Forecasting with Temporal Hierarchies
This presentation was given to the practitioner track of the conference and its aim was to introduce the basic idea of temporal hierarchies for forecasting and highlight the forecasting and business challenges they can help address. You can read more details in the abstract
Temporal hierarchies, a new development in time series modelling aims to address all the issues above. The principal idea is to model time series at multiple levels of temporal aggregation (for example: weekly, monthly, quarterly and annually) and combine the resulting predictions. There is a two-fold motivation for this. First, resulting forecasts are able to extract better the information captured in time series, as temporal aggregation allows attenuating or strengthening different components of the time series. Second, by modelling time series in this way we can ensure that short-term forecast (constructed at disaggregate levels, e.g. weekly) and long-term forecasts (constructed at aggregate levels, e.g. yearly) are aligned and can support decision making at all different levels. This provides a statistically sound way to achieve the so called `one-number’ forecast. This presentation will introduce temporal hierarchies, show how to build and use them and demonstrate their advantages with real case studies.
Measuring Forecasting Performance: A Complex Task
This research presentation introduces a new error metric to evaluate forecast performance. Although there has been substantial research on accuracy metrics, there has been very limited work on bias metrics that is an equally important dimension of performance. An interesting aspect of the proposed metric are the informative visualisations of accuracy and bias. You can read more details in the abstract