Nicolas Debarsy, Chargé de recherche au CNRS

Jeudi 25 janvier 2024 nous avons le plaisir d'accueillir en séminaire Nicolas Debarsy, Chargé de Recherche au CNRS, affecté au laboratoire LEM de Lille.

Titre de sa présentation : Semiparametrically efficient of linear models with spillovers, article co-écrit avec Vincenzo Verardi et Catherine Vermandele.

Résumé : The maximum likelihood (ML) estimator of regression models that explicitly account  for cross-sectional dependence between observations is generally based on the assumption of normally distributed errors. This method yields efficient estimators when the distribution is Gaussian. Lee (2004) suggests a Quasi-Maximum Likelihood (QML) estimator for the model with endogenous (and contextual) effects, which accounts for departures from the normal distribution assumption. More generally, when the distribution is not correctly specified, the Generalized Method of Moments (GMM) proposed by Liu et al. (2010) may provide more efficient estimators than their ML or QML counterparts. In this paper, we propose a semiparametrically efficient estimator, based on Local Asymptotic Normality theory of Le Cam (1960) that does not assume any specific distribution for the error term (the distribution should only be strongly unimodal). The performed Monte Carlo simulations indicate that our proposed estimator is much more efficient than both QML and GMM when we depart from the
Gaussian distribution of the error term. To illustrate the usefulness of the proposed methodology, relying on Behrens et al. (2012), we present a trade regression and show how results might change drastically when the Gaussian distribution is not imposed.

Le séminaire a lieu à 13h30 en salle 227.