The cost of ensembling: Is it always worth combining global retail forecasting models?

Marco Zanotti, Flavio A. Ziegelmann · 2026

Read The cost of ensembling: Is it always worth combining global retail forecasting models? by Marco Zanotti, Flavio A. Ziegelmann (2026) 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/PMC13623108.txt.

Description

Marco Zanotti and Flavio A Ziegelmann ask whether combining many retail forecasting models is worth its computing cost. Retailers forecast demand for huge product ranges, and blending model outputs, called ensembling, usually improves accuracy. Bigger blends, though, burn far more computer time, and the trade-off is rarely measured.

The pair tested ten base models and eight blend designs on two large retail data sets, scoring both single-value and probability forecasts at several retraining schedules. Blends did forecast better, above all for probability forecasts, but the largest blends cost much more to run. Retraining less often cut costs sharply with little accuracy loss. Small blends of two or three models usually came close to the best scores at a fraction of the cost. The authors frame their guidance as a path to forecasting that is both scalable and lighter on energy use.

Catalog record

CREATORMarco Zanotti, Flavio A. Ziegelmann
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
LICENSEhttps://creativecommons.org/licenses/by/4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/article/MED/PMC13623108

Browse the full library · About the Tower of Babel