-
Notifications
You must be signed in to change notification settings - Fork 2.9k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ENH: heteroscedasticity, variance function with quantile regression - nonlinearity #3302
Labels
Comments
parking a script, AFAIR, I got quite large quantile crossings in the center for some parameters, all quantiles are bunched together in this area but overall shape of the regression curve is dominated by range with large heteroscedasticity.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
(another random thought, a quick google search shows that there is some information, but I haven't read anything.)
In the illustrative quantile regression example, I used heteroscedasticity to get quadratic quantile functions. In general, outside of i.i.d. or non-normal (?) samples, the quantile function might be a non-linear function of the parameters. We can handle this, for example, with splines, however splines might run easily into crossing problems (example ?). If one source of the nonlinearity of the quantile function is in systematic, exog dependent heteroscedasticity, then a pre-weighting as in HETWLS might remove most of the non-linearity.
Question: Can we add a HetQuantileRegression that does the variance scaling before computing the Quantile Regression and then move back to original space? (pre-whiten and recolor add variance function scaling to objective function. (viewed as an M-estimator with scaling, analog to RLM)
aside: RLM doesn't allow for prior variance functions or heteroscedasticity either.
The text was updated successfully, but these errors were encountered: