Concepts#
This section explains the core abstractions in macroforecast, one page per
workflow stage. Each page describes the abstraction in plain language, shows the
key public callables, and links to the corresponding reference documentation.
Workflow Stages#
Data
Loading FRED-MD, FRED-QD, FRED-SD, and custom panels into a DataBundle.
Preprocessing
Stationarity transforms, outlier rules, EM imputation, and standardization.
Features
F / X / MARX / MAF / Level feature families and the FeatureSpec abstraction.
Models and Arms
ModelSpec, Arm, and how one arm becomes one contender in the evaluation.
Windows
Expanding, rolling, and no-validation windows; retrain and retune cadence.
Running
run and run_pipeline: direct vs path-average forecast policy.
Evaluation
RMSE, relative MSE, relative RMSE, DM/CW tests, and the Model Confidence Set.