Forecasting stock prices using deep learning techniques: An application in BIST

Stock market investors should be able to make appropriate moves by predicting the future consistently in order to make a profit. For this purpose, investors should analyze the information they have correctly and make successful predictions. It is getting harder and harder to make future price predictions for financial markets, which are becoming more and more global day by day. For example, sometimes, for the price prediction of stock in Turkey, not only the factors in Turkey but also other factors in the world should be taken into consideration. Therefore, simple statistical calculations are insufficient in this case. As a solution to this inadequacy, computer programs and algorithms and techniques in these programs are used. Deep learning is one of the most advanced tools of artificial intelligence the foundations of which were laid for the first time about 80 years ago, and deep learning's popularity has increased in the last 10 years and has found a lot of place in studies in finance. In the first part of the study, information is given about the stock market and energy resources in Turkey. In the second part, the definition, history, techniques, and usage areas of deep learning are introduced. In the third section, which is the last section, there are the introduction of the data set and the variables, previous similar studies, the steps and visuals of the application. In this study, LSTM (Long-Short Term Memory) and GRU (Gated Recurrent Unit) techniques of deep learning, which are among the latest advanced technologies, were applied in the Google Colab program for stock price predictions. The data set used in the study was taken from Yahoo Finance and covers the dates between 02.01.2013 and 30.12.2022. Forecast models were created by considering 5 companies belonging to the XELKT (Borsa Istanbul Electricity Index) index, which is part of BIST (Borsa Istanbul). Afterward, the success of these prediction models were tested with the calculated model performance criteria and it was aimed to determine whether the techniques used were successful in stock price prediction. In addition, based on the results of MSE (Mean Squared Error) and MAPE (Mean Absolute Percentage Error) among the calculated model performance criteria, the techniques used were compared with each other and it was aimed to determine which of these techniques made predictions with less error. Then, from the analyzes made by taking 4 different days, it was tried to predict which day made the most successful predictions. As a last step, it is aimed to find a model with the least error by technique, epoch number, and number of day forecast for both MSE and MAPE on the basis of stocks. Since the model performance criteria outputs obtained as a result of these analyzes are below 1 for MSE and below 5% for MAPE, it can be said that both techniques show successful price predictions. As a result, as a result of the comparison of these two techniques with each other, it is seen that the LSTM technique is more successful than the GRU technique with a slight difference.