Bi-Level Chaotic Fusion Based Graph Convolutional Network for Stock Market Prediction Interval

2026-05-19 · Source: cs.LG updates on arXiv.org · Field: Finance & Economics — Capital Markets & Investment Management, FinTech & Digital Financial Services · Depth: Expert, short

Summary

A new spatio-temporal graph-based model, the Bi-Level Chaotic Fusion Based Graph Convolutional Network, has been developed to address the limitations of point predictions in financial market forecasting by providing prediction intervals. This model employs separate nonlinear transformation functions for estimating the interval center and width, incorporating a volatility-aware gating mechanism to adapt predictions to market regimes. It also considers temporal dependencies through embedded graph structures and sequential modeling, trained using a Lower-Upper Bound Estimation (LUBE) objective. Experimental results, based on data from 2016 to 2026 covering 43 leading companies across eight sectors of the NSE, demonstrate significant improvements over baselines like LSTM, GRU, GCN, and HGNN. The model achieved a Winkler score of 0.0778, a Prediction Interval Average Width (PIAW) of 0.1407, and a Prediction Interval Coverage Probability (PICP) of 96.6%, with all improvements being statistically significant (p < 0.001) according to the Diebold-Mariano test.

Key takeaway

For research scientists developing financial forecasting models, you should prioritize prediction interval generation over single-value point predictions to provide more robust, uncertainty-aware outputs. Consider integrating spatio-temporal graph networks and volatility-aware mechanisms, as demonstrated, to improve interval calibration and sharpness, which is critical for practical risk management in dynamic markets.

Key insights

Prediction intervals, not just point forecasts, are crucial for risk-aware financial market decision-making.

Principles

• Uncertainty quantification is vital for financial forecasts.
• Market regimes influence prediction accuracy.
• Spatio-temporal relationships enhance forecasting.

Method

The model uses bi-level chaotic fusion with separate nonlinear functions for interval center/width, a volatility-aware gating mechanism, and LUBE objective for training.

In practice

• Use prediction intervals for risk assessment.
• Incorporate market volatility into models.
• Model asset relationships with graph structures.

Topics

• Stock Market Prediction
• Prediction Intervals
• Graph Convolutional Networks
• Bi-Level Chaotic Fusion
• Spatio-Temporal Graph Modeling

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Learning Long Range Spatio-Temporal Representations over Continuous Time Dynamic Graphs with State Space Models — CTDG-SSM unifies state-space modeling with graph topology to capture long-range spatio-temporal dependencies in continuous-time dynamic graphs.
• Probabilistic Salary Prediction with Graph Attention Networks and a Mixture Density Network — Probabilistic salary prediction benefits from modeling attribute relationships and output uncertainty using graph neural networks and mixture density networks.
• When the Chart is Lying. — Integrating NLP with GNNs allows trading bots to react to real-world news before market charts reflect price changes.
• Including Node Textual Metadata in Laplacian-constrained Gaussian Graphical Models — Fusing node signals with textual metadata significantly enhances graph learning in Gaussian Graphical Models.
• Optimal Spatio-Temporal Decoupling for Bayesian Conformal Prediction — SA-BCP optimally balances temporal adaptability and structural stability in non-stationary time series using spatio-temporal decoupling.
• The Boosting Paradox — Machine learning struggles in finance due to market adaptivity, adversarial competition, and irreducible uncertainty, despite theoretical power.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.