基于多因素的国际碳市场价格预测研究

发布时间：2018-01-25 15:57

本文关键词： 碳价预测最优变分模态分解核函数极限学习机广义弹性网多因素模型　出处：《合肥工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：碳排放权是一种不可或缺的商品,其价格波动会影响企业生产经营及碳交易市场的稳定发展。碳市场价格预测是碳市场参与者进行风险管理的关键问题。因此,研究国际碳市场价格预测具有重要的理论和现实意义。由于国际碳市场价格呈现非线性、非平稳、多频率等不规律特征,传统单一模型难以全面刻画碳价波动特征,多频组合预测模型能够深入挖掘碳价在不同频率上隐含的多种内在规律,从而更好地把握碳价波动规律。论文采用最优变分模态分解方法(OVMD)将欧盟碳排放配额(EUA)现货价格分解为多个模态分量;为使分量包含的碳价信息更为集中,基于游程判定法将其重构为碳价低频、中频和高频序列。针对碳价各频率序列受不同因素影响而呈现不同波动特征,而现有研究在预测碳价各频率项时,仅考虑其自身历史数据,未完全涵盖所有预测碳价的信息,可能影响碳价预测的准确度。论文在碳价各频率预测模型构建中,不仅考虑其自身历史数据,还考虑碳市场同类替代品价格、能源价格和宏观经济等因素对碳价各频率序列的影响。针对已有研究运用BP神经网络或SVM等方法预测碳价各频率序列,导致算法不稳定或程序运行较慢等问题,考虑到核函数极限学习机(KELM)方法预测性能稳定且计算效率较高,论文首次引入相应核函数的KELM预测碳价各频率序列,以得到最终的碳价预测结果。研究发现:(1)引入影响因素的碳价多频组合模型预测结果优于仅考虑碳价各频率时间序列的模型,表明在碳价低频项预测中引入CER现货价格、EUA期货价格和煤炭价格等因素,在碳价中频项预测中引入CER现货价格因素,能够为碳价预测提供更多有用的信息,提升碳价预测精度,引入影响因素的多频组合模型更加贴合碳市场实际情况;(2)引入相应的KELM预测碳价各频率序列,能够充分发挥各核函数的优势,很好地解决了已有研究预测碳价时采用BP预测性能不稳定或SVM计算速度有待提高的问题;(3)采用OVMD分解碳价有效解决了EMD分解碳价存在的模态混叠问题,使碳价的变化规律和层次特性更加清晰;(4)碳价多频组合模型预测效果较好,解决了单一模型难以全面刻画碳价波动特征的问题。论文的研究贡献在于拓展了国际碳市场价格预测方法的理论研究,能够为投资者作出合理的国际碳市场投资决策和规避碳市场风险提供有益的理论参考。
[Abstract]:Carbon emission right is an indispensable commodity, its price fluctuation will affect the stable development of enterprise production and operation and carbon trading market. Carbon market price prediction is the key issue of risk management for carbon market participants. It is of great theoretical and practical significance to study the price prediction of international carbon market. Because the price of international carbon market is nonlinear, non-stationary, multi-frequency and other irregular characteristics. Traditional single model can not fully describe the characteristics of carbon price fluctuations, multi-frequency combination prediction model can deeply explore the different inherent laws of carbon price in different frequencies. In this paper, the optimal variational mode decomposition method (OVMD) is used to decompose the spot price of EU carbon emission quota (EUA) into multiple modal components. In order to make the carbon price information contained in the component more concentrated, it is reconstructed into low frequency, middle frequency and high frequency sequences based on run-length decision method. Different fluctuation characteristics are presented for each frequency sequence of carbon price affected by different factors. However, the existing research only considers its own historical data when predicting the frequency terms of carbon price, and does not fully cover all the information of predicting carbon price. This paper considers not only its own historical data, but also the price of similar substitutes in carbon market. The influence of energy price and macroeconomic factors on the frequency series of carbon price. BP neural network or SVM are used to predict the frequency series of carbon price. The algorithm is unstable or the program is running slowly, considering that the kernel function extreme learning machine (KELM) method has stable prediction performance and high computational efficiency. In this paper, KELM with corresponding kernel function is introduced for the first time to predict each frequency sequence of carbon valence. In order to obtain the final carbon price prediction results. It is found that the carbon price multi-frequency combination model with influence factors is superior to the model which only considers the time series of carbon price. The results show that the CER spot price and coal price are introduced into the low frequency carbon price prediction, and the CER spot price factor is introduced in the carbon price intermediate frequency prediction. It can provide more useful information for carbon price prediction, improve the accuracy of carbon price prediction, and introduce the multi-frequency combination model of influencing factors to fit the actual situation of carbon market. 2) introducing corresponding KELM to predict each frequency sequence of carbon price can give full play to the advantages of each kernel function. The problem of using BP to predict carbon price is unstable or the speed of SVM calculation needs to be improved. (3) using OVMD to decompose carbon valence effectively solves the modal aliasing problem of EMD decomposition carbon valence, and makes the variation law and hierarchical characteristic of carbon price more clear. The carbon price multi-frequency combination model has good prediction effect, which solves the problem that it is difficult to describe the fluctuation characteristics of carbon price in a single model. The contribution of this paper is to expand the theoretical research of the international carbon market price forecasting method. It can provide a useful theoretical reference for investors to make reasonable investment decisions in the international carbon market and to avoid the risks of the carbon market.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：X196;F224;F831.5

【相似文献】