Forecast Reconciliation is a a post-forecasting process aimed to improve the accuracy and align forecasts for a system of linearly constrained (e.g. hierarchical/grouped) time series. The FoReco package provides a comprehensive set of classical (bottom-up, top-down and middle-out), and modern (optimal and heuristic combination) forecast reconciliation procedures in different frameworks including cross-sectional, temporal, or cross-temporal settings.
The core functions for reconciliation categorized by framework are as follows:
| Reconciliation | Cross-sectional | Temporal | Cross-Temporal |
|---|---|---|---|
| Classical reconciliation | |||
Top-down: *td() |
cstd() |
tetd() |
cttd() |
Bottom-up: *bu() |
csbu() |
tebu() |
ctbu() |
Middle-out: *mo() |
csmo() |
temo() |
ctmo() |
| Regression‑based reconciliation | |||
Least squares: *rec() |
csrec() |
terec() |
ctrec() |
LCC: *lcc() |
cslcc() |
telcc() |
ctlcc() |
Additionally, FoReco provides various functions for different aspects of forecast reconciliation, including aggregating time series, balancing hierarchies, computing projection and covariance matrices, and more.
You can install the stable version on R CRAN
install.packages("FoReco")You can also install the development version from Github
# install.packages("devtools")
devtools::install_github("daniGiro/FoReco")To get started with using the FoReco package, refer to the documentation for detailed information on how to apply different forecast reconciliation procedures to your data.
If you encounter any bugs or have suggestions for improvements, please feel free to report them on GitHub Issues page. Contributions are also welcome!