These methods can be used, only when the residuals are normal and uncorrelated. Taylor
(2007) argued that the Gaussian assumption is unrealistic and shape of the distribution of sales
data are often not constant over time. Instead of point forecast based prediction intervals, he
suggested the use of quantiles in forecasting. The classical mean regression model describes
how mean of response variable ðyÞ changes with vector of covariates
(x) based on conditional mean function E(y). Koenker and Hallock