Fund comparison

Skill cxcscmu/SkillLearnBench/skills/b4-skill-creator-claude-sonnet-4-6/financial-analysis/fund-comparison

Compare hedge fund holdings between two quarters to identify changes in investment positions. Use this skill when analyzing Q-over-Q changes in fund portfolios, finding top increased/decreased positions, or tracking specific fund activity across reporting periods. Triggers on questions about "change from Q2 to Q3", "increased investment", "top buys/sells", or any cross-quarter 13F comparison.From its SKILL.md

Install

npx -y skills add cxcscmu/SkillLearnBench --skill fund-comparison

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

3.6 KB, ~1.0k tokens by cl100k_base, as published. Nobody here has run it

Fund Quarter-over-Quarter Comparison Skill

Overview

Compare a fund's holdings between two quarters to find the biggest changes in investment positions by dollar value or share count.

Step-by-Step Workflow

Step 1: Get Accession Numbers for Both Quarters

import pandas as pd

def find_fund_accession(quarter_dir, search_term):
    cover = pd.read_csv(f'{quarter_dir}/COVERPAGE.tsv', sep='\t', dtype=str)
    mask = cover['FILINGMANAGER_NAME'].str.contains(search_term, case=False, na=False)
    results = cover[mask][['ACCESSION_NUMBER', 'FILINGMANAGER_NAME', 'ISAMENDMENT']]
    # Prefer non-amended filing, or latest amendment
    return results

q2_acc = find_fund_accession('/root/2025-q2', 'berkshire')
q3_acc = find_fund_accession('/root/2025-q3', 'berkshire')
print("Q2:", q2_acc)
print("Q3:", q3_acc)

Step 2: Load Holdings for Both Quarters

q2_info = pd.read_csv('/root/2025-q2/INFOTABLE.tsv', sep='\t', dtype=str)
q3_info = pd.read_csv('/root/2025-q3/INFOTABLE.tsv', sep='\t', dtype=str)

q2_holdings = q2_info[q2_info['ACCESSION_NUMBER'] == q2_accession][['CUSIP', 'NAMEOFISSUER', 'VALUE', 'SSHPRNAMT']].copy()
q3_holdings = q3_info[q3_info['ACCESSION_NUMBER'] == q3_accession][['CUSIP', 'NAMEOFISSUER', 'VALUE', 'SSHPRNAMT']].copy()

Step 3: Convert Numeric Columns

q2_holdings['VALUE'] = pd.to_numeric(q2_holdings['VALUE'], errors='coerce').fillna(0)
q3_holdings['VALUE'] = pd.to_numeric(q3_holdings['VALUE'], errors='coerce').fillna(0)
q2_holdings['SSHPRNAMT'] = pd.to_numeric(q2_holdings['SSHPRNAMT'], errors='coerce').fillna(0)
q3_holdings['SSHPRNAMT'] = pd.to_numeric(q3_holdings['SSHPRNAMT'], errors='coerce').fillna(0)

Step 4: Aggregate by CUSIP (handle duplicate rows per CUSIP)

q2_agg = q2_holdings.groupby('CUSIP').agg({'VALUE': 'sum', 'SSHPRNAMT': 'sum', 'NAMEOFISSUER': 'first'}).reset_index()
q3_agg = q3_holdings.groupby('CUSIP').agg({'VALUE': 'sum', 'SSHPRNAMT': 'sum', 'NAMEOFISSUER': 'first'}).reset_index()

Step 5: Merge and Calculate Changes

merged = pd.merge(
    q3_agg.rename(columns={'VALUE': 'v_q3', 'SSHPRNAMT': 's_q3'}),
    q2_agg.rename(columns={'VALUE': 'v_q2', 'SSHPRNAMT': 's_q2', 'NAMEOFISSUER': 'name_q2'}),
    on='CUSIP', how='outer'
).fillna(0)

# Use Q3 name where available, else Q2
merged['NAME'] = merged['NAMEOFISSUER'].where(merged['NAMEOFISSUER'] != 0, merged['name_q2'])
merged['value_change'] = merged['v_q3'] - merged['v_q2']
merged['share_change'] = merged['s_q3'] - merged['s_q2']

Step 6: Get Top 5 Increases

# By dollar value increase
top5 = merged.nlargest(5, 'value_change')[['CUSIP', 'NAME', 'v_q2', 'v_q3', 'value_change']]
print(top5)

# Answer: list of CUSIPs
cusips = top5['CUSIP'].tolist()
print("Top 5 CUSIPs:", cusips)

Notes

VALUE is in thousands of USD in INFOTABLE
Some holdings may appear in only one quarter (new buys or complete sells)
outer join ensures stocks held in only one quarter are included (value=0 for missing quarter)
When a fund has amendments, typically use the most recent (highest amendment number)
Berkshire Hathaway files under "BERKSHIRE HATHAWAY INC" — search for "berkshire"

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.