Can Competition Enhance the Proficiency of Agents Powered by Large Language Models in the Realm of News-driven Time Series Forecasting?

TL;DR

This work tackles news-driven time-series forecasting with large language models by embedding a competition-based multi-agent framework. It introduces mechanisms for information asymmetry (IA), multi-indicator evaluation (MIE), and survival of the fittest (SF), augmented by a multi-stage reflection (MSR) module that uses a fine-tuned small LLM to suppress misleading logics. Experimental results across four datasets show that competition improves forecasting accuracy, with especially strong effects on variance-sensitive metrics like MSE and RMSE, and reveal a characteristic U-shaped relation between competition intensity and performance. The study demonstrates the potential of competition-aware, LLM-based multi-agent systems for more robust and adaptive news-driven forecasting, while outlining directions for scalability, theoretical grounding, and integration with multivariate time-series methods.

Abstract

Multi-agents-based news-driven time series forecasting is considered as a potential paradigm shift in the era of large language models (LLMs). The challenge of this task lies in measuring the influences of different news events towards the fluctuations of time series. This requires agents to possess stronger abilities of innovative thinking and the identifying misleading logic. However, the existing multi-agent discussion framework has limited enhancement on time series prediction in terms of optimizing these two capabilities. Inspired by the role of competition in fostering innovation, this study embeds a competition mechanism within the multi-agent discussion to enhance agents' capability of generating innovative thoughts. Furthermore, to bolster the model's proficiency in identifying misleading information, we incorporate a fine-tuned small-scale LLM model within the reflective stage, offering auxiliary decision-making support. Experimental results confirm that the competition can boost agents' capacity for innovative thinking, which can significantly improve the performances of time series prediction. Similar to the findings of social science, the intensity of competition within this framework can influence the performances of agents, providing a new perspective for studying LLMs-based multi-agent systems.

Figures (11)

• Figure 1: The framework of the proposed model.
• Figure 2: This section compares logic similarity between models with and without Information Asymmetry (IA), using electricity and Bitcoin datasets. Higher logic similarity indicates less innovative thinking.
• Figure 3: This figure compares logic update degree and MAPE across epochs for models with and without IA, using electricity and Bitcoin datasets.
• Figure 4: Comparison of Logic Update Degree ($rank$, $top$, $ave$) across three epochs in IA and no IA contexts, demonstrating the impact of competition on innovative thinking.
• Figure 5: This figure shows the relationship between MAPE and the competitive degrees of different agents. The U-shaped trend indicates that MAPE gets its optimal value when when competition is at a moderate level. HHI is the Herfindahl-Hirschman Index.