Market-Derived Financial Sentiment Analysis: Context-Aware Language Models for Crypto Forecasting

TL;DR

This work reframes financial sentiment analysis by replacing subjective, human-annotated sentiment labels with market-derived labels that reflect actual short-term price movements. It introduces Triple Barrier Labeling to create robust, volatility-aware labels and enhances language-model predictions through prompt tuning that injects market context and temporal information. The context-aware models (CA, and its variants) demonstrate strong improvements in both tweet-level trend classification (F1 up to ~89.5%) and end-to-end trading signal generation, yielding favorable backtested Sharpe ratios across diverse market regimes. This approach reduces reliance on fusion or separate price-prediction models and demonstrates the practical viability of LLMs as standalone short-term market predictors in crypto markets. Overall, the study provides a scalable, interpretable, and market-grounded framework for real-time financial forecasting using language models, with clear implications for digital-asset trading and risk management.

Abstract

Financial Sentiment Analysis (FSA) traditionally relies on human-annotated sentiment labels to infer investor sentiment and forecast market movements. However, inferring the potential market impact of words based on their human-perceived intentions is inherently challenging. We hypothesize that the historical market reactions to words, offer a more reliable indicator of their potential impact on markets than subjective sentiment interpretations by human annotators. To test this hypothesis, a market-derived labeling approach is proposed to assign tweet labels based on ensuing short-term price trends, enabling the language model to capture the relationship between textual signals and market dynamics directly. A domain-specific language model was fine-tuned on these labels, achieving up to an 11% improvement in short-term trend prediction accuracy over traditional sentiment-based benchmarks. Moreover, by incorporating market and temporal context through prompt-tuning, the proposed context-aware language model demonstrated an accuracy of 89.6% on a curated dataset of 227 impactful Bitcoin-related news events with significant market impacts. Aggregating daily tweet predictions into trading signals, our method outperformed traditional fusion models (which combine sentiment-based and price-based predictions). It challenged the assumption that sentiment-based signals are inferior to price-based predictions in forecasting market movements. Backtesting these signals across three distinct market regimes yielded robust Sharpe ratios of up to 5.07 in trending markets and 3.73 in neutral markets. Our findings demonstrate that language models can serve as effective short-term market predictors. This paradigm shift underscores the untapped capabilities of language models in financial decision-making and opens new avenues for market prediction applications.

Figures (6)

• Figure 1: Overall scheme of the proposed approach
• Figure 2: Visualization of the proposed labeling method. The green and red lines represent the upper and lower barriers respectively, while the markers denote assigned labels for the given trend windows (denoted with vertical dotted lines).
• Figure 3: SHAP values plot for a sample tweet. The plot shows the contribution of each word to the model's prediction, highlighting the significant impact of the RSI feature on the Bearish prediction.
• Figure 4: Precision-recall scatter plot comparing performance across different methods for Bullish and Bearish predictions. The majority and mean aggregation methods and the fusion method demonstrate their precision, recall, and F1-scores.
• Figure 5: Yearly periods selected to represent three distinct market regimes: Bullish, Bearish, and Neutral.