Stock market pattern prediction using perceptron multilayer artificial neural networks

Dissertation of Master of Information Technology Engineering, Information Systems Management

Abstract

In today's world, due to the change in lifestyle, people are looking for a way to improve and improve their economic situation, one of the most important ways to improve their financial situation is to increase their income.  One of the easiest ways is investment, which has different dimensions.  In Iran, due to the drastic changes in the coin and currency market, more people turned to the stock market.  One of the most attractive issues in the stock market is finding a way to increase capital and reduce losses as much as possible.  This problem led to the investigation of algorithms and methods of artificial intelligence in stock market management.  In this research, various methods have been examined and the reasons for the success and failure of these methods have been examined.  Based on the studies, it can be said that neural networks are the best and most widely used method.  If this method is combined with statistical methods, it will be possible to achieve better results, and if this is possible, it will lead to an improvement in the economic situation of companies and as a result, an improvement in the economic situation of the country.  

In this project, an attempt has been made to predict the stock market with the help of perceptron neural network.  After the investigations, it was revealed that less work has been done on the fundamental data of the stock exchange.  Therefore, fundamental and technical data and fundamental and technical combination have been examined and compared.  Then it is shown that combined data are more suitable for prediction.  In order to improve forecasting, the head and shoulders pattern has been added to the data used in the stock market. By examining and comparing normal data without head and shoulder patterns and data using this pattern with the help of the MATLAB neural network toolbox, it has been shown that this pattern will improve prediction. Chapter 1 Introduction Introduction In today's world due to lifestyle changes, people are looking for a way to improve their economic situation, which is one of the most important ways to improve their situation. Finance can be referred to as an increase in income.  One of the easiest ways is investment, which has different dimensions.  In Iran, due to the drastic changes in the coin and currency market, more people have turned to the stock market.  One of the most attractive issues in the stock market is finding a way to increase capital and reduce losses as much as possible.  Shares in the stock market is a risky purchase or sale in which there is either a lot of profit or a person incurs a loss. 

From 1409 AD, the fields of exchange of goods, money and other means of payment were created in Finland [1].  Money changers used to gather every day in Tubores Square in front of the house of the famous merchant Vander Bors to trade money.  On this occasion, these groups took the short name of Bors.  The first real stock exchange was launched in 1460 in Belgium.  Tehran Stock Exchange was established in 1346, this organization started its activity on the 15th of February of that year by making several transactions on the shares of Industrial and Mineral Development Bank.  Finally, electronic stock trading was implemented in Iran on 9/24/1982 by Saderat Development Bank to reduce the gap between people and modern technology.   Since 1997, many people have been looking for a way to advance in the stock market.  The most important issue for traders and stock buyers was to find a way to predict the stock market.  They did not know when to hold and when to sell.  Buying shares was another issue in this field.  If we look at the issue more precisely, the fall and bankruptcy in the stock market is not related to one person because the amount of profit and loss is small and this issue cannot cause problems for the country's economy.  This issue is very important for businessmen and important economic companies in the country because their bankruptcy will cause crises and economic problems and stagnation in the international stock market.  In fact, it can be said that forecasting stock indices is one of the main pillars, but it is very difficult.  Volatility of data, many changes in the market and lack of a clear rule that the data should follow are some of the reasons for the difficulty.  In this regard, artificial intelligence algorithms can be mentioned.  Because in order to find a solution, one should seek to discover a pattern in non-linear and irregular systems.Artificial intelligence tools are very useful for forecasting in environments with extended data.

If you look at the stock market from a managerial point of view, it can be seen that humans have always interacted with the concept of decision-making since they knew themselves.  In the meantime, some scientists of management science have gone as far as to declare that a manager has nothing to do with decision-making.  Decision making can be defined as a choice issue.  In such problems, predicting the decision parameters is very complicated and in many cases impossible.  One of the decisions that financial managers are related to is the decisions related to cash dividends and related functions.  Based on the importance of decisions related to cash dividends of stocks, there have always been many concerns to predict this factor and models have been presented for it, but like other prediction models, a fundamental question in this field is what variables affect cash dividends of stocks and in fact, what independent variables should be included in the model that is presented for forecasting in order to be able to predict the dependent variable of cash dividends of stocks.

In practice, when the number of input variables of a system and communication between the input variables will increase exponentially, the predicted results will show more deviation from the actual outputs.  There are many tools and methods for analyzing these connections and finally, predicting the existing results, among which one of the most powerful can be mentioned, i.e. artificial neural networks.  The use of artificial neural networks for financial forecasts and dividend forecasting also has a history, but there are always problems in selecting the variables that are entered as input to the model, which are mentioned below.  This is not easily possible without the help of financial markets, especially a large and efficient capital market.  Investing in shares offered on the stock exchange is one of the most profitable options in the capital market.  However, evaluating and forecasting stocks or any other securities has a historical process and requires special expertise.  Different theories have been proposed regarding stock market evaluation and forecasting in organized markets. At the beginning of the 20th century, a group of experts with experience in evaluating securities firmly believed that it is possible to provide an image for predicting the future price of stocks through the study and analysis of the historical trend of stock price changes. More scientific studies, emphasizing the accurate identification of stock price behavior, led to the trend towards stock price valuation models.  But these methods were not successful.  In an efficient capital market, it is believed that the stock price is a reflection of the current information related to that stock, and stock price changes do not have a specific predictable pattern.  The theories proposed until the 1980s were good determinants of stock price behavior in the market, until the developments of the New York stock market in 1987 severely questioned the validity of the assumptions of the efficient capital market and models such as the randomness of prices [2]. In the 1990s and after, most of the experts' attention was directed to a chaotic behavior with order, and the effort to design nonlinear models to predict stock prices became increasingly important.   With these theories, among the techniques that became highly important were intelligent systems, because assuming the linearity of the market structure, many models can be easily designed.  However, it is very difficult to show the behavior of complex groups such as the capital market in a modern economic system completely in a set of simple and linear equations.  The main advantage of intelligent systems such as artificial neural networks and fuzzy neural networks is in modeling and predicting irregular and non-linear sets. According to many researchers, another tool such as the genetic algorithm can be fruitful in reducing the time to reach the answer and even optimizing predictions in artificial neural networks and fuzzy neural networks. In this field, machine learning techniques are useful for stock market prediction.   Stock market forecasting is a challenging task that is a part of time series forecasting.  In this research, the strengths and weaknesses of this technique will be investigated.  Also, the articles and researches that have done useful works in the field of forecasting will be examined further.