Quantile regression - error message for Q(1)

[Pieter vdspuy]

Dear group

This is my first time that I post a question.

Command: qreg Y1 X1 X2 X3, quantile (?)

I run the command for quantile regression above on cross-sectional data. This yields results for quantile (0.1) and upwards to quantile(0.9), but when I specify the 100^th quantile (1) -then I get the following error-warning: “VCE computation failed; try a different bandwidth or bsqreg”

Could someone perhaps tell me why?
Thank you so much
regards
Pieter

[Sven-Kristjan Bormann]

Quantiles can not be 0 or 1. With the 100th quantile you have all observations within this quantile and no observations above this quantile. Hence it is not possible to calculate the quantile regression. Maybe you meant the 99th quantile instead.

[FernandoRios]

In addition to Sven comments,
the command qreg tries to be smar on this regard. when you type
qreg y1 x1 x2 x3, quantile(1)
it does not try to estimate the 100th quantile regression (since that is not possible), but estimate the 1st quantile. Some other community-contributed commands have adopted only the first syntax (using any number q such that 0<q<1), whereas others use q with 0<q<100.
Now, the reason why the command fails is because standard errors are a function that is inversely related to the density at the qth quantile. If you use a quantile that is too large (say 99 or higher) or too small (1 or smaller), the density at that point is almost zero, and the VCE cannot be estimated.
HTH