A Comprehensive Forecasting Framework based on Multi-Stage Hierarchical Forecasting Reconciliation and Adjustment
Zhengchao Yang, Mithun Ghosh, Anish Saha, Dong Xu, Konstantin Shmakov, Kuang-chih Lee · 2024
Read A Comprehensive Forecasting Framework based on Multi-Stage Hierarchical Forecasting Reconciliation and Adjustment by Zhengchao Yang, Mithun Ghosh, Anish Saha, Dong Xu, Konstantin Shmakov, Kuang-chih Lee (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-14718.txt.
Description
Zhengchao Yang and co-authors describe the demand forecasting system behind Walmart's advertising business in this 2024 paper. Planners must predict ad demand at many levels at once, from the whole company down to single products, and the levels must add up. Standard reconciliation keeps the sums consistent but often dulls accuracy and flattens each series' own seasonal rhythm.
Their framework starts with an ensemble of models tuned by Bayesian search. A multi-stage process then aligns the forecasts: top-down shaping first, a trace-minimizing fix at the upper levels, and harmonic alignment plus stratified scaling at the bottom. Error fell sharply against both the plain ensemble and leading rival methods across Walmart's data and three public datasets. Forecasts stayed consistent at every level. The system now runs in production, guiding the company's ads, sales, and operations planning.
Catalog record
| CREATOR | Zhengchao Yang, Mithun Ghosh, Anish Saha, Dong Xu, Konstantin Shmakov, Kuang-chih Lee |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-14718