基于图结构自适应Lasso的碳排放权价格影响因素分析  

Influencing Factors Analysis for the Price of China’s Regional Carbon Emissions Based on Graph-Adaptive Lasso

作　　者：王小燕[1] 周思敏 徐晓莉 周四军[1] WANG Xiao-yan;ZHOU Si-min;XU Xiao-li;ZHOU Si-jun(School of Finance and Statistics,Hunan University,Changsha 410012,China;School of Mathematics,Sun Yat-sen University,Guangzhou 510000,China)

机构地区：[1]湖南大学金融与统计学院,湖南长沙410012 [2]中山大学数学学院,广东广州510000

出　　处：《统计与信息论坛》2022年第4期73-83,共11页Journal of Statistics and Information

基　　金：国家社会科学基金重点项目“中国能源高质量发展的统计监测研究”(19ATJ007);湖南省社会科学成果评审委员会资助项目“数据融合视角的高维机器学习方法及其应用研究”(XSP22YBZ003)。

摘　　要：碳交易作为实现低碳经济的一种途径,既具有环境效益,又具有经济效益。为了研究碳排放权价格的影响因素,选取广州碳排放权交易所的碳配额价格收盘价(GDEA)为研究对象,从6个维度构建了24个指标:国际碳价、国内外经济指标、国外能源指标、国内能源指标、气候环境和宏观政策,并将指标间复杂的相关关系纳入模型来改进指标筛选效果。首先基于复杂网络理论构建了24个指标的图结构,表示它们的复杂联动关系,再建立图结构自适应Lasso方法(G-AdLasso)进行影响因素识别。研究发现:指标之间存在无可忽视的中等或高度相关,依据两两相关关系建立图结构时,上述24个指标可被分为6个团体,体现了指标的内部关系。同时G-AdLasso选择出了10个因素,其中欧盟核证减排量收盘价影响最为显著,欧盟EUA收盘价、迪拜原油现货价、美元兑人民币中间价4个因素对GDEA有正向影响;欧盟CER收盘价、NYMEX天然气期货收盘价、欧洲三港DES ARA动力煤指数、广州工业天然气市场价、广州日最高气温、银行间7日同业拆借平均利率、欧元兑人民币中间价7个因素对GDEA有负向作用;这些因素在上述6个维度上均有涉及,且它们在图结构中具有较高的度,说明G-AdLasso可识别出图结构中较重要的指标。对比不带图结构的自适应Lasso和Lasso方法,G-AdLasso方法选择更少的指标,说明该方法可优化和精简模型。As a way to realize low-carbon economy,carbon trading has both environmental and economic benefits.In order to identify the influencing factors of carbon emission price,this study takes the Carbon Emission Allowance price of Guangzhou Carbon Emission Exchange as the response(referred to as GDEA).There are 24 possible indexes,which describe the influencing factors of GDEA from the following six aspects:international carbon price,domestic and foreign economic influence,foreign energy,domestic energy,climate environment,and macro policies.To select the important factors that may have significant effect of GDEA,adopt the adaptive Lasso penalty,which can realize factors selection and simultaneously conduct model estimation.It shows Oracle property in the existing literature and thus is widely used in variable selection problems.To improve the accuracy of factors selection,incorporate the complex correlations among the indexes via a network structure.The network among factors shows that all the 24 indicators can be divided into six modules,which can be summarized as high temperature effect,trade markets between Shanghai/Guangdong and Europe,energy and economic market,Carbon market in Europe and America,European coal market,and exchange rate of US dollar.This structure also indicates that correlations among factors cannot be ignored.The empirical analysis shows that adaptive Lasso with network structure(referred to as G-AdLasso)selects 10 important factors.Among them,EUA has the most significant effect on GDEA.EUA closing price,Dubai crude spot price,and the central parity of the Chinese Yuan against the US dollar present positive effects on GDEA.CER(European Union)closing price,NYMEX natural gas futures closing price,DES ARA Thermal Coal Index in Three European ports,Guangzhou Industrial Natural gas Market price,daily maximum temperature in Guangzhou,average inter-bank offered rate for 7 days,and the central parity of the Chinese Yuan against euros have negative effects on GDEA.The identification results indicate a rea

关 键 词：碳排放权价格 变量选择 Lasso 复杂网络 低碳经济 

分 类 号：F222[经济管理—国民经济]