Statistics, School of Mathematical and Physical Sciences, University of Newcastle, Newcastle, Australia.
We present a parametric regression method that allows great flexibility in the shape of the error distribution by modelling the errors with the generalised lambda distribution (gld). Our approach allows accurate estimates of 100(1-p)% prediction intervals in the face of severe departures from symmetry and with a wide variety of tailweights. The gld is a distribution that allows a great variety of shapes within one distributional form.
Parameter estimation is via the starship method. Estimates of sampling variation are obtained via a parametric bootstrap method.
We demonstrate the implementation of the method in an R package and illustrate its use with real datasets.
This presentation is based on a paper currently in preparation, and on a University of Newcastle Technical Report.
The method will implemented in the R package, gld (available soon, email for more details)
publication and conference paper details