Performance Evaluation of Nonlinear Auto- Regressive with Exogenous Input (narx) in the Foreign Exchange Market

Foreign exchange rate prediction is a stimulating research area from past decade. There are several statistical and machine learning methods already have been proposed by the researchers for foreign exchange rate prediction which provide better results. These models performed a vital role in future financial decision making which is taken by financial department administration of that country and organization. The aim of this paper is to introduce a methodology that uses Nonlinear AutoRegressive with eXogenous inputs network with Levenberg-Marquardt learning algorithm to improve the prediction of exchange rate fluctuation compared with other researches. This study is being performed on Indian Rupee/ USD Dollar currency pair. For more accuracy to NARX mode, we have provided some exogenous inputs as BSE Indictors like BSE Index, BSE GAS &OIL Index and BSE IT Index, Interest Rate, GDP Growth, Inflation, Gross Savings, Total Reserves and Bank capital to assets ratio. IND/USD exchange rates and exogenous variable are gathered from 2000 to 2012 period. After collecting these parameters, we get 3228 time stamp sample. The NARX artificial neural network model is built with 10 hidden neurons and 1 delay into the Matlab 2012b tool environment. This model is trained with Levenbergmarquardt back propagation algorithm. The performance of the NARX ANN model is evaluated through simulation study and the results have been compared with other forecasting model. It is noticed that the NARX ANN forecasting model outperforms than other forecasting models in literature. KeywordsForeign Exchange Rate, Nonlinear Autoregressive with Exogenous Inputs (NARX) Model, LevenbergMarquardt learning algorithm, Performance Analysis, Mean Square Error (MSE) and Regression Coefficient (R).