Skip to content

Multi-objective probabilistic forecast combination for inventory demand

Home / Papers / Multi-objective probabilistic forecast combination for inventory demand

Multi-objective probabilistic forecast combination for inventory demand

Authors: Shengjie Wang, Yanfei Kang, Evangelos Spiliotis, Fotios Petropoulos Date: 2026-06-03 Paper ID: openalex:2606.04900

Summary

This paper addresses the disconnect between statistical forecast accuracy and operational decision performance in inventory management. The authors introduce a multi-objective probabilistic forecast combination framework that optimizes across both accuracy metrics and inventory-specific costs. By identifying Pareto-optimal combinations, the method allows decision-makers to navigate trade-offs between competing operational targets. Extensive empirical validation on Walmart and Royal Air Force datasets confirms that this integrated approach outperforms traditional single-objective combination techniques.

Key Contributions

  • Proposes a multi-objective optimization framework for combining probabilistic forecasts that balances statistical accuracy with inventory-specific operational objectives.
  • Formulates forecast combination as a multi-objective problem to derive Pareto-optimal solutions for complex inventory cost structures.
  • Demonstrates superior, more robust inventory decision performance compared to standard single-objective combination methods on real-world retail and spare parts datasets.

Key Concepts

Archivist Review

I approved the core framework concept as it effectively formalizes the tension between predictive accuracy and operational decision utility, which is a foundational problem in time-series forecasting. The datasets were rejected because they are not provided as public, named benchmarking artifacts that would warrant a persistent, searchable entry in the vault.

Approved Concepts

  • Multi-objective probabilistic forecast combination: It addresses the common misalignment between statistical forecast accuracy and operational decision-making efficiency in high-stakes inventory management.

Rejected Candidates

  • [dataset] Walmart retail data (walmart-retail-data) - low_impact: Generic retail dataset description, not a specific, named, and public benchmark artifact.
  • [dataset] Royal Air Force spare parts data (royal-air-force-spare-parts-data) - low_impact: Proprietary or generic domain-specific data without public distribution or standard benchmark status.

Metadata & Links

url
https://arxiv.org/abs/2606.04900
paper_id
2606.04900
paper_source
openalex
domain
time-series
tags
forecastingoptimizationdecision-making
architectures
datasets
concept_slugs
multi-objective-probabilistic-forecast-combination
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-06-06T06:13:32Z