Wrds Data

wrds-pgdata.wharton.upenn.edu:9737:wrds:USERNAME:PASSWORD

If the user is working from a local checkout of this plugin and wants the helper script, it lives at:

plugins/wrds-data/skills/wrds-data/scripts/wrds_setup.py

Do NOT proceed with queries until --check passes.

Step 1: Identify the Dataset

If the user specifies a library/table, proceed directly. Otherwise, help them find it.

Browse libraries and tables

import wrds
db = wrds.Connection()

# List all accessible libraries
libs = db.list_libraries()

# List tables in a library
tables = db.list_tables(library='crsp')

# Describe a table (columns, types)
schema = db.describe_table(library='crsp', table='dsf')

# Approximate row count
count = db.get_row_count('crsp', 'dsf')

# Preview data
sample = db.get_table('crsp', 'dsf', rows=5)

For common datasets (CRSP, Compustat, IBES, etc.), consult the reference:

• Common datasets and query recipes — libraries, tables, standard filters, and variable glossary

Step 2: Build and Execute the Query

Use db.raw_sql() for all queries — it is the most flexible method.

data = db.raw_sql(
    "SELECT permno, date, ret FROM crsp.dsf WHERE date >= '2020-01-01'",
    date_cols=['date']
)

Key parameters

Parameterized queries (prevent SQL injection)

params = {'tickers': ('AAPL', 'MSFT'), 'start': '2023-01-01'}
data = db.raw_sql("""
    SELECT a.permno, a.date, a.ret, b.ticker
    FROM crsp.dsf a
    JOIN crsp.stocknames b ON a.permno = b.permno
        AND a.date >= b.namedt AND a.date <= b.nameendt
    WHERE b.ticker IN %(tickers)s
        AND a.date >= %(start)s
""", params=params, date_cols=['date'])

Large downloads

For datasets exceeding ~1M rows, use chunked iteration:

chunks = db.raw_sql(
    "SELECT * FROM crsp.dsf WHERE date >= '2000-01-01'",
    chunksize=500000, return_iter=True
)
for i, chunk in enumerate(chunks):
    chunk.to_parquet(f'data/crsp_dsf_{i}.parquet')

Or download in one shot and save:

data = db.raw_sql("...", date_cols=['date'])
data.to_parquet('data/output.parquet')
# or
data.to_csv('data/output.csv', index=False)

Step 3: Save Output

Default conventions:

• Save to Data/ directory in the project root (create if needed)
• Prefer Parquet for large datasets (faster, smaller, typed)
• Use CSV if the user requests it or for small datasets
• Name files descriptively: crsp_daily_2020_2023.parquet, compustat_annual.csv

Always close the connection when done:

db.close()

Error Handling

• NotSubscribedError — user lacks subscription to this library. Inform them to request access via their institution's WRDS coordinator.
• SchemaNotFoundError — library name is wrong. Use db.list_libraries() to find the correct name.
• Timeout on large queries — add date filters or use chunked downloads.

Notes

• WRDS uses PostgreSQL under the hood; any valid PostgreSQL SQL works in raw_sql().
• Always include date filters to avoid accidentally downloading entire multi-decade datasets.
• The wrds package connects to wrds-pgdata.wharton.upenn.edu:9737 over SSL.
• Credentials in ~/.pgpass are never exposed to the LLM — the wrds package reads them directly.

Resources

scripts/

• wrds_setup.py — Check environment and interactively configure WRDS credentials

references/

• common_datasets.md — Common WRDS libraries, tables, query recipes, and variable glossary