Abstract:The integration of Artificial Intelligence (AI) in the financial domain has opened new avenues for quantitative trading, particularly through the use of Large Language Models (LLMs). However, the challenge of effectively synthesizing insights from diverse data sources and integrating both structured and unstructured data persists. This paper presents TradeExpert, a novel framework that employs a mix of experts (MoE) approach, using four specialized LLMs, each analyzing distinct sources of financial data, including news articles, market data, alpha factors, and fundamental data. The insights of these expert LLMs are further synthesized by a General Expert LLM to make a final prediction or decision. With specific prompts, TradeExpert can be switched between the prediction mode and the ranking mode for stock movement prediction and quantitative stock trading, respectively. In addition to existing benchmarks, we also release a large-scale financial dataset to comprehensively evaluate TradeExpert's effectiveness. Our experimental results demonstrate TradeExpert's superior performance across all trading scenarios.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to effectively integrate multiple data sources (including news articles, market data, alpha factors, and fundamental data) in financial trading, and conduct stock price trend prediction and quantitative trading strategy formulation through large language models (LLMs). Specifically, the paper proposes a new framework named TradeExpert. This framework adopts the Mixture of Experts (MoE) approach, using four specialized LLMs to analyze different types of financial data respectively, and then synthesizes the insights of these specialized LLMs through a general - purpose expert LLM to make the final prediction or decision. The paper also proposes a new ranking mode, enabling TradeExpert to act as a comparator to select Top - K - ranked stocks for trading in a loose ranking algorithm. In addition, the paper also releases a large - scale financial data set to comprehensively evaluate the effectiveness of TradeExpert. Experimental results show that TradeExpert performs excellently in all trading scenarios.
### Main problems solved by the paper:
1. **Effective integration of multiple data sources**: How to effectively extract useful information from news articles, market data, alpha factors, and fundamental data, and integrate it into a unified prediction model.
2. **Improve prediction accuracy**: By using the Mixture of Experts (MoE) method, combine multiple specialized LLMs to improve the accuracy of stock price trend prediction.
3. **Optimize trading strategies**: By introducing a ranking mode, enable the model to select the best stock portfolio in actual trading, thereby optimizing trading strategies.
4. **Publication of data sets**: Provide a large - scale financial data set so that other researchers can use it to evaluate and improve related models.
### Specific tasks:
1. **Stock price trend prediction**: Based on multi - modal data sources, predict the rise and fall of stock prices in the next day.
2. **Stock trading simulation**: Through simulating the buy - and - hold strategy, evaluate the profitability and risk of the model in the actual market.
### Method overview:
- **Mixture of Experts (MoE) method**: Use four specialized LLMs to process different data sources respectively, and each LLM focuses on a specific type of data (such as news articles, market data, alpha factors, and fundamental data).
- **General - purpose expert LLM**: Synthesize the outputs of the four specialized LLMs to generate the final prediction or decision.
- **Ranking mode**: In the trading strategy, use the general - purpose expert LLM as a comparator, and select Top - K - ranked stocks through a loose ranking algorithm.
### Experimental results:
- **Stock price trend prediction**: On multiple benchmark data sets, the performance of TradeExpert is better than that of existing hybrid models and large language models.
- **Stock trading simulation**: The back - testing results on the DOW 30 stock pool show that TradeExpert is significantly better than traditional models and deep - learning models in multiple indicators such as annualized return rate, Sharpe ratio, annualized volatility, and maximum drawdown.
### Conclusion:
The TradeExpert framework proposed in the paper effectively integrates multiple financial data sources through the Mixture of Experts method, improves the accuracy of stock price trend prediction and the optimization effect of trading strategies. The experimental results verify the superior performance of this framework in financial trading.