When forecasting economic variables, one is often faced with the choice of either fore- casting an aggregate directly or forecasting its components and then summing the component forecasts. This is frequently encountered when forecasting inflation, where prices are commonly available for a large number of components series in addition to the aggregate price index. The aggregation issue is a major practical consideration when it comes to forecasting key economic indicators but frequently forecasters are in the dark in terms of which approach is likely to yield the best results. There is a con- siderable set-up cost when estimating models on disaggregate data if there are a large number of component series so researchers are understandably reluctant to pursue this strategy unless it is likely to yield benefits.
Arguably, the literature on forecast aggregation is at an impasse. The early con- tributions focussed on deriving theoretical results but this approach was eventually abandoned as the underlying assumptions were too restrictive. Empirical papers tend to focus on a specific application. Competing sets of forecasts are constructed for a given country or set of countries to see whether forecast aggregation helps. With the exception of Hubrich (2003), few papers offer potential explanations of why the forecast aggregation strategy was a success or failure so there is little guidance to forecasters faced with the option of combining disaggregate forecasts.
We conduct empirical exercises but relate our findings back to the properties of the dataset and the models used. The exercise is conducted on both United States (US) and Euro Area (EA) inflation. Although both datasets relate to inflation, these datasets have distinct characteristics and we tailor the model to the properties of the data. In contrast to most previous studies, we consider multiple levels of aggregation for each dataset. We find that, once the appropriate model is found for a dataset, forecast aggregation always leads to improvements in forecast accuracy - the critical issue is to find the appropriate model. Frequently, the forecast based on the aggregate results in the worst forecast performance. This story is consistent with the theoretical literature. By providing a detailed explanation for main factors driving results for both datasets, we provide a greater understanding of the key issues relative to other empirical papers. In the next section, we provide a summary of the main contributions in both the theoretical and empirical side of the literature. Section 3 describes the data used for the empirical exercises. Section 4 outlines the notation and the models used in the paper with the results reported in section 5. Section 6 concludes the paper.