After having done the heteroscedasticity test, and having confirmed its existence, I want to correct the model. To correct it, and proceed with the transformation of the data, I must identify the form of heteroskedasticity, that is to say how the sigma square error variance function is written. In the works, it is never mentioned how to identify the form of heteroskedasticity, they assume only a few forms.
can you give me an example of the type of regression to use?
In many cases it is not necessary to do transformations on the data itself if assumptions for OLS regression are not met. Most of the time (at least in my experience) moving to a generalized linear model often helps. What's most important is to understand the nature of your outcome variable, e.g. is the data continuous, strictly positive continuous, bounded between zero and one, binary, integers (counts), etc.
If none of this helps, then moving to robust regression or quantile regression may help (as suggested by @PeterFlom). Another option are non-linear regression or generalized additive models.
In the end it all comes down to what are you trying to model and from there identify techniques that are most appropriate.
