What is the Optimal Strategy of Aggregation for Forecasting Sales? Time Series Forecast Reconciliation by Region, Product Category, and Channel

Informações

Código: MKT396
Divisão: MKT - Marketing
Tema de Interesse: Tema 06 - Estratégia e Desempenho em Marketing e Vendas

Autores

Carla Freitas Silveira Netto (Prog de Pós-Grad em Admin/Esc de Admin – PPGA/EA/UFRGS - Universidade Federal do Rio Grande do Sul) carla.netto@gmail.com
Rob J. Hyndman (Department of Econometrics & Business Statistics/Monash University) Rob.Hyndman@monash.edu
Vinicius Andrade Brei (Prog de Pós-Grad em Admin/Esc de Admin – PPGA/EA/UFRGS - Universidade Federal do Rio Grande do Sul) brei@ufrgs.br

Resumo

While some companies still struggle to gather, store and analyze data necessary to make better predictions, others are worried about increased requirements for data minimization and anonymization. This scenario raises questions about which variables are important to gather, and the resources necessary to do so. An important topic for academia and practice is how to improve forecasts, having limited access to data. In this paper, we consider such difficulties and propose forecasting strategies based on sales data in different aggregations criteria and structures. We compare aggregation criteria using both hierarchical and grouped time series structures, applying data that most companies already have access, marketing mix variables. Our paper indicates whether product category, channel type or region (geographic location) works best alone or combined when using the optimal reconciliation approach. This research suggests how to run sales forecasting more efficiently, using open-source tools. The method is also generalizable to all types of goods.

Abrir PDF