Paper: Integrable Elasticity via Neural Demand Potentials (arXiv 2505.17388)

Summary

This forum post introduces the paper "Integrable Elasticity via Neural Demand Potentials" (arXiv:2505.17388) by Carlos Heredia and Daniel Roncel, published on 2025-05-23 in the machine learning field. The paper proposes the Integrable Context-Dependent Demand Network (ICDN), a demand-first neural model for multi-product retail demand forecasting. ICDN learns log-demand as a smooth, context-conditioned function of log-prices, which allows elasticities to be derived exactly from the learned demand surface rather than estimated separately. This integrability-by-construction approach ensures that price elasticities remain internally consistent across products and price points. On the Dominick's beer dataset, the authors report improved out-of-sample generalization over log-log benchmarks, along with more stable and economically plausible elasticity estimates, particularly for weakly identified cross-price effects, which are notoriously difficult to estimate reliably in multi-product settings. The post includes the paper summary and a link to the arXiv preprint.