Application of time series models and machine learning methods for stock returns prediction

In this diploma thesis we applied time series models and machine learning methods on real stock data to predict future returns movements. The time series models include ARMA and GARCH models. We also used several machine learning algorithms such as logistic regression, gradient boosted trees and Long Short-Term Memory neural networks. The models were used to predict future daily returns movements. Then strategies based on predictions were evaluated by their profitability and riskiness. Also we measured the quality of the models from the point of accuracy and discrimination power. The main goal of the thesis is to compare results of the models application with the Buy & Hold strategy and between each other.