Modeling and forecasting stock market returns

In this thesis, five models are introduced, discussed, and applied – autoregressive integrated moving average (ARIMA), artificial neural network (ANN), convolutional neural network (CNN), long short-term memory (LSTM) and hybrid CNN+ANN+LSTM. For the analysis, data about daily adjusted closing prices of six pharmaceutical stocks from 1st January till 3rd June 2022 were obtained, transformed into logarithmic returns, and split into training, validation, and testing sets. Models were estimated on the training data, evaluated on the validation set and their performance was compared on the test data. To compare models’ predictive power, the forecast accuracy metric Root Mean Square Error (RMSE) was used. The primary aim of this thesis is to study and explain the concept of introduced models, as well as to construct own models and compare their predictive power in order to answer given research questions. The research questions investigated throughout this thesis are as follows: 1) Is it possible to predict stock market returns with a decent forecast accuracy? 2) Do machine learning algorithms provide with more accurate forecasts than classical ARIMA models?