Accurate medium-term wind power forecasting in a censored classification framework

We provide a wind power forecasting methodology that exploits many of the actual data's statistical features, in particular both-sided censoring. While other tools ignore many of the important stylized facts or provide forecasts for short-term horizons only, our approach focuses on medium-term forecasts, which are especially necessary for practitioners in the forward electricity markets of many power trading places; for example, NASDAQ OMX Commodities (formerly Nord Pool OMX Commodities) in northern Europe. We show that our model produces turbine-specific forecasts that are significantly more accurate in comparison to established benchmark models and present an application that illustrates the financial impact of more accurate forecasts obtained using our methodology.