← Back to list
report2026-04-135 min read

Trading Agents (Tauric Research) — Detailed Report

pythonclaude-apibloombergequity-researchmulti-agent

Project Overview

Project Name: Trading Agents (Tauric Research)

Purpose: Build a multi-agent institutional investment analysis system that mirrors the workflow of a full equity research team, integrating Bloomberg terminal data to produce institutional-grade research reports with buy/sell recommendations.

Architecture: 15 specialized AI agents coordinated through an orchestration layer

Data Integration: 26 Bloomberg data codes for real-time market data, financials, and consensus estimates

Output: Institutional-grade equity research reports and IC (Investment Committee) memos for 6 equities

System Architecture

15-Agent System Design

The system is architected around 15 specialized agents, each responsible for a distinct analytical domain:

Data Collection Agents:

  • Bloomberg data retrieval agent — pulls financial statements, consensus estimates, price history using 26 Bloomberg field codes
  • Market data aggregation agent — normalizes and structures raw Bloomberg data
  • News and sentiment collection agent — gathers relevant news flow and market sentiment

Fundamental Analysis Agents:

  • Financial statement analysis agent — income statement, balance sheet, cash flow decomposition
  • Valuation agent — DCF, comparable company analysis, precedent transactions
  • Earnings quality agent — accrual analysis, revenue recognition patterns
  • Growth trajectory agent — revenue drivers, margin expansion potential

Technical Analysis Agents:

  • Price action agent — support/resistance, trend analysis, chart patterns
  • Volume and flow agent — institutional flow, options activity, short interest
  • Momentum agent — relative strength, sector rotation signals

Sentiment and Macro Agents:

  • News sentiment agent — NLP-based sentiment scoring of news articles
  • Social media sentiment agent — retail investor sentiment tracking
  • Macro overlay agent — interest rate impact, sector cycle positioning

Synthesis Agents:

  • Report compiler agent — assembles all agent outputs into coherent research reports
  • Risk assessment agent — identifies key risks, downside scenarios, and position sizing recommendations

Bloomberg Integration

26 Bloomberg data codes integrated for comprehensive coverage:

  • Price and Volume: PX_LAST, PX_VOLUME, PX_HIGH, PX_LOW, VWAP
  • Fundamentals: IS_COMP_EPS, BEST_EPS, BEST_SALES, EBITDA, NET_INCOME
  • Valuation: PE_RATIO, PB_RATIO, EV_TO_EBITDA, DIVIDEND_YIELD
  • Consensus: BEST_TARGET_PRICE, BEST_ANALYST_RECS, BUY/HOLD/SELL counts
  • Risk: VOLATILITY_30D, BETA_ADJ_OVERRIDABLE, SHORT_INT_RATIO
  • Financials: TOT_DEBT_TO_EQUITY, RETURN_ON_EQUITY, FREE_CASH_FLOW

Research Output

Equities Analyzed

The system produced comprehensive research reports for 6 equities spanning US-listed Chinese ADRs and US tech/growth stocks:

  1. BABA (Alibaba Group) — Full equity research report (Traditional Chinese)
  2. TSLA (Tesla) — Full equity research report (Traditional Chinese)
  3. TIGR (UP Fintech / Tiger Brokers) — Full equity research report (Traditional Chinese)
  4. NVDA (NVIDIA) — IC Memo format (Traditional Chinese)
  5. FUTU (Futu Holdings) — IC Memo format v2 (Traditional Chinese)
  6. CRCL (Circle Internet Group) — IC Memo format v3 (Traditional Chinese)

Report Formats

Full Equity Research Reports (~30-50 pages):

  • Executive summary with clear buy/sell/hold recommendation
  • Company overview and business model analysis
  • Industry and competitive landscape
  • Financial analysis with historical trends and projections
  • Valuation analysis (DCF + comps)
  • Technical analysis with chart annotations
  • Risk factors and bear case scenarios
  • Price target derivation and sensitivity analysis

IC Memos (~10-15 pages):

  • Concise investment thesis
  • Key catalysts and timeline
  • Valuation summary
  • Risk/reward assessment
  • Position sizing recommendation
  • Key metrics dashboard

Development Timeline

Phase 1 — Agent Architecture and Bloomberg API Integration

  • Designed the 15-agent coordination framework
  • Built Bloomberg data retrieval pipeline supporting 26 field codes
  • Established data normalization and caching layer
  • Created agent communication protocol for passing analysis between agents

Phase 2 — Fundamental and Technical Analysis Agents

  • Implemented financial statement decomposition logic
  • Built valuation models (DCF with multiple scenario assumptions, trading comps)
  • Developed technical analysis modules (trend detection, support/resistance, momentum indicators)
  • Integrated sentiment analysis from news and social media sources

Phase 3 — Report Synthesis and Recommendation Engine

  • Built the report compiler agent that assembles all analytical outputs
  • Developed the recommendation engine weighing fundamental, technical, and sentiment signals
  • Created PDF generation pipeline for institutional-quality output
  • Implemented Traditional Chinese localization for all reports
  • Iterative quality improvement through multiple report generations

Technical Highlights

Multi-Agent Coordination

The 15-agent system coordinates through a structured pipeline where each agent operates independently on its domain, then feeds results to synthesis agents. This mirrors how an institutional equity research team operates — sector analysts, quant analysts, and portfolio strategists each contribute their expertise before a chief strategist assembles the final view.

Bloomberg Data Pipeline

The Bloomberg integration handles real-time and historical data across 26 field codes, with automatic data validation, missing data handling, and caching to minimize API calls. The system normalizes data across different fiscal year-ends and reporting currencies.

Institutional-Grade Output

Reports are formatted to match the standards of major sell-side research houses — consistent formatting, proper citation of data sources, clear methodology disclosure, and professional chart/table presentation. All reports generated in Traditional Chinese to serve the fund's primary audience.

Business Impact

Delivered a multi-agent research system that mirrors the workflow of a full equity research team, dramatically reducing analysis turnaround time. What would take a junior analyst 2-3 weeks to produce (comprehensive equity initiation report) can now be generated in hours, with consistent methodology and comprehensive data coverage.

The system demonstrates that AI agents can handle the full research workflow — from data collection through analysis to report generation — while maintaining the analytical rigor expected in institutional investment research.

Demo Reports

The following research outputs are available as PDF demonstrations:

  • BABA Full Equity Research Report (Traditional Chinese)
  • TSLA Full Equity Research Report (Traditional Chinese)
  • TIGR Full Equity Research Report (Traditional Chinese)
  • NVDA IC Memo (Traditional Chinese)
  • FUTU IC Memo v2 (Traditional Chinese)
  • CRCL IC Memo v3 (Traditional Chinese)

Report Date: April 13, 2026