【作者】 马超；

【导师】 李颖；

【作者基本信息】 山东大学 ， 金融（专业学位）， 2017， 硕士

【摘要】 2005年7月人民币汇率制度改革,中国人民银行宣布我国实行以市场供求为基础、参考一篮子货币进行调节、有管理的浮动汇率制度,自此人民币汇率告别盯住美元,形成了更富弹性的汇率定价机制。汇率弹性的加大和波动性的增强给国际经济中的企业和个人决策带来挑战,也凸显了人民币汇率预测的重要性。许多传统理论模型和线性预测方法在新的经济环境下表现出的预测效果不理想,推动了非参数汇率预测方法的兴起,以反映汇率变动呈现的非线性特征。但是,以人工神经网络为代表的一类非参数汇率预测方法虽然在人民币汇率预测中取得了良好效果,但该方法过度重视样本内汇率数据的拟合效果,过度拟合容易影响汇率样本外数据的拟合质量,其理论基础和模型结构选择在实际运用中也存在局限性。因此通过分析以往汇率预测模型的不足,并在传统汇率决定理论的基础上构建新型的汇率预测指标体系,使其更加符合开放经济环境下的人民币汇率变动特征,为汇率预测提供一种科学有效的方法,正是本文研究的意义和创新之处。支持向量机方法建立在统计学习理论的结构风险最小原理基础之上,其理论基础完备且在金融时间序列预测领域应用广泛。本文详细梳理了有关汇率预测的前人研究和相关理论,在此基础上将汇率预测的传统理论与支持向量机方法相结合,通过构建新型的汇率预测指标体系,对人民币汇率进行了实证预测和分析。文章结构分为五大部分,第一章介绍选题背景、研究意义、汇率预测国内外研究现状、文章主要内容、研究方法及文章创新点等;第二章阐述传统汇率决定理论和支持向量机原理;第三章着重介绍汇率预测模型构建;第四章对汇率预测模型的实证效果进行对比和评价;第五章对基于支持向量机的汇率预测方法进行总结并作研究展望。本文运用支持向量机方法对人民币汇率进行实证预测,以自回归模型和BP神经网络模型为对照,结果表明基于支持向量机的汇率预测方法不仅在预测精度和稳定性方面优于对照模型,在模型理论基础方面也更加完备,更能体现汇率行为所反映的经济因素。更多还原

【Abstract】 In July 2005,the People’s Bank of China announced that China’s implementation of managed floating exchange rate regime based on market supply and demand with reference to a basket of currencies which was the big reform of RMB exchange rate system.China’s exchange rate regime has gradually formed a more flexible exchange rate pricing mechanism after the ending of pegging solely to the USD.The volatility and non-linear characteristics of exchange rate are becoming more and more obvious.Uncertainties of exchange rate changes bring challenges to trade decision-making,and affect the formulation and adjustment of relevant policies such as exchange rate policy and interest rate policy.Forecasting RMB exchange rate accurately becomes increasingly important but difficult in the meantime.Most traditional statistical model models and linear forecasting methods perform less well in the new economic environment.While the non-parametric exchange rate forecasting method,which is represented by artificial neural network,has achieved good results in the RMB exchange rate forecasting.However,the theoretical basis and the choice of model structure have limitations which lead to over-fitting and poor forecasting results of out-of-sample data of RMB exchange rate in practice.Therefore,to overcome the shortcomings of the previous exchange rate forecasting model,it is of great practically importance to construct a RMB exchange rate forecasting model in the more open environment based on the exchange rate determination theory,which is the core significance and innovation of the paper.Based on the theory of statistic learning and minimum structure risk principle,support vector machine(SVM)method has a more complete theoretical basis,and it is widely used in financial time series forecasting field.Based on the previous study of exchange rate forecasting,this paper combines the exchange rate determination theory with the support vector machine(SVM)method.By reconstructing the exchange rate index system,the autoregressive model and the BP neural network model are used as the control models to evaluate and analyze the RMB exchange rate forecasting results.The paper is divided into five parts.In the first part,it combs the previous literatures and research results,briefly introduces the content structure and innovation of the article.The second part elaborates the basic theory of exchange rate determination theory and support vector machine(SVM),which is intended to find the theoretical basis for exchange rate forecasting.Then in the third part,it mainly introduces the steps of the exchange rate forecasting model construction.The fourth part is the empirical part of this paper which analyzes the exchange rate of the RMB against the US dollar by applying the exchange rate forecasting model based on the support vector machine(SVM)method.The last part summarizes the method of forecasting the exchange rate based on support vector machine(SVM)and research of the method is discussed.Comparative analysis are made from the perspective of the overall error and stability of the forecasting error using autoregressive and BP neural network as control models.The comparison results show that the method based on support vector machine is not only better than the control models in forecasting accuracy and stability,but also more comprehensive in the theoretical basis of the model,which contains the economic factors reflected by exchange rate behavior.更多还原