Research Of RMB Exchange Rate Forecasting Based On NDF And NARX Network

With the great econimic globalization today, the position of excange rate has come to a high degree in international economic. It is the bridge that connects each country. The purpose of this study is to seek a new way of exchange rate forecasting, in order to avoid the risks from changes of exchange rate, which have a great significance to both national and foreign-ralated enterprises.During our study of research and explorations about decision theorys of exchange rate, factors that may affect exchange rate. and methods used for exchange rate forecasting, Non Deliverable Forwards, which is known as NDF for short, comes into our view of research. NDF is a kind of forward exchange transactions, is a kind of derivative financial instruments. We found that there is a strong connection between the NDF derivatives and exchange rate, and in this case we try to find a new nonlinear method for forecasting different from the previous theoretical models and linear forecasting methods. Add NDF this kind of economic variables, in order to improve the accuracy of forecasting, and provide theory and method to avoid exchange rate risk for national and foreign-related enterprises.In this paper we selected Nonlinear Auto Regressive neural network with exogenous inputs, NARX network for short, and established a network to forecast RMB exchange rate.Because of its quick reaction of the market when policies published to a certain extent, NDF can be used as the external input of NARX network to improve the performance of exchange rate forecasting network when emergency policies anouced.Using or using not NDF as an external input does not have a great influnce to the result of forecasting when no policies published, and NDF values are correlated with changes in the exchange rate significantly. In a relatively short period after policies published, forecasting network with NDF has advatages to network without NDF. In long term, the error of NARX forecasting network is less than NAR network without NDF as a external input.In this case, the introduction of NDF into NARX network for exchange rate forecasting is effectively feasible. In this paper, we select exchange rate data before and after the reform of RMB exchange rate to do the forecasting and achieved good results.After the effectiveness of the NDF in the forecasting of exchange rate is determined, we try to find out what type of NDF is the most suitable one for RMB exchange rate forecasting. From the experimental results,1-year NDF which is often seleced to research the correlation between NDF and spot market because of its largest trading volume and highest liquidity is not the one that has the strongest interaction relationship with spot market. Trading volume and liquidity of the NDF market have a certain effect on the interactive relationship with spot market, but not decisive. Five different kind of NDF with different period of contract are all have good result when using in exchange rate forecasting, and further verify the effectiveness of the NDF in exchange rate forecasting.This paper measures the process of RMB exchange rate market-oriented in seven years after reform, and trying to seek a quantitative method to illustrate the process of RMB marketization. We use Krean Won NDF as external input of the previously established NARX network, and found that Krean Won NDF can fit the trend of RMB exchange rate well in forecasting, the effect of forecasting is good. We get a further verification that the process of RMB exchange rate marketization is gradually and smoothly.