军事知识服务平台
基于MS-BSMOTE-SVM算法的碳市场智能风险预警研究
Research on Intelligent Risk Early Warning in Carbon Market Based on MS-BSMOTE-SVM Algorithm
摘要 维护碳排放权交易市场的安全对保障经济可持续性、促进政策执行和稳定金融环境至关重要.鉴于碳价格受金融和能源市场影响可能呈现多重分形波动,本文基于主要碳市场日收益率,在挖掘出多重分形特征基础上,构建多重分形谱刻画真实风险状态,并结合BSMOTE采样与SVM算法,建立MS-BSMOTE-SVM碳市场风险预警模型.实证结果表明,相较GARCH模型,引入多重分形谱划分碳市场风险状态并结合BSMOTE采样优化非均衡数据集,能够显著提升各类预警模型的预测性能.本研究为碳市场风险管理提供了科学依据,也为此类缺乏明确风险标签的市场开发了一套高效的风险识别与预警方法.
Abstract:
As an emerging trading system under environmental regulation,the carbon market faces higher uncertainty and systemic risk than traditional financial markets.The drastic fluctuation of carbon price seriously affects the stability of the market and the compliance cost of enterprises.Existing risk management mechanisms have two shortcomings:linear models are difficult to characterize nonlinear dynamics,and machine learning models are limited by sample imbalance and lack of risk labels.In this study,an intelligent early warning framework is constructed to provide a scientific basis for risk management in the carbon market,and to develop a set of efficient risk identification and early warning methods for this kind of market that lacks clear risk labels. This study proposes a hybrid MS-BSMOTE-SVM model to solve the difficult problem of risk early warning in the carbon market,and the specific methods and data are as follows:the model quantifies the risk state through multifractal spectrum,and utilizes the multifractal spectrum width as the risk threshold instead of the traditional manually set threshold;the BSMOTE algorithm effectively solves the sample imbalance problem through the bounded-sample synthesis technique,and at the same time avoids the noise introduction;PSO optimization algorithm is used to adjust SVM parameters so that the model can adapt to the dynamic evolution of different markets.The research data come from the five major carbon markets in China from 2016-2024,and the 20-dimensional raw features are downscaled to 13 principal components by PCA,which retains the key signals of policy sensitivity. The empirical results show that the MS-BSMOTE-SVM model significantly outperforms the GARCH model in risk state classification,and the MS-BSMOTE-SVM model outperforms the benchmark model in abnormal state identification.The innovation of this model is that it solves the classification problem of unlabeled data by labeling and improves the ability of SVM to recognize a few categories by the BSMOTE algorithm.These findings provide a new technical path for risk management in the carbon market:regulators can dynamically adjust their policies according to the thresholds;emission control enterprises can optimize their production plans by using risk warnings;and investment institutions can design hedging strategies based on the warnings.The framework develops an efficient risk identification and early warning methodology for such markets that lack clear risk labels.
刊 名 工程管理科技前沿 ISTICPKU
年,卷(期): 2026, 45(5)
分类号: F124.5
机标分类号: F270 F832.51 TP301.6
在线出版日期: 2026年10月16日
基金项目: 2022M720545:中国博士后科学基金资助项目71903017:国家自然科学基金2023NSFSC0523:四川省自然科学基金资助项目,