How to estimate a mixture of quantile regressions in SAS 9.4?

[DaLack]

I am interested in estimating a mixture of quantile regressions similar to the mixture of linear regressions using proc fmm. I have checked the one-component-case in SAS (proc quantreg) but can't find any solution for the case of a mixture. A mixed model is not possible due to the unkown group membership of the observations.

Thanks in advance.

[Rick_SAS · Accepted solution]

Interesting idea. I've never heard of this before, but I see that there is a recently published (2016) article by Wu and Yao.

I can't think of any procedure that does this automatically. I'd guess you need to formulate it as an optimization problem. Given that it require optimization for the quantile regression AND optimization for the mixing parameters, I'd conjecture that you'd need to attack this with PROC IML, although maybe someone can show how to do it with PROC NLIN. Either way, I suspect this will require some challenging (but fun!) programming.

[DaLack]

Thanks for your answer. Yes, I have read this article and was wondering whether it is already available in SAS. But I agree, programming seems to be the only option to perform it...