شبکه عصبی

Predicting the future trends in financial markets stands as a critical task for both investors and researchers, given its pivotal role in enabling well-informed decision-making processes and effective risk management strategies. Nevertheless, the realm of stock market dynamics is fraught with inherent complexities and uncertainties, posing a formidable challenge when it comes to achieving accurate predictions. A wide array of predictive modeling techniques have been meticulously investigated, spanning from conventional statistical methodologies to more sophisticated machine learning algorithms. The primary focus of this research endeavor revolves around the predictive analysis of the Tehran Stock Exchange (TSE) Composite Index, wherein a novel hybrid neural network framework is employed. This approach seamlessly integrates multiscale temporal features, with the ultimate objective of bolstering prediction precision and offering profound insights into prevailing market trends and dynamics.

The hybrid neural network architecture that has been put forward integrates the unique capabilities of convolutional neural networks (CNNs) in the realm of feature extraction with the effectiveness of long short-term memory (LSTM) networks in capturing temporal dependencies. The dataset used in this study consists of daily historical data pertaining to the TSE Composite Index, covering a substantial period from the year 1998 to 2022, which has been meticulously gathered, preprocessed, and subsequently partitioned into distinct sets for training and validation purposes. Within the framework of this hybrid neural network model, a sophisticated approach is adopted to harness multiscale temporal features derived from the input data, enabling the generation of highly accurate predictions regarding the future trends of the index. Moreover, to further enhance the performance and resilience of the model, sophisticated feature engineering methodologies are implemented to optimize its overall functionality.

The results of the study reveal that while the hybrid neural network model, integrating CNN and LSTM components, demonstrates promising capabilities in predicting the TSE Composite Index, its accuracy falls short compared to competing models, particularly at weekly and monthly time scales. Conversely, the standalone CNN model exhibits superior performance, yielding more accurate predictions of the index's movements. These findings challenge the prevailing notion regarding the efficacy of hybrid neural network models in financial market prediction, highlighting the importance of evaluating alternative modeling approaches based on their specific strengths and limitations.

Despite the potential of hybrid neural network models, as demonstrated in previous research, the findings of this study suggest that simpler neural network architectures, such as CNNs, may offer better prediction performance in certain scenarios. To address the limitations identified, future research endeavors could explore alternative model configurations, ensemble methods, or hybrid architectures that combine the strengths of different predictive models. Additionally, incorporating additional market indicators and exploring intraday data sources could further enhance prediction accuracy and robustness. This abstract encapsulates the key findings and implications of the research, providing valuable insights for investors, researchers, and practitioners in the field of financial market prediction.

In Iran's capital market, only European-style options are bought and sold. In the case of global markets, other trading options such as American option, Asian option, barrier option, lookback option, etc. are bought and sold. Since warrants are a suitable and widely used tool for hedging risk and have leverage properties, diversity in the warrants market warms up the financial market and attracts many interested parties. There are many models for option pricing, the most important of which is the classic Black-Scholes model. But this model is far from the real market models due to its weaknesses. Therefore, it needs developed models. For example, the study of time series shows that the price of the underlying asset has an inverse relationship with its volatility. The constant elasticity of variance model is a suitable model to show this inverse relationship. In this research, the constant elasticity of variance model is considered for the valuation of American options, whose volatility is a function of two parameters of volatility elasticity and the scale parameter fixing the initial instantaneous volatility. Studies show that non-parametric methods such as neural network and machine learning are better suited for economic modeling than other methods. Therefore, in this research, the neural network method will be used for numerical valuation of American options.

