جستجوی مقالات مرتبط با کلیدواژه « شبکه عصبی GMDH » در نشریات گروه « اقتصاد »

Natural gas because of its advantages plays a key role in economy of Iran. Given the continuously rising natural gas consumption, planning in the gas sector by taking into account predicted demand is critical for ensuring sustainable development. In this study we develop models for predicting natural gas demand in residential-commercial, industrial and power plant sectors in Iran using ARIMA (Autoregressive Integrated Moving Average) and Neural Network GMDH(Group Method of Data Handling) approaches. The time series data of natural gas demand, natural gas price and air temperature are used as the model variables. the RMSE, MSE and Prediction Error Percentage indexes are used for comparing this models. The prediction accuracy percentage index for residential-commercial, industrial and power plant sectors in ARIMA method are 93.8, 98.3 and 87 percent and those in Network GMDH method are 96.4, 99 and 98.2 percent respectively. The results indicate better performance and accuracy for the GMDH approach compared to the ARIMA model in predicting natural gas demand.

In this research, bankruptcy of companies is predicted using the Hybrid Expert System (HES) approach. Factors affecting bankruptcy and the extent of their impact are saved as rules in a Rule-Based Expert System, and together with financial ratios they are considered as inputs in the GMDH neural networks. The impact of each of these factors on the accumulated profit or loss to capital ratio is evaluated in order to isolate the impact of external shocks on bankruptcy. Variables and the following four financial ratios: Current Ratio, Gross Profit Margin Ratio, Net Profit to Current Debt Ratio and Return of Assets are used as inputs in the GMDH neural networks. The results are compared with those of artificial neural networks for the year of bankruptcy, the preceding year, two years earlier and the average of these three years. In all of these cases, the HES approach produces better outcome than those produced by neural networks. The differences were more marked for 1384 and 1385 when there were large numbers of corporate bankruptcies. Hence the hypothesis of higher accuracy of Expert System compared to neural networks in predicting bankruptcies is confirmed.

In this research we are going to develop a model for evaluating the credit risk and credit ranking through customers in Parsian Bank by the help of Logit and Probit Regression and GMDH neural network methods. This model will be based on the qualitative and financial data of a random sample of 400 customers receiving credit facilities. After analyzing the credit files of each customers, we identified 11 explanatory variables including qualitative and financial aspects as follows: type of security, type of the workplace owner ship, cooperation background, capital, current ratio, quick ratio, the ratio of current asset to total assets, total asset turn over, turnover, current capital turnover, dept ratio and stock holder equity ratio that have significant impact on credit risk. Findings of the study corroborate the economic and financial theories of effective factors influencing credit risk and indicate that neural network model produce more efficient and precise results than the other popular economic models like Logit and Probit. Also, money explanatory variables such as the type of collateral and dept ratio have the most effects and the cooperation background, current ratio, stock holder equity have a usual effect and the rest variables are less effective.

This paper examines the relative importance of alternative asset prices and macro variables in the movements of stock prices in Iran, by applying GMDH (Group Method of Data Handling) neural network with genetic learning algorithms. The results imply that CPI, base money, oil price and house price play an important role in explaining the fluctuations of TEPIX index, whereas exchange rate and gold prices do not appear to have considerable effect on stock prices.

In this paper, we use GMDH neural network based on Genetic Algorithm to model and forecast the price of gasoline using two approaches; Deductive Method and Technical Analysis. The results of deductive method indicate that the accuracy of prediction could reach up to 96% and in technical analysis could reach up to 99%. Furthermore the comparison reveals that the GMDH neural networks model consistently outperforms the regression model used in this study.

Conventionally, regression and time series analyses have been employed in modeling water demand forecasts. In recent years, the relatively new technique of neural networks (NNS) has been proposed as an efficient tool for modeling and forecasting. The objective of this study is to investigate the relatively new technique of GMDH – Type neural networks for the use of forecasting long – term urban water demand in Tehran city. The data employed in this study includes water consumption (per capita), water price, average household income and the annual average air temperature for the city of Tehran, Iran. The neural networks model, regression model, and time series model have been estimated and compared. The comparison reveals that the neural networks model consistently outperformed the regression and time series models developed in this study.