1

I have just learned about the statsmodels module by searching about quantile regression. I was wondering if there is a way to use statsmodels for quantile regressoin and take the errorbars of the data into account, like in sklearn.linear_model.LinearRegression?

http://scikit-learn.org/stable/modules/generated/sklearn.linear_model.LinearRegression.html

Thanks for looking!

vestland
  • 55,229
  • 37
  • 187
  • 305
astromath
  • 302
  • 1
  • 3
  • 13
  • I'm not sure I understand what you are talking about, but statsmodels' quantile regression does not allow for any kind of weights. – Josef Feb 02 '17 at 02:04
  • Thank you very much for your answer! I've been reading in the Cade et al. 1999 paper that there are some weights taken into account, depending on the residuals being ether positive or negative. Is this why statsmodels' quantile regression does not allow for any kind of weights? Again, thank you very much.I enjoy practising on Statsmodels. – astromath Feb 02 '17 at 21:05
  • Weights are used internally as part of the optimization algorithm. But they are not the reason that user specified weights are not supported. The main reason for that is that `weights` are a difficult topic, e.g. https://github.com/statsmodels/statsmodels/issues/2848 For example, I don't know how variance weights affect the quantile regression results. Just one detail: the standard errors for the estimated parameters are based on a local kernel density estimate, and I don't know how that would be affected by heteroscedasticity. – Josef Feb 02 '17 at 22:54

0 Answers0