Developing algorithmic trading strategies for Borsa Istanbul (BIST) using artificial intelligence

Stock market is a place where shares of public listed companies are traded and buying and selling activities of publicly-held companies that takes place. From buying and selling activities of companies in stock markets, a profit can be made or a loss of money can be happen. Therefore, stock market prediction is remarkable subject because successful prediction of stock prices or creation of trading systems that offer buy-sell points on time may promise pleasing benefits. However it is a discouraging and challenging task to find out which is the best way to maximize the trading profit. In this thesis, the branches of artificial intelligence have been studied and the papers about stock market prediction have examined. After that, a novel stock trading system for offering buy-sell points, based on a feed forward neural network and self-organizing maps for technical analysis indicators optimization is proposed. Surprisingly, to the best of our knowledge, no related research has been investigated that uses the combination of self-organizing maps for technical analysis indicators optimization and feedforward neural network. Firstly, the developed model uses self-organizing maps to optimize RSI and MACD technical indicators with the buy-sell trigger signals of the financial time series data. Secondly, optimized values are passed to feedforward neural networks for the improvement of the buy-sell offers. Stocks from BIST (Istanbul Stock Exchange) are used as a case study but the developed model can be used universally and can be applied to global stock markets. The results show that the developed model performance are better than the buy-hold strategy, traditional RSI and MACD strategies in the most of the stocks in BIST.