forecast_combo¶

Purpose¶

forecast_combo combines point forecasts from multiple sources. It copies a ForecastData instance from forecast_evaluation, estimates weights over real-time vintages, stores combined forecasts in the copy, and supports optional plots and Shiny dashboards.

Features¶

• Equal-weight, error-based, and regression-based forecast combinations.
• Rolling windows, exponential discounting, and period filters.
• Hierarchical combinations built from nested ComboSpec objects.
• Partial-source handling when a model is unavailable for a target or horizon.
• Weight visualisations and dashboards for combined forecasts.
• Outturn-maturity controls for vintage-aware estimation.

ForecastCombo copies the supplied ForecastData before fitting, so combined forecasts do not alter the caller's data. It stores the results in combo.forecast_data and makes them available to the evaluation workflow.

Quick start¶

import forecast_evaluation as fe
import forecast_combo as fc

data = fe.ForecastData(load_fer=True)
combo = fc.ForecastCombo(forecast_data=data)

combo.fit(
    sources=["mpr", "compass conditional", "bvar unconditional"],
    variables=["gdpkp", "cpisa"],
    method=["average", "rmse", "constrained_least_squares"],
    training_start="2016-01-01",
    metric="pop",
)

# Visualise weights and launch the joint forecast dashboard
fc.heatmap_by_vintage(combo.weights, method="rmse", variable="gdpkp")
combo.run_forecast_dashboard()

Repository¶

Read the implementation and full API reference in the forecast-combo repository.