Research & Analysis
Constructing New Power System Stock Indices via a Compound AI System
A Hierarchical RAG and Knowledge Graph Approach for Dynamic Financial Indexing
Authors: Tao Ma, Bingcai Liu (Central University of Finance and Economics), Siqi Yang
Published: ICCSMT '25, 01 April 2026
Executive Impact at a Glance
Our analysis reveals significant improvements and strategic advantages offered by the novel AI-driven approach for financial index construction.
0%
Peak 5-Year Return Rate
0%
Improved Classification Accuracy
0%
Reduction in Hallucination
0%
Faster Index Adaptation
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The Dynamic Knowledge Injection Pipeline (DKIP)
This research introduces a novel **Dynamic Knowledge Injection Pipeline (DKIP)**, a compound AI system designed to overcome challenges in extracting and classifying unstructured data. DKIP leverages a Hierarchical Retrieval-Augmented Generation (H-RAG) architecture that integrates a Large Language Model (LLM) with a graph database (Neo4j). This allows the system to anchor semantic inferences within strict and complex taxonomies, significantly mitigating the "hallucination" issues common in traditional LLM-based approaches.
The system automates the construction of a rigorous knowledge graph from regulatory documents, processing complex textual regulations into executable logic gates. This ensures structural integrity and allows for dynamic updates in response to frequent policy changes without human intervention.
Innovating Financial Indexing with AI
The global energy system's profound transition necessitates new financial instruments. This study applies the DKIP framework to construct a "New Power System Stock Index" using Chinese A-share data. The objective is to capture the financial value of businesses related to the new power system, which is complex due to unstructured policy data and fragmented industry boundaries.
The DKIP methodology performs **semantic analysis** of firms' revenue descriptions, classifying them into five functional dimensions (source, grid, load, storage, intelligence). This enables intelligent recognition of industry boundaries and supports dynamic updates, allowing for a strategic stock index that is replicable, transparent, and responsive to market changes.
Superior Performance & Strategic Market Relevance
The proposed "New Power System Stock Index" demonstrably **outperforms traditional keyword matching** in classification accuracy and yields higher returns than existing benchmarks over five years. Experimental results show the index consistently ranks at the top in cumulative returns under various weighting adjustments (e.g., up to 158% peak 5-year return).
While exhibiting higher volatility—common in dynamic renewable energy sectors—the index maintains moderate but stable risk-adjusted returns (Sharpe ratios between 0.602 and 0.655). This confirms that LLM-based knowledge injection is a viable technological approach for designing precise and robust financial instruments that adapt to evolving market dynamics and policy shifts.
Enterprise Process Flow: Dynamic Knowledge Injection Pipeline (DKIP)
Policy Documents & Data
→
LLM Extraction & Logic Parsing
→
Neo4j Knowledge Graph Construction
→
Business Descriptions
→
Agentic Reasoning & LLM Filter
→
Semantic Classification & Report
158%
Peak 5-Year Cumulative Return for New Index
The proposed 'Core Business Activity Description' index consistently delivered the highest 5-year cumulative returns among all benchmarks, showcasing the superior ability of the DKIP framework to identify and track high-growth opportunities in the New Power System sector.
| Feature | DKIP (Proposed AI System) | Traditional Methods |
|---|---|---|
| Data Processing |
|
|
| Semantic Understanding |
|
|
| Adaptability |
|
|
| Accuracy & Hallucination |
|
|
| Output |
|
|
| Investment Value |
|
|
Case Study: New Power System Stock Index (Chinese A-share)
Challenge: Developing a robust financial index for the rapidly evolving 'New Power System' sector, characterized by complex, unstructured policy data, fragmented industry boundaries, and the need for dynamic semantic mapping to ensure investment relevance and replicability.
Solution: Implemented the Dynamic Knowledge Injection Pipeline (DKIP), leveraging a Hierarchical Retrieval-Augmented Generation (H-RAG) architecture integrated with a Neo4j knowledge graph. This AI system enabled intelligent identification and dynamic updating of business boundaries by anchoring semantic inferences within strict taxonomies.
Outcome: The resulting 'New Power System Stock Index' significantly outperformed traditional benchmarks in cumulative returns over five years. It provides investors with a transparent and efficient tool to observe and invest in the energy transition sector, demonstrating how advanced AI can create precise, policy-aligned financial instruments for dynamic markets.
Calculate Your Potential AI ROI
See how our enterprise AI solutions can deliver tangible efficiency gains and cost savings for your organization.
Estimated Annual Savings
$0
Productive Hours Reclaimed Annually
0
Your AI Implementation Roadmap
A structured approach to integrating advanced AI, from initial assessment to continuous optimization.
Discovery & AI Blueprint
Comprehensive assessment of your current data landscape, identification of key challenges, and definition of AI objectives. Design the optimal DKIP architecture tailored to your enterprise needs.
Knowledge Graph & H-RAG Development
Building the Neo4j knowledge graph, implementing LLM extractors for policy documents, and developing the Hierarchical RAG for robust semantic understanding and query processing.
Index Prototyping & Validation
Construct initial power system stock indices, perform rigorous backtesting against historical data, and refine semantic mapping and weighting schemes based on performance analytics.
Deployment & Integration
Deploy the DKIP into your production environment, integrate with existing financial systems, and establish monitoring frameworks for performance and compliance. Training for your team.
Continuous Optimization & Expansion
Ongoing fine-tuning of LLM models (e.g., QLORA), exploration of advanced features like real-time sentiment analysis, and expanding the framework to new data sources and financial products.
Ready to Transform Your Financial Data?
Connect with our AI specialists to explore how a custom DKIP solution can revolutionize your enterprise's data analysis and decision-making.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat