Modeling Spatial Externalities: A Panel Data Approach

Abstract

In this paper we argue that the Spatial Durbin Model (SDM) is an appropriate framework to empirically quantify different kinds of externalities. Besides, it is attractive from an econometric point of view as it nests several other models frequently employed. Up to now the SDM has been applied in cross-sectional settings only, thereby ignoring individual heterogeneity. This paper extends the SDM to panel data allowing for non-spherical disturbances and proposes an estimator based on ML techniques. Results from a Monte Carlo study reveal that the estimator has satisfactory small sample properties also in cases when neither heteroskedasticity nor serial correlation is present in the data generating process. Moreover, we show that conventional testing procedures may wrongly reject the existence of spatial externalities. In particular, we show that the incidence of a type II error increases as the spatial weight matrix becomes denser.

Keywords: Spatial panel data, Spatial Durbin Model, Maximum Likelihood, AR(1) and heteroskedastic errors, Monte Carlo simulation

JEL Classification: C21, C23