kairusama/ris-one

FDIC Research Information System (RIS)

Curated Parquet snapshots of the FDIC Research Information System (RIS) — the public