Online Program Home
My Program

Abstract Details

Activity Number: 233
Type: Topic Contributed
Date/Time: Monday, August 1, 2016 : 2:00 PM to 3:50 PM
Sponsor: Business and Economic Statistics Section
Abstract #318969
Title: Hierarchical Time Series Forecasting
Author(s): Julie Novak* and Beatriz E. Etchegaray Garcia and Yasuo Amemiya
Companies: IBM and IBM Research and IBM Research
Keywords: Time Series ; Hierarchical Modeling ; Heterogeneous Loss Function ; Business Metric Forecasting

An important task for any large-scale business is to prepare forecasts of business metrics, such as revenue, cost, and event occurrences, at different time horizons. Often these businesses are structured in a hierarchical manner by line of business, division, geography or a combination thereof. Projections for these business metrics may have been obtained independently, and there is no guarantee that forecasts are aggregate consistent according to the hierarchical structure, while remaining as accurate as possible. Also, it is often important to achieve accurate forecasts at certain levels of the hierarchy. We propose a Bayesian hierarchical method that treats the "base" forecasts (those initially provided) as observed data, which are then updated and obey the hierarchical organizational structure. In addition, by leveraging the prior covariance matrix, we set up a heterogeneous loss function to obtain higher accuracy at the levels prescribed by the user. We develop a novel approach to hierarchical forecasting that provides an organization with optimal forecasts that reflect their preferred levels of accuracy while maintaining the proper additive structure of the business.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2016 program

Copyright © American Statistical Association