Improving Underlying Scenarios for Aggregate Forecasts: A Multi-level Combination Approach

Abstract In some situations forecasts for a number of sub-aggregations are required for analysis in addition to the aggregate itself. In this context, practitioners typically rely on bottom-up methods to produce a set of consistent forecasts in order to avoid conflicting messages. However, using this approach exclusively can mean that forecasting accuracy is negatively affected when compared to using other methods. This paper presents a method for increasing overall accuracy by jointly combining the forecasts for an aggregate, any sub-aggregations, and the components from any number of models and measurement approaches. The framework seeks to benefit from the strengths of each of the forecasting approaches by accounting for their reliability in the combination process and exploiting the constraints that the aggregation structure imposes on the set of forecasts as a whole. The results from the empirical application suggest that the method is successful in allowing the strengths of the better-performing approaches to contribute to increasing the performance of the rest