.g., Stata, R...).
Here, you have a way of implementing FGLS in Stata, jountly with a brief explanation.
http://www3.grips.ac.jp/~yamanota/Lecture_Note_10_GLS_WLS_FGLS.pdf
Dear Sajjad,
I have to say here that I have adopted quite a number of methods in empirical analysis, but I v not seriously looked at FGLS techique.
However, I feel I could make a little contribution. I think your interest should not be into looking at how FGLS controls for heteroscedasticity, however, I feel you should understand the FGLS and provide enough background information as to how it is an improvement on previous methods.
I feel the model implicitly controls for heteroscedasticity. I feel strongly that looking through the model you may come up with some interesting things.
Why would you want to eliminate the heteroscedasticity - it may be substantively very interesting!
see
Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
and
Article Fixed and Random effects models: making an informed choice
Dear Sajjad Sargolzaei ,
While some special attention needs to be equally given to panel dimensions N and T sizes, the feasible generalized least squares (FGLS) estimator is more efficient than the ordinary least squares (OLS) in the presence of heteroskedasticity, as well as cross sectional and serial correlation. To deal with these problems (heteroskedasticity, and cross sectional and serial correlations) in the error terms of a panel regression model, the common two approaches used are the:
(1). Ordinary least squares (OLS) estimator with a robust standard error that is robust to heteroskedasticity and correlations (White, 1980; Liang & Zeger, 1986; Newey & West, 1987; Arellano, 1987; Driscoll & Kraay, 1998; Hansen, 2007; Vogelsang, 2012; and Bai, Choi & Liao, 2020), and
(2). Generalized least squares estimator (GLS) that directly takes into account heteroskedasticity (as well as cross-sectional and serial correlations in the estimation) – GLS estimator regarded as more efficient than the OLS estimator.
I agree that applying robust standard errors or implementing GLS estimation are two common approaches of dealing with heteroscedasticity.
As a third option, in case the issue is rooted in a skewed outcome variable, non-linear transformation might offer a solution. Depending on the form of the heteroscedasticity, an appropriate transformation (e.g., logarithmic, cubic, square root) can sometimes stabilize the error variance.
However, depending on the data, certain transformations may fail to improve heteroscedasticity. In fact, they can even worsen it (Glen B. offers a good illustration of this issue here: https://stats.stackexchange.com/questions/336315/will-log-transformation-always-mitigate-heteroskedasticity).
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics