We propose a nonparametric test for the linearity of a network interaction function. Such functions arise commonly in the spatial and network literature, either as a practitioner imposed specification or as a result of optimizing behaviour by economic agents with quadratic utility functions, to name but two situations. Our test is based on a series expansion of the potential nonparametric component of the interaction function and is based on the Lagrange Multiplier principle. This is particularly appealing as it entails estimation only under the null hypothesis, which imposes a familiar linear spatial autoregressive (SAR) model. We focus on test statistics implemented with the Instrumental Variables/Two Stage Least Squares approach, but also explore the form of the statistic when using a (Gaussian) Quasi Maximum Likelihood Estimation objective function instead.
Apr 18, 2022, 12:30 pm – 1:30 pm
101 - Sherrerd Hall