Hello everyone, could anyone provide working documents on incorporating exogenous variables in hierarchical forecasting? Alternatively, any key points or guidance on how to work with them would also be appreciated.

Hi @Sai krishna Sirikonda 👋, Great question! The

HierarchicalForecast

library is primarily designed for *forecast reconciliation*—ensuring coherence across different levels of a hierarchy—rather than generating forecasts itself. To incorporate exogenous variables in hierarchical forecasting, here’s a step-by-step guide: Step-by-Step Guide 1. Prepare Your Data using `HierarchicalForecast`:

hierarchy_levels = [['top_level'],
                    ['top_level', 'middle_level'],
                    ['top_level', 'middle_level', 'bottom_level']]
Y_hier_df, S_df, tags = aggregate(df=bottom_df, spec=hierarchy_levels)

•

Y_hier_df

: DataFrame with your time series at all hierarchical levels. •

S_df

: Structural matrix encoding the aggregation constraints. •

tags

: Dictionary identifying the hierarchy levels of each series. 2. Generate Base Forecasts: Use any model that supports exogenous variables. Options include: • StatsForecast: e.g.,

AutoARIMA

• MLForecast: e.g.,

LightGBM

• NeuralForecast: e.g.,

NBEATSx

Make sure to include exogenous features (

X_df

) aligned with your target series. 3. Reconcile Forecasts: Use a reconciliation method from

HierarchicalForecast

, such as: •

bottom_up

•

top_down

•

min_trace

This ensures that forecasts across all levels of the hierarchy are coherent. Helpful Resources • HierarchicalForecast Tutorial • https://nixtlaverse.nixtla.io/hierarchicalforecast/examples/mlframeworksexample.html • Using Exogenous Variables ◦ StatsForecast: https://nixtlaverse.nixtla.io/statsforecast/docs/how-to-guides/exogenous.html ◦ MLForecast: https://nixtlaverse.nixtla.io/mlforecast/docs/how-to-guides/exogenous_features.html ◦ NeuralForecast: https://nixtlaverse.nixtla.io/neuralforecast/docs/capabilities/exogenous_variables.html

Thank you, Max 💯.