Designing a hybrid computational intelligence model for predicting exchange rates in Iran

Dissertation of the master course in the field of economic sciences

February 2013

Abstract

Prediction is effective tools and solutions in order to plan and formulate financial methods. Prediction accuracy is one of the most important factors in choosing a prediction method. Today, despite the numerous forecasting methods, accurate forecasting of the exchange rate is still not an easy task, and most researchers are trying to use and combine different methods in order to achieve more accurate results. Cumulative moving average autoregressive patterns are one of the most important and widely used time series patterns. Their most important limitation is the linear assumption of the model. Artificial neural networks are among the most important and accurate current methods for non-linear modeling of data. But despite all the advantages of neural networks, this type of network cannot be considered in all cases and as a general model suitable for all cases. The fuzzy regression model is a suitable model in forecasting conditions with little data. But their performance in general is not very satisfactory. Therefore, in the present research, artificial neural networks and fuzzy regression have been used, respectively, in order to remove the linear limitations and the number of data required in the cumulative moving average autoregression method and improve the results. The information used in this research includes 115 weekly exchange rate data (US dollar against Iranian rial) from 12/01/2013 to 21/03/2013. In this research, in order to measure the prediction performance of the presented model, various indices such as the absolute mean of error ( ), mean square of error ( ), sum of square of error ( ), root mean square of error ( ), mean absolute percentage of error ( ) and mean error ( ) have been used. The results show that the combined computational intelligence model provides more accurate results in predicting the exchange rate (dollar vs. rial) compared to other models.

Key words

Exchange rate, combined computational intelligence, Iranian economy.

Introduction

Forecasting is one of the successful management tools and a key element in It is considered economic management and planning. The exchange rate as a very important macro-economic variable that affects various domestic and foreign economic sectors of a country, as well as the balance of payments and the strength of international competition, plays a decisive role in economic policies. Exchange rate changes affect different sectors of a country's economy. Therefore, modeling and predicting the future trend of this variable seems to be necessary for providing economic policies and guidelines, but this becomes doubly important considering the economic structure of Iran. Since most of the country's foreign exchange income is provided through the sale of crude oil, and the main source of government income is the sale of crude oil, for this reason, changes in the exchange rate can have many effects on the country's economic structure and domestic markets. Considering the above, it is not surprising that a huge amount of economic literature has been devoted to the modeling and forecasting of exchange rates.

The review of the literature on the subject of forecasting in financial markets shows that examining the behavior of the exchange rate using a model is difficult to predict and forecasting the exchange rate entails inherent problems (Preminger and Frank [1], 2007).

To­ Using hybrid methods or combining different methods is a common way to overcome the limitations of single methods and improve the accuracy of predictions. The basic idea in the combination of methods is based on the fact that none of the existing methods is a comprehensive method for forecasting and cannot be used in any situation and any type of data. Therefore, by combining different methods, the weaknesses of one method can be improved by using the strengths of another method (Chen[2], 1996).

Therefore, in this research, by using the basic concepts and unique advantages of each of the autoregressive models of cumulative moving average, artificial neural network and fuzzy regression, a combined method is presented in order to achieve more accurate results for forecasting the exchange rate (US dollar vs. Iranian rial).

1-1. Statement of the problem

Prediction of effective tools and solutions for the purpose of planning and formulating financial methods. Prediction accuracy It is one of the most important and effective components in choosing a forecasting method. Forecasting of economic variables is done in two ways: qualitative forecast[3] and quantitative forecast[4]. Qualitative forecasting depends on the experience and abilities of people, and quantitative forecasting depends on the probability distribution function of each phenomenon. Gujarati[5] considers forecasting to be an important part of econometric analysis, and for some researchers, forecasting is the most important part of econometrics. Friedman [6] believes that the only appropriate test for the validity of a model is to compare its prediction with experience. Pindike and Rubinfeld[7] consider the main purpose of building regression models to be prediction.

The exchange rate is an economic variable whose prediction is of interest to many economic activists. These economic activists can be divided into three groups. The first category is economic policymakers and central banks. Under a floating currency system, central banks intervene in the currency market in order to smooth market fluctuations. Their reasons for this intervention can include more than usual exchange rate fluctuations and as a result its effects on economic activities. Therefore, forecasting the exchange rate on behalf of this group is necessary for such an intervention. The second category is companies active in the field of trade and foreign direct investment, which with the globalization of the economy, this type of investment and consequently the risks associated with these international activities have increased. One of the most important risks related to these activities is the risk related to the exchange rate; Because the exchange rate changes affect the income, cost and consequently the profit of the companies in order to gain more profit. Finally, the third category is the currency market speculators, who, considering the importance of this market in the international arena, can be considered as the most interested in currency exchange rate forecasting (Musa[8], 2000).

Rate forecasting Currency is of fundamental importance for activists and forecasters in the exchange rate market. Despite this, some believe that it is not possible to predict the exchange rate and the evolution of any type of exchange rate follows the efficient market hypothesis (9). Based on this hypothesis, the best method to predict the exchange rate of the future day is to rely on its current rate, and the real exchange rate follows the random step hypothesis ( ) [10]. This pessimism in exchange rate forecasting came about after the publication of Mace and Rogoff's article[11]. In their study, they showed that no single-equation method is better than the random step method for predicting the exchange rate. Because all the methods examined by these researchers are linear methods, while this fact is accepted by many researchers that the exchange rate changes are non-linear (Pailbeam [12], 1998). The results of examining the efficiency or inefficiency of the financial markets in Iran indicate the inefficiency of the foreign exchange market in Iran (Salami, 1380).

Using time series methods in order to forecast financial markets, improve decision-making and investments has become an undeniable necessity in today's world. Many efforts have been made in the last few decades to develop and improve time series forecasting models. One of the most important and widely used time series patterns is the cumulative moving average autoregressive patterns. These patterns have been of great interest in the last few decades due to their simplicity in understanding and application, but their application is generally limited. The most important limitation of these models is their linearity