Probabilistic forecasts quantify the uncertainty associated with predictions about the future. They are useful in decision-making, and essential when the user’s objective is risk management, or optimisation with asymmetric cost functions. Probabilistic forecasts are widely utilised in finance and weather services, and increasingly by the energy industry, to name a few applications. The R package ProbCast provides a framework for producing probabilistic forecasts using a range of leading predictive models, plus visualisation, and evaluation of the resulting forecasts. It supports both parametric and nonparametric density forecasting, and high-dimensional dependency modelling based on Gaussian Copulas. ProbCast enables a simple workflow for common tasks associated with probabilistic forecasting, making leading methodologies more accessible then ever before. These features are described and then illustrated using an example from energy forecasting, and the first public release of the package itself accompanies this paper.
|Publication status||Accepted/In press - 11 May 2020|
|Event||PMAPS 2020 - the 16th International Conference on Probabilistic Methods Applied to Power Systems - Online, Liege, Belgium|
Duration: 18 Aug 2020 → 21 Aug 2020
|Conference||PMAPS 2020 - the 16th International Conference on Probabilistic Methods Applied to Power Systems|
|Period||18/08/20 → 21/08/20|
- probabilistic forecasting
- uncertainty quantification
Browell, J., & Gilbert, C. (Accepted/In press). ProbCast: Open-source production, evaluation and visualisation of probabilistic forecasts. Paper presented at PMAPS 2020 - the 16th International Conference on Probabilistic Methods Applied to Power Systems, Liege, Belgium.