---
title: Dynamic Spatio-Temporal Graph Attention Networks with Nested Clustered Optimization for Multi-Asset Portfolios
url: https://www.ml-quant.com/papers/arxiv/2611.19903/
site: ML-Quant (https://www.ml-quant.com)
updated: 2026-11-20
license: Summaries CC BY 4.0; links go to the original sources
index: https://www.ml-quant.com/llms.txt
identifier: arXiv:2611.19903
source_url: https://arxiv.org/abs/2611.19903
featured: 2026-11-20
citations: unknown
topic: Portfolio & Allocation
---


# Dynamic Spatio-Temporal Graph Attention Networks with Nested Clustered Optimization for Multi-Asset Portfolios

A graph attention network learns time-varying downside risk contagion to guide nested clustered optimization, achieving 5.32% annualized returns and 0.30 excess Sharpe ratio across 13 asset classes.

- Source: https://arxiv.org/abs/2611.19903
- Identifier: arXiv:2611.19903
- Released: 2026-11-17
- First featured: Quant Letter No. 140 (2026-11-20): https://www.ml-quant.com/issues/2026-11-20/
- Citations (Semantic Scholar): not tracked
- Published in: not yet
- Topic: Portfolio & Allocation
- Authors: Supriadi

## Abstract (arXiv, CC0)

Nested Clustered Optimization (NCO) insulates asset allocation from Markowitz covariance ill-conditioning. However, classical tree clustering relies on static pairwise Pearson correlation, failing to capture time-varying, asymmetric downside risk contagion across multi-asset markets. We propose a unified deep graph machine learning architecture: Dynamic Spatio-Temporal Graph Attention Networks with Nested Clustered Optimization (ST-GAT-NCO) for multi-asset tactical allocation. The framework extracts temporal momentum and downside risk dynamics via causal dilated 1D-convolutions, then deploys a multi-head Graph Attention Network (GAT) over dynamic market topologies to learn latent spatial node representations. These learned embeddings define an angular metric distance space that guides two-stage downside NCO optimization under institutional constraints and turnover smoothing. Evaluated over a 73-month non-anticipative walk-forward out-of-sample window (January 2020 to January 2026) across 13 global asset classes net of 10 bps transaction costs, ST-GAT-NCO achieves a 5.32% CAGR and 37.06% cumulative net return, delivering an excess Sharpe ratio of 0.30 and a Sortino ratio of 0.48. This substantially outperforms classical Minimum Variance (3.68% CAGR, 0.08 Sharpe), Hierarchical Risk Parity (2.65% CAGR), and Plain Symmetric NCO (4.11% CAGR, 0.15 Sharpe), while restricting maximum drawdown to -13.07% (versus -19.44% for Equal-Weight and -16.36% for Downside NCO) under a controlled one-way turnover of 15.31%/month. Regime analysis confirms active capital preservation across the 2020 COVID-19 crash, 2022 monetary tightening, and 2023-2026 disinflation rallies.

## Related

- [From Headlines to Holdings: Deep Learning for Smarter Portfolio Decisions](https://www.ml-quant.com/papers/arxiv/2509.24144/): A new deep learning framework for portfolio optimization, which combines LSTM networks, Graph Attention Networks, and financial news sentiment analysis, outperforms traditional benchmarks by directly learning portfolio weights, resulting in higher cumulative returns and Sharpe ratios.
- [Winners vs. Losers: Momentum-based Strategies with Intertemporal Choice for ESG Portfolios](https://www.ml-quant.com/papers/arxiv/2505.24250/): The study introduces a new momentum framework that adjusts to changing ESG sentiment, showing that portfolios with poor ESG performance outperform those with good ESG performance in pro-ESG regimes due to market overreaction, challenging traditional ESG investment beliefs.
- [Clearing Sections of Lattice Liability Networks](https://www.ml-quant.com/papers/arxiv/2503.17836/): A novel mathematical framework has been created to improve comprehension of complex financial networks, offering a unified basis for analyzing systemic risk, resource allocation, and network stability.
- [The Concentration Risk Indicator: Raising the Bar for Financial Stability and Portfolio Performance Measurement](https://www.ml-quant.com/papers/arxiv/2408.07271/): The Concentration Risk Indicator (CRI) is a new tool designed to assess risks associated with concentrated portfolios, useful in areas such as insurance risk and product portfolio mixes, especially where wealth is concentrated in few tokens.
- [Set risk measures](https://www.ml-quant.com/papers/arxiv/2407.18687/): The study presents set risk measures, which extend traditional risk measures to sets of random variables, and establishes an axiom scheme for them, demonstrating their use in systemic risk, portfolio optimization, and decision-making under uncertainty.
- [Short Communication: On the Separability of Vector-Valued Risk Measures](https://www.ml-quant.com/papers/arxiv/2407.16878/): The study argues that convex vector-valued risk measures are not suitable for defining capital allocation rules in multi-asset markets for a variety of financial applications, including systemic risk measures.
