Conditional Spatial Quantile: Characterization and Nonparametric Estimation

Conditional quantiles are required in various economic, biomedical or industrial problems. Lack of objective basis for ordering multivariate observations is a major problem in extending the notion of quantiles or conditional quantiles (also called regression quantiles) in a multidimensional setting. We first recall some characterizations of the unconditional spatial quantiles and the corresponding estimators. Then, we consider the conditional case. In this work, we focus our study on the geometric (or spatial) notion of quantiles introduced by Chaudhuri (1992a, 1996). We generalize, in the conditional framework, the Theorem 2.1.2 of Chaudhuri (1996), and we present algorithms allowing the calculation of the unconditional and conditional spatial quantile estimators. Finally, these various concepts are illustrated using simulated data.