| Abstract [eng] |
The effectiveness of combining machine learning methods with Modern Portfolio Theory (MPT) to forecast stock prices and enhance investment portfolios is investigated in this thesis. In particular, it looks into how well the Random Forest, ARIMA, LSTM, and Transformers machine learning models perform when it comes to predicting the stock prices of 20 components of the OMXBBGI index. In accordance with the forecasts, the thesis builds a portfolio with the goal of maximizing returns for a specific degree of risk using Markowitz's Modern Portfolio Theory. Using historical stock data, the machine learning models were trained and verified. Based on these forecasts, a portfolio was subsequently constructed using MPT principles, with an emphasis on maximizing the risk-return trade-off. Important financial measures including Jensen's Alpha, Value at Risk (VaR), and the Sharpe Ratio were used to evaluate the performance of the portfolio. The results show that portfolios built using machine learning models' forecasts performed better than the benchmark index, offering higher returns for the same amount of risk. This enhancement demonstrates how machine learning may be used to improve investment strategies and stock market analytics. By demonstrating the useful advantages of fusing analytical methods with traditional financial theories to enhance risk management and investment choices, the thesis adds to the body of knowledge in the field of finance. |
