Scholarlink

<p>In this paper, we compare the predictive performance of six machine learning methods against the AR and VAR models in forecasting global oil prices, aiming to identify the optimal approach. We focus on the European market, Chinese market, and US market, spanning from January 2019 to January 2024. Our findings indicate that the AR model demonstrates the most outstanding performance in forecasting tasks across various markets. However, the RF, XGBoost, LightGBM, CatBoost and LSTM model exhibit statistically equivalent outstanding performance in forecasting tasks to that of the AR model. Besides, to address the potential issue of data leakage associated with the Singular Spectrum Analysis (SSA) method to handle raw sequences in existing literature, we adopt a standardized approach for SSA decomposition and find that the impact is limited. Meanwhile, emergencies such as the COVID-19 pandemic and the Russia-Ukraine conflict can lead to significant fluctuations in oil prices, thereby reducing the predictive accuracy of models. However, we find that popularity trends of emergencies related keywords can improve the predictive modeling.</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

To Hybridize, or not To Hybridize: A Study on Hybrid Models for International Oil Prices Prediction

  • 期刊论文

  • 发表日期：

    2025年12月30日

查看全文
  • Wenxian Li,
  • Kefei Han,
  • Qi Zhang

摘要

In this paper, we compare the predictive performance of six machine learning methods against the AR and VAR models in forecasting global oil prices, aiming to identify the optimal approach. We focus on the European market, Chinese market, and US market, spanning from January 2019 to January 2024. Our findings indicate that the AR model demonstrates the most outstanding performance in forecasting tasks across various markets. However, the RF, XGBoost, LightGBM, CatBoost and LSTM model exhibit statistically equivalent outstanding performance in forecasting tasks to that of the AR model. Besides, to address the potential issue of data leakage associated with the Singular Spectrum Analysis (SSA) method to handle raw sequences in existing literature, we adopt a standardized approach for SSA decomposition and find that the impact is limited. Meanwhile, emergencies such as the COVID-19 pandemic and the Russia-Ukraine conflict can lead to significant fluctuations in oil prices, thereby reducing the predictive accuracy of models. However, we find that popularity trends of emergencies related keywords can improve the predictive modeling.