The Neural Network and Exchange Rate Modeling
A. M. M. Jamal · 2005
This paper applies a neural network model for representing the exchange rate of the United States with two of its major trading partners, Canada and the Euro countries. A one-period lagged exchange rate was used as the explanatory variable. The results show that this model was quite successful in predicting the exchange rates studied. They also indicate that models of the relationship of an exchange rate with its explanatory variable may be updated as often as necessary for estimation and forecasting. I. Introduction The value of international currencies traded daily is approximately 3 trillion dollars (Bank of International Settlements). In comparison, the value of goods and services would be less than 5 percent of that amount (Froot and Thaler, 1990). Consequently, the relationship between an exchange rate and its explanatory variables continues to be of interest to the researchers. Most studies use a regression model to estimate the relationship between a bilateral exchange rate and variables such as relative interest rates, money supply, industrial production, inflation and other macroeconomic variables. The models generally assume a linear or a logarithmic relationship between the dependent and explanatory variables. Some researchers have found that a random walk model best explained the exchange rate behavior (for example, Meese and Rogoff, 1983). Others such as Hooper and Motion (1982), Frankel( 1983), Boughton ( 1987) and Fuhrer and Weiler ( 1991 ) concluded that models based on economic factors were more successful in explaining exchange rate movements. Frankel and Rose (1995) and Taylor(1995) provide comparison of various of various models. Some have analyzed the effect exchange intervention and monetary policy on an exchange rate (Kim 2003). Most of these models, however, do poorly in predicting exchange rates in the short run. Since a daily or even hourly or more frequent trade is based on short-term factors, it is crucial to explore other means for modeling and predicting exchange rates. Neural networks may be used to model nonlinear relationships between an exchange rate and its explanatory variables. The purpose of this paper is to apply a neural network model to forecast the daily bilateral exchange rate between Canada and the Euro countries with the United States. The model could also be used to generate more frequent forecasts if needed. The rest of the paper is organized as follows. Section II describes the neural network model used in this study. Section III discusses the empirical results and concluding remarks are made in section IV. II. Neural Network Model Neural network models, inspired by the neural architecture of the brain, have been successfully applied in numerous pattern recognition and nonlinear estimation problems'. A neural network is a set of computational elements, neurons, which are interconnected. The neurons are often grouped into three layers: an input layer, a hidden layer, and an output layer. The input layer accepts input from the outside and transmits them to the hidden layer. Each hidden layer of neurons receives a weighted linear combination of the input and applies a nonlinear transformation before transmitting the information to the output layer. The output neurons similarly transform the input received, before providing the output signals. The neural network representation model used in this study is known as the backpropagation model. The estimation process in this case is essentially one of mapping from one set of vectors to another set of vectors {vector space X ) [arrow right] [W] [arrow right] {vector space Y ) or Y = WX. This means that for a set of inputs X, the network searches for a matrix W, such that it produces a vector of outputs which is as close a possible to the target output Y. The process is known as network training. The initial value of the weight matrix can be randomly chosen. In backpropagation, the error of the output relative to the target output is propagated backward through the network in order to adjust the weights. …