Time series and forecasting · insight-0215
Reconcile hierarchical forecasts so route, region and network numbers agree
- Evidence type
- documented fact
- Difficulty
- medium
- Roles
- Data scientist, Data analyst
- Industries
- aviation, retail, logistics
Summary: situation, action, reason, boundary
Situation
Route-level and network-level demand forecasts are produced separately, and the routes do not add up to the network total.
Action
Generate base forecasts at all levels, then apply a reconciliation method (bottom-up as a baseline, MinT when you can estimate error covariances) and evaluate accuracy at every level after reconciliation.
Reason
Independently produced forecasts are generally not coherent. Reconciliation enforces the aggregation constraints, and optimal combination methods can also improve accuracy.
Boundary
Product, geography, channel or route hierarchies; grouped series.
Symptom or risk
Planning and finance use different numbers, meetings argue about whose forecast is right, and nobody trusts either.
Exceptions
If only one level is ever used for decisions, coherence matters less than accuracy at that level.
Trade-offs
MinT needs a reliable covariance estimate; with many series, use shrinkage or simpler weights.
Pitfalls
Reconciling point forecasts but leaving intervals unreconciled.
Lessons and concepts
Claims and their sources
Base forecasts produced independently for each series in a hierarchy are generally not coherent; reconciliation methods make them add up.
MinT reconciliation minimises the trace of the reconciled forecast-error covariance matrix.