Evaluate the Ability of Artificial Neural Network to Predict the Stock Price of Non-Metallic Mineral Products Industry in Tehran's Stock Exchange

Transcription

1 Evaluate the Ability of Artificial Neural Network to Predict the Stock Price of Non-Metallic Mineral Products Industry in Tehran's Stock Exchange Mohammad Sarchami, Department of Accounting, College Of Accounting, Haji Abad Branch, Islamic Azad University, Iran. Mohammad Heydari Narmandi, Department of Law, College Of Law, Haji Abad Branch, Islamic Azad University, Iran. Abstract Developing of computer science and its application in other sciences, such as accounting, artificial neural networks created increasing use in forecasting stock prices. The purpose of this study is evaluating of the ability of artificial neural network to predict the stock price in non-metallic mineral products industry. Multi-Layer Perceptron neural network with error back-propagation algorithm and sigmoid transfer function for 35 companies in the 11 financial years is designed with entries contains the monthly average of price of an ounce of gold, the monthly average of price of oil (OPEC), the monthly average of inflation rates (base year 2005), The monthly average of exchange reference rate (America dollars), the monthly average of interest rate (the interest rate on account of short-term deposits) and output prices of different companies. Thus the conclusion shows confirming the hypothesis. Field of Research: accounting 1

2 1. Introduction Investment and capital accumulation has an important role in the economic development of the country. The importance of this factor and the role can be clearly seen in the capitalist countries. No doubt stock is the best place to attract capital of small and using them to develop a company in macro level as well as the personal growth of the individual investor. Therefore the most important factor in this case is buying stock in the lower price and selling it at a higher price. This means forecasting the stock price (Toloi Ashalghi, 2005). Investors usually invest with 2 purposes: temporary investment of surpluses cash and longterm investment in order to increase income in securities Often investors has surpluses cash that may not need to the money and instead of keeping it, they invest in useless account as short time.or they invest in surpluses to creating commercial relationship which increase profitable ability directly or indirectly (Nazari, 2009).So one of the main purposes of investment in surpluses stock is getting the profit, which the investors need to predict the stock price for achieving to this aim. Therefore predicting the stock price is so important for investors. The predicting the stock price is not easy despite complexity of the market because there are many marketing factors which could not take these factors merely into account in technical analysis. On the other hand stock market is a disturbing and non-linear system that is affected by political, economical and psychological situation, thus we could use systems based on artificial neural network to predict the stock price. The present study is focused on using the artificial intelligence and machine learning methods to test the ability of artificial neural network for predicting the stock price in non-metallic mineral products industry. 2. Literature review Moshiri and Morovat (2007) were examined to predict the general index of Tehran stock output by using linear and nonlinear models. In this study, general index of Tehran stock output were are estimates and predicted by using of daily and weekly data in the period from 1377 to 1382 and using different forecast methods such as ARIMA, ARIMA, GARCH models and artificial neural network (ANN). Comparison of predicting accurate of mentioned models by the predicting standards such as RMSE, MAE and U-THIEL show that ANN models for predicting of daily and weekly index has the better performance to the other models. But the statistical comparison of different of predicting accurate of these models by using statistical Dibbled Mariano does not show a significant difference between the predictions accurate of the models. However, according to the obtained results, it can be recommended to the researchers to use ARFIMA non-linear models and ANN to predict Tehran stock prices. 2

