人民币汇率长记忆性与风险度量算法研究
发布时间：2018-11-28 19:44
【摘要】：随着我国汇率在两次汇改之后由盯住美元制度走向更为市场化的浮动外汇管理制度，对其市场风险的研究显得尤为必要。本文以美元兑人民币汇率作为研究对象，内容包括以下四个方面。 首先，利用长记忆性参数估计方法发现人民币美元汇率存在显著的长记忆特征。从结构突变、频率结构和时间聚合三个角度考察该市场长记忆特征的根源，得到以下结论：（1）结构突变是引起人民币汇率市场长记忆特征的一大原因；（2）低频序列是支撑长记忆特征的根源，高频意味着随机扰动；（3）长记忆特征存在期限结构特征，即不同周期的数据呈现不同的长记忆动态特征。 其次，为了寻找最优的人民币汇率风险预测模型，本文以回测检验方法作为评价方法，介绍6种刻画波动聚集性、非对称性或长记忆性的风险模型，同时引入FHS技术、EVT技术和EVT-SKT等进行全面分析。结果表明：FIGARCH类模型通过检验且检验值处于低位，，但回测方法无法有效区别波动模型的优劣性以及优劣程度。 再次，基于传统回测方法的不足，本文综合监管成本和超额亏损冲击风险的概念，构造一个新的风险评价指标Indic。应用结果表明：Indic指标可以提供更为详细的局部动态信息，且结合SPA检验方法能有效区别精选长记忆波动模型的优劣性和优劣程度，对风险度量模型择优具有一定的指导意义。 最后，目前的风险度量算法缺乏灵活与可控性，模型之间的选择非此即彼。本文基于Boorstraping算法与模型样本重构的思想，将FHS技术和EVT技术相结合，获得关于VaR的稳健统计分布，同时设置分位数水平为滑点，构造一类改进免疫遗传算法--Indic-SPA检验方法寻找最优的风险度量模型。结果表明：改进的算法能以最低的Indic成本获得风险规避效用。
[Abstract]:With the change of exchange rate from dollar pegging system to more market-oriented floating exchange rate management system, it is necessary to study its market risk. This paper takes the exchange rate of US dollar to RMB as the object of study, including the following four aspects. First, the long memory parameter estimation method is used to find that the RMB dollar exchange rate has significant long memory characteristics. From the three angles of structural mutation, frequency structure and time aggregation, the causes of long memory characteristics in this market are investigated, and the following conclusions are obtained: (1) structural mutation is one of the major causes of long memory characteristics in RMB exchange rate market; (2) low frequency sequence is the root of supporting long memory feature, high frequency means random disturbance; (3) long memory feature exists term structure feature, that is, the data of different period presents different long memory dynamic feature. Secondly, in order to find the best prediction model of RMB exchange rate risk, this paper introduces six risk models which describe volatility aggregation, asymmetry or long memory, and introduces FHS technology. EVT technology and EVT-SKT were comprehensively analyzed. The results show that the FIGARCH model is tested and the test value is low, but the backmeasure method can not effectively distinguish the merits and demerits of the volatility model. Thirdly, based on the deficiency of the traditional method, this paper synthesizes the concepts of supervision cost and excess loss impact risk, and constructs a new risk evaluation index, Indic.. The application results show that the Indic index can provide more detailed local dynamic information, and the combination of SPA test method can effectively distinguish the advantages and disadvantages of the selected long-memory volatility model, which has a certain guiding significance for the risk measurement model to choose the best. Finally, the current risk measurement algorithm lacks flexibility and controllability, and the choice between models is either or. In this paper, based on the idea of Boorstraping algorithm and model sample reconstruction, the robust statistical distribution of VaR is obtained by combining FHS technique with EVT technique, and the quantile level is set as the sliding point. An improved immune genetic algorithm (Indic-SPA test) is constructed to find the best risk measurement model. The results show that the improved algorithm can obtain risk aversion utility at the lowest Indic cost.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP301.6;F832.6
本文编号：2364077
[Abstract]:With the change of exchange rate from dollar pegging system to more market-oriented floating exchange rate management system, it is necessary to study its market risk. This paper takes the exchange rate of US dollar to RMB as the object of study, including the following four aspects. First, the long memory parameter estimation method is used to find that the RMB dollar exchange rate has significant long memory characteristics. From the three angles of structural mutation, frequency structure and time aggregation, the causes of long memory characteristics in this market are investigated, and the following conclusions are obtained: (1) structural mutation is one of the major causes of long memory characteristics in RMB exchange rate market; (2) low frequency sequence is the root of supporting long memory feature, high frequency means random disturbance; (3) long memory feature exists term structure feature, that is, the data of different period presents different long memory dynamic feature. Secondly, in order to find the best prediction model of RMB exchange rate risk, this paper introduces six risk models which describe volatility aggregation, asymmetry or long memory, and introduces FHS technology. EVT technology and EVT-SKT were comprehensively analyzed. The results show that the FIGARCH model is tested and the test value is low, but the backmeasure method can not effectively distinguish the merits and demerits of the volatility model. Thirdly, based on the deficiency of the traditional method, this paper synthesizes the concepts of supervision cost and excess loss impact risk, and constructs a new risk evaluation index, Indic.. The application results show that the Indic index can provide more detailed local dynamic information, and the combination of SPA test method can effectively distinguish the advantages and disadvantages of the selected long-memory volatility model, which has a certain guiding significance for the risk measurement model to choose the best. Finally, the current risk measurement algorithm lacks flexibility and controllability, and the choice between models is either or. In this paper, based on the idea of Boorstraping algorithm and model sample reconstruction, the robust statistical distribution of VaR is obtained by combining FHS technique with EVT technique, and the quantile level is set as the sliding point. An improved immune genetic algorithm (Indic-SPA test) is constructed to find the best risk measurement model. The results show that the improved algorithm can obtain risk aversion utility at the lowest Indic cost.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP301.6;F832.6
【参考文献】
相关期刊论文 前3条
1 华仁海,陈百助;我国期货市场期货价格收益及波动方差的长记忆性研究[J];金融研究;2004年02期
2 张卫国;胡彦梅;陈建忠;;中国股市收益及波动的ARFIMA-FIGARCH模型研究[J];南方经济;2006年03期
3 吕亚芹,何晓群,汤果;FIGARCH模型的参数估计与检验[J];统计研究;1999年S1期
本文编号：2364077
本文链接：https://www.wllwen.com/guanlilunwen/bankxd/2364077.html