In this article, the goal is to design and present a model to determine the effect of macroeconomic and banking variables on the occurrence of asset freezing in the country's banking system using meta-heuristic models. The current research is applied in terms of purpose, in terms of research method, correlation analysis type and in terms of overall research design, post-event and retrospective. In order to answer the research questions, the annual data of macroeconomic and banking variables, during the period of 1399-1390, were collected and using the test of regression models in EViews, Smart PLS software and also the neural network model. It was estimated in SPSS Modeler software. The estimation results of the regression model of the first hypothesis in EViews software showed that the economic variables of GDP, unemployment rate and interest rate, consumer price index, currency strength at the error level of one percent and the economic growth rate variable at the error level of ten percent have a significant relationship. They have a dependent variable (asset freezing). Also, the estimation results of the structural model of the first hypothesis in the PLS software are significantly aligned with the output of the Eviuse software. So; The first research hypothesis is confirmed. Also, the results of the regression model estimation of the second hypothesis in EViews software showed that the intra-bank variable of the bank size ratio, return on equity, and the amount of liquidity at the error level of ten percent, and the variables of capital adequacy, return on assets, bank capital, at the error level of one percent. The percentage has a significant relationship with the dependent variable (asset freezing). Also, the estimation results of the structural model of the second hypothesis in the PLS software are significantly aligned with the output of the Eviuse software. So; The second research hypothesis is also confirmed.

Interval-valued forecasting is related to predicting an interval that is determined by two random variables. In the present study, using the neural networks method, the interval related to the lowest and highest daily prices is predicted and then based on it, a daily scalping trading system is formed, including buying and selling in the forecasted amounts. To reduce the risk of the trading system and increase the number of trading positions, the optimal daily scalping trading portfolio is developed in the mean-variance framework. The sample portfolio includes five shares of the Tehran Stock Exchange in a 190-day period, taking into account trading costs, shows that the average daily return is 0.0028 and the Sharpe ratio is 0.6379, which is better than the Sharpe ratio of individual daily scalping trading of portfolio assets. The daily average of the total index in the research period is 0.001 and the Sharp ratio is 0.0835, which shows that the trading system has a much better performance than the buy and hold strategy.

Stock market manipulation has gained significant attention in recent years. The fact that markets can be manipulated has great implications for designing trading rules and market efficiency. Market manipulation has been developing since the beginning. Thoughtful and accurate manipulation techniques adopted by manipulators may not easily be detected. This research examines stock price manipulation in Tehran Stock Exchange and addresses the identification and detection of stock price manipulation by mathematical algorithm (spoof trading). This study examines transaction-based manipulation in which manipulators place bid and ask orders to control the stock prices. More specifically, this paper studies manipulated stocks by looking at market microstructure (intraday transactions) and aims to introduce an algorithm for detecting suspicious trades so that regulators can detect them beforehand. The ultimate goal of this paper is to help the “Securities and Exchange Organization” (SEO) of Iran to reform market regulations in order to prevent the occurrence of suspicious trades in the market.
This research relies on price data placed by investors in the market. Transaction data consists of two levels; the first level includes executed bid-ask orders available to the public, such as open, close, high, and low prices as well as trading volumes in a specific period. The second level includes the first layer plus all buying and selling orders whether they are executed or not. The latter can only be observed by the regulator and is not available to the public. This paper focuses on the second level in which characteristics and patterns of manipulated stocks are examined by utilizing the intraday transactions of 50 manipulated companies from 2013 to 2016. Panel data analysis and F-limer, Hausman Test, Heteroskedasticity Test, Cointegration Test, Variance Inflation Factors Test, as well as econometric tests relating to price manipulation including stationary, autocorrelation, kurtosis, skewness, run and duration dependence test, were applied to the considered data. Finally, an artificial neural network was used to test the effectiveness of the designed algorithm. The obtained results were illustrated in a confusion matrix.
The results of the mentioned econometric tests were consistent with the results of the designed algorithm. The study’s hypotheses were accepted.
The results of the designed algorithm indicated that the efficiency of the utilized algorithm for detecting suspicious trades is equal to 90.4%, which is an excellent level of performance for accepting an algorithm. Additionally, this research shows that the price and volume of buy and sell orders in the first two lines of the ticker screen are effective in detecting stock price manipulation.

