Paper Overview
- Research Area: Machine Learning (ML)
- Authors: Carlos Heredia, Daniel Roncel
- Release Date: 2025-05-23
- arXiv: 2505.17388
- Demand-first design: The neural network predicts demand while embedding economic structure (integrability) directly into its architecture.
- Exact elasticity computation: Because log-demand is modeled as a smooth, differentiable function of log-prices, price elasticities—both own-price and cross-price—can be derived analytically from the learned surface rather than estimated via post-hoc approximation.
- Improved out-of-sample performance: On the Dominick's beer dataset, ICDN outperforms standard log-log regression baselines in generalization.
- More stable cross-price estimates: The model is particularly valuable for weakly identified cross-price effects, producing elasticity estimates that are economically more plausible and less noisy than baseline approaches.
- Practical relevance: Useful for retail analytics tasks such as price optimization and assortment planning, where reliable price-sensitivity estimates are critical.
- arXiv link: https://arxiv.org/abs/2505.17388
Chinese Abstract (translated)
We propose the Integrable Context-Dependent Demand Network (ICDN), a demand-first neural model for multi-product retail demand forecasting. The model learns log-demand as a smooth, context-conditioned function of log-prices, enabling elasticities to be derived exactly from the learned demand surface. On the Dominick's beer dataset, ICDN improves out-of-sample generalization over log-log baselines and produces more stable, economically more reasonable elasticity estimates, especially for weakly identified cross-price effects.
Original Abstract
*Not provided in the source post.*
Key Points
Source
#paper #arXiv #ML