Abstract: This study is presented the feasibility of cross-referencing of exchange rates to estimate exchange rates on a short-term basis. The cross-referencing technique suggested herein was used to predict EURO currency based on the exchange rate relations modeled by using Artificial Neural Networks. Foreign exchange rates namely UK Pound (GBP), Switzerland Francs (CHF), Canadian Dollar (CAD) and Singaporean Dollar (SGD) have been selected to estimate the EURO currencies based on the data collected from the past 10 years from 1999 to 2008. The main objective this paper is to estimate EURO currency trend based on the cross-referenced relations with the other four currencies by using Artificial Neural Networks. Promising result is shown that the Artificial Neural Networks has been found to be appropriate for modeling and simulation in the data assessments. The paper gives detailed results regarding the use of Artificial Neural Networks for modeling EURO trends in terms of other currencies.
INTRODUCTION
There are two common approaches to estimate exchange rates. They are usually utilized in conjunction with one another which is good. Unfortunately, they are often expressed in highly technical, very specialized and frequently in mathematical language.
In literature, there are several methods to estimate the foreign exchange currency. Gradojevic and Yang (2000) determined the exchange rates using microstructure variable into a set of daily observations of macroeconomics variables to explain Canada/Dollar exchange rate movements. ANN has proven to be efficient and profitable in estimating financial time series. In particular, RNN, in which activity patterns pass through the network more than once before they generate an output pattern, can learn extremely complex temporal sequences (Andersen and Bollerslev, 1998). Ashok and Mitra (2002) in tried to use a hybrid artificial intelligence method, based on neural network and genetic algorithm for modeling daily foreign exchange rates . A detailed comparison of the proposed method with non-linear statistical models is also performed. The results indicate superior performance of the proposed method as compared to the traditional non-linear time series techniques and also fixed-geometry neural network models (Ashok and Mitra, 2002). According to Huang et al. (2004), several design factors significantly impact the accuracy of neural network estimates. These factors include the selection of input variables, preparing data and network architecture.
There is no consensus about the factors. In different cases, various decisions have their own effectiveness (Kamruzzaman and Sarker, 2004). They also described the integration of ANNs with other methods and report the comparison between performances of ANNs and those of other estimating methods and finding mixed results. Zhang and Hu (1998) stated the effects of the number of input and hidden nodes as well as the size of the training sample on the in-sample and out-sample performance. In their paper, they used British Pound/US Dollar for detailed examinations. It was found that ANN outperforms linear models, particularly when the estimate horizon is short. In addition, the number of inputs nodes has a greater impact on performance than the number of hidden nodes while a larger number of observations do reduce estimate errors (Zhang and Hu, 1998). Kamruzzaman and Sarker (2004) investigated artificial neural networks based prediction modeling of foreign currency rates using three learning algorithms, namely, Standard Backpropagation (SBP), Scaled Conjugate Gradient (SCG) and Backpropagation with Bayesian Regularization (BPR). The models were trained from historical data using five technical indicators to predict six currency rates against Australian dollar. The estimate performance of the models was evaluated using a number of widely used statistical metrics and compared.
Results show that significantly close prediction can be made using simple technical indicators without extensive knowledge of market data (Gencay, 1999).
The approach that appears at first sight to be the less scientific, sometimes called qualitative, looks at such factors as the trade balance, the money supply situation and its likely effect on inflation and employment. It also considers the currency reserves and changes in their size, overseas assets and oversea debts, the level of investment and the need for modernization of change in industry, the strictness of and the politically possible changes to exchange control and the world economic situation and its likely impact on domestic production and consumption. Short term flows of capital, government philosophy, electoral prospects, market feelings, interest rates and land prices and wage rates, social and economic pressures, the age and ability of management in industry and in banking and the efficiency and effect of trade union structures are also taken into account.
The other method, associated with charts and models and largely dependent on the use of computers, is often called econometric. An initial decision as to which of the economic indicators to include in the model and what weight to attach to them is essential. By now most estimate accept that this method can only work if it also takes into account the political and psychological factors which play so great and often so unpredictable a part both in the decisions of government and in the reactions of markets.
Whatever method or combinations of methods is used to estimate exchange rates, one fundamental fact must never be forgotten; an exchange rate requires a comparison between two national currencies. Any evaluation of the situation and of trends affecting the situation favorably or adversely must take into account the essentially comparative nature of exchange rates. The final outcome is not at the objective value, nor comparing today with yesterday but instead comparing today here with today over there and tomorrow here with tomorrow over there.
MATERIALS AND METHODS
Fundamentals of artificial neural networks: Artificial Neural Networks are a neurobiological inspired paradigm that emulates the functioning of the brain. They are based on neuronal function, because neurons are recognized as the cellular elements responsible for the brain information processing. Neural networks are innate candidates for the estimating domain due to advantages that they have such as capability to learn from nonlinear data trend and provide noise tolerance.
| Fig. 1: | A simple model of the biological neural networks |
| Fig. 2: | Multi Layer Perceptron (MLP) |
The construction and functioning of an Artificial Neural Networks follows all the stages of a Connectionist Systems (CS), starting with the design of the network architecture, followed by the training, testing and execution phases. A simple model of the biological neural networks is presented in Fig. 1.
In particular, the multilayer Perceptron is the most popular neural architecture, where neurons are grouped in layers and only forward connections exist, providing a powerful base learner with advantages such as nonlinear learning ability and also noise tolerance. Figure 2 shows the architecture model of Multi Layer Perceptron (MLP).
Since late 90s, applications of Artificial Neural Networks solutions are growing more sophisticated and no doubt in the coming years our skills for engineering these computing devices will improve. Already, though, there is vast array of exciting developments. The application base for Neural Networks is enormous: credit card fraud detection, stock market estimating, credit scanning, optical character recognition, human health monitoring and diagnosis, machine health monitoring, road vehicle autopilots, learning to and damaged aircraft, etc.
