Time series forecasting is ubiquitous in the modern world. Applications range from health care, astronomy and include climate modelling, to financial trading and monitoring of critical engineering equipment. To offer robust value over this wide range of activities, models must not only provide accurate forecasts, but also quantify and adjust their uncertainty over time. In this work, we directly tackle both tasks with a novel, fully end-to-end deep learning method. By recasting time series prediction as an ordinal regression task, we develop a principled methodology to assess long-term predictive uncertainty and describe the rich, multi-modal, non-Gaussian behaviour which arises regularly in many problem domains.Notably, our framework is a wholly general-purpose approach that requires little to no user intervention to be used. We showcase this key feature in a large-scale benchmark test with 45 datasets drawn from both a wide range of realworld applications and a comprehensive list of synthetic data. This wide comparison uses, as benchmark methods, state-ofthe-art models from both Machine Learning and Statistics literature, such as the Gaussian Process. We find that our approach not only provides excellent predictive forecasts and associated uncertainty bounds, but also allows us to infer