A1638
Title: Structural estimation of spatial equilibrium models: A nonlinear extension of spatial autoregressive models
Authors: Shunsuke Segi - Kobe University (Japan) [presenting]
Abstract: A Full Information Maximum Likelihood (FIML) approach is proposed for the structural estimation of logit-type spatial equilibrium models with endogenous accessibility. Recent quantitative spatial models typically rely on natural experiments or instrumental variables to address endogeneity, but such identification strategies are often constrained by data availability. The logit-based spatial equilibrium structure can be interpreted as a nonlinear generalization of the spatial autoregressive (SAR) model, subject to a sum-to-one constraint on population shares. Analogous to the ML estimator for SAR models, the proposed estimator exploits the Jacobian term of the Likelihood function to correct for endogeneity bias without requiring external instruments. Using Monte Carlo simulations, FIML is compared with ordinary least squares and aggregate logit estimators, the latter being mathematically equivalent to the Poisson pseudo-Maximum Likelihood. The results show that FIML yields unbiased parameter estimates and recovers parameter values that correspond to dynamically stable equilibria under logit dynamics. These findings suggest that nonlinear extensions of SAR estimation offer an instrument-free route to structural identification in spatio-temporal equilibrium models.
