Neural Functionally Generated Portfolios

2025-06-24

AI Summary

Conventional functional generation portfolio (FGP) design relies on estimates of asset drift and covariance, making it difficult to simultaneously satisfy self-financing and pathwise decomposition properties. Method: We propose an end-to-end neural learning framework that parameterizes the generating function (G(cdot)) as a differentiable neural network, trained directly on synthetic or real market data to maximize cumulative log return relative to the market portfolio—eliminating the need for explicit parameter estimation. The architecture is theoretically guaranteed to enforce strict self-financing and pathwise decomposition. Results: Empirical evaluation demonstrates that the proposed neural FGP consistently outperforms classical FGP benchmarks across diverse market regimes, achieving superior returns, robustness, and adaptability. This work establishes a novel, model-free, data-driven paradigm for portfolio construction.

Technology Category

Natural Language Processing: GenerationMachine Learning: Deep Generative Models & AutoencodersSearch and Optimization: Learning to Search

Application Category

Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applicationsGraph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphsSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

We introduce a novel neural-network-based approach to learning the generating function $G(cdot)$ of a functionally generated portfolio (FGP) from synthetic or real market data. In the neural network setting, the generating function is represented as $G_θ(cdot)$, where $θ$ is an iterable neural network parameter vector, and $G_θ(cdot)$ is trained to maximise investment return relative to the market portfolio. We compare the performance of the Neural FGP approach against classical FGP benchmarks. FGPs provide a robust alternative to classical portfolio optimisation by bypassing the need to estimate drifts or covariances. The neural FGP framework extends this by introducing flexibility in the design of the generating function, enabling it to learn from market dynamics while preserving self-financing and pathwise decomposition properties.

Problem

• Learn generating function for portfolios using neural networks
• Maximize investment return relative to market portfolio
• Bypass drift and covariance estimation in portfolio optimization

Innovation

• Neural network learns portfolio generating function
• Maximizes return relative to market portfolio
• Preserves self-financing and decomposition properties