ResidualBoostingForecaster

Residual boosting forecast fitting one forecaster on residuals of another.

Residual boosting can be used for:

• improving forecasts from one forecaster with another, by using either as base_forecaster or residual_forecaster

• adding exogenous capability to a forecaster, by using it as residual_forecaster, and fitting it on the residuals of an exogenous capable base_forecaster

• adding probabilistic forecasting capability to a forecaster, by using it as base_forecaster, and adding probability forecasts from a probabilistic forecaster used as residual_forecaster

In fit: fits base_forecaster to y and X, computes in-sample residuals, and fits residual_forecaster to the residuals and X.

In predict, it predicts with both base_forecaster and residual_forecaster, and returns the sum of the two.

Probabilistic forecasts are obtained by shifting quantiles of the residuals forecast by residual_forecaster by the point forecast of the base_forecaster. This requires residual_forecaster to support probabilistic forecasts, but not base_forecaster.