Learning to Manage Investment Portfolios beyond Simple Utility Functions

TL;DR

The paper tackles the challenge of modeling fund manager behavior without predefined, explicit utility functions. It introduces a conditional GAN framework that learns the joint distribution of market states, holdings, and latent strategy representations, enabling realistic portfolio generation and analysis of implicit objectives. Key contributions include formulating strategy learning as conditional generation, integrating Carhart-factor-based market modeling, demonstrating superior reconstruction and latent-space interpretability on 1,436 mutual funds, and showing that learned strategies transfer across market regimes. This data-driven approach supports more realistic market simulations, strategy attribution, and regulatory oversight by capturing the diversity and transferability of real-world investment decisions. The work highlights the practical impact of moving beyond simple risk-return models to capture the full spectrum of manager behavior in financial markets.

Abstract

While investment funds publicly disclose their objectives in broad terms, their managers optimize for complex combinations of competing goals that go beyond simple risk-return trade-offs. Traditional approaches attempt to model this through multi-objective utility functions, but face fundamental challenges in specification and parameterization. We propose a generative framework that learns latent representations of fund manager strategies without requiring explicit utility specification. Our approach directly models the conditional probability of a fund's portfolio weights, given stock characteristics, historical returns, previous weights, and a latent variable representing the fund's strategy. Unlike methods based on reinforcement learning or imitation learning, which require specified rewards or labeled expert objectives, our GAN-based architecture learns directly from the joint distribution of observed holdings and market data. We validate our framework on a dataset of 1436 U.S. equity mutual funds. The learned representations successfully capture known investment styles, such as "growth" and "value," while also revealing implicit manager objectives. For instance, we find that while many funds exhibit characteristics of Markowitz-like optimization, they do so with heterogeneous realizations for turnover, concentration, and latent factors. To analyze and interpret the end-to-end model, we develop a series of tests that explain the model, and we show that the benchmark's expert labeling are contained in our model's encoding in a linear interpretable way. Our framework provides a data-driven approach for characterizing investment strategies for applications in market simulation, strategy attribution, and regulatory oversight.

Figures (2)

• Figure 1: Architecture of the generative adversarial network for learning investment strategies. The framework consists of four main components: (1) A market model that generates synthetic stock universes $(\mathbf{\hat{X}}, \mathbf{\hat{r}})$ based on the Carhart four-factor model; (2) A strategy encoder that maps observed portfolio allocations to latent strategy representations $\boldsymbol{\phi}$' (3) A portfolio allocator (decoder) that generates realistic portfolio weights $\mathbf{\hat{w}}$ conditioned on market states and strategy encodings; and (4) A discriminator that distinguishes between real and generated portfolio-market data tuples. Blue boxes represent input data, red boxes show latent representations and synthetic recreations, beige nodes indicate neural network components, and green nodes denote training objectives and evaluation metrics.
• Figure 2: Two-dimensional visualization of learned strategy representations in latent space for December 2020. Each point represents a fund's strategy encoding $\boldsymbol{\phi}_a$, with coordinates detrended relative to the S&P 500 index (positioned at origin). The separation between groups of portfolios with the same label demonstrates that the learned representations successfully capture known investment styles without supervision.