The Multivariate Discription And Forecasting Of Renminbi Exchange Rate Based On Nonliear Dependency Analysis

The focus on the description and prediction of the exchange rate behavior has along history, from the perspective of the historical process of human social andeconomic development. Especially under the floating exchange rates regimes,understanding and grasp of the exchange rate changes behavior have rich theoretical andpractical significance, with the development of the economic and financial globalizationand the modern international foreign exchange markets. At the same time, Renminbi isbecoming more and more important base on the Chinese economy’s contribution to theworld economic development, when Chinese economy has achieved a stable growingand the degree of opening of China as the biggest developing country has beenimproved. So under this background, accurately and effectively multivariate descriptionand forecasting of the RMB exchange rate on Chinese market can not only helpevaluating the rationality and validity of the RMB exchange rate policy, but alsoprovides a feasible analysis tool of the future reform direction, by considering thedifferences between data correlation, market information flows and transmissionefficiency, the linear and nonlinear dependency structure between RMB exchange ratesequences on different international foreign exchange market.In this paper, base on the above reality and theoretical background, the study isarranged to seven parts following the general research idea from the theoretical analysisto the empirical test and then to the theory inference: firstly we review the internationalforeign exchange market and summarize practical experience of the development ofChinese foreign exchange market, then distinguish the definition of exchange ratebehavior research, and compare the differences between the two kinds of basic researchparadigm and the research train of thought. Based on the above, the exchange ratedetermination theories and relative models are analyzed in the three major evolvingcurrency systems. Secondly the methods of statistical learning theory associated to thedescription and prediction of the exchange rate behavior is introduced, and we carry onempirical test on the different RMB exchange rate data from four exchange markets tofind the nonlinear dependency between them. We take a series of empirical test onnonlinear dependences and low-dimensional chaotic characteristics of the RMBexchange rate sequences from Chinese, British, American and Singapore markets, underthe two different assumptions about where the nonlinear dependency come from, and on the basis of phase space reconstruction theory and surrogate data method. In thefollowing part, on the basis of the empirical test on the relative correltions betweendifferent RMB exchange rate sequences, a multiple dynamic Gauss radial basis functionneural network model is proposed and constructed. The in sample description andgeneralization ability of the different empirical models are compared, and sometheoretical hypothesis,such as the openness of Chinese exchange market, the flexibilityand chaotic characteristics of the RMB exchange rate and so on, are also be tested. Thefinal part is the conclusions of the entire study.The main empirical results of the reaserch are obtained as follows: Firstly, theresults of the stationary test, Self-autocorrelation coefficients, and BDS test show thatthe RMB exchange rates in the different research period from four markets don t followthe simple random walk process, because of the nonlinear dependencies embedded inthe exchange rate data. Secondly, we can capture and explain the conditionalheteroskedasticity in the form of volatility clustering and leverage effect in the RMBexchange rate return series. Thirdly there are some general nonlinear dependencies inthe exchange rate system, and all the nonlinear dependency structures in the data showthe unstable in the present of episodic and transient features. Fourthly, the multipledynamic Gauss radial basis function neural network model is better than the otherempirical models both in description and forecasting abilities. Fifthly, base on the aboveempirical results, we can draw the theoretical assumption conclusions that there aredefinitely some other form nonlinear dependencies in the exchange rate data except forGARCH type, the chaos assumption is in some content reasonable, the RMB exchangerate in Chinese market can be predicted by the other data from different markets, andthe openness, correlation, efficiency of information transfer degree of the Chinesemarket show different structural characteristics.