CFE 2019: Start Registration
View Submission - CMStatistics
Title: Parametric modelling of quantile functions Authors:  Paolo Frumento - Karolinska Institute (Sweden) [presenting]
Matteo Bottai - Karolinska Institute (Sweden)
Abstract: Quantiles can be used to describe complexity and diversity, and to capture features beyond location and scale parameters. In traditional quantile regression, no parametric structure is imposed and different quantiles are estimated one at a time. While this is generally seen as an advantage, it also presents a number of important drawbacks: (i) it is statistically inefficient; (ii) it increases the chance of quantile crossing; (iii) it generates a large amount of information, making it difficult to summarise and interpret the results; (iv) it does not allow using identifying assumptions; and (v) it makes it difficult to apply quantile regression to censored, truncated, or longitudinal data. Describing the quantile function by a parsimonious parametric model represents an obvious, yet unexplored alternative. A new, broad family of models and estimators has been designed and implemented. An R package 'qrcm' provides all the necessary functions for inference, plotting, prediction, and goodness-of-fit assessment.