Information Overload in Equity Research is Real
A common challenge when valuing equities is: processing vast amounts of information from multiple sources to make informed investment decisions.
Traditional approaches involve:
- Gathering financial data from various platforms.
- Reading through multiple reports, news and other articles.
- Building and maintaining complex spreadsheet models.
- Synthesizing this information into something actionable.
Holding and managing all of these inputs all at once often starts looking like this:
It’s no secret, AI stands as the most viable solution, making it easy to process large datasets efficiently.
However, most large language models (LLMs) available to the public today still can’t perform detailed deep dives into real-time stock data with much accuracy.
- ChatGPT and Claude have context date cutoffs.
- Perplexity is great for real-time information but limited when it comes to analytical tasks.
- ChatGPT Search is still underwhelming.
For detailed research and equity analysis, we need something more precise and able to work with structured datasets well.
In fact, what if we could combine it all? What if we could leverage AI agents that can navigate the news, Google searches, finacial datasets and coding tasks all in one system?
A superpowered, and always on equity analyst.
Enter the AI Agent Solution
To solve this issue, I developed a system that orchestrates multiple specialized AI agents, each handling specific aspects of stock analysis.
The Agent Team
- Sentiment Analysis Agent (GPT-4o)
- Processes news and market sentiment
- Uses Google Search to gather recent developments
- Provides sentiment scoring and trend analysis
- Financial Analysis Agents (Claude 3.5)
- Different agents for basic stock data and historical data
- Analyzes company financials and metrics
- Performs valuation calculations (DCF, public comparables, fundamental analysis)
- Evaluates key performance indicators
- Quantitative Analysis Agent (Claude 3.5 Sonnet or Haiku)
- Executes Python code for technical analysis
- Processes large datasets and other Agents’ outputs
- Generates visualizations and statistical insights
- Executive / Portfolio Manager Agent (Claude 3.5 Sonnet or Haiku)
- Acts like a portfolio manager designed to synthesize data, optional routing.
- Aggregates all the information collected by other tool-wielding agents and provides a BUY, SELL or HOLD recommendation
Phidata: The Agent Builder
Phidata is an AI Agent framework that enables developers to:
- Build Agents with memory, knowledge and external connections
- Build a team of agents that can work together
- Monitor, evaluate and optimize Agents
They also offer an intutive and user-friendly Agent UI where users can test agents in playground settings. Check it out here
Phidata allows us to integrate several powerful tools out of the box:
- Yahoo Finance API for real-time price data and historical financials
- Google Search for news and sentiment analysis
- Python tools for AI-directed code execution and quantitative analysis (use with caution)
- Custom Quant functions for data processing and visualization (Optional)
- Note: Code Execution Agents need tight controls and stringent prompt engineering
- Code Agents can run into several errors such as recursive functions or saving and reading files
Getting Started
Want to try it yourself? The complete code is available in the Google Colab notebook.
You'll need:
- API keys for OpenAI and Anthropic (We use different models for different agents but you could opt for a more unified workflow)
- Phidata framework installed, check out the documentation here
- Some basic Python knowledge to follow along
Required Packages
!pip install phidata openai anthropic yfinance googlesearch-python pycountry -qInstantiate Libraries & API Keys
## Libraries
from phi.agent import Agent
from phi.model.openai import OpenAIChat
from phi.model.anthropic import Claude
from phi.tools.yfinance import YFinanceTools
from phi.tools.googlesearch import GoogleSearch
## API Keys
import requests
from google.colab import userdata
OPENAI_API_KEY = userdata.get('OPENAI_API_KEY')
ANTHROPIC_API_KEY = userdata.get('ANTHROPIC_API_KEY')
import os
os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY
os.environ["ANTHROPIC_API_KEY"] = ANTHROPIC_API_KEYDefine an Agent - Sentiment Analysis Example
# Sentiment Analysis Agent
sentiment_agent = Agent(
name="Sentiment Analysis Agent",
role="Search and interpret news articles",
model=OpenAIChat(id="gpt-4o"),
## define the tools for the Agent's use
tools=[GoogleSearch(), YFinanceTools(company_news=True)],
instructions=[
"Find relevant news articles for each company and critically analyze the news sentiment.",
"Provide sentiment scores from 1 (negative) to 10 (positive) with reasoning and sources."
