PostForecasts.jl

Julia package for postprocessing forecasts

PostForecasts.jl provides structures and functions that allow to easily postprocess point forecasts into predictive distributions. Postprocessing methods use only the past performance of a given point forecasting model (or ensemble of models) to build probabilistic forecasts conditional on point predictions.

PostForecasts.jl:

• Relies on deterministic models for reliable and repeatable results

• Does not require hyperparameter tuning

• Leverages forecaster diversity via averaging

We believe that following these three principles allowed us to develop a robust tool for computing probabilistic forecasts that combines ease of use, high accuracy, fast results and good interpretability. This makes PostForecasts.jl an attractive choice for both academic and industrial applications.

Quick start

Dedicated types

PostForecasts.jl introduces PointForecasts and QuantForecasts types for storing time series data along with point and probabilistic forecasts respectively.

Postprocessing models

The package provides interface to four selected models for probabilistic forecasting:

• Normal: Normal error distribution
• CP: Conformal Prediction (and Historical Simulation)
• IDR: Isotonic Distributional Regression
• QR: Quantile Regression

All the models belong to the PostModel supertype. They have corresponding train(m, X, Y) methods for calibrating a model m to inputs X and targets Y, and predict(m, input, quantiles) methods that yield the specified quantiles of the predictive distribution from m conditional on input.

Easy postprocessing

The core functionality of PostForecasts.jl is building probabilistic forecasts from point predictions. For easy postprocessing, use point2quant function to turn PointForecasts into QuantForecasts with one of the implemented models.

In addition to methods for postprocessing point forecasts, the package provides conformalize function, allowing to correct quantile forecasts using historical errors.

Installation

PostForecasts.jl is a registered Julia package, to install it just run the following lines in the Julia REPL:

julia> using Pkg
julia> Pkg.add("PostForecasts")

Alternatively, you can use the Pkg REPL mode:

pkg> add PostForecasts

Citing

Check out the original software publication on PostForecasts.jl in SoftwareX. If you use the package in your research, then please cite it as:

@article{lipiecki:weron:2025,
    title = {PostForecasts.jl: A Julia package for probabilistic forecasting by postprocessing point predictions},
    journal = {SoftwareX},
    volume = {31},
    pages = {102200},
    year = {2025},
    issn = {2352-7110},
    doi = {https://doi.org/10.1016/j.softx.2025.102200},
    author = {Arkadiusz Lipiecki and Rafał Weron}
}

Research papers

Below is the list of research papers using PostForecasts.jl:

• A. Lipiecki, B. Uniejewski & R. Weron Postprocessing of point predictions for probabilistic forecasting of day-ahead electricity prices: The benefits of using isotonic distributional regression Energy Economics, 139 (2024) 107934