A Bayesian approach to the problem of  forecast reconciliation

Giorgio Corani (IDSIA Lugano, Svizzera)

Abstract: Often time series are organized into a hierarchy. For example, the total visitors of a country can be divided into regions and the visitors of each region can be further divided into sub-regions. Forecasts of hierarchical time series should be coherent; for instance, the sum of the forecasts of the different regions should equal the forecast for the total. The forecasts are incoherent if they do not satisfy such constraints.
Reconciliation methods proceed in two steps. First, base forecasts are computed by fitting an independent model to each time series and ignoring the sum constraints. Then, the base forecasts are adjusted to become coherent; this step is called reconciliation. Besides being coherent, reconciled forecasts are generally more accurate than the base forecasts, as they benefit from information coming from multiple time series.
In this talk, we show how to forecast reconciliation can be addressed by taking a Bayesian viewpoint.