Endogeneité et autocorrélation spatiale : quelle utilité pour le modèle de Durbin ?

This article focuses on the endogeneity problem in a spatial framework. We analyze the case where endogeneity comes from an omitted variable, which is spatially autocorrelated. We show, both theoretically and with Monte-Carlo simulations, that one way to decrease the omitted variable bias is to estimate a spatial Durbin model, which includes an endogenous spatial lag and exogenous spatial lag variables. The simulations also show that the bias and RMSE of the estimators obtained with the spatial Durbin model remain relatively low in the case of quasi non-stationarity.