Nonparametric Additive Exchange Rate Predictive Models

The exchange rate is one of the most important indicators in the national macroeconomic operation index system.Its value identifies the exchange ratio between two different currencies.Its fluctuation reflects the variety of supply and demand of one currency,providing reference and analysis for macroeconomic regulation.Since the collapse of the Bretton Woods system based on the US dollar and gold in 1973,countries around the world have adopted a floating exchange rate system.Exchange rate forecasting has been widely concerned and used in the financial field.More accuracy of exchange rate forecasts can be used for banks and enterprises' decision guidance,especially in the trading that involving foreign exchange.From the perspective of risk aversion,the accurate prediction of the exchange rate can improve the risk aversion ability and reduce the probability of damage to related financial assets.From the perspective of maximize the benefits in economics,increase the accuracy of the predicted exchange rate can provide a basis and assistance to the decision-making process of maximizing.There are many literatures on the study of exchange rate forecasting.Different exchange rate decision theory and exchange rate forecasting methods are constantly seeking better predictive effects,and relevant research on exchange rate forecasting is continuing.In order to find new exchange rate prediction methods and obtain higher prediction accuracy,based on the nonparametric modeling theory,we study exchange rate volatility and its forecasting issue.From the perspective of methodology,the nonparametric method is more flexible than any parameter or semiparametric method,but in order to overcome the so-called "the curse of dimensionality",we propose a nonparametric additive model to study the exchange rate forecasting.The nonparametric unknown functions will be approximated and estimated by their truncated orthogonal series developments in the proper function space.Therefore,the nonparametric additive model should perform better than the parametric model and random walk model in the literature.The innovation of this paper lies in the selection of the function space,consider a Hilbert speace:L2{[a,b],?(x)} space,which includes all the square integrable functions in the closed interval[a,b](measured by?(x)dx).Using a set of orthogonal bases of Hilbert space L2{[a,b],?(x)} to expand the nonparametric unknown functions,the unknown functions can be well approximated in some sense.Hence,the functions are estimable.Empirical study using the proposed model is divided into two parts to verify the prediction effect.The first part selects the exchange rate data widely used in the exchange rate forecasting literature.After using the data to obtain the model solution,let the Nonparametric Additive Exchange Rate Predictive Models do the out-of-sample prediction,calculate the loss function and compare the predicted values with the real values.The results show that the proposed model outperforms all the benchmark models in the literature,the root mean square errors of the proposed model are minimal.On the one hand,it tells that the proposed model has better out-of-sample predictive power.On the other hand,it is also a strong proof that the random walk model is not really difficult to surpass(The Meese and Rogoff Puzzle,see the text).The second part of the empirical study predicts the latest RMB exchange rate data,calculates the predicted loss function and compares the predicted values with the real values.It also obtains an ideal prediction effect,indicating the model has a nice adaptability and predictive power for the RMB exchange rate in the current economic environment.This study contributes to the literature a novel methodology for exchange rate prediction,and helps to eliminate the pessimistic tone of unpredictable exchange rate.