<p>Time series forecasting is a key tool in financial markets, helping to predict asset prices and guide investment decisions. In highly volatile markets like Bitcoin (BTC) and Ethereum (ETH), forecasting is challenged by sharp price swings driven by sentiment, technological shifts, and regulatory changes. Traditionally, forecasting relied on statistical methods, but as markets became more complex, deep learning models like LSTM, BiLSTM, and the newer FinBERT-LSTM emerged to capture intricate patterns. Building upon these advancements and addressing the volatility inherent in cryptocurrency markets, we propose a novel hybrid model that combines Bidirectional Long Short-Term Memory (BiLSTM) networks with FinBERT. In volatile markets like cryptocurrencies, where price movements reflect both immediate reactions and delayed responses to sentiment, the FinBERT-BiLSTM model offers a clear advantage over unidirectional approaches. We evaluate the model against traditional baselines (LSTM, BiLSTM), sentiment-aware variants (FinBERT-LSTM, FinBERT-TCN), and state-of-the-art models, including Transformer-based (Informer, TCN, TFT) and GPT-based (TimeGPT) architectures. These evaluations span short-term (intra-day and one-day-ahead) and long-term (30-day-ahead) forecasting horizons, augmented with a realistic trading simulation. Experimental results show that FinBERT-BiLSTM achieves low Mean Absolute Percentage Error values across all horizons—2.03% (BTC) and 2.52% (ETH) for intra-day, 2.20% (BTC) and 2.77% (ETH) for one-day-ahead, and 1.62% (BTC) and 5.10% (ETH) for 30-day-ahead prediction. Moreover, it outperforms all competing models in simulated trading profitability, demonstrating both statistical and practical value for sentiment-informed cryptocurrency forecasting.</p>