Scholarlink

<p>Forecasting cotton prices enables farmers, traders, and policymakers to make informed planting, buying, and risk-management decisions by anticipating market trends and potential price shocks. Accurate price forecasts also support efficient supply-chain planning and help mitigate financial losses through timely use of hedging instruments. This study proposes an innovative forecasting framework leveraging Gaussian process regression (<i>GPR</i>) to predict cotton prices in the Chinese market. Model hyperparameters are inferred through a Bayesian estimation scheme, allowing the system to adapt in real time to underlying market dynamics and abrupt structural changes. By integrating these evolving features, the method more accurately captures shifts in price trajectories. The empirical analysis employs daily observations from July 19, 2004 to April 10, 2025, encompassing phases of regulatory reform, industrial restructuring, and macroeconomic transition. Validation on an out-of-sample period from March 1, 2021 to April 10, 2025 yields a relative root mean square error (<i>RRMSE</i>) of 0.1532%, an <i>RMSE</i> of 26.4548, a mean absolute error (<i>MAE</i>) of 18.0736, and a correlation coefficient (<i>CC</i>) of 0.99994. To our knowledge, this work constitutes the first deployment of a Bayesian-tuned <i>GPR</i> model for cotton-price forecasting. Beyond its contribution to theoretical advances in machine-learning-based price prediction, the approach offers a versatile analytical framework applicable to analogous commodity time-series forecasting tasks.</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

Machine Learning Cotton Price Forecasts Based on Gaussian Process Regression Models with Hyper-parameters Tuned Through Bayesian Optimization and Cross Validation

  • 期刊论文

  • 发表日期：

    2025年10月24日

查看全文
  • Bingzi Jin,
  • Xiaojie Xu

摘要

Forecasting cotton prices enables farmers, traders, and policymakers to make informed planting, buying, and risk-management decisions by anticipating market trends and potential price shocks. Accurate price forecasts also support efficient supply-chain planning and help mitigate financial losses through timely use of hedging instruments. This study proposes an innovative forecasting framework leveraging Gaussian process regression (GPR) to predict cotton prices in the Chinese market. Model hyperparameters are inferred through a Bayesian estimation scheme, allowing the system to adapt in real time to underlying market dynamics and abrupt structural changes. By integrating these evolving features, the method more accurately captures shifts in price trajectories. The empirical analysis employs daily observations from July 19, 2004 to April 10, 2025, encompassing phases of regulatory reform, industrial restructuring, and macroeconomic transition. Validation on an out-of-sample period from March 1, 2021 to April 10, 2025 yields a relative root mean square error (RRMSE) of 0.1532%, an RMSE of 26.4548, a mean absolute error (MAE) of 18.0736, and a correlation coefficient (CC) of 0.99994. To our knowledge, this work constitutes the first deployment of a Bayesian-tuned GPR model for cotton-price forecasting. Beyond its contribution to theoretical advances in machine-learning-based price prediction, the approach offers a versatile analytical framework applicable to analogous commodity time-series forecasting tasks.