Title: Probabilistic forecasting of hierarchical time series data
Authors: Souhaib Ben Taieb - University of Mons (Belgium) [presenting]
James Taylor - University of Oxford (United Kingdom)
Rob Hyndman - Monash University (Australia)
Abstract: Time series can often be naturally represented in a hierarchical or grouped structure. For example, a manufacturing company can disaggregate total demand for their products by country of sale, retail outlet, product type, package size, and so on. As a result, there can be a large number of individual time series to forecast at the most disaggregated level, plus additional series to forecast at higher levels of aggregation. In order to allow consistent decisions over different levels of the hierarchy, the forecasts for the disaggregated series are usually required to add up exactly to the forecasts of the aggregated series, a constraint known as aggregate consistency. Computing mean aggregate consistent forecasts involves fitting a linear regression model where the design matrix has one column for each of the series at the most disaggregated level. We will discuss some algorithms to generate probabilistically aggregate consistent forecasts with an application to electricity demand forecasting using smart meter data.