Scholarlink

<p>This paper proposes and evaluates a new measure of latent economic structure for the concept of economic dependency within the world-system. We propose Dynamic Dependency Index (DDI), derived from a Long Short-Term Memory (LSTM) autoencoder, for capturing the complex, non-linear, and path-dependent nature of national economic development. We apply this model to the panel data set of 60 countries over 26 years (1995–2020), and create a two-dimensional dependency space describing the global core-periphery-semiperiphery hierarchy. The LSTM autoencoder demonstrates improved clustering performance in capturing the dynamic structural dependency compared to other methods such as Principal Component Analysis (PCA) and static autoencoders. The results obtained with a Silhouette score of 0.545 (higher than 0.464 for PCA) indicate more coherent and well-separated clusters. The resulted latent dimensions (Structural Complexity &amp; Productivity and Financial vulnerability &amp; Dependency) are strongly and significantly related to established economic structure such as manufacturing exports (ρ = 0.883) and R&amp;D expenditure (ρ = 0.747). This paper contributes three main contributions: (1) a novel application of LSTM autoencoders to quantify a theory-laden concept (economic dependency) from large-N panel data, demonstrating that established deep learning architectures can serve as powerful measurement instruments in computational social science, (2) a significant empirical contribution mapping the global dependency structure and evolution for 60 countries outside small-N data sets, and (3) evaluation of the proposed DDI as a heoretically-grounded measure of structural economic positions. Our results are consistent with the structural predictions of world-systems analysis and provide a reproducible, data-driven measurement framework for tracking the evolution of global economic positions over time.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Quantifying economic dependency through computational structural economics: an LSTM autoencoder approach

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Ping Chen

摘要

This paper proposes and evaluates a new measure of latent economic structure for the concept of economic dependency within the world-system. We propose Dynamic Dependency Index (DDI), derived from a Long Short-Term Memory (LSTM) autoencoder, for capturing the complex, non-linear, and path-dependent nature of national economic development. We apply this model to the panel data set of 60 countries over 26 years (1995–2020), and create a two-dimensional dependency space describing the global core-periphery-semiperiphery hierarchy. The LSTM autoencoder demonstrates improved clustering performance in capturing the dynamic structural dependency compared to other methods such as Principal Component Analysis (PCA) and static autoencoders. The results obtained with a Silhouette score of 0.545 (higher than 0.464 for PCA) indicate more coherent and well-separated clusters. The resulted latent dimensions (Structural Complexity & Productivity and Financial vulnerability & Dependency) are strongly and significantly related to established economic structure such as manufacturing exports (ρ = 0.883) and R&D expenditure (ρ = 0.747). This paper contributes three main contributions: (1) a novel application of LSTM autoencoders to quantify a theory-laden concept (economic dependency) from large-N panel data, demonstrating that established deep learning architectures can serve as powerful measurement instruments in computational social science, (2) a significant empirical contribution mapping the global dependency structure and evolution for 60 countries outside small-N data sets, and (3) evaluation of the proposed DDI as a heoretically-grounded measure of structural economic positions. Our results are consistent with the structural predictions of world-systems analysis and provide a reproducible, data-driven measurement framework for tracking the evolution of global economic positions over time.