Analysis and Prediction of the Impact of Macroeconomic Indicators on the Price of Bitcoin Using Machine Learning

Bachelor's thesis focuses on the design and backtesting of a trading strategy on the Bitcoin market in response to the release of macroeconomic indicators in the United States. Specifically, it examines the Consumer Price Index and Nonfarm Payrolls. The analyzed period span is from September 2017 to March 2025. The thesis compares classification algorithms such as logistic regression, random forest, and XGBoost, with XGBoost achieving the best result with an accuracy of 61,54% in predicting the 60-minute price movement following CPI announcements. In the next phase, rule-based trading strategies were proposed and tested, where the most successful variant with a fixed risk-reward ratio achieved a return of 23,9 R out of a total of 91 CPI events. In the final chapter of the practical part, the most successful model was a reinforcement learning agent trained over 350 000 steps, achieving a return of 21,43 R on the training data and 14,09 R on the test data. Out of a total of 91 events, split into 54 for training and 37 for testing, the agent chose to open 34 trades from the training set and 17 trades from the test set. The results also show that models consistently perform better when trained and tested on the same type of macroeconomic event. When attempting to generalize the model to a different type of data (applying a CPI-trained model to NFP events), a significant drop in performance was observed. The outcome of the thesis is a validated trading strategy and a methodological framework that enables repeatable testing and development of predictive trading models for macroeconomic event-driven scenarios.