Identification of Spatial Durbin Panel Models

Abstract

Summary This paper considers identification of spatial Durbin dynamic panel models under 2SLS and ML estimations. We show that the parameters are generally identified via 2SLS moment relations or expected log‐likelihood or quasi‐likelihood functions. Monte Carlo experiments suggest that omitting relevant Durbin terms can result in significant biases in regression estimates, while including an irrelevant Durbin term causes no obvious loss of efficiency. Empirical illustration of the international spillover of economic growth through bilateral trade shows that inclusion of Durbin terms can be important. Copyright © 2015 John Wiley & Sons, Ltd.

Analysis of the impact of school performance on income inequality in the long run: An application to Chilean municipalities

“…Furthermore, the model in Equation () is endogenous because of the presence of the spatial lag; therefore, the log‐likelihood function is highly nonlinear. This estimation is performed by using the quasi likelihood method for panel proposed by Millo and Piras (2012), which provides consistent results (Lee & Yu, 2016).…”

Quantifying Urban Sprawl and Its Driving Forces in China

“…Therefore, compared to other spatial models, the SDM is a more generalized form. However, for making sure the applicability of SDM to certain regression analyses, it is necessary to perform relevant statistical tests, and the Wald and likelihood ratio (LR) test shall be carried out for confirming if the SDM can be reduced to a SLM or SEM [50]. The Hausman test helps the study to confirm that which effect is adopted by the spatial econometric model, fixed effect or random effect [51].…”

On the Contagion Effect in the US Banking Sector

“…Specifically, we allow banks to be treated the same by the public irrespective of the distance between them, for example, banks are allowed to have interbank connections around the whole banking sector. Note that the identification in the spatial models is an ongoing issue (see, e.g., Gibbons and Overman , Gibbon, Overman and Patacchini , and Lee and Yu among others) . Gibbons and Overman (, p. 180) notes that, “in theory if W is not idempotent, identification is possible, but the parameters in the spatial Durbin model are likely only weakly identified in practice.” Note that spatial Durbin model is a combination of the SAR (Spatial Autoregressive Model) and the SLX (Spatial Lag of X) model.…”

“…In a later paper, Gibbons, Overman, and Patacchini () also discuss this issue in detail and note that spatial models are identified when the spatial‐weights matrix is nonblock diagonal. Lee and Yu () also address this issue and show that spatial specifications are identified under likelihood estimation. In our analysis, the equal weights W matrix with zero diagonal elements is not idempotent.…”