Scholarlink

<p>Global supply chains during the post-COVID19 period faced with unprecedented volatility, characterized by structural regime shifts and greater inter-dependency among energy, shipping, and logistics sectors. Additionally, the traditional univariate forecasting models fail to capture these spatial spillovers across the economic network. Thus, this study proposes a novel spatio-temporal framework for predicting supply chain price dynamics. We first use Economic Network Analysis (ENA) to map the shifting topology of global interdependencies, identifying structural changes among the pre- and post-COVID19 periods. Further, these inputs are integrated into a Graph Attention Network (GAT) coupled with a Long Short-Term Memory (LSTM), producing a walk-forward validation mechanism to handle post-pandemic non-stationarity. Our results reveal a structural break in the post-COVID19 period, with Systemic Risk Index (SRI) for Crude oil increasing by over 341%, representing a primary risk transmitter. The proposed GAT-LSTM framework is benchmarked against Vanilla LSTM, Stacked LSTM, and Transformer models using recursive walk-forward forecasting. Notably, the proposed hybrid model achieved a 7.92% improvement in directional accuracy (Hit Ratio) for gold relative to the Vanilla LSTM benchmark, while also demonstrating comparatively stronger forecasting effectiveness for key logistics and freight-related assets, including MAERSK, UPS, and BDRY. Furthermore, the results demonstrate that supply chain forecasting is comparatively enhanced when modeled as a graph rather than a series of independent lines. The novel integration of spatial attention allows the framework to autonomously identify the super-transmitter nodes, offering a robust, early-warning tool for proactive risk management in a highly interdependent global economy.</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 systematic risk and price dynamics in disrupted supply chains: a dynamic GAT-LSTM approach based on economic network topologies

  • 期刊论文

  • 发表日期：

    2026年6月1日

查看全文
  • Bilal Ahmed Memon

摘要

Global supply chains during the post-COVID19 period faced with unprecedented volatility, characterized by structural regime shifts and greater inter-dependency among energy, shipping, and logistics sectors. Additionally, the traditional univariate forecasting models fail to capture these spatial spillovers across the economic network. Thus, this study proposes a novel spatio-temporal framework for predicting supply chain price dynamics. We first use Economic Network Analysis (ENA) to map the shifting topology of global interdependencies, identifying structural changes among the pre- and post-COVID19 periods. Further, these inputs are integrated into a Graph Attention Network (GAT) coupled with a Long Short-Term Memory (LSTM), producing a walk-forward validation mechanism to handle post-pandemic non-stationarity. Our results reveal a structural break in the post-COVID19 period, with Systemic Risk Index (SRI) for Crude oil increasing by over 341%, representing a primary risk transmitter. The proposed GAT-LSTM framework is benchmarked against Vanilla LSTM, Stacked LSTM, and Transformer models using recursive walk-forward forecasting. Notably, the proposed hybrid model achieved a 7.92% improvement in directional accuracy (Hit Ratio) for gold relative to the Vanilla LSTM benchmark, while also demonstrating comparatively stronger forecasting effectiveness for key logistics and freight-related assets, including MAERSK, UPS, and BDRY. Furthermore, the results demonstrate that supply chain forecasting is comparatively enhanced when modeled as a graph rather than a series of independent lines. The novel integration of spatial attention allows the framework to autonomously identify the super-transmitter nodes, offering a robust, early-warning tool for proactive risk management in a highly interdependent global economy.