AI Uses News Sentiment for Bitcoin and Tesla Active Trading

Researchers developed an AI system for the CLEF 2026 FinMMEval Lab, focusing on active trading strategies for Bitcoin and Tesla. The system leverages deep reinforcement learning algorithms, including Policy Gradient, PPO, DQL, and DDPG, to make daily trading decisions. A key innovation is the integration of daily news sentiment scores, generated by LLaMA 3.2, alongside technical indicators and cyclical calendar encodings. To enhance robustness and align with the goal of exceeding buy-and-hold returns, the system introduces an alpha reward based on excess market return and randomizes episode start dates during training. Hyperparameter optimization was extensive, using Ray Tune across 180 trials per algorithm-asset pair. On the test set, DDPG demonstrated the best overall performance, significantly outperforming buy-and-hold for both Tesla (54.96% vs 16.45%) and Bitcoin (1.58% vs -34.27%). However, the study also noted a substantial generalization gap between validation and test periods, indicating challenges in transferring policies from bull to bear market conditions.