TradeFM: A Generative Foundation Model for Trade-flow and Market Microstructure

Paper: arXiv 2602.23784

Authors: Maxime Kawawa-Beaudan, Srijan Sood, Kassiani Papasotiriou, Daniel Borrajo, Manuela Veloso

Abstract

Foundation models have transformed domains from language to genomics by learning general-purpose representations from large-scale, heterogeneous data. We introduce TradeFM, a 524M-parameter generative Transformer that brings this paradigm to market microstructure, learning directly from billions of trade events across >9K equities. To enable cross-asset generalization, we develop scale-invariant features and a universal tokenization scheme that map the heterogeneous, multi-modal event stream of order flow into a unified discrete sequence – eliminating asset-specific calibration. Integrated with a deterministic market simulator, TradeFM-generated rollouts reproduce key stylized facts of financial returns, including heavy tails, volatility clustering, and absence of return autocorrelation. Quantitatively, TradeFM achieves 2-3x lower distributional error than Compound Hawkes baselines and generalizes zero-shot to geographically out-of-distribution APAC markets with moderate perplexity degradation. Together, these results suggest that scale-invariant trade representations capture transferable structure in market microstructure, opening a path toward synthetic data generation, stress testing, and learning-based trading agents.

Complexity vs Empirical Score

• Math Complexity: 5.5/10
• Empirical Rigor: 8.5/10
• Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: TradeFM applies large-scale foundation model paradigms to market microstructure using billions of event tokens across thousands of equities. It combines strong empirical rigor via extensive validation and out-of-distribution APAC zero-shot testing with innovative scale-invariant tokenization and closed-loop market simulation.

Research Flowchart

  flowchart TD
    A["Research Goal: Build a universal generative foundation model for market microstructure"] --> B["Input Data: Billions of trade events across >9K equities"]
    B --> C["Methodology: Scale-invariant features & universal discrete tokenization"]
    C --> D["Model Training: 524M-parameter Transformer for order flow sequence modeling"]
    D --> E["Simulation & Eval: Deterministic market simulator rollouts & zero-shot OOD testing"]
    E --> F["Outcomes: 2-3x lower error than Hawkes baselines, zero-shot APAC generalization, and reproduction of return stylized facts"]