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

Area of Science:

• Econometrics
• Computational Economics
• Regional Science

Background:

• Accurate regional input-output (IO) tables are crucial for economic analysis and policy.
• Data limitations restrict the availability of these essential economic tools.
• Existing methods like RAS have limitations in accuracy and handling complex data.

Purpose of the Study:

• To develop a novel hybrid framework for estimating regional input-output tables.
• To integrate deep learning (GANs) with matrix balancing techniques.
• To outperform existing methods, specifically improved RAS, in accuracy and structural preservation.

Main Methods:

• A hybrid framework combining Generative Adversarial Networks (GANs) with residual boosting.
• Integration of hard marginal constraints via an IPF layer within the GAN generator.
• Correction of systematic errors using residual boosting.
• Validation using National Input-Output Tables (NIOT) and World Input-Output Tables (WIOT).

Main Results:

• The GAN+Boost framework significantly outperforms the improved RAS technique.
• Achieved [Formula: see text] improvements of 8.9 pp on NIOT and 1.2 pp on WIOT.
• Demonstrated substantial gains in structural metrics: 160% improvement in diagonal correlation (NIOT) and 73-82% better preservation of distributional structure.

Conclusions:

• The proposed GAN+Boost framework offers a superior method for estimating regional IO tables.
• The approach enhances economic interpretability and statistical performance.
• Enables more reliable regional economic analysis, reducing data collection costs and aiding policy-making.