The Bitcoin market is defined by extreme price swings, often driven by investor sentiment rather than clear fundamentals. Standard quantitative trading strategies struggle in this environment. While many studies confirm a link between social media mood and Bitcoin's price, a gap exists in developing and testing practical trading models based on this data . This thesis aims to close that gap by building and evaluating such a strategy to determine if it can deliver superior risk-adjusted returns.

This Thesis analyzed public data from Reddit for the years 2021-2024. A robust sentiment signal was created by combining two distinct methods: the fast, lexicon-based VADER and the context-aware machine learning model FinBERT. This composite signal was used to drive a rule-based strategy that tactically adjusts its Bitcoin allocation weekly based on the prevailing market mood . The prototype was rigorously backtested with a strict out-of-sample validation for the 2024 year.

The sentiment-driven strategy proved highly effective and significantly outperformed passive benchmarks. In the 2024 out-of-sample test, the primary strategy delivered a total return of 143%, compared to 121% from a simple buy-and-hold approach. The risk-adjusted performance was even more impressive. The strategy achieved a Sharpe ratio of 2.70, nearly double the benchmark's 1.49, demonstrating a far more efficient return for the risk taken. The key success factor was superior risk management: the strategy successfully preserved capital by halving the maximum drawdown from 26% to just 14%. Further analysis confirmed that the composite signal of VADER and FinBERT was more effective than either method used in isolation, and the strategy's performance remained robust even when tested with higher transaction costs and simulated execution delays.