Nonparametric estimation of conditional quantiles
In the first part of this talk, we state sufficient conditions for asymptotic normality of convergent estimates of conditional quantiles irrespective of data dependence. We consider the case of an ?-mixing stationary process and a kernel estimate of the conditional quantiles. The theorem allows us to forecast a time series using a nonparametric estimate of the conditional median and to build confidence bands. We apply those results to a real example : forecast daily 03 concentration in Rennes, France.