quartercall

Read QuarterCall Data — the NCUA Form 5300 Call Reports, republished as Parquet at data.quartercall.com — into pandas. No account or API key is needed.

pip install quartercall

import quartercall

quartercall.quarters()

quarters() returns one row per published quarter, newest first:

No quarter is final: NCUA accepts an amendment to a cycle of any age, and a dormant quarter becomes open again when one arrives. A recent quarter is published while many credit unions have yet to file.

Read a quarter

import quartercall as qc

# Assets and members for one institution, from its latest filing for the quarter.
accounts = qc.accounts("2026Q2", codes=["A010", "A083"], cuid=5163)

# The ten largest credit unions by total assets.
largest = qc.accounts("2026Q2", codes="A010").nlargest(10, "account_value")

profiles = qc.profiles(
    "2026Q2",
    columns=["quarter", "cu_number", "submission_id", "cuid", "credit_union_name", "state"],
)

# Non-numeric values, such as the filed net-worth classification.
text = qc.accounts_text("2026Q2", codes="A700")

Each reader returns a pandas DataFrame with the published column names and Arrow-backed types, as does quarters(), so their frames merge on quarter and cycle_date. A missing value is <NA>, never zero. account_value is a float64, which arithmetic, ranking, plotting and export expect; exact=True keeps the published decimal128(24, 8), whose values are Python Decimal objects, for reconciling to the cent. Nullable integers stay integers. Dates and timestamps retain their Parquet types, and NCUA profile strings stay strings, including dates, leading zeros and Yes/No values.

Filters combine with AND, and an empty sequence selects no rows. Filtering works even when its column is absent from columns. A missing account or value is never filled with zero. A numeric NULL means published as text only when a matching accounts_text row exists. A published quarter without a text file returns a typed empty DataFrame.

Use cuid to follow an institution across charter conversions; cu_number identifies a charter. The text table has no cuid, so filter it by charter. Profile names and profile_charter_number reflect when NCUA supplied the profile and do not necessarily describe the historical quarter.

Some filings have no profile. Keep them with a left join on the filing's keys:

named_accounts = accounts.merge(
    profiles,
    on=["quarter", "cu_number", "submission_id"],
    how="left",
    suffixes=("", "_profile"),
    validate="many_to_one",
)

Each call reads its own manifest, so separate calls may see different amendments. Including submission_id in the join prevents attaching another filing's profile. For a text join, use those same keys plus account_code.

Downloads and caching

Readers follow paths in the manifest and download the whole quarter file before applying filters and column selection. Selection reduces the returned data, not the network download. They check the file's SHA-256 and byte count before reading Parquet. If a publication changes mid-read, they fetch a fresh manifest and retry, at most three attempts total.

assets = qc.accounts("2026Q2", codes="A010", cache_dir=".quartercall-cache")
# Reuses the verified quarter file if its manifest digest has not changed.
members = qc.accounts("2026Q2", codes="A083", cache_dir=".quartercall-cache")

The manifest is fetched on every call, even for dormant quarters. Cached files are checked before reuse and retained by digest; old versions are not automatically removed. The cache can be deleted when no reader is using it. It does not provide offline access.

Schema 1.2 and later require digests and byte counts. Older 1.0/1.1 manifests without them remain readable, but those files cannot be verified against the manifest and are not cached.

• DataUnavailableError: the requested quarter or table has not been published.
• DataIntegrityError: invalid digest metadata, an incomplete download, or a mismatch that persists after retries. Files are replaced before the manifest that names them is published, so a mismatch most often means the quarter is being republished: try again in a few minutes.
• UnsupportedSchemaError: a major data schema version this package cannot read.
• ValueError / TypeError: invalid quarter, filters or column selection.
• OSError: a network or HTTP failure, as urllib.error.URLError (including HTTPError), or as TimeoutError when a response stalls. A missing file that the manifest says exists raises an error; it is never returned as an empty table.

Malformed Parquet raises the underlying PyArrow error. Filesystem errors from an explicitly requested cache are also propagated.

QuarterCall is a registered trademark of QuarterCall Holdings Inc. The code is licensed under the Apache License 2.0.