Testing Random Eects in Two Way Spatial Panel Data Models在双向空间面板数据模型中测试随机效应.pdfVIP

Testing Random Effects in Two-Way Spatial Panel Data Models Nicolas Debarsy∗ May 27, 2010 Abstract This paper proposes an alternative testing procedure to the Hausman test statistic to help the applied researcher to determine the adequacy of the random effects specification in a two-way spatial autoregressive panel data model. Our model includes both individual and time specific effects to deal with unobserved heterogeneity. Moreover, we allow for interactions between cross-sectional units through the presence of an endogenous spatial lag (SAR). We suggest a two-step procedure that first partial out the spatial component of the model and then generalizes the original Mundlak approach to both individual and time effects. Three tests are proposed to find out the best way to model individual and time effects in the specification. This approach has the advantage of simplicity since all tests are based on FGLS estimations and standard Wald statistics can be applied. Some Monte Carlo simulations show the properties of this procedure in small sample. Keywords: Spatial Autocorrelation, Panel Data, Random effects, Mundlak JEL: C12; C21; C23; C52 1 Introduction Spatial autoregressive (SAR) panel data models are of high interest since they allow capturing individual, temporal and interactive heterogeneity. The first two types of heterogeneity come from individual or time characteristics and are easily dealt with using either a fixed or random effects panel data framework. Interactive heterogene- ity is due to differentiated feedback effects, originating from cross-section interactions between individuals.1 It cannot be dealt with traditional panel data