Optimal combination forecasts for hierarchical time series
By Rob J Hyndman, R. Ahmed, George Athanasopoulos, Han Lin Shang. 2011
543 citations recorded by the source corpus.
In many applications, there are multiple time series that are hierarchically organized and can be aggregated at several different levels in groups based on products, geography or some other features. We call these ''hierarchical time series''. They are commonly forecast using either a ''bottom-up'' or a ''top-down'' method. In this paper we propose a new approach to hierarchical forecasting which provides optimal forecasts that are better than forecasts produced by either a top-down or a bottom-up approach. Our method is based on independently forecasting all series at all levels of the hierarchy and then using a regression model to optimally combine and reconcile these forecasts. The resulting revised forecasts add up appropriately across the hierarchy, are unbiased and have minimum variance amongst all combination forecasts under some simple assumptions. We show in a simulation study that our method performs well compared to the top-down approach and the bottom-up method. We demonstrate our proposed method by forecasting Australian tourism demand where the data are disaggregated by purpose of travel and geographical region.
Subfields: Management Science and Operations Research, Signal Processing, Time-Series & Forecasting
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Management Science and Operations Research
- Rob J Hyndman
- Fotios Petropoulos
- Konstantinos Nikolopoulos
- Nikolaos Kourentzes
- Vassilios Assimakopoulos
- Spyros Makridakis
- Christoph Bergmeir
- Evangelos Spiliotis
In Signal Processing
- Rob J Hyndman
- Vassilios Assimakopoulos
- Fotios Petropoulos
- Evangelos Spiliotis
- Spyros Makridakis
- Konstantinos Nikolopoulos
- George Athanasopoulos
- Tim Januschowski
In Time-Series & Forecasting
- Fotios Petropoulos
- Rob J Hyndman
- Konstantinos Nikolopoulos
- Christoph Bergmeir
- Jan Gasthaus
- George Athanasopoulos
- Tim Januschowski
- Spyros Makridakis
Citations in the corpus
- DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks (2017, background): “To prevent confusion we avoid the ambiguous terms “past” and “future” and will refer to time ranges [1 , t 0 − 1] and [ t 0 , T ] as the conditioning range and prediction range, respectively.”
- Forecasting with temporal hierarchies (2017, background): “Finally Section 7 demonstrates their use and benefits in a case study of Accident & Emergency departments, followed by concluding remarks.”
- Optimal Forecast Reconciliation for Hierarchical and Grouped Time Series Through Trace Minimization (2018, unknown): “This implies that the base forecasts are unbiased, that is, Then a set of reconciled forecasts will also be unbiased iff SPS = S or equivalently PS = I n , as then from Equation (2), we have Hyndman et al. (2011) proposed the “optimal combination” approach, based on the regression model”
- Forecasting hierarchical and grouped time series through trace minimization (2015, methodology): “Our approach is to extend the work of Hyndman, Lee, and Wang (2016) and Hyndman et al. (2011) and frame the problem in terms of finding a set of minimum variance unbiased estimates of future values of all time series across the entire collection.”
- Energy Forecasting: A Review and Outlook (2020, background): “Over several publications, computation issues that hinder the large-scale applications of hierarchical forecasting have been addressed [46]–[48].”
- Hierarchical forecasts for Australian domestic tourism (2009, methodology): “Therefore the simple ANOVA method presented in equations (12) and (13) in Hyndman et al. (2007) can be applied for producing the revised forecasts for the optimal combination approach.”
- Forecasting: theory and practice (2020, methodology): “This adjustment is achieved by a matrix that maps the base forecasts into new bottom-level forecasts which are then added up (Hyndman et al., 2011).”
- Review of low voltage load forecasting: Methods, applications, and recommendations (2021, background): “It is worth noting here that although aggregate forecasts can be generated by simply summing up the corresponding forecasts at the lower level, it has been shown that this bottom-up approach produces poor results [109].”