NetWeaver is a research and analytics module within the NovoXpert ecosystem, designed to model structural relationships, dependencies, and information flow across financial instruments using graph neural networks.
Important: NetWeaver is not an investment advisory system. It does not provide buy/sell instructions, allocations, target prices, or performance promises. All outputs are intended for market structure analysis, monitoring, and risk awareness only.
NetWeaver leverages Graph Attention Networks (GATs) to learn how instruments are connected across sectors and time.
Its primary goal is to identify influence patterns, lead–lag behavior, and systemic coupling, not to recommend trades.
The project is inspired by academic research (e.g., FinGAT) and has been extended for diagnostic, monitoring, and governance-oriented use cases.
- Graph-based modeling of inter-asset relationships
- Identification of influence-out / influence-in dynamics
- Detection of lead–lag dependencies (probabilistic, historical)
- Sector-level and cross-sector connectivity analysis
- Structural stress and correlation diagnostics
All outputs are descriptive, not prescriptive.
NetWeaver consists of three conceptual layers:
-
Temporal Encoding
Encodes historical time-series behavior using recurrent or attention-based encoders. -
Graph Attention Layer
Learns weighted relationships between nodes (assets) based on:- Sector proximity
- Historical co-movement
- Correlation structure
-
Analytical Heads
Produce scores used for:- Relative salience
- Dependency strength
- Network diagnostics
These scores are not directions and should not be interpreted as expected returns.
- Historical OHLCV data
- Sector or category metadata
- Graph edge definitions (intra-sector / inter-sector)
Optional:
- External signals for experimental research
NetWeaver produces structural metrics, including:
- Dependency strength between instruments
- Influence centrality (in / out)
- Lead–lag likelihood (probabilistic, historical)
- Network density and clustering diagnostics
No output implies:
- Allocation
- Trade timing
- Profit expectation
- Market structure analysis
- Risk committee diagnostics
- Research on systemic coupling
- Stress and contagion monitoring
- Educational and academic experimentation
This repository is provided for research and informational purposes only.
Any financial interpretation or use of results is the sole responsibility of the user.
Past data analysis does not imply future outcomes.
This project is intended for:
- Academic research
- Internal analytics
- Experimental system design
Commercial or regulated use requires independent legal and compliance review.
Inspired by academic research on financial graph neural networks, including FinGAT, with significant adaptations for non-advisory, risk-focused analysis.
© NovoXpert — Graph Intelligence for Market Structure