Scholarlink

<p>This paper develops a machine learning–based decision-support framework to evaluate the trade-off between continuing vessel operations and recycling for residual steel value. We compare the forecasting performance of Long Short-Term Memory (LSTM) networks with traditional econometric models (ARIMA, VAR/VECM, and GARCH/CGARCH) in predicting dry bulk freight rates and steel scrap prices. As vessel classes differ in cargo markets, volatility structures and steel-yield economics, models are estimated separately by segment, a heterogeneity central to scrapping incentives and forecast accuracy. Using a proprietary weekly dataset (1996–2025) from Clarksons Research and a major demolition firm, the LSTM consistently outperforms econometric benchmarks across classes and horizons. Incorporating macroeconomic variables and exogenous shocks further strengthens predictive robustness. Embedded in a real-time simulation environment, the framework enables shipowners to test scenario-based fleet-renewal decisions. Defining resilience as the ability to adjust renewal timing under uncertainty, the study offers a scalable tool for optimizing scrapping strategies.</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

Key drivers of the shipping sector: comparing time-series econometrics and machine learning artificial intelligence for forecasting freight rates and steel scrap prices

  • 期刊论文

  • 发表日期：

    2025年12月16日

查看全文
  • Xakousti Afroditi Merika,
  • Rafaela Theodora Merika,
  • Theodore Syriopoulos

摘要

This paper develops a machine learning–based decision-support framework to evaluate the trade-off between continuing vessel operations and recycling for residual steel value. We compare the forecasting performance of Long Short-Term Memory (LSTM) networks with traditional econometric models (ARIMA, VAR/VECM, and GARCH/CGARCH) in predicting dry bulk freight rates and steel scrap prices. As vessel classes differ in cargo markets, volatility structures and steel-yield economics, models are estimated separately by segment, a heterogeneity central to scrapping incentives and forecast accuracy. Using a proprietary weekly dataset (1996–2025) from Clarksons Research and a major demolition firm, the LSTM consistently outperforms econometric benchmarks across classes and horizons. Incorporating macroeconomic variables and exogenous shocks further strengthens predictive robustness. Embedded in a real-time simulation environment, the framework enables shipowners to test scenario-based fleet-renewal decisions. Defining resilience as the ability to adjust renewal timing under uncertainty, the study offers a scalable tool for optimizing scrapping strategies.