Bayesian Transfer Learning of Spatial Lag Stochastic Frontier Analysis

Soziale Passagen · Springer Nature · 2026-10-07 · DOI: 10.1007/s11067-026-09793-4

Аннотация

This study tackles the challenge of estimating parameters in spatial lag stochastic frontier models with limited data by proposing a novel Bayesian transfer learning framework. We develop estimation methods utilizing shared-parameter models, accompanied by tailored MCMC sampling algorithms. Simulation results demonstrate that our approach can drastically reduce parameter estimation MSE by over 90% in high-similarity, small-sample scenarios, though “negative transfer” risks emerge with increasing domain disparity. An empirical application on Chinese high-tech enterprises confirms that transfer learning not only enhances estimation precision but also effectively aligns target-domain parameters with more representative source-domain information, offering a robust new methodology for small-sample spatial econometric analysis in fields like regional economics.

Читать у издателя →

Похожие статьи