Crisis-Resilient Portfolio Management via Graph-based Spatio-Temporal Learning

2025-10-23

AI Summary

Financial time-series forecasting faces a core challenge during crises: inter-asset dependency structures undergo abrupt, mechanism-driven shifts (e.g., credit contagion, pandemic shocks, inflation-driven selloffs), yet prevailing graph neural network (GNN) approaches rely on predefined static graph topologies, severely limiting generalizability. To address this, we propose CRISP—a novel end-to-end framework that jointly learns dynamic sparse graph structures and spatiotemporal dependencies. CRISP employs multi-head graph attention to adaptively infer asset relationships, integrating graph convolution for spatial modeling with bidirectional LSTM and self-attention for temporal modeling—without any prior graph assumptions. Empirically evaluated on high-inflation markets (2022–2024), CRISP achieves a Sharpe ratio of 3.76—representing a 707% improvement over the equal-weighted benchmark and a 94% gain over static-graph baselines. Notably, during crises, CRISP’s attention weights significantly increase for defensive assets, demonstrating both superior predictive performance and inherent interpretability.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Financial time series forecasting faces a fundamental challenge: predicting optimal asset allocations requires understanding regime-dependent correlation structures that transform during crisis periods. Existing graph-based spatio-temporal learning approaches rely on predetermined graph topologies--correlation thresholds, sector classifications--that fail to adapt when market dynamics shift across different crisis mechanisms: credit contagion, pandemic shocks, or inflation-driven selloffs. We present CRISP (Crisis-Resilient Investment through Spatio-temporal Patterns), a graph-based spatio-temporal learning framework that encodes spatial relationships via Graph Convolutional Networks and temporal dynamics via BiLSTM with self-attention, then learns sparse structures through multi-head Graph Attention Networks. Unlike fixed-topology methods, CRISP discovers which asset relationships matter through attention mechanisms, filtering 92.5% of connections as noise while preserving crisis-relevant dependencies for accurate regime-specific predictions. Trained on 2005--2021 data encompassing credit and pandemic crises, CRISP demonstrates robust generalization to 2022--2024 inflation-driven markets--a fundamentally different regime--by accurately forecasting regime-appropriate correlation structures. This enables adaptive portfolio allocation that maintains profitability during downturns, achieving Sharpe ratio 3.76: 707% improvement over equal-weight baselines and 94% improvement over static graph methods. Learned attention weights provide interpretable regime detection, with defensive cluster attention strengthening 49% during crises versus 31% market-wide--emergent behavior from learning to forecast rather than imposing assumptions.

Problem

• Predicting optimal asset allocations during financial crises
• Adapting to changing market correlation structures dynamically
• Filtering noise while preserving crisis-relevant asset relationships

Innovation

• Graph Convolutional Networks encode spatial asset relationships
• BiLSTM with self-attention captures temporal market dynamics
• Multi-head Graph Attention learns sparse crisis-relevant dependencies