Forecasting share prices using artificial intelligence techniques: An application in BIST

The stock market is one of the largest financial markets in the world, with companies listed on major stock exchanges worldwide, with a market value of approximately $115 trillion in 2021. Forecasting stock prices has long been a topic of interest to researchers, institutional investors, fund managers, and individual investors. It is a dynamic, non-linear process that is affected by various factors such as stock prices, domestic and foreign economic environment, international situations, expectations in the sector, internal and external factors such as financial data, exchange rates, interest rate, inflation, total economic activities or global price indices. It is a difficult subject to predict. The stable development of stock markets is vital for a country and is deeply influenced by national monetary and cost policies. In this sense, the correct estimation of the share price will reduce the investor's losses and provide returns above the inflation or market average return. In this study, models were established to predict monthly stock prices by using 16 technical indicators, 8 stock market performance ratios and 11 macroeconomic indicator variables belonging to the stocks of companies traded in BIST 30 between 2008 and 2017, and these models were established between 2018-2020 period. It is aimed to obtain the future forecast values of the shares with the help of in order to predict the price of each stock, with the help of PSO, ANT and GA application, 3, 4, 5, ….30 variables were selected and the best prediction models with the lowest RMSE value were created among the possible 20 different prediction models. R2, MSE, MAPE, RRMSE and VK statistical performance indicators were used to evaluate the success of the optimal forecasting models performed with the data set. Wilcoxon signed-rank test was applied to determine whether there is a statistically significant difference between the estimation results of the applied algorithms and the observed stock values. CRITIC, CODAS, MAIRCA and TOPSIS methods were used to rank the algorithms used in the research according to the statistical performance criteria. As a result of the study, it was revealed that PSO showed more successful prediction performance than ANT and GA applications. It has been seen that it may be appropriate to use the estimation model we recommend for investors and analysts in determining the stock price. Key words: stock price prediction, particle swarm optimization, genetic algorithm, ant colony algorithm.