MEME: Modeling the Evolutionary Modes of Financial Markets

TL;DR

The paper reframes financial market analysis from asset- or pool-centric perspectives to a logic-oriented view that treats price movements as manifestations of evolving investment logics or Modes of Thought. MEME operationalizes this view through three stages: (1) multi-agent argument extraction to produce Investment Arguments from multimodal data, (2) Gaussian Mixture Model–based identification and temporal alignment of Modes of Thought, and (3) a mode-evaluation mechanism that weights signals by historically robust narratives to construct daily portfolios. Across three Chinese stock pools (SSE 50, CSI 300, CSI 500) from 2023 to 2025, MEME consistently outperforms seven SOTA baselines on multiple metrics (IC, ICIR, AR, SR) and demonstrates robustness via ablation studies, sensitivity analyses, and a data-leak–free out-of-sample test. A case study visualizes mode lifecycles, illustrating how effective investment narratives emerge, endure, or decay with market regimes, underscoring MEME’s potential to capture durable market wisdom beyond transient anomalies.

Abstract

LLMs have demonstrated significant potential in quantitative finance by processing vast unstructured data to emulate human-like analytical workflows. However, current LLM-based methods primarily follow either an Asset-Centric paradigm focused on individual stock prediction or a Market-Centric approach for portfolio allocation, often remaining agnostic to the underlying reasoning that drives market movements. In this paper, we propose a Logic-Oriented perspective, modeling the financial market as a dynamic, evolutionary ecosystem of competing investment narratives, termed Modes of Thought. To operationalize this view, we introduce MEME (Modeling the Evolutionary Modes of Financial Markets), designed to reconstruct market dynamics through the lens of evolving logics. MEME employs a multi-agent extraction module to transform noisy data into high-fidelity Investment Arguments and utilizes Gaussian Mixture Modeling to uncover latent consensus within a semantic space. To model semantic drift among different market conditions, we also implement a temporal evaluation and alignment mechanism to track the lifecycle and historical profitability of these modes. By prioritizing enduring market wisdom over transient anomalies, MEME ensures that portfolio construction is guided by robust reasoning. Extensive experiments on three heterogeneous Chinese stock pools from 2023 to 2025 demonstrate that MEME consistently outperforms seven SOTA baselines. Further ablation studies, sensitivity analysis, lifecycle case study and cost analysis validate MEME's capacity to identify and adapt to the evolving consensus of financial markets. Our implementation can be found at https://github.com/gta0804/MEME.

Figures (6)

• Figure 1: The overall architecture. MEME first mitigates structural noise by distilling raw multimodal data into structured Investment Arguments via multi-agent extraction. To uncover latent consensus, it identifies representative Modes of Thought through GMM-based probabilistic clustering and models market evolution via temporal alignment of these modes across consecutive days. Finally, MEME filters logical noise by evaluating the historical profitability of each mode through an alignment-based scoring mechanism, ensuring that the portfolio is guided by robust and enduring investment narratives.
• Figure 2: Backtest on the CSI 300. We compare MEME with all baselines and the CSI 300 Index.
• Figure 3: MEME exhibits a moderate sensitivity.
• Figure 4: Temporal evolution of normalized performance for different categories of Modes of Thought on the CSI 300. The stack plot illustrates the dynamic lifecycle and shifting prevalence of modes across different market conditions.
• Figure 5: MEME exhibits a moderate sensitivity to changes in hyperparameters on RIC.