基于NDF与NARX网络的人民币汇率预测研究

发布时间：2018-07-10 04:16

本文选题：人民币汇率 + 预测　；参考：《大连理工大学》2013年博士论文

【摘要】：在经济高度全球化的今天,汇率在国际经济中的地位越来越重要,越来越深刻的影响着各国之间的经济与贸易往来。本文研究的目的就是为汇率预测寻求一种新的方法,以此规避汇率变动所带来的风险,这对国家和涉外经济体都有着十分重要的意义。在对汇率的决定理论,可能影响汇率的因素,以及汇率预测的方法等进行研究和探索时,我们发现无本金交割远期外汇(Non Deliverable Forwards,简称NDF)这种金融衍生物与汇率之间存在着很大的联系,所以我们试图寻找一种新的,不同于以往理论模型和线性预测方法的非线性预测方法,加入NDF这种经济变量,以提高预测的精度,并对国家和涉外企业等提供良好的规避汇率风险的理论和方法。本文选用有外部输入的非线性自回归神经网络(Nonlinear Auto Regressive Neural Network with Exogenous Inputs),简称NARX网络,建立NARX人民币汇率预测网络,由于NDF在一定程度上可以代表政策出台时市场的反应,所以它可以作为NARX网络的外部输入,以改善在突发政策时,NARX网络在汇率预测方面的性能。在无政策出台时,使用NDF作为外部输入和不使用NDF的预测结果基本一致,NDF值与汇率变动值存在较明显的相关性。在出台政策的较短时间内,有NDF加入网络的性能优于无NDF加入的网络。长时间来看,预测人民币汇率时,引入NDF的NARX网络的误差小于无NDF的NAR网络,所以引入NDF的NARX网络用于汇率预测是有效的可行方案。选取人民币两次汇改前后数据对人民币汇率进行了预测,取得了良好的效果。在确定了NDF在汇率预测中的有效性之后,我们尝试找出究竟哪种NDF用于人民币汇率预测时的效果最好。从数据实验结果发现,NDF期限越短,与即期市场的互动关系越强,而常被以往文献用来研究与即期市场关联性的1年期NDF,其并不是与即期市场互动关系最强的。NDF合约的交易量、流动性对其与即期市场的互动关系有一定的影响,但不是决定性的。五个不同合约期限的NDF数据用于汇率预测的效果都很好,进一步验证了NDF在汇率预测中的有效性。本文衡量了汇改后人民币汇率市场化进程,试图从定性和定量两个角度来衡量人民币市场化进程。我们将韩币NDF作为外部输入加入到前文所建立的NARX网络中,发现韩币NDF可以很好地对人民币汇率走势进行预测,预测效果良好。说明人民币汇率走势已同国际外汇市场有效接轨,并且可以再一次验证,人民币汇率市场化进展顺利。
[Abstract]:In today's highly globalized economy, the exchange rate plays a more and more important role in the international economy and has a profound impact on the economic and trade exchanges between countries. The purpose of this study is to find a new method to avoid the risk of exchange rate change, which is of great significance to both countries and foreign economies. In the course of studying and exploring the theory of exchange rate determination, the factors that may affect exchange rate, and the methods of exchange rate prediction, We found that there was a strong link between the non-deliverable forwards (NDF), a financial derivative, and the exchange rate, so we tried to find a new one. Different from the previous theoretical models and linear forecasting methods, NDF is added as an economic variable to improve the accuracy of prediction, and to provide a good theory and method to avoid exchange rate risk for countries and foreign enterprises. In this paper, nonlinear Auto recurrent neural network with inputs (NARX network) is used to establish the NARX RMB exchange rate forecasting network. To some extent, NDF can represent the market reaction when the policy is introduced. So it can be used as the external input of NARX network to improve the performance of NARX network in exchange rate prediction in case of sudden policy. In the absence of policy, the results of using NDF as external input and not using NDF are basically consistent. There is a significant correlation between NDF and exchange rate change. In a short period of time, the performance of NDF joining the network is better than that of the network without NDF. In a long time, the error of NARX network with NDF is less than that of NAR network without NDF, so it is feasible to use NARX network of NDF to predict exchange rate. The data before and after the two RMB exchange rate reforms are selected to predict the RMB exchange rate, and good results have been obtained. After determining the effectiveness of NDF in exchange rate forecasting, we try to find out which NDF is the most effective in RMB exchange rate forecasting. It is found from the data experiment that the shorter the NDF period, the stronger the interaction with spot market. However, the one-year NDFs, which are often used to study the relationship between NDF and spot market, are not the trading volume of .NDF contract, which has the strongest interaction with spot market. Liquidity has a certain influence on the interaction with spot market, but it is not decisive. Five NDF data with different contract duration have good effect on exchange rate forecasting, which further verifies the effectiveness of NDF in exchange rate forecasting. This paper measures the marketization process of RMB exchange rate after the exchange rate reform, and tries to measure the marketization process of RMB from two aspects: qualitative and quantitative. We add Korean NDF as external input to the NARX network, and find that Korean NDF can predict the trend of RMB exchange rate very well, and the forecast effect is good. It shows that the trend of RMB exchange rate has been effectively connected with the international foreign exchange market, and it can be verified once again that the marketization of RMB exchange rate is progressing smoothly.
【学位授予单位】：大连理工大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：F832.6;F224

【参考文献】