---
title: "Sentiment-Augmented Deep Reinforcement Learning for Active Trading: An Alpha-Reward Approach for CLaC@FinMMEval 2026"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.16028v1"
permalink: "https://feed.astralor.com/items/itm_7548a6a8ba9448db090789ca"
published: "2026-07-17T15:01:35.000Z"
score: 77
channel: "finance"
tags: ["CLEF 2026","DDPG","DQL","Sentiment Analysis","Deep Reinforcement Learning","Quantitative Trading"]
---

# Sentiment-Augmented Deep Reinforcement Learning for Active Trading: An Alpha-Reward Approach for CLaC@FinMMEval 2026

## AI commentary

> This paper is valuable for both quantitative trading practitioners and financial AI researchers. It not only demonstrates how to integrate LLM-based sentiment scores with deep RL on real market data, but also honestly reveals the generalization gap when transferring strategies from bull to bear markets, offering direct insights for risk management and backtesting frameworks in production. I recommend paying attention to the alpha reward design and the performance differences between DDPG and DQL across different assets.

## At a glance

This paper presents the system for CLEF 2026 FinMMEval Lab Task 3, using daily news sentiment scores from LLaMA 3.2 1B combined with technical indicators and calendar encodings to build deep reinforcement learning trading agents for Bitcoin and Tesla. Four algorithms—PG, PPO, DQL, and DDPG—are compared, with a novel alpha reward based on excess market return to encourage outperforming buy-and-hold. DDPG achieves the strongest test-set performance; DQL is selected a priori for the live endpoint due to the highest validation Sharpe ratio. For TSLA, DDPG and DQL yield cumulative returns of 54.96% and 52.62%, respectively, versus 16.45% for buy-and-hold. For BTC, DDPG returns +1.58% while buy-and-hold suffers -34.27%. The results also highlight a significant validation-to-test generalization gap when transferring bull-market policies to a bear regime.

## Key points

- Daily news sentiment scores from LLaMA 3.2 1B are integrated into the deep RL state space. 
- An alpha reward function based on excess market return is introduced to incentivize outperforming the buy-and-hold baseline.
- DDPG and DQL significantly beat buy-and-hold on both Tesla and Bitcoin tests, with positive returns even in bearish Bitcoin markets.
- A substantial generalization gap between validation and test periods is identified, warning against transferring bull-market policies to bear regimes.

## Full text

This paper presents our system for Task 3 of the CLEF 2026 FinMMEval Lab, which requires daily long, flat, or short trading decisions for Bitcoin (BTC) and Tesla (TSLA) using news and historical market data. We formulate the problem as a discrete-action Markov Decision Process and compare four deep reinforcement learning algorithms: Policy Gradient (PG), Proximal Policy Optimization (PPO), Deep Q-Learning (DQL), and Deep Deterministic Policy Gradient (DDPG). The agents use technical indicators, cyclical calendar encodings, and daily news sentiment scores produced by LLaMA 3.2 1B. To reduce overfitting and align training with the objective of outperforming buy-and-hold, we introduce an alpha reward based on excess market return and randomize episode start dates. Hyperparameters are optimized with Ray Tune over 180 trials per algorithm-asset pair, with early stopping and model selection based on validation Sharpe ratio. On the CLEF Task 3 test set, DDPG achieves the strongest overall performance. DQL was selected a priori for the live endpoint because it obtained the highest validation Sharpe ratio, with selection performed without access to the test period. For TSLA, DDPG and DQL achieve cumulative returns of 54.96% and 52.62%, respectively, compared with 16.45% for buy-and-hold. For BTC, DDPG achieves a positive return of 1.58% while buy-and-hold declines by -34.27%. The results also reveal a substantial validation-to-test generalization gap, highlighting the difficulty of transferring policies selected in bull-market conditions to a bear-market regime.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_7548a6a8ba9448db090789ca