A0899
Title: Quantile regression with Bernstein polynomials
Authors: Santiago Pereda-Fernandez - Universidad de Cantabria (Spain) [presenting]
Abstract: An alternative to quantile regression is proposed to estimate conditional quantiles. To do so, conditional quantiles are modeled using Bernstein polynomials, which are a nonparametric smoother related to series estimation. With this model, it is possible to write the conditional density function in terms of the Bernstein coefficients and the data, allowing for the use of maximum likelihood for estimation. Moreover, the estimator has several desirable features, including much fewer quantile crossings, no need to interpolate between quantiles in a grid, the estimated coefficients are differentiable, and simple functions of the coefficients are integrable with respect to the quantile index.
