We propose a new simulation-based estimation method, adversarial estimation, for structural models. The estimator is formulated as the solution to a minimax problem between a generator (which generates synthetic observations using the structural model) and a discriminator (which classifies if an observation is synthetic). The discriminator maximizes the accuracy of its classification while the generator minimizes it. We show that, with a sufficiently rich discriminator, the adversarial estimator attains parametric efficiency under correct specification and the parametric rate under misspecification. We advocate the use of a neural network as a discriminator that can exploit adaptivity properties and attain fast rates of convergence. We apply our method to the elderly’s saving decision model and show that including gender and health profiles in the discriminator uncovers the bequest motive as an important source of saving across the wealth distribution, not only for the rich.

More on this topic

BFI Working Paper·Nov 10, 2026

When Food Stamps Became Free: Impacts and Implications of Removing the Purchase Requirement

Amy Finkelstein, Matthew Notowidigdo, and Charlie Rafkin
Topics: Economic Mobility & Poverty
BFI Working Paper·Oct 21, 2026

What Can (and Cannot) Be Learned about Systemic Racial Discrimination from Randomized Controlled Trials?

Dania V. Francis, Rob Gillezeau, Damon Jones, and Angelino C. G. Viceisza
Topics: Economic Mobility & Poverty
BFI Working Paper·Oct 14, 2026

Indirect Variational Inference: Applications to Earnings Dynamics

Neele L. Balke, Stéphane Bonhomme, and Thibaut Lamadon
Topics: Economic Mobility & Poverty, Employment & Wages