Title: Revisiting estimation methods for spatial econometric interaction models
Authors: Lukas Dargel - Universite Toulouse 1 Capitole (France) [presenting]
Thibault Laurent - Toulouse School of Economics (France)
Abstract: Spatial interactions describe phenomena as diverse as international trade flows, migration flows, or passenger flows in public transport. Explaining the main causes of such origin-destination flows is, therefore, a goal that unites empirical researchers and practitioners from various industries and scientific disciplines. The best-known interaction model is the gravity model, but it leads to inconsistent parameter estimates if there is some dependence between the flows. Techniques from the field of spatial econometrics solve this problem by including explicit measures of network dependence in the model. However, the parameters of this extended model cannot be estimated by ordinary least squares, and we have to adopt more sophisticated estimation methods, such as Maximum Likelihood (MLE), spatial two-stage least squares (S2SLS) or Bayesian Markov chain Monte Carlo (MCMC). We develop improved calculations for these three estimators, and our R package makes them operational for a broader public. Our simulations show that we can estimate the model parameters consistently and that the MLE is superior to S2SLS and MCMC estimation in terms of computational performance.