Weak ties in large-scale networks facilitate the diffusion of information and resources, generating spillover effects. While geographic proximity may generate straightforward networks, a non predetermined spillover framework better accommodates socioeconomic interactions. In such cases, it is important to test the endogeneity of a spatial weight matrix before full estimation to ensure valid and efficient model selection. I propose a robust score test to assess this endogeneity in the weak-tied spatial dynamic panel data model, emphasizing its asymptotic optimality and computational efficiency. First, I correct score function biases to address the incidental parameters problem. Second, I adjust the score functions to enhance robustness against local parametric misspecifications. Monte Carlo simulations demonstrate favorable finite sample properties. In my application, I illustrate how this test helps model selection for the Solow-Swan growth model, considering spillovers from geographic and knowledge proximities. The estimation results reveal endogenous spillover effects with weakly positive impacts on real income per worker, while savings show positive internal effects but negative spillover effects, possibly due to competition for investment. Notably, the labor growth rate positively influences knowledge accumulation and innovation, leading to beneficial effects on real income per worker through both internal and spillover channels.
Read full abstract