Decision-informed Neural Networks with Large Language Model Integration for Portfolio Optimization

TL;DR

This work targets the gap between forecast accuracy and portfolio performance by introducing DINN, a decision-informed neural network that integrates Large Language Model (LLM) representations with a differentiable portfolio optimization layer. The architecture employs an attention mechanism to fuse cross-asset relationships, temporal dynamics, and macro variables, and is trained with a hybrid loss $\mathcal{L}_{loss}=\beta \mathcal{L}_{MSE}+(1-\beta)\mathcal{L}_{Decision}$ where $\beta=0.4$, enabling end-to-end learning that aligns predictions with investment decisions. Empirical results on the S&P 100 and DOW 30 demonstrate that DINN outperforms strong Transformer- and LLM-based baselines across eight metrics, including higher annualized return and superior risk-adjusted performance, while gradient analyses show that high-sensitivity assets receive more targeted learning. The work also shows that prob-sparse cross-attention improves efficiency and interpretability, providing a robust framework that bridges forecasting and portfolio optimization for context-aware decision-making in finance.

Abstract

This paper addresses the critical disconnect between prediction and decision quality in portfolio optimization by integrating Large Language Models (LLMs) with decision-focused learning. We demonstrate both theoretically and empirically that minimizing the prediction error alone leads to suboptimal portfolio decisions. We aim to exploit the representational power of LLMs for investment decisions. An attention mechanism processes asset relationships, temporal dependencies, and macro variables, which are then directly integrated into a portfolio optimization layer. This enables the model to capture complex market dynamics and align predictions with the decision objectives. Extensive experiments on S\&P100 and DOW30 datasets show that our model consistently outperforms state-of-the-art deep learning models. In addition, gradient-based analyses show that our model prioritizes the assets most crucial to decision making, thus mitigating the effects of prediction errors on portfolio performance. These findings underscore the value of integrating decision objectives into predictions for more robust and context-aware portfolio management.

Figures (1)

• Figure 1: Schematic of the proposed Decision-Informed Neural Network (DINN) architecture for unified return forecasting and portfolio selection. The entire system is trained end-to-end to align predictive accuracy with decision quality.