Financial time series forecasting using artificial intelligence methods

The methods provided by the field of computer science have considerable advantages in domains of reaching, storing and processing data to satisfy the requirements of finance field compared to humans. A subdomain of computer science; artificial intelligence provides methods such as neural networks, genetic algorithms and machine learning to find effective solutions to financial problems / goals such as trend prediction, portfolio management, fraud detection, risk management and stock prediction. This multidisciplinary thesis aims to teach certain artificial intelligence algorithms the financial time series data, provide future forecasts and compare these forecasts to original values to examine their performance. Seven different artificial intelligence algorithms have been programmed for this thesis. A dataset of 775 business days between 2014-2016 consisting of closing prices of companies that have İstanbul Stock Exchanges 30 highest trading volume and market value are used. Due to the expectation of different algorithms to provide different performance depending on the number of learning / forecasting days, firstly %80 of the data equaling 603 days have been used for training, and the remaining %20 of the data equaling 152 days have been forecasted by the algorithms. Simulating "Long Term Forecasts", "Fast Forest Quantile Regression" algorithm has shown the least error percentage. Secondly %90 of the data equaling 680 days have been used for training, and the remaining %10 of the data equaling 75 days have been forecasted by the algorithms. Simulating "Medium Term Forecasts", "Boosted Decision Forest" algorithm has shown the least error percentage. Lastly %99 of the data equaling 747 days have been used for training, and the remaining %1 of the data equaling 8 days have been forecasted by the algorithms. Simulating "Short Term Forecasts", "Poisson Regression" and "Neural Network Regression" algorithms has shown the least error percentage. The thesis concludes by confirming artificial intelligence algorithms can be used as effective tools for financial time series forecasting.