Quant Orchestration (CONCEPT:AU-KG.research.research-pipeline-runner)¶

Overview¶

The KG-Native Quant Orchestration layer transforms agent-utilities into a master quantitative trading framework. It introduces a multi-agent debate architecture (Bull vs Bear) for consensus-driven trading, regime-aware signal fusion, and complete unified market data abstraction. All state, lifecycle, and provenance of hypotheses and backtests are persisted as first-class nodes in the Knowledge Graph.

Key Capabilities¶

1. TradingAgents Debate Architecture¶

A structured peer-review mechanism for trading hypotheses inspired by the TradingAgents paper. - Debate Engine: Orchestrates a multi-round debate between specialized SwarmRole agents (e.g., BULL_RESEARCHER and BEAR_RESEARCHER). - Risk Veto: After the debate rounds, a Risk Manager evaluates the arguments' confidences and issues a final RiskVeto approving or rejecting the trade, complete with position sizing limits. - Research Autopilot Integration: Hypotheses generated by the autopilot are automatically run through a Debate peer review before running computationally expensive backtests.

2. Unified Market Data Abstraction¶

An extensible protocol-based data provider system with automatic fallback chains. - Normalizes OHLCV data across different asset classes and geographies. - Providers Include: - YFinanceProvider: Global equities, ETFs, indices. - AKShareProvider: Chinese equities (A-shares) and index data. - CryptoProvider: Global cryptocurrency exchange data via CCXT (e.g., Binance).

3. Bayesian Signal Fusion & Regime Detection¶

Regime Detector: Uses moving averages and volatility heuristics (extensible to HMMs) to classify the market state (e.g., bull_market, high_volatility).
Bayesian Fusion: Combines discrete signals from multiple agents (fundamental, technical, sentiment) into a unified directional conviction using Bayesian inference based on the historical accuracy of the sources.

Knowledge Graph Ontology Integration (FIBO-Aligned)¶

The framework registers multiple new OntologySchema types into the core KG engine: - Schema:TradingHypothesis - Schema:BacktestResult - Schema:TradingStrategy - Schema:MarketRegime - Schema:AlphaFactor - Schema:TradingSignal - Schema:Order - Schema:Portfolio

Implementation Details¶

Source Code:
agent_utilities/domains/finance/quant_ontology.py
agent_utilities/domains/finance/market_data.py
agent_utilities/domains/finance/debate_engine.py
agent_utilities/domains/finance/trading_swarm.py
agent_utilities/domains/finance/research_autopilot.py
agent_utilities/domains/finance/signal_fusion.py
agent_utilities/domains/finance/regime_detector.py
agent_utilities/domains/finance/quant_mcp_tools.py
Pillar: KG (Pillar 2) & ECO (Pillar 4 MCP Tools)

CONCEPT:AU-KG.research.research-pipeline-runner - Domain: Finance (Base Layer)
CONCEPT:AU-KG.query.object-graph-mapper - Active Knowledge Graph