Signature-Informed Transformer for Asset Allocation

2025-10-03

AI Summary

Deep learning predictors suffer from insufficient robustness in quantitative asset allocation due to objective mismatch and error propagation. To address this, we propose an end-to-end risk-aware investment strategy framework. Methodologically, we innovatively integrate path signatures with the Transformer architecture: path signatures capture the geometric structure and lead-lag relationships of asset price paths; we design a signature-enhanced attention mechanism and formulate a differentiable end-to-end model that directly optimizes risk-sensitive portfolio objectives—bypassing intermediate prediction tasks. Empirical evaluation on S&P 100 daily data demonstrates that our framework significantly outperforms classical mean-variance optimization, factor models, and state-of-the-art deep learning baselines, delivering superior risk-adjusted returns and robustness—particularly under volatile market conditions. Our key contribution is the first incorporation of path signatures into the Transformer architecture for geometric modeling of financial time series and end-to-end strategy optimization.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Intelligent Robots: Learning & Optimization for ROB
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Security and Privacy: Data transparency and provenance

Abstract

Robust asset allocation is a key challenge in quantitative finance, where deep-learning forecasters often fail due to objective mismatch and error amplification. We introduce the Signature-Informed Transformer (SIT), a novel framework that learns end-to-end allocation policies by directly optimizing a risk-aware financial objective. SIT's core innovations include path signatures for a rich geometric representation of asset dynamics and a signature-augmented attention mechanism embedding financial inductive biases, like lead-lag effects, into the model. Evaluated on daily S&P 100 equity data, SIT decisively outperforms traditional and deep-learning baselines, especially when compared to predict-then-optimize models. These results indicate that portfolio-aware objectives and geometry-aware inductive biases are essential for risk-aware capital allocation in machine-learning systems. The code is available at: https://github.com/Yoontae6719/Signature-Informed-Transformer-For-Asset-Allocation

Problem

• Addresses objective mismatch in deep-learning financial forecasting models
• Develops end-to-end allocation policies optimizing risk-aware financial objectives
• Incorporates geometric asset dynamics and financial biases into transformer architecture

Innovation

• Uses path signatures for geometric asset dynamics representation
• Integrates signature-augmented attention with financial inductive biases
• Learns end-to-end allocation via direct risk-aware objective optimization