Liquidity of stocks is an important challenge in the capital market. Identifying the factors affecting liquidity helps to predict the stock liquidity situation and thus stock risk management. The purpose of this study is to find the factors affecting the liquidity of stocks. For this purpose, in the first stage, using the research literature and experts, the influencing factors are identified and using the methods of minimum redundancy and maximum correlation (MRMR) and genetic algorithm, the effective variables are selected. In this research, using Excel software and existing raw data, the required data was created and then using support software and neural network toolbox and support vector machine was created. . Finally, the extracted variables using MRMR include stock market value, intensity of product market competition, GDP growth, equity returns, stock returns, inflation rate and family ownership, and using the financial model of financial leverage, government ownership, Equity returns, GDP growth, share buoyancy percentage, market type and board (on the stock exchange and OTC), the intensity of competition in the product market were selected.

Achieving the optimal capital structure in order to gain maximum profitability and value while having minimum cost of capital is one of the important topics studied by financial professionals. In the field of the relationship between financing decisions and corporate profitability, there exists four major theories of Miller and Modigliani, agency theory, static balance theory (all observing the positive relationship between debt ratio and profitability) and hierarchical theory (negative relationship between debt ratio and Profitability). In this study, using data from 161 companies listed on the Tehran Stock Exchange in five industries (including ceramic tiles, cement, metals, oil and gas, and pharmaceuticals) between 2010 and 2017, the problem of estimating the profitability function by the capital structure has been addressed. In order to estimate the mentioned function, regression analysis as well as a combination of wavelet neural network algorithm and the Imperialistic Competition Algorithm have been used. The results indicate a negative relationship between debt ratios and return on asset (confirmation of hierarchical theory) in ceramic tile, cement, metals, and pharmaceutical companies. However, in the oil and gas industry, no significant relationship has been found between these variables. In addition, the performance of the wavelet neural network optimized by the imperialistic competition algorithm has shown to be more desirable than simple linear regression for all the investigated industries.

Discussions about forecasting Stock returns in developed countries has long been regarded as one of the most interesting scientific topics.However,due to many problems,the correct prediction of stock returns has remained a matter of strengthTtherefore,the researcher seeks to provide an accurate,practical and effective model for predicting stock returns for investors.The statistics sampel of research is consist of 138 active companies in Tehran Stock Exchange from 2008 to 2017 wich are selected by the systematic removal method . ANFIS,MGGP, regresion and neural network and different statistics tests are used for data analysis. For impelement of these techniques MATLAB and GenXproTools software are used respectively.The result of the study showed that in oreder to predict stock returns.the use of a meta –heuristic Hybrid models is more accurate and faster than other meta huristic models.Because ,first the most optimal input variables are selected through the ANFIS technique and then predicted using theMGG meta heuristic model.Therefore,due to the correct choice of input variables,predicting stock returns is both more accurate and faster.In addition ,the mathematical model is used to predict.

We live in an age characterized by the very rapid rate of financial innovation. The study of the historical evolution of progress and economic development of developed and industrialized countries shows that one of the main factors in the emergence of rapid and massive growth has been the existence of financial reforms in these countries. There are different incentives for individuals and active enterprises in the financial system to perform financial innovation, which is one of the most important incentives, the introduction of tools and methods to reduce, eliminate or manage existing risks. One of the most important tools the current situation that can help banks in the optimal management of consumption and prevention of claims. Designing and applying credit risk assessment models in granting facilities. The purpose of this study is to provide a suitable model for financial innovation based on credit risk measurement of SMEs in commercial banks. In this regard, effective indicators on the credit risk of SMEs were identified by using the genetic algorithm method and logit, neural network and fuzzy expert system were evaluated. The results show that the using the hybrid model has more accurate results in the assessment the credit risk of SMEs.

