Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adventures in Demand Analysis Using AI

Philipp Bach, Victor Chernozhukov, Sven Klaassen, Martin Spindler, Jan Teichert-Kluge, Suhas Vijaykumar

TL;DR

This paper addresses how consumer demand analysis can be enhanced by AI-generated, multimodal product representations that fuse text, images, and tabular data. It builds a causal inference framework around a dynamic SEM, using CAPE and CACE with orthogonal, cross-fitted estimates to quantify price effects while accounting for product-specific heterogeneity. The key contributions are (i) a practical workflow for generating and evaluating transformer-based embeddings from multimodal data, (ii) evidence that these embeddings improve predictions of prices and quantities, and (iii) robust findings that price elasticity is highly heterogeneous across products and is moderated by product features and popularity. The approach provides a methodological bridge between AI representations and econometric causal analysis, with implications for more credible policy and business decisions in digital marketplaces.

Abstract

This paper advances empirical demand analysis by integrating multimodal product representations derived from artificial intelligence (AI). Using a detailed dataset of toy cars on \textit{Amazon.com}, we combine text descriptions, images, and tabular covariates to represent each product using transformer-based embedding models. These embeddings capture nuanced attributes, such as quality, branding, and visual characteristics, that traditional methods often struggle to summarize. Moreover, we fine-tune these embeddings for causal inference tasks. We show that the resulting embeddings substantially improve the predictive accuracy of sales ranks and prices and that they lead to more credible causal estimates of price elasticity. Notably, we uncover strong heterogeneity in price elasticity driven by these product-specific features. Our findings illustrate that AI-driven representations can enrich and modernize empirical demand analysis. The insights generated may also prove valuable for applied causal inference more broadly.

Adventures in Demand Analysis Using AI

TL;DR

This paper addresses how consumer demand analysis can be enhanced by AI-generated, multimodal product representations that fuse text, images, and tabular data. It builds a causal inference framework around a dynamic SEM, using CAPE and CACE with orthogonal, cross-fitted estimates to quantify price effects while accounting for product-specific heterogeneity. The key contributions are (i) a practical workflow for generating and evaluating transformer-based embeddings from multimodal data, (ii) evidence that these embeddings improve predictions of prices and quantities, and (iii) robust findings that price elasticity is highly heterogeneous across products and is moderated by product features and popularity. The approach provides a methodological bridge between AI representations and econometric causal analysis, with implications for more credible policy and business decisions in digital marketplaces.

Abstract

This paper advances empirical demand analysis by integrating multimodal product representations derived from artificial intelligence (AI). Using a detailed dataset of toy cars on \textit{Amazon.com}, we combine text descriptions, images, and tabular covariates to represent each product using transformer-based embedding models. These embeddings capture nuanced attributes, such as quality, branding, and visual characteristics, that traditional methods often struggle to summarize. Moreover, we fine-tune these embeddings for causal inference tasks. We show that the resulting embeddings substantially improve the predictive accuracy of sales ranks and prices and that they lead to more credible causal estimates of price elasticity. Notably, we uncover strong heterogeneity in price elasticity driven by these product-specific features. Our findings illustrate that AI-driven representations can enrich and modernize empirical demand analysis. The insights generated may also prove valuable for applied causal inference more broadly.
Paper Structure (20 sections, 3 theorems, 23 equations, 9 figures, 9 tables, 3 algorithms)

This paper contains 20 sections, 3 theorems, 23 equations, 9 figures, 9 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Using AI to Understand and Represent Products
  3. The Data and Measurement of Prices and Quantities
  4. Using AI to Represent Products
  5. Evaluating the Embeddings
  6. Qualitative Assessment
  7. Quantitative Assessment
  8. Estimating Price Effects
  9. Initial Approach and Challenges
  10. The Causal Dynamic Model and Regressions
  11. Empirical Models
  12. Orthogonal Inference of Causal Effects
  13. Empirical Results
  14. Homogeneous Elasticity Model
  15. Heterogeneous Elasticity Model
  16. ...and 5 more sections

Key Result

Proposition 1

If the SEM model eq:causal1--eq:causal3 holds, then the CACE is identified by the CAPE: almost surely, provided that both exist and are finite. Then the ACE is identified by the APE, $\mathrm{E}\bigl[\alpha_t(S_{it})\bigr] = \mathrm{E}\bigl[\delta_t(S_{it})\bigr]$, again provided these expectations exist and are finite.

Figures (9)

  • Figure 1: A product example with an image and text description in the "toys" category.
  • Figure 2: Price and sales rank series for two example products.
  • Figure 3: 3d representation of product embeddings (with image) and five clusters
  • Figure 4: 3d representation of product embeddings (no image) and five clusters
  • Figure 5: A directed acyclic graph for the dynamic model.
  • ...and 4 more figures

Theorems & Definitions (8)

  • Remark 1
  • Proposition 1: Identification of CACE
  • Remark 2: Orthogonalization
  • Remark 3: Causal Fine-Tuning
  • Remark 4: Robustness of DML to Estimation Noise in Embeddings
  • Proposition 2
  • proof
  • Corollary 1