"Cite your sources. Be specific, crtical and provide relevant links."
],
show_tool_calls=True,
markdown=True,
)Define an Agent - Basic Stock Info Example
# Financial Analyst Agent
basic_stock_agent = Agent(
name="Basic Financial Data Agent",
role="Retrieve basic company financial data and expertly interpret trends and data using a rigourously analytical approach",
model=Claude(id="claude-3-5-sonnet-latest"), ## swapped claude for gpt-4o
## define the tools for the Agent's use
tools=[YFinanceTools(stock_price=True, company_info=True, stock_fundamentals=True, analyst_recommendations=True)],
instructions=[
"Retrieve stock prices, analyst recommendations, and key summary financial data.",
"Focus on company funamentals and trends, presenting the data in tables with key insights."
],
show_tool_calls=True,
markdown=True,
)Users can choose to add more Agents based on their unique requirements, including visualization Agents and other formats
Build Your Agent Team
With just few lines of code, we can set up a powerful multi-agent system with specific roles that collaborate to analyze a stock.
The agent_team acts as a conductor, picking the right agent work flow and ensuring that each agent contributes its insights and that the final output is well-structured, data-driven, and easily understandable
agent_team = Agent(
model=Claude(id="claude-3-5-sonnet-latest"),
team=[sentiment_agent, basic_stock_agent], ## Add more Agents here if desired.
instructions=[
"Combine the expertise of all the agents part of your team to provide a cohesive, highly analytical and well-supported response",
"Always include references and dates for all data points and sources. Your context is current as of [insert date here]"
"Present data in structured tables for clarity, ensuring all conclusions are data-driven."
"Explain the methodlogy and always determine a BUY, SELL, or HOLD recommendation."
],
show_tool_calls=True,
markdown=True,
)Start Chatting with Your Agent Team 🤖
You’re now ready to start querying your Agent team! Here’s what a sample prompt and output should look like:
Sample Prompt:
agent_team.print_response(
"Perform a public comparable companies analysis for CALM Cal Maine Foods. Identify it's main competitors and market positioning. Assess each comp in a structrued table format",
stream=True
)Sample Output:
┏━ Response (59.0s) ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ ┃
┃ I'll help you perform a comprehensive comparative analysis for Cal-Maine Foods (CALM). I'll break this down ┃
┃ systematically by using our agents to gather the necessary information. ┃
┃ ┃
┃ First, I'll transfer tasks to our agents to collect the required data: ┃
┃ ┃
┃ • Running: transfer_task_to_basic_financial_data_agent(task_description=..., expected_output=A list of 3-4 ┃
┃ direct public competitors with ticker symbols, additional_information=Focus on companies in egg production, ┃
┃ processing, and distribution) ┃
┃ ┃
┃ Now, I'll gather some additional market positioning information: ┃
┃ ┃
┃ • Running: transfer_task_to_comprehensive_financial_data_agent(task_description=..., expected_output=Detailed ┃
┃ analysis of CALM's market share, unique positioning, and competitive strengths, additional_information=Focus ┃
┃ on production capacity, distribution network, and market differentiation) ┃
┃ ┃
┃ Let me compile a comprehensive comparative analysis in a structured format: ┃
┃ ┃
┃ ┃
┃ Public Comparable Companies Analysis for Cal-Maine Foods (CALM) ┃
┃ ┃
┃ Comparative Company Overview ┃
┃ ┃
┃ ┃
┃ Company Symbol Market Cap (USD) Key Focus Egg Production Emphasis ┃
┃ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ┃
┃ Cal-Maine Foods CALM $4.96B Specialized Egg Producer Primary Business ┃
┃ Post Holdings POST $6.66B Packaged Foods Secondary Business ┃