The purpose of this research is designing a credit rating model for issuers and tools for financing Islamic securities in the Iranian capital market. To do this, three major steps were taken. The first step was to identify the evaluation criteria or the risks associated with the Islamic securities, which was carried out by the experts and a review of theoretical basics. The second step, is modeling of Islamic securities using adaptive-network-based fuzzy approach in which the mean error of the training of all main and subset models was below the threshold. The third step is to apply adaptive fuzzy neural network modeling in credit rating of Islamic securities. In order to do this, the issuer’s ranking was used in the first stage and the results of the research showed that the issuer of the government had the least risk and private companies had the highest risk. In the second stage, for ranking financial instruments, the results showed that for issuer of government, treasury bonds had the lowest risk and forward bonds had the highest risk. For the issuer of state-owned companies, the forward bonds had the highest risk and lease bonds had the lowest risk.

This study aims to present an intelligent model for predicting financial opulence in the security companies as a system that supports the decisions. For this reason, by investigating background of the seventeen numbers of variables as a predictor variable for predicting the class of financial opulence from valid sources of central security Site G.A.A during the years 1390- 1395 had been extracted. For conducting this investigation, there had been used of the data of Security Industry, during the years 1390 to 1395. In this investigation, first, we compare the results of applying different models of prediction based on Data Mining and in the second stage, we investigate the ranking of predicting algorithms. The finding results of this investigation showed that the financial opulence with acceptable precision is predictable and the extracted model by using the decision tree has very high precision and capability.

This study presents the new hybrid network of GARCH family and an artificial neural network to predict the Tehran Stock Exchange index during the period of 2008-2017. The existence of long-term memory in the conditional variance of the Tehran stock returns causes use in addition GARCH and EGARCH models with short- memory, long-term memory models. In addition to long-term memory models, considering the better performance of hybrid models in predicting financial data of the Garch family models (short and long-term) are combined with the artificial neural network. Using hybrid models the return of stock index was forecast for the next 10 days and its accuracy was evaluated using the evaluation criteria. The results showed that the hybrid FIEGARCH with the student-t distribution model was more efficient in forecasting return of stock and had a lower forecast error than others models

Profit as one of the most important indicators of measuring the performance of the economic unit is one of the important accounting issues that has a high status due to the competitive environment and the importance of quick and proper decision making by managers. Therefore, it is important to analyze the index, factors affecting it and predict profitability. In this regard, the present study was conducted by selecting a sample of 124 observations for the period from 1387 to 1395, based on the basic information of the companies financial statements; the effect of 34 variables on the accuracy of predicting the profitability of the accepted companies by Tehran stock exchange, has been investigated. Tree C5 method was used to determine the significant variables in predicting profitability due to the high ease of understanding of the model. Finally, after determining the effective variables and identifying 8 variables, the accuracy of the predictions was measured using the neural network technique, the C5 decision tree and the backup vector machine (SVM), and the results from these three algorithms were compared. The results of the comparison show that using the c5 decision tree and the 8 variables have the best prediction with accuracy of 93.54%, and then the neural network model is 81.45% more accurate than the supported vector machine (69.35%) and has an error.

Valuation and pricing of securities the process of estimating the value of securities is one of the initial shares of companies. Because, on the one hand, investors need to know in a conscious investment that knows the true value of the stock they are interested in investing in and on the other hand, the owners of companies that are going to sell their securities have to evaluate and value their assets in a proper manner. Therefore, the purpose of this study is to explain the optimal model of evaluation and pricing of the initial public supply of shares of companies accepted in Tehran Stock Exchange using fuzzy multi-criteria decision-making techniques, stepwise regression, neural network and genetic algorithm. To this end, data on 421 companies were collected that during the years 2006 to 2018 launched a public offering of shares on the Tehran Stock Exchange. Fuzzy AHP method, forward regression, neural network and genetic algorithm are also used to analyze the data. The results of the research showed that the genetic algorithm model is the optimal pricing model and initial stock valuation.

