Keeping up with the regions: a hybrid machine learning framework for estimating regional input-output tables

Abstract

Accurate regional input-output (IO) tables are indispensable for economic analysis and policy-making, yet their availability remains limited due to data constraints. This paper develops a novel hybrid framework combining Generative Adversarial Networks (GANs) with residual boosting to estimate regional IO tables, and compares it to improved RAS methods. To our knowledge, this is the first approach to embed hard marginal constraints via an IPF layer inside the GAN generator, combined with residual boosting to correct systematic errors. Our approach integrates deep learning with matrix balancing, maintaining economic interpretability while significantly improving statistical performance. Through validation on real-world data from the National Input-Output Tables (NIOT) and World Input-Output Tables (WIOT), we demonstrate that our GAN+Boost framework substantially outperforms the improved RAS technique, achieving improvements of 8.9 and 1.2 percentage points on NIOT and WIOT respectively. Structural metrics show even larger gains: diagonal correlation improves by 160% on NIOT, while Gini-based inequality metrics demonstrate 73–82% better preservation of distributional structure across both datasets. The framework enables more reliable regional economic analysis while reducing data collection costs, with potential applications in regional development planning and economic impact assessment.