[Paper Review] Nested Pseudo-GMM Estimation of Demand for Differentiated Products

Aguirregabiria, Victor, Hui Liu+1 more|arXiv (Cornell University)|Feb 4, 2026

Consumer Market Behavior and Pricing0 citations

TL;DR

The paper introduces a fast Nested Pseudo-GMM estimator for the Berry, Levinsohn, and Pakes demand model that avoids repeated high-cost demand inversions and exploits a closed-form, product-separable inner step.

ABSTRACT

We propose a fast algorithm for computing the GMM estimator in the BLP demand model (Berry, Levinsohn, and Pakes, 1995). Inspired by nested pseudo-likelihood methods for dynamic discrete choice models, our approach avoids repeatedly solving the inverse demand system by swapping the order of the GMM optimization and the fixed-point computation. We show that, by fixing consumer-level outside-option probabilities, BLP's market-share to mean-utility inversion becomes closed-form and, crucially, separable across products, yielding a nested pseudo-GMM algorithm with analytic gradients. The resulting estimator scales dramatically better with the number of products and is naturally suited for parallel and multithreaded implementation. In the inner loop, outside-option probabilities are treated as fixed objects while a pseudo-GMM criterion is minimized with respect to the structural parameters, substantially reducing computational cost. Monte Carlo simulations and an empirical application show that our method is significantly faster than the fastest existing alternatives, with efficiency gains that grow more than proportionally in the number of products.

Motivation & Objective

• Motivate computational inefficiency in standard NFXP estimation of the BLP model and seek a faster alternative.
• Develop a Nested Pseudo-GMM framework that reduces fixed-point solves by treating incidental parameters as fixed objects.
• Provide a theoretically sound estimator with consistency and asymptotic normality under standard identification, adaptable to high- and infinite-dimensional settings.
• Demonstrate computational gains via Monte Carlo simulations and an empirical LCBO wine-demand application.

Proposed method

• Swap the order of the fixed-point inversion and the GMM optimization to avoid repeated high-cost inversions.
• Introduce a regression-like representation where outside-option shares are incidental parameters, yielding a closed-form, product-separable mean-utility inversion.
• Define a pseudo-GMM criterion Q that depends on incidental parameters lambda, enabling a simple closed-form solution for theta in the inner step.
• Prove that GMM and NP-GMM are distinct estimators with different first-order conditions and mappings.
• Establish consistency and asymptotic normality of NP-GMM under standard identification, and provide conditions with JT growing large (markets and products).
• Show through Monte Carlo simulations and LCBO wine data that NP-GMM substantially outperforms alternatives in speed, especially with more products.

Experimental results

Research questions

• RQ1 Can a Nested Pseudo-GMM approach match the identification of GMM in the BLP model while reducing computational burden?
• RQ2 Does treating outside-option shares as incidental parameters enable closed-form, product-by-product mean-utility inversions and allow scalable estimation?
• RQ3 What are the asymptotic properties of NP-GMM in settings where the dimensionality grows with sample size?
• RQ4 How does NP-GMM perform relative to ABLP and NFXP in terms of computation time and accuracy in large product-differentiated demand systems?
• RQ5 Is the method robust to parallelization and multithreading in practical implementations?

Key findings

• NP-GMM delivers large computational speedups over existing methods, with efficiency gains increasing with the number of products.
• The gradient of the NP-GMM criterion is simple and explicit, facilitating parallel and multithreaded computation.
• NP-GMM uses a closed-form, product-separable mean-utility inversion conditional on incidental outside-option shares, avoiding repeated fixed-point solves.
• The method provides consistency and asymptotic normality under identification assumptions similar to GMM, despite using a different optimization mapping.
• Empirical application to LCBO wine data demonstrates scalability to a large number of products and shows substantial timing advantages over the fastest existing estimator.