Economic Implications of Enhanced Forecast Accuracy: The Case of Photovoltaic Feed-In Forecasts

Abstract

Forecasts are usually evaluated in terms of accuracy. With regard to application, the question arises if the most accurate forecast is also optimal in terms of forecast related costs and risks. Combining insights from research and practice, we show that this is indeed not necessarily the case. Our analysis is grounded in the dynamic field of short-term forecasting of solar electricity feed-in. A clear sky model is implemented and combined with a linear model, an autoregressive model, and an artificial neural network. These models are applied to a portfolio of ten large-scale photovoltaic systems in Germany. We compare the different models in order to quantify the connection between errors and costs. We find that apart from accuracy, correlation with market prices is an important characteristic of forecasts when economic implications are considered as important.