Integrable Elasticity via Neural Demand Potentials: The ICDN Model (arXiv 2505.14484)

Forum topic · 小凯 · 2026-05-25

Summary

Researchers Carlos Heredia and Daniel Roncel propose the Integrable Context-dependent Demand Network (ICDN), a demand-first neural model for multi-product retail demand. The model learns log-demand as a smooth, context-conditioned function of log-prices, allowing elasticities—including cross-price effects—to be derived exactly from the learned demand surface via integration rather than estimated directly. Evaluated on the Dominick beer dataset, ICDN improves out-of-sample generalization over a directed log-log baseline and produces more stable, economically plausible elasticity estimates, particularly for weakly identified cross-price effects. The work, listed under machine learning on arXiv as 2505.14484, bridges neural demand modeling with economic integrability constraints, addressing a common weakness of directly parameterized elasticity models in retail pricing applications.

Paper Overview

Research Area: Machine Learning Authors: Carlos Heredia, Daniel Roncel Published: 2026-05-25 arXiv: 2505.14484

Abstract

The authors propose the Integrable Context-dependent Demand Network (ICDN), a demand-first neural model for multi-product retail demand.

Key ideas:

The model learns log-demand as a smooth, context-conditioned function of log-prices.
Because the demand surface is learned directly, elasticities can be derived exactly from the learned demand function via integration, rather than being estimated as free parameters.
This integrability-by-construction approach yields more stable and economically plausible elasticity estimates, especially for weakly identified cross-price effects.

Empirical Results

On the Dominick beer dataset, ICDN:

Improves out-of-sample generalization compared to a directed log-log baseline.
Produces more stable, economically reasonable elasticity estimates.