3 Raai and Chavoshi (2004) were studied to predict stock returns in Tehran Stock Exchange with artificial neural network model and multi-factor model. To test this problem, daily stock price of Behshahr Industrial Development Corp has been selected as a sample. Independent variables are (inputs) research, five macro-economic variables means Tehran Stock Price Index, Currency (USD) on the open market, the price of oil and the gold price. To evaluate the factor is used in the multiple linear regression models and neural network model MLP architecture with back propagation learning algorithm. The results show the success of two models to predict of stock output and also preference of performance in artificial neural network to multiple- factor models. Motavaseli and Taleb Kashefi (2007) have done a comparative study of the neural network with the input parameters of technical analysis to predict the stock price. In this study, forty-company stock price in the next ten days in Tehran Stock Exchange is predicted by using three different methods. In the first method, the price is predicted by using one-layer with learning algorithm Lon berg Marco at and the performance of mean square error to market value input. Then in addition of market value input, five-days, ten and twenty days and movement mean and twelve-day RSI and ROC was introduced as the input to the network and the prediction was made. Stock prices by using ARIMA models were predicted for all the companies. Using analysis of variance was used to compare three methods of forecasting. Since about 30 companies price forecasting models ARIMA significantly higher than the neural network models better results presented, it can be stated that the linear model is better than the models of nonlinear-neural networks have complex time series Price stock analysis and are used to predict the stock price. Heydari Zare and Kordloee (2010) examined to predict the stock price using an artificial neural network of the company's four-year period Ghadir investment company Independent variable input including general price index of Tehran Stock Exchange, Currency (USD) on the open market, the price of oil and gold prices are. 80% of the training data, 10% and 10% of the test are confirmed. Network design is and type conversion functions in different layers tested and the best conversion functions were obtained hyperbolic tangent function for the logistics function for the output layer and the layers, The optimal amount of training iterations stop when the method is that the best answer is replicated in Azar et al (2007) examined comparison to classical methods and artificial intelligence to predict the stock price index and design of hybrid models. For the purpose of evaluating the performance of classical methods is investigated such as exponential, trend analysis, ARIMA, artificial intelligence, such as neural networks and phase neural networks. The third scenario, the design of a hybrid model of ARIMA, Neural networks and phase neural networks is 3

4 studied. In this study, the input variables to the network, including exchange rates, oil prices, the ratio P / E, volume, inflation and economic indices (CPI, PPI,...), Respectively. To select a function of the intermediate layer, a variety of functions is selected such as sigmoid function, Hyzloly tangent, sigmoid colon, sigmoid logarithmic, linear, and. Learning algorithm used in this study, propagation algorithm is a repeat of the 3000 time. The results show that the error rate of artificial intelligence techniques such as neural networks and phase neural networks is less than the classical methods. The combination of all methods of artificial intelligence techniques and classical error is less. An-sing al (2003) was reviewed application of neural networks in emerging financial markets: forecasting and stock exchange index of Taiwan in the period from 1982 to Input variables include short-term interest rate, index of output, consumption, GDP and GDP, consumer prices and production levels and the output variable is the stock index returns. The result show that strategies of investment based on artificial neural networks to the tested strategies of investment in this study gain the higher output. Lendasse (2000) has done to predict index using neural networks, the entered data into networks was two types exogenous and endogenous which exogenous economic data include international stock price indices, exchange rates (dollar /mark/yen) and interest rate (threemonth and Treasury interest rate). The endogenous contain the index date.the results indicate that the neural network work well than linear methods. 3. Methodology After entering the data in EXCEL software and perform the necessary calculations and normalizing data, Neural network software in MATLAB 2009 Prespetron multi-layer backpropagation algorithm is trained by using sigmoid transfer function design and artificial neural network and by comparing the results with the actual values of the deviation calculation network and accurately predict stock prices of various industries Tehran Stock Exchange is tested using artificial neural network. The type of study is Cross-functional which of the inferential statistics (mean square error and mean absolute error) is used to confirm or reject the hypothesis. 3.1 Hypothesis Artificial neural network is able to predict the stock price is non-metallic mineral products industry. 3.2 Variables Predictor variables (independent) 1. The monthly average price per ounce of gold 2. The average monthly price of oil (OPEC) 4

5 3. The average monthly inflation rates (base year 2005). 4. The monthly average exchange reference rate (USD America) 5.mtvst monthly interest rate (the interest rate of short-term deposit account) Respondents predicted variable (dependent) Stock prices 3.3 Population and sample The research companies, non-metallic mineral products industry are listed on the Tehran Stock Exchange, which has been active since 2004 to 2014 in stock. To select a representative sample suitable for the intended target population is used of exclusion method. For this purpose, there is one criterion. While a company achieved this criterion, it would be selected as one Sample Company. These criteria are as follows: Company should investigate the period from 2004 to 2014 is beginning to be actively involved in the exchange. Table 1: Population and sample Industry Type The number listed on the Stock Exchange by the end of 2014 The number of eligible sample members Non-metallic mineral products Findings For each company, from the incomplete data at the beginning and end of the series period when the stock price was skipped, intermediate incomplete data were estimated using linear interpolation. Using the time series of obtained stock prices and other five series, the time series of monthly average price per ounce of gold, the average monthly time series of oil price time series of monthly average inflation rate were constructed the average monthly time series Reference Currency (USD America) interest rate and monthly time series data sets. In other words neural network have learned with 30 inputs and one made output by using this data series. 70% of the samples for training, testing, and 15% for testing and 15% were used for validation. Before using the neural network data is linearly normalized in the interval [0,1] are scale. Neural networks which are built for each company has a hidden layer with 12 neurons respectively. The results are listed in the following table: 5

6 Non-metallic mineral products Proceedings of the Second Middle East Conference on Global Business, Economics, Finance and Banking Table 2: Mean square error and mean absolute error in prediction of stock prices of companies, non-metallic mineral products industry Company No MAE MSE Company Name Name Industry Plaster of Iran Pars Ceram wool of Iran Cement of Ardebil and Lime 4 Azarshahr Urmia Cement Esfahan cement Bojnoord Cement Bahbahan cement Cement of Tehran Khash Cement Khazar cement Dorud Cement Sepahan Cement Shahrood cement Shargh Cement Shomal Cement Sufi cement Gharb cement Fars and Khuzestan Cement Ghaen cement Kerman Cement Mazandaran Cement HegmatanCement Qazvin Glass Hamadan Glass tiles of Esfahan Iran porcelain Clay industry tile of Alvand Ahvaz Farsyt Azar Refractories Pars Refractories Dorud Faryst Pars Tile ceramic tile of Sina porcelain's of Iran Average 5. Conclusion One hypothesis was tested "artificial neural network be able to predict the stock price of the automotive industry and parts manufacturing." To test the hypothesis, it was used MATLAB software and statistical methods of square error average and absolute error average. Mean square error and mean absolute error in prediction of stock price of non-metallic mineral products industry by using artificial neural network were respectively and

7 One hypothesis was tested "artificial neural network to predict the stock price of nonmetallic mineral products industry has". To test the hypothesis of MATLAB and statistical methods mean square error and mean absolute error were used. The main reason for the influence of these factors in various industries due to fluctuations in gold prices, oil prices, inflation, exchange rates and interest rates, At a glance it is clear that factors such as the average monthly price of an ounce of gold, the average monthly price of oil, the average monthly inflation rate, the monthly average exchange rate and the monthly average interest rate are effective in the prediction accuracy of different industries of Tehran Stock Exchange share prices. The main reason for the influence of these factors in various industries is due to fluctuations in gold prices, oil prices, inflation, exchange rates and interest rates. The severity of the impact on industries that are related to these factors are more and conversely, the industries that are less affected by the aforementioned factors appear less severe. Nonmetallic mineral products industry companies that are less affected by mentioned factors fluctuations has the mean squared error and mean absolute error lower. In other words, an artificial neural network is able to predict the stock price in non-metallic mineral products industry companies. References Azar, A., Afsar, A Modeling to predict the stock price with fuzzy neural network approach, Journal of Business Research vol. 40, pp An-Sing C., Mark T., Hazem D Application of Neural Networks to an Emerging Financial Market: Forecasting and Trading the Taiwan Stock Index, Computer & Operation Researc vol.30, pp Heidari Zare, B., Kordloee, H.R Using artificial neural network to predict the stock price, Research Journal vol. 17, pp Lendasse A Non-Linear Financial Time Series ForecastingApplication to Bell 20 Stock Market Index, Eur. J. Econom vol. 1, no. 1, pp Moshiri, S., Morovat, H Predictive Index Tehran stock returns using linear and nonlinear models, Journal of Business Research vol. 41, pp Motavaseli, M., Talib Kashefi alasl, B Comparison of the input parameters of technical analysis with artificial neural networks to predict the stock price, Economic Research Journal vol. 1, pp Nazari, R Accounting for investments in shares and other securities. Accounting and Auditing Research Center, National Audit Office, Eleventh Edition. Pp

8 Raee, R., Chavoshi, K Predicting stock returns in Tehran Stock Exchange: artificial neural network model and multi-factor model, Journal of Financial Research vol.5,no.15,pp Toloi Ashalghi, A., Hagh Dost, SH modeling the stock price forecasting using neural network and comparison with the predictions of the mathematical, mag. Economic Journal vol. 25, no. 2, pp