A method for spatial downscaling of input-output tables

Abstract

Input-output tables with fine spatial resolution (e.g., municipal input-output tables) are expected to become increasingly important data sources for designing policies aimed at revitalizing regional economies. However, due to constraints in human and financial resources, most basic municipalities have not yet developed such tables. Therefore, non-survey methods capable of generating input-output tables at low cost are gaining attention, although prediction accuracy remains a key challenge.
In this study, we propose a method for constructing regional input-output tables with fine spatial resolution that is both cost-effective and highly accurate. To this end, we introduce an ensemble learning-based approach for predicting production values and other relevant metrics. By using prefectural-level data to predict municipal production values, we demonstrate that our method outperforms conventional approaches in terms of accuracy. Furthermore, we examine the characteristics of the municipal input-output tables generated using our proposed method.