┃ Tyson Foods TSN $20.65B Meat & Food Products Minor Business Line ┃
┃ Vital Farms VITL $1.64B Pasture-Raised Eggs Specialty Segment ┃
┃ Seaboard Corporation SEB $2.34B Diversified Conglomerate Minor Business Line ┃
┃ ┃
┃ ┃
┃ Comparative Financial Metrics ┃
┃ ┃
┃ ┃
┃ Metric CALM POST TSN VITL SEB ┃
┃ ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ┃
┃ P/E Ratio 11.59 16.11 13.80 31.31 N/A ┃
┃ EPS $8.73 $5.64 $2.25 $1.12 -$5.30 ┃
┃ Revenue Growth (YoY) 71.1% Varies Moderate High Varies ┃
┃ Gross Margin 28.02% 15-20% 12-15% 25-30% Varies ┃
┃ ┃
┃ ┃
┃ Cal-Maine Foods' Competitive Positioning ┃
┃ ┃
┃ Unique Strengths ┃
┃ ┃
┃ 1 Market Leadership in Egg Production ┃
┃ 2 Diversified Product Portfolio ┃
┃ 3 Strong Regional Distribution Network ┃
┃ 4 Focus on Specialty Egg Segments ┃
┃ 5 Robust Financial Performance ┃
┃ ┃
┃ Key Differentiators ┃
┃ ┃
┃ • Brands: Egg-Land's Best, Land O' Lakes, Farmhouse Eggs ┃
┃ • Product Range: Conventional, organic, free-range, pasture-raised eggs ┃
┃ • Geographic Coverage: Southwestern, Southeastern, Midwestern, Northeastern, Mid-Atlantic regions ┃
┃ ┃
┃ Market Recommendations ┃
┃ ┃
┃ Comparative Assessment: ┃
┃ ┃
┃ • Cal-Maine Foods appears to be the most focused and pure-play egg production company ┃
┃ • Strong financial metrics compared to diversified competitors ┃
┃ • Potential for continued growth in specialty egg market segments ┃
┃ ┃
┃ Recommendation: STRONG BUY ┃
┃ ┃
┃ • Solid fundamentals ┃
┃ • Market leadership ┃
┃ • Adaptability to consumer trends ┃
┃ • Attractive valuation compared to peers ┃
┃ ┃
┃ Note: This analysis is based on current market data and is subject to change. Always conduct personal due ┃
┃ diligence before making investment decisions. ┃
┃ ┃
┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛And that’s it. You’re on your way to build your personal equity analyst to help you supercharge your investment process.
Real-World Results
Testing the system on various stocks has shown promising results:
- Cost Reduction: May eliminate the need for multiple financial database subscriptions by aggregating data automatically
- Research Efficiency: Cuts equity research time and can handle multiple companies simultaneously, enabling broader coverage without additional headcount
- Competitive Analysis: Rapidly generates peer comparisons and industry positioning
- Financial Analysis: Reduces reliance on junior analysts for data gathering, allowing them to focus on higher-value analysis
- Multiple Model Support: Plug in Claude, GPTs, Groq, HF models or any other LLM provider that best suits your needs
⚠️ Limitations
While powerful, the system has some limitations:
- Limited ability to analyze data beyond LLM training cutoff date
- Reliance on external API data quality and availability
- Higher computational costs with additional tools and agents
- Need for precise prompt engineering
- Complex decisions still require human oversight
🛠️ Potential Enhancements
Future enhancements could include:
- Additional data sources and APIs
- More sophisticated inter-agent communication
- Enhanced visualization and code execution capabilities
- Machine learning models for pattern recognition
Conclusion
This multi-agent system is a significant step forward in automated financial analysis and it's available to everyone for free. By combining the strengths of different AI models and tools, we can create more robust and comprehensive analysis workflows.
Remember: While AI can greatly enhance our analytical capabilities, it should complement, not replace, human judgment in investment decisions.
Disclaimer: This post is for educational purposes only and should not be considered financial advice. Always conduct your own research before making investment decisions.