Seminario nº 12:   Interaction matrix selection in spatial econometric  models: Application to the Schumpeterian Growh  model with worldwide interactions

Información

Cem Ertur, especialista de prestigio  en econometría espacial (y coautor de varios profesores de la linea 9 del programa, especializados en esta rama de la econometría) hizo una visita  a la UMU en junio. Gracias a la cooperación entre la UMU y la UNED, pudimos disfrutar de un seminario muy didáctico y conocer nuevos resultados teóricos y empíricos en esta disciplina. El seminario se retransmitió en directo y se grabó.

Abstract: The interaction matrix or the spatial weight matrix is the fundamental tool to model cross-sectional interdependence between observations in spatial econometric models. However, it is most of the time not derived from theory, as it should be ideally, but chosen on an ad hoc basis. At best some limited robustness or sensitivity analyses are proposed. Recently, some testing procedure for non-nested models have been extended to the spatial econometric framework (Keleijian, 2008; Kelejian and Piras, 2011; Burridge, 2012 ; Jin and Lee, 2012) and may be used to formally select the interaction matrix. In this paper, we propose a modified version of the Kelejian and Piras J-test (2011), based on the application of the RGMM estimation method, which is robust versus unknown heteroskedasticity, proposed by Lin and Lee (2010). Results of Monte-Carlo experiments to asses size and power of the various test statistics in small samples are presented. However the common problem faced by all the tests for non-nested models is that it may happen that no conclusion can be reached, when permuting the null and alternative hypotheses. A new testing procedure for non-nested models, labeled the MJ test, is proposed by Hagemann (2012) to overcome this problem. We propose an extension of this testing procedure to spatial econometric models. An application is then presented for the Schumpeterian growth model with worldwide interactions (Ertur and Koch, 2011) using four different types of interaction matrix: geographic distance, genetic distance, linguistic distance and an interaction matrix based on trade flows.