Retrieval-Augmented Diffusion for Stochastic Discount Factor Portfolios

Summary

This work proposes RADAR, a retrieval-augmented diffusion framework for learning market state representations used in portfolio optimization under the stochastic discount factor framework. Its motivation is that markets shift between regimes, financial inputs such as prices and news are noisy, and conventional diffusion models may impose uniform Gaussian noise that does not reflect changing uncertainty across states.

RADAR retrieves similar historical regimes to create context-dependent noise distributions, then uses conditional diffusion to denoise multimodal representations. It also initializes diffusion from empirical statistics to represent state-dependent uncertainty. The authors report better risk-adjusted metrics and economically meaningful signals for returns and correlations. The excerpt does not name the datasets, baselines, evaluation periods, or specific metrics, so the strength and portability of those results cannot be independently assessed from this description.

Key ideas

• RADAR learns market representations for stochastic discount factor portfolio optimization.
• Retrieval of similar historical regimes informs context-dependent noise distributions.
• Conditional diffusion is used to denoise multimodal market representations.
• The method initializes diffusion from empirical statistics to reflect state-dependent uncertainty.
• Reported results include improved risk-adjusted measures, though evaluation details are not supplied in the excerpt.

Tags

# Retrieval-Augmented Diffusion Modeling for Stochastic Discount Factor Portfolios


# Retrieval-Augmented Diffusion Modeling for Stochastic Discount Factor Portfolios









In this work, we study portfolio optimization under the stochastic discount factor (SDF) framework by learning market state representations that capture the underlying risk structures of financial data. This is challenging due to several factors: financial markets exhibit non-stationary dynamics with shifting regimes, multimodal inputs such as price and news data often contain stochastic noise, and existing diffusion-based approaches, while effective for modeling stochastic dynamics, rely on assumptions such as isotropic Gaussian noise that fail to capture the state-dependent nature of financial uncertainty. To address these challenges, we introduce RADAR, a retrieval-augmented diffusion framework that learns market representations by conditioning on similar historical regimes. RADAR leverages retrieval to construct context-dependent noise distributions, applies conditional diffusion to denoise multimodal representations, and initializes the diffusion process using empirical statistics to reflect state-dependent uncertainty. Experiments show that RADAR achieves state-of-the-art performance on key risk-adjusted metrics while producing economically meaningful signals on asset returns and correlations.

Shown in full with attribution under the source's licence. Licence: abstract CC0

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.