Time series forecasting applies to a large variety of problems. In order to forecast future values of a time series, it is frequently more robust to use an estimator based on the median or, more generally, based on a quantile. In this talk, we develop strategies for non-parametric sequential quantile forecasting. We prove the convergence of our strategies under weak assumptions when considering an expert-aggregation strategy relying on Nearest Neighbors experts. To conclude, those strategies are empirically evaluated against real world data - a call center call volume data set.