13f fund search

13F Fund Search Skill

Overview

This skill enables searching the SEC 13F COVERPAGE data to locate specific hedge funds by name, returning their accession numbers needed for detailed analysis.

When to Use

Use this skill whenever you need to:

• Find a fund's accession number by its name
• Locate multiple funds (e.g., Q2 and Q3 versions of the same fund)
• Verify fund names and get their correct identifiers

Data Location

The 13F filings are stored in:

• /root/2025-q2/ - Q2 2025 13F filings
• /root/2025-q3/ - Q3 2025 13F filings

Each quarter folder contains subdirectories with COVERPAGE files containing fund metadata.

Search Process

Step 1: Explore Directory Structure

ls -la /root/2025-q2/
ls -la /root/2025-q3/

Look for CSV or JSON files containing COVERPAGE information. These typically contain:

• Fund name
• Accession number
• Filing date
• CIK (Central Index Key)

Step 2: Fuzzy Search Implementation

For searching funds by name, use Python with fuzzy matching:

import os
import pandas as pd
from fuzzywuzzy import fuzz
from fuzzywuzzy import process

def search_fund_in_quarter(quarter_path, search_term, threshold=80):
    """
    Search for fund by fuzzy name matching in a specific quarter.

    Args:
        quarter_path: Path to quarter data (e.g., '/root/2025-q2/')
        search_term: Fund name to search (e.g., 'Renaissance Technologies')
        threshold: Fuzzy match threshold (0-100)

    Returns:
        Dictionary with best matches containing fund name and accession_number
    """
    # Find COVERPAGE files
    results = []
    for root, dirs, files in os.walk(quarter_path):
        for file in files:
            if 'COVERPAGE' in file and (file.endswith('.csv') or file.endswith('.json')):
                file_path = os.path.join(root, file)
                try:
                    if file.endswith('.csv'):
                        df = pd.read_csv(file_path)
                    else:
                        df = pd.read_json(file_path)

                    # Look for name column (may be named differently)
                    name_col = None
                    for col in df.columns:
                        if 'name' in col.lower() or 'fund' in col.lower():
                            name_col = col
                            break

                    if name_col is None:
                        continue

                    # Fuzzy search
                    for idx, row in df.iterrows():
                        fund_name = str(row[name_col])
                        score = fuzz.token_sort_ratio(search_term.lower(), fund_name.lower())

                        if score >= threshold:
                            accession = row.get('accession_number') or row.get('ACCESSION') or row.get('accession')
                            results.append({
                                'fund_name': fund_name,
                                'accession_number': accession,
                                'match_score': score
                            })
                except Exception as e:
                    continue

    # Sort by match score
    results.sort(key=lambda x: x['match_score'], reverse=True)
    return results

# Usage
q2_results = search_fund_in_quarter('/root/2025-q2/', 'Renaissance Technologies')
q3_results = search_fund_in_quarter('/root/2025-q3/', 'Renaissance Technologies')

if q2_results:
    print(f"Q2 Match: {q2_results[0]}")
if q3_results:
    print(f"Q3 Match: {q3_results[0]}")

Output Format

Returns a list of matches sorted by fuzzy match score:

{
    "fund_name": "Renaissance Technologies",
    "accession_number": "0001234567-25-000123",
    "match_score": 95
}

Use the accession_number with the 13f-fund-analysis skill.

Common Fund Names to Search

• "Renaissance Technologies" (Jim Simons)
• "Berkshire Hathaway" (Warren Buffett)
• Full names or abbreviated versions both work

Tips

• Use token_sort_ratio from fuzzywuzzy for name variations
• Set threshold around 80-90 for good precision
• Multiple results indicate possible matches—use the highest score
• If searching for Q2 and Q3, run separately as accession numbers will differ