Run2 13f data analysis

Skill cxcscmu/SkillLearnBench/skills/b2-self-feedback-claude-opus-4-6/financial-analysis/run2_13f-data-analysis

[COLM'26] SkillLearnBench is the first benchmark for evaluating continual learning methods that automatically generate agent skills.

Install

npx -y skills add cxcscmu/SkillLearnBench --skill run2_13f-data-analysis

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

What its author says it does

Copied from the file, not written here

Comprehensive guide for analyzing SEC 13-F quarterly filings data (TSV format) including fund lookup, AUM, holdings comparison, and stock investor analysis.

SKILL.md

3.0 KB, as published. Nobody here has run it

SEC 13-F Data Analysis (Improved)

Data Files

Each quarter folder contains TSV files:

File	Purpose	Key Columns
COVERPAGE.tsv	Fund manager metadata	`ACCESSION_NUMBER`, `FILINGMANAGER_NAME`
INFOTABLE.tsv	Individual stock holdings (~300MB)	`ACCESSION_NUMBER`, `NAMEOFISSUER`, `CUSIP`, `VALUE` (x1000), `SSHPRNAMT` (shares)
SUMMARYPAGE.tsv	Aggregate stats per filing	`ACCESSION_NUMBER`, `TABLEENTRYTOTAL`, `TABLEVALUETOTAL` (x1000)

Important Notes

VALUE is in thousands of dollars — multiply by 1000 for actual dollar amounts
TABLEVALUETOTAL is also in thousands in SUMMARYPAGE
CUSIP casing: INFOTABLE may have mixed-case CUSIPs (e.g., 69608A108 vs 69608a108). Always normalize: df['CUSIP'] = df['CUSIP'].str.upper()
Multiple accession numbers: A single filer (e.g., Berkshire Hathaway) may have multiple filings. Use SUMMARYPAGE to identify the main one (largest TABLEVALUETOTAL).
Low memory: INFOTABLE is large. Use low_memory=False or specify dtype={'CUSIP': str}.

Workflow Patterns

1. Fuzzy Search for a Fund

from difflib import SequenceMatcher
cover = pd.read_csv('COVERPAGE.tsv', sep='\t', dtype=str)
cover['score'] = cover['FILINGMANAGER_NAME'].apply(
    lambda x: SequenceMatcher(None, search.lower(), str(x).lower()).ratio()
)
best = cover.sort_values('score', ascending=False).iloc[0]
accession = best['ACCESSION_NUMBER']

2. Get AUM (from SUMMARYPAGE)

summary = pd.read_csv('SUMMARYPAGE.tsv', sep='\t')
row = summary[summary['ACCESSION_NUMBER'] == accession]
aum_thousands = row['TABLEVALUETOTAL'].values[0]

3. Get Number of Holdings

Use TABLEENTRYTOTAL from SUMMARYPAGE or count unique CUSIPs from INFOTABLE (usually identical).

4. Compare Holdings Across Quarters

q2_h = info_q2[info_q2['ACCESSION_NUMBER']==acc_q2].groupby('CUSIP')['VALUE'].sum()
q3_h = info_q3[info_q3['ACCESSION_NUMBER']==acc_q3].groupby('CUSIP')['VALUE'].sum()
diff = q3_h.subtract(q2_h, fill_value=0).sort_values(ascending=False)
top5_increased = diff.head(5).index.tolist()  # CUSIPs

5. Find Top Investors for a Stock

# Find CUSIP by issuer name
matches = info[info['NAMEOFISSUER'].str.contains('NAME', case=False, na=False)]
cusip = matches['CUSIP'].str.upper().mode()[0]
# Aggregate by filing
holders = info[info['CUSIP'].str.upper()==cusip].groupby('ACCESSION_NUMBER')['VALUE'].sum()
top = holders.sort_values(ascending=False).head(3)
# Map to names via COVERPAGE

Disambiguation Tips

When multiple accession numbers match the same filer, the one with higher TABLEVALUETOTAL or more TABLEENTRYTOTAL is typically the main 13F-HR filing.
Some funds file separate confidential treatment requests — filter by ISCONFIDENTIALOMITTED='N' if needed.