Unbiased Estimation of Multi-Way Gravity Models
By: Lucas Resende , Guillaume Lecué , Lionel Wilner and more
Potential Business Impact:
Fixes math problems for better predictions.
Maximum likelihood estimators, such as the Poisson Pseudo-Maximum Likelihood (PPML), suffer from the incidental parameter problem: a bias in the estimation of structural parameters that arises from the joint estimation of structural and nuisance parameters. To address this issue in multi-way gravity models, we propose a novel, asymptotically unbiased estimator. Our method reframes the estimation as a series of classification tasks and is agnostic to both the number and structure of fixed effects. In sparse data environments, common in the network formation literature, it is also computationally faster than PPML. We provide empirical evidence that our estimator yields more accurate point estimates and confidence intervals than PPML and its bias-correction strategies. These improvements hold even under model misspecification and are more pronounced in sparse settings. While PPML remains competitive in dense, low-dimensional data, our approach offers a robust alternative for multi-way models that scales efficiently with sparsity. The method is applied to estimate the effect of a policy reform on spatial accessibility to health care in France.
Similar Papers
Optimal Estimation for General Gaussian Processes
Statistics Theory
Makes computer predictions more accurate and reliable.
Revisiting Penalized Likelihood Estimation for Gaussian Processes
Methodology
Improves computer predictions when data is scarce.
Debiased Inference for High-Dimensional Regression Models Based on Profile M-Estimation
Methodology
Makes computer predictions more trustworthy and faster.