| Abstract [eng] |
Gavronskytė, Gabija (2025). Quantitative Risk Analysis of Cryptocurrency Portfolios Using Machine Learning Methods. MBA Graduation Paper. Kaunas: Vilnius University, Kaunas Faculty, Institute of Social Sciences and Applied Informatics. 77 p. Since the inception of cryptocurrencies in 2009, when the first cryptocurrency, Bitcoin (BTC), was launched, this market has grown to a $1.13 trillion market. It is predicted that by 2030 this market should exceed 4 trillion US dollars with an average annual growth of 27.8%. With the growing popularity of cryptocurrencies, risk analysis of cryptocurrency portfolios using machine learning methods is becoming relevant in the scientific community as well. This line of research allows you to evaluate and understand what factors influence the dynamics of the crypto currency market and provides new insights into investment effectiveness of strategies. Compared to traditional risk assessment methods, machine learning risk assessment methods enable the management of large and complex data sets and the discovery of non-linear relationships. Unlike traditional static models, machine learning algorithms are constantly learning and adapting to new data, making them highly responsive to volatile cryptocurrency markets. Additionally, machine learning techniques can integrate real-time sentiment analysis from multiple sources, increasing predictive accuracy. These capabilities allow for a more accurate and dynamic assessment of risk in the rapidly evolving cryptocurrency environment. Based on the analysis of literature, it was found that DMLP machine learning technique together with sentiment analysis was used only on stock prices prediction and risk evaluation, so this model was chosen to be tested on cryptocurrencies. Before forecasting cryptocurrencies prices methods like sentiment analysis, correlation and regression analysis were carried out. To evaluate DMLP models’ with sentiment analysis results, they were compared to the results received from Monte Carlo simulation, DMLP and SVR models. Predictions were calculated for 4 periods. For periods when market was consolidated DMLP model with sentiment analysis provided the most accurate results and when there was bull and bear markets, DMLP method had the best results. This paper consists of 68 pages, 42 tables and 11 illustrations. |
