Clicky

Nested Pseudo-Likelihood Estimation of Dynamic Discrete Games: A Replication of Aguirregabiria and Mira (2007)

Jason R. Blevins and Minhae Kim
Working paper.

Versions and Availability:

Abstract. We replicate the Monte Carlo experiments of Aguirregabiria and Mira (2007, Econometrica) using a new, independent MATLAB implementation. The nested pseudo-likelihood (NPL) estimator they introduced for dynamic discrete games remains the basis for newer methods. All six experiments replicate closely. We then extend their study in three directions. First, we revisit the non-convergence findings of Pesendorfer and Schmidt-Dengler (2010). We compute the spectral radius of the NPL operator for each experiment and find that all six lie in the stable region, although Experiment 3 is borderline and its estimates are bimodal across replications. Second, we evaluate the finite-sample performance of standard errors and confidence intervals implied by their asymptotic variance formula. We find that these are accurate, with near-nominal coverage in convergent cases at moderate sample sizes, and that a cross-sectional bootstrap performs comparably. Third, we extend the experiments to a larger-scale entry/exit model with permanent unobserved heterogeneity—the setting in which the NPL estimator was originally applied. In addition to the well-known source of instability arising from the strength of strategic interaction, we find a second form of instability from the initialization of the mixture pseudo-likelihood. Our implementations are open source, with detailed annotations.

Keywords: Dynamic discrete games, multiple equilibria, pseudo maximum likelihood estimation, nested pseudo likelihood, entry and exit in oligopoly markets, replication.

JEL Classification: C13, C35, C73, L11, L13.

BibTeX Record:

@TechReport{blevins-kim-2026-wp,
    author      = {Jason R. Blevins and Minhae Kim},
    title       = {Nested Pseudo-Likelihood Estimation of Dynamic Discrete Games: A Replication of Aguirregabiria and Mira (2007)},
    type        = {Working Paper},
    institution = {The Ohio State University},
    year        = 2026
}