Hierarchical time-series models for forecasting oilseeds and pulses production in India

Abstract

Hierarchical time-series, which are multiple time-series that are hierarchically organized and can be aggregated at several different levels in groups based on geographical locations or some other features, has many practical importance. There are certain specialized strategies, viz. top-down, bottom-up, middle-out and optimal approaches which take care of predicting future values for such multi-level data. The top-down approach at first provides forecasting for the aggregated series at the top most level of the hierarchy, then disaggregating the forecasts in the lower levels based on historical and forecasted proportions. The bottom-up method provides forecasting for the most disaggregated series at the bottom level of the hierarchy and then aggregates these forecasts to obtain the forecasts at the top level of the hierarchy. The middle-out approach is a combination of bottom-up and top-down approaches. The optimal combination approach involves forecasting all series at all levels in the hierarchy, and then using a regression model to obtain the optimally combined forecasts. As an example, forecasting of oilseeds as well as pulses production in India is attempted using hierarchical time-series models.