Searcharxiv⌕ Search

arXiv · 2611.19903

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

Abstract

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.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Supriadi. 2026-11-15. Dynamic Spatio-Temporal Graph Attention Networks with Nested Clustered Optimization for Multi-Asset Portfolios. https://searcharxiv.com/abs/2611.19903

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Mean--Variance Portfolio Selection by Continuous-Time Reinforcement Learning: Algorithms, Regret Analysis, and Empirical Study

We study continuous-time mean--variance portfolio selection in markets where stock prices are diffusion processes driven by observable factors that are also diffusion processes, yet the coefficients of these processes are unknown. Based on the recently developed reinforcement learning (RL) theory for diffusion processes, we present a general data-driven RL approach that learns the pre-committed investment strategy directly without attempting to learn or estimate the market coefficients. For multi-stock Black--Scholes markets without factors, we further devise an algorithm and prove its performance guarantee by deriving a sublinear regret bound in terms of the Sharpe ratio. We then carry out an extensive empirical study implementing this algorithm to compare its performance and trading characteristics, evaluated under a host of common metrics, with a large number of widely employed portfolio allocation strategies on S\&P 500 constituents. The results demonstrate that the proposed continuous-time RL strategy is consistently among the best, especially in a volatile bear market, and decisively outperforms the model-based continuous-time counterparts by significant margins.

q-fin.PM↗

A Second Order Transport Scheme for Dynamic Portfolio Subspaces under a Varying Risk Metric

We analyze the transport of portfolio exposures between moving subspaces under a varying positive definite risk metric. The reference evolution is the induced connection obtained from the ambient connection of minimum norm in fixed asset coordinates. An explicit expansion of risk weighted Procrustes identifies a leading local error that couples metric deformation with covariant motion normal to the subspace. A rank two construction with an exact continuous solution shows that this term can reduce global accuracy to first order even when the metric commutes with its derivative. Geometric mean alignment and a symmetric Procrustes construction cancel the term and yield reversible updates with second order global accuracy. We establish uniform error bounds on nonuniform partitions, characterize the cases in which the Procrustes defect vanishes, and distinguish discretization error from errors in estimated geometry. We also characterize the connection induced by principal square root whitening and its relation to the selected transport. Manufactured examples confirm the predicted defect and convergence rates. A historical geometry stress test then estimates the defect from preceding observations and finds that it predicts the subsequent separation between risk weighted Procrustes and both second order updates, while the two second order schemes remain much closer to each other. Portfolio comparisons separately evaluate target persistence, turnover, and economic implementation.

q-fin.PM↗

Revisiting Equity-Bond Diversification under Tail Risk and Structural Market Change: Evidence from Global ETFs and Model-Implied Option Prices

This study examines thirty equity--bond and multi-asset exchange-traded funds (ETFs) listed in the United States, Canada, and Australia using daily U.S.-dollar returns over 2020--2024. A rolling-window framework compares minimum-variance and Conditional Value-at-Risk (CVaR) portfolio optimization under long-only and long--short constraints against an equally weighted benchmark. Performance is assessed using risk-adjusted measures, downside risk, extreme-tail estimation, and conditional volatility models. Optimized portfolios exhibit lower empirical VaR and CVaR than the benchmark, with long--short strategies generally providing greater downside-risk reduction. However, the equally weighted benchmark achieves the highest reported Sharpe, Sortino, and Rachev ratios. GARCH and FIGARCH comparisons favor GARCH based on information criteria. A historically estimated Normal Inverse Gaussian model further generates European option prices and model-implied volatility surfaces under a martingale-corrected pricing specification. The findings highlight trade-offs between risk reduction, risk-adjusted performance, and practical implementation costs.

q-fin.PM↗