Study design: All the data input for this study are taken from the statistical resources from the Central Bank of Malaysia. The input data consists of four major currencies that are GBP, CHF, CAD and SGD, stands for UK Pound Sterling, Switzerland Francs, Canadian Dollar and Singaporean Dollar.
| Table 1: | Four input elements for training and testing purpose to estimate the EURO output |
| Fig. 3: | Estimating model of Neural Network |
The amount of data was taken enormous which was from January 4, 1999 to until March 23, 2009 that almost 10 years time frame and containing 2535 set of data. This is to give more training to the networks which is hoped to yield more precise output. From these inputs, the study was carried on with the aim to estimate the EURO currency data trend.
For the first time study conducted, there are total of 1617 inputs of elements, to be trained for predicting output result for 2nd half of 2005 using Artificial Neural Networks. From the inputs, first 1617 sets of data was taken for training purpose while set from 1 618 to 1743 used for testing purpose. The output result started from 1st July 2005 to 31st Dec 2005. For every currency, all the elements are given in four decimal value in order to create precision and accuracy. Table 1 below shows the matrix figure of input elements for this study.
Estimating model: Estimating model of Neural Network is presented in Fig. 3. Study model constraints; Types of currencies: GBP, CHF, CAD and SGD:
| • | ANN software used: Neunet Pro |
| • | No. of hidden nodes: Four |
| • | No. of iterations: 1000±500 |
RESULTS AND DISCUSSION
The prediction value of EURO is based upon four input values. GBP, CHF, CAD and SGD stand for UK Pound, Switzerland Francs, Canadian Dollar and Singaporean Dollar, respectively.
| Table 2: | Tabulated results for coefficient correlation of EURO upon four major currencies |
This four currencies represent major current currency in the forex market.
In this study, 2535 inputs of data were considered. The foreign exchange dataset was taken from the Central Bank of Malaysia http://www.bnm.gov.my/currency statistical webpage. The data range from 4th January 1999 to 23rd March 2009 with total of 2535 rows of data. The whole dataset was divided into two sections as training and testing data for validation and estimation purpose. The testing data was never presented to Artificial Neural Networks and they were mainly used for testing the performance of the networks in order to provide acceptable results. Table 2 shows the tabulated results for the correlation coefficient of predicted value EURO against the actual by using ANN technique. The output value is depending on four input major currencies GBP, CHF, CAD and SGD. The correlation coefficient a concept from statistics is a measure of how well trends in the predicted values follow trends in actual values. It is a measure of how well the predicted values from an estimate model "fit" with the real-life data.
The correlation coefficient is a number between 0 and 1. If there is no relationship between the predicted values and the actual values the correlation coefficient is 0 or very low (the predicted values are no better than random numbers). As the strength of the relationship between the predicted values and actual values increases so does the correlation coefficient. A perfect fit gives a coefficient of 1.0. Thus the higher the correlation coefficient the better. Global financial crisis of 2008-2009 emerged in September 2008 with the failure, merger or conservatorship of several large US-based financial firms and spread with the insolvency of additional companies, governments in Europe, recession and declining stock market price around the globe.
The financial crisis of 2007-2009 began in July 2007 when a loss of confidence by investors in the value of securitized mortgages in the US resulted in a liquidity crisis that prompted a substantial injection of capital into financial markets by the US Federal Reserve Bank. In September 2008, the crisis deepened, as stock markets worldwide crashed and entered a period of high volatility and a considerable number of banks, mortgage lenders and insurance companies failed in the following weeks.
| Fig. 4: | The prediction performance vs. actual output data |
| Fig. 5: | Comparison of actual and predicted ANN values for USD currency using ANN Backprop type |
Beginning with failures caused by misapplication of risk controls for bad debts, collateralization of debt insurance and fraud, large financial institutions in the United States and Europe faced a credit crisis and a slowdown in economic activity. The crisis rapidly developed and spread into a global economic shock, resulting in a number of European bank failures, declines in various stock indexes and large reductions in the market value of equities and commodities. At the end of October a currency crisis developed, with investors transferring vast capital resources into stronger currencies such as the yen, the dollar and the Swiss franc, leading many emergent economies to seek aid from the International Monetary Fund. If we feed this currency to our proposed model during this crisis period, the predicted outcome will be very poor (near to 0.00 correlation of coefficient).
Figure 4 shows the comparison between actual and predicted value for EURO using ANN Backprop. Figure 5 shows the line series graph of the actual results of USD currency in comparison to the prediction value. As in figure 2, the correlation of coefficient was calculated as 0.67 for the four input values and the predicted output value.
The aim of this study was to model the estimate value of EURO currency based on four major currencies GBP, CHF, CAD and SGD. For the study the output value was predicted with the correlations coefficients varies from 0.00 (no relation) to 0.91 (very strong relation) for the estimate value of EURO currency. The result shows that this prediction model is suitable to use during stable market condition. During this period the correlation between predicted and actual value show strong relationship. The weakness of this model is that the model is unsuitable to use while in global financial crisis.
CONCLUSION
This study is intended to offer new alternatives for estimating foreign exchange data by using Artificial Neural Networks technique. The benefit of this study is to estimate the expected value of EURO currency before entering foreign exchange market. This will give some estimation to speculators before hitting the FX market. This study also offer additional assist in problem solving inside financial and economical sectors. It is believed that by dealing with more physical factors may be useful to predict EURO currency more precisely. Therefore, further research may be carried out for such purpose. Multi Layer Perceptron with Backprop equation chosen as the ANN model for the estimating work. Lastly, further enhancement can be concentrated on predicting currencies based on the relations between more vector inputs and prediction output.