Handling multicollinearity in quantile regression through the use of principal component regression

Abstract In many fields of applications, linear regression is the most widely used statistical method to analyze effect a set explanatory variables on response variable interest. Classical least squares focuses conditional mean response, while quantile extends view quantiles. Quantile very convenient, whereas classical parametric assumptions do not hold and/or when relevant information lies in tails and therefore interest modeling distribution at locations different from mean. A situation common applications presence strong correlations between predictors. This leads well-known problem collinearity. While effects collinearity estimates are well investigated, this case for estimates. paper aims explore regression. First, simulation study analyses concerning degrees various distributions. Then proposes using latent components as possible solution Finally, shows assessment quality service highly correlated