Title: Continuously updated indirect inference
Authors: Maria Kyriacou - University of Southampton (United Kingdom) [presenting]
Peter CB Phillips - Yale University (United States)
Francesca Rossi - University of Southampton (United Kingdom)
Abstract: Spatial units are often heterogeneous as they vary in many of their observed characteristics and so the assumption of homoskedasticity may not hold in practice. In the presence of unobserved heterogeneity of the disturbance term, standard methods based on the (quasi-)likelihood function produce, in general, inconsistent estimates of both the spatial parameter and the exogenous regressors coefficients. There is an evident lack of estimation methods that account for the presence of heteroskedasticity of unknown form. A robust generalized methods of moments estimator as well as a modified likelihood method have been previously proposed to address this issue. We propose an indirect inference methodology which relies on a simple Ordinary Least Squares (OLS) procedure as its starting point, which is computationally simple, robust to unobserved heterogeneity, allows for a general range of weight matrix structures and has excellent finite sample performance. Our proposed Continuously Updated Indirect Inference (CUII) estimator is derived using a binding function with a continuously-updated diagonal variance-covariance matrix. Simulation results reveal that our proposed estimator is effective in reducing both bias and MSE compared to competitor estimators.