One of the most important issues of banking managers as an influential variable on the banking industry is the knowledge of the status of bank deposits that the bank depends on a large extent on it. Therefore, bank managers are keen to know how much the total bank deposits will be at a given time in the future. Predicting the amount of deposits, changes and fluctuations of these deposits can help banks in planning and decision making. In this research, using statistical techniques and approach of artificial neural network models, we have tried to introduce a model with the highest estimation power and the least amount of error to predict the amount of deposits or the same sources of finance by their different types for the desired bank. To test the hypotheses, one private bank information was used during the period of 1387-1396. In this research, we compared the predictive power of ARIMA and artificial neural network method. To assess the accuracy of forecasting the bank's resources, the ARIMA method used Coopiff and Christopherson tests.  The results of the research on the amount of bank deposits monthly showed that the neural network method provides better estimates than the ARIMA method.

One of the most important issues of financial markets is the prediction of price and stock returns. In this paper, we try to find the best model and stock price prediction approach based on the mean square error (MSE), root-mean-square error (RMSE), R-squared, standard deviation (SD), Mean absolute error and the mean absolute percent error (MAPE) for the Fama and French five-factor model. For this purpose, after the formation of a portfolio based on the Fama and French model during the period from 2009 to 2017, stock price is estimated by econometric model, neural network and Fuzzy Neural Networks, so the accuracy of each approach was compared. The results of the prediction the efficiency of the generated portfolios show that the prediction accuracy of the radial base function network (RBF) is very high compared to other ARMA models and other neural networks.

The stock exchange is one of the financial instruments of countries around the world. The recession in this market can have important effects, for example reducing liquidity, reducing the profitability of companies admitted to the stock exchange, and reducing economic growth. In this paper, we are looking for extraction and prediction of time cycles in the stock market. Initially, using the total stock index and the MSI (3) AR (2) model, three cycles of recession, medium prosperity and high prosperity are extracted in the stock market. Then the most important predictor variables are determined by using the integration of the NSGA (II) algorithm and the three types of neural network models and predicted the market situation for the next three months. Finally, the performance of three types of multilayer perceptron neural network, radial basis and probable network were compared in terms of feature selection and prediction of future market situation. The results indicate that all three models have acceptable error rates, accuracy, and Kappa coefficients, and the probable network model has lower error rate, more accuracy and kappa coefficient than other models.

Exchange rate prediction is an important economic variable of interest to the economic actors. Technical approach is one of the commonly used approaches to forecasting, which uses the past behavior of the exchange rate for prediction. However, given the chaotic and non-linear structure of financial markets, the market forecasting cannot be done using a certain and simple method obtained by combining different technical tools and more sophisticated methods are required. In recent decades, neural networks have been employed as one of the most widely used methods in classification, pattern recognition and prediction of complex time series. In this research, a multilevel neural network model was provided to predict the euro-dollar exchange rate, which predicts the price on the next day with an appropriate accuracy by utilizing the data and variables derived from the technical analysis. The results demonstrated the proper function of this method versus other conventional methods of technical analysis and neural network.

The purpose of this study was to compare the predictive power of fuzzy neural network with fuzzy wavelet neural network in predicting stock prices of banks in Tehran Stock Exchange. The period of this research was from 2011 to 2016. In this research, the fuzzy logic system with the use of a multi-layer neural network system with an error-optimized back-propagation optimization structure and a Maximum Overlapping Discrete Wavelet Transform for exchange rate variables, opec oil, each ounce of gold, the total stock index as well as the volume of trades were used in order to predict stock prices.The results of the model were done by using the updated cost function. The results of the research in comparison of fuzzy wavelet network and fuzzy neural network showed that the reliability of banks with fuzzy wavelet neural network is over 90% and with fuzzy neural network above is 80%. As a result, fuzzy wavelet neural network is more reliable than fuzzy neural network

The purpose of this study is identifying factors affecting the probability of loan default and forecasting default probability of non-corporate (natural) customers of Pasargad bank by means of neural networks method (NNM). Variables influencing creation of default were identified through investigating background studies and literature review. At the next step, data related to 470 customers were collected from a statistical population of 25342 people who received loans from Pasargad bank in Tehran region from 2013 to 2014. Results show that NNM could accurately forecast 92% of applicants default probability. According to NNM results, bad financial history or type of collateral have had more significant effect on default probability than the other input variables.