Computer Science > Neural and Evolutionary Computing
[Submitted on 3 Sep 2026]
Title:CompEvo: Competition-Induced Evolution for Multi-Agent in News-Driven Time Series Forecasting
View PDF HTML (experimental)Abstract:News-driven time series forecasting uses evolving textual events together with historical observations to predict future values, supporting applications such as market risk monitoring and resource scheduling. In multi-agent settings, two challenges still remain. The first is degeneration of thought, where agents converge to similar evidence-seeking behaviors. The second is insufficient theoretical grounding, where strategy updates are often heuristic and lack a principled formulation. To address the above challenges, we propose CompEvo, a competition-induced evolution framework for multi-agent news-driven time series forecasting. For theoretical grounding, we introduce an evolutionary game formulation to guarantee equilibrium existence and optimization convergence. Building on this formulation, we construct a trainable multi-agent evolution framework that integrates strategy execution, fitness-based differentiable selection, and competition-induced strategy evolution. CompEvo enables heterogeneous agents to explore diverse news evidence, converts forecasting feedback into differentiable influence weights, and evolves agent strategies under competitive pressure to preserve effective logic while maintaining diversity. Experiments on four real-world datasets show that CompEvo reduces RMSE by 27.3% and MAPE by 26.2% on average over strong baselines. Further analysis indicates that CompEvo successfully maintains diverse and specialized agent behaviors.
References & Citations
Loading...
Bibliographic and Citation Tools
Bibliographic Explorer (What is the Explorer?)
Connected Papers (What is Connected Papers?)
Litmaps (What is Litmaps?)
scite Smart Citations (What are Smart Citations?)
Code, Data and Media Associated with this Article
alphaXiv (What is alphaXiv?)
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub (What is DagsHub?)
Gotit.pub (What is GotitPub?)
Hugging Face (What is Huggingface?)
ScienceCast (What is ScienceCast?)
Demos
Recommenders and Search Tools
Influence Flower (What are Influence Flowers?)
CORE Recommender (What is CORE?)
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Facts Only
* CompEvo is a competition-induced evolution framework.
* The framework is designed for multi-agent news-driven time series forecasting.
* The submission date is 3 September 2026.
* The system utilizes an evolutionary game formulation.
* Components include strategy execution, fitness-based differentiable selection, and competition-induced strategy evolution.
* The framework enables heterogeneous agents to explore news evidence.
* Forecasting feedback is converted into differentiable influence weights.
* Evaluation was conducted on four real-world datasets.
* Results show a 27.3% average reduction in RMSE.
* Results show a 26.2% average reduction in MAPE.
* The framework aims to solve degeneration of thought and insufficient theoretical grounding in multi-agent settings.
Executive Summary
News-driven time series forecasting integrates textual event data with historical observations to predict future values for applications like resource scheduling and market risk monitoring. Current multi-agent approaches to this problem suffer from "degeneration of thought," where agents converge on identical behaviors, and a lack of principled theoretical grounding for strategy updates.
CompEvo addresses these issues by applying an evolutionary game formulation to ensure optimization convergence and equilibrium. This framework employs a trainable system where agents compete, and their strategies evolve based on differentiable selection and fitness. By converting forecasting feedback into influence weights, the system maintains agent diversity and specialization. Experimental data across four real-world datasets indicate significant performance improvements, specifically reducing RMSE by 27.3% and MAPE by 26.2% compared to existing strong baselines.
Full Take
The methodology centers on transforming a forecasting problem into a competitive ecosystem. By treating agent strategies as evolving species in a game-theoretic landscape, the researchers attempt to solve the common AI failure mode of mode collapse (degeneration of thought). The use of differentiable selection is a sophisticated bridge between discrete evolutionary logic and gradient-based optimization, which is a sound technical approach for maintaining heterogeneity in neural ensembles.
However, a rigorous peer review would flag the absence of specific dataset descriptions and baseline identities in the abstract. While the percentage drops in RMSE and MAPE are numerically impressive, the magnitude of "improvement" is relative to the strength of the baselines. Without knowing if these baselines are state-of-the-art or rudimentary, the claim of superiority remains provisional. Furthermore, the "equilibrium existence" guarantee is a theoretical claim that requires verification against the actual stochastic nature of real-world news data, which rarely follows stable game-theoretic rules.
The real-world implication is a shift toward "adversarial collaboration" in AI, where diversity is not just a byproduct but a forced architectural requirement. If this holds, forecasting systems will move away from seeking a single "correct" model toward maintaining a portfolio of specialized, competing perspectives.
To falsify these claims, one would need to test CompEvo on "black swan" events where historical news patterns break down entirely; if the competitive pressure leads to over-specialization on noise, the framework may fail in high-volatility environments.
Bridge Questions:
1. Does the drive for agent diversity create a "competition for the sake of competition," or does each specialized behavior map to a distinct, identifiable market logic?
2. How does the system distinguish between an agent that is "diverse but wrong" and one that is "diverse and insightful"?
Sentinel — Human
This text exhibits the characteristic structure and vocabulary of peer-reviewed machine learning research but lacks the generic flow often seen in pure synthetic content, suggesting human authorship or high-quality AI synthesis based on source material.