Diffolio: A Diffusion Model for Multivariate Probabilistic Financial Time-Series Forecasting and Portfolio Construction

2025-11-10

AI Summary

This paper addresses the joint optimization of probabilistic forecasting for multivariate financial time series and dependence-aware portfolio construction. Methodologically, it proposes a diffusion-based probabilistic forecasting framework featuring a hierarchical attention architecture that integrates asset-level and market-level features; incorporates a correlation-guided regularization term; employs robust correlation matrix estimation to enhance cross-sectional dependence modeling; and jointly forecasts asset returns by incorporating both asset-specific and systemic covariates. Experiments on daily excess returns across 12 industry sectors demonstrate statistically significant improvements in forecast accuracy (e.g., lower CRPS) over state-of-the-art baselines, alongside enhanced out-of-sample portfolio performance—measured by higher Sharpe ratios and certainty-equivalent returns—validating both statistical efficacy and economic value. The core contribution lies in the principled unification of diffusion modeling, hierarchical attention, and correlation regularization, enabling high-fidelity, cross-asset dependence characterization and end-to-end investment performance optimization for the first time.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Computer Vision: Diffusion Models for Vision
• Machine Learning: Feature Construction/Reformulation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Architectures and workflows that use LLMs for crowd work

Abstract

Probabilistic forecasting is crucial in multivariate financial time-series for constructing efficient portfolios that account for complex cross-sectional dependencies. In this paper, we propose Diffolio, a diffusion model designed for multivariate financial time-series forecasting and portfolio construction. Diffolio employs a denoising network with a hierarchical attention architecture, comprising both asset-level and market-level layers. Furthermore, to better reflect cross-sectional correlations, we introduce a correlation-guided regularizer informed by a stable estimate of the target correlation matrix. This structure effectively extracts salient features not only from historical returns but also from asset-specific and systematic covariates, significantly enhancing the performance of forecasts and portfolios. Experimental results on the daily excess returns of 12 industry portfolios show that Diffolio outperforms various probabilistic forecasting baselines in multivariate forecasting accuracy and portfolio performance. Moreover, in portfolio experiments, portfolios constructed from Diffolio's forecasts show consistently robust performance, thereby outperforming those from benchmarks by achieving higher Sharpe ratios for the mean-variance tangency portfolio and higher certainty equivalents for the growth-optimal portfolio. These results demonstrate the superiority of our proposed Diffolio in terms of not only statistical accuracy but also economic significance.

Problem

• Forecasting multivariate financial time-series with probabilistic methods
• Modeling cross-sectional dependencies for portfolio construction
• Enhancing forecast accuracy and portfolio performance economically

Innovation

• Diffusion model for multivariate financial forecasting
• Hierarchical attention with asset and market layers
• Correlation-guided regularizer using target matrix