Pseudoprospective analyses of forecasters and backcasters really test the featurefulness, reliability, computational performance, reproducibility, and fit quality of quantile regression routines. In my testing:
Featurefulness: generally, quantgen + gurobi/glpk > quantreg + fn/br/etc.; main thing was noncross. Maybe some lasso lambda fitting and handling as well, though quantreg does have a method that allows for a lasso penalty.
Reliability-wise (in terms of successfully producing a fit vs. generating an error), quantreg with method="fn" (producing a fit in all cases I've tested) > quantreg with the default "br" method > glpk > gurobi.
Pseudoprospective analyses of forecasters and backcasters really test the featurefulness, reliability, computational performance, reproducibility, and fit quality of quantile regression routines. In my testing:
noncross. Maybe some lasso lambda fitting and handling as well, though quantreg does have amethodthat allows for a lasso penalty.quantregwithmethod="fn"(producing a fit in all cases I've tested) >quantregwith the default"br"method > glpk > gurobi.It be very cool to have a way in quantgen to use quantreg with a given
method(or just"fn") if it can handle the requested parameter settings.