Hierarchical Time Series Forecasting
Hierarchical time series forecasting addresses forecasting problems for time series with hierarchical aggregation structure, requiring forecasts at all levels to satisfy sum consistency constraints. It is widely applied in areas such as revenue forecasting and supply chain demand forecasting.
Timeline (last 90 days)
The core difficulty of hierarchical time series forecasting is that the forecasts at different aggregation levels must remain consistent (e.g., the forecast of a group's total revenue equals the sum of the forecasts of its subsidiaries).
Existing hierarchical forecasting approaches fall roughly into two routes: two-stage pipelines (independent forecasting followed by ex-post reconciliation) and end-to-end models.
Classical hierarchical time series forecasting is expressed with a summing constraint matrix S: the aggregate-level values of the bottom-level series y_t can be written as S y_t, and consistency requires the forecasts to lie in the column space of S.
All Facts (3)
The core difficulty of hierarchical time series forecasting is that the forecasts at different aggregation levels must remain consistent (e.g., the forecast of a group's total revenue equals the sum of the forecasts of its subsidiaries).
50%UnverifiedExisting hierarchical forecasting approaches fall roughly into two routes: two-stage pipelines (independent forecasting followed by ex-post reconciliation) and end-to-end models.
50%UnverifiedClassical hierarchical time series forecasting is expressed with a summing constraint matrix S: the aggregate-level values of the bottom-level series y_t can be written as S y_t, and consistency requires the forecasts to lie in the column space of S.
50%