Quantile Regression with statsmodels 

Thank you very much , you helped me a lot ! very useful .

Thanks, that helped a lot :)

Doing data.plot.scatter(x="income", y="foodexp") early on would make it obvious that you need to take the log of both variables first. When you do log vs log, you'll get quantile estimates that *do* mostly fall within the OLS confidence band. I guess that's happening because you have a very heteroskedastic dataset, where e = kf(x) approximately: obviously the quantile slopes will vary!

can I just ask how the code of visualizing the results part would be if you had 2 or more independent variables? thank you in advance!