Publications
Home
Activities
Publications
- Title
- Diffolio: A diffusion model for multivariate probabilistic financial time-series forecasting and portfolio construction
- KIAS Author
- Ban, Kayoung
- Journal
- INFORMATION FUSION, 2026
- Archive
-
arXiv:2511.07014 [cs.CE]
- 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 novel diffusion-based model specifically designed for multivariate financial time-series forecasting and portfolio construction. Unlike existing diffusion-based financial time-series models that typically process financial covariates without structural distinction, Diffolio employs a denoising network with a hierarchical attention architecture, comprising both asset-level and market-level layers. This structure effectively extracts salient features not only from historical returns but also from asset-specific and systematic covariates in a manner inspired by asset pricing theory, and significantly enhances the performance of forecasts and portfolios. Furthermore, to rigorously reflect cross-sectional correlations, we introduce a correlation-guided regularizer informed by a stable estimate of the target correlation matrix. By explicitly aligning the attention map with this target correlation, our approach enforces that the learned dependency structures are well-conditioned and economically meaningful. Experimental results on the daily excess returns of 12 industry portfolios show that Diffolio outperforms various probabilistic forecasting baselines in both statistical 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 remarkably higher Sharpe ratios for the mean-variance tangency portfolio and higher certainty equivalents for the growth-optimal portfolio. These results underscore the superiority of our proposed Diffolio in terms of not only statistical accuracy but also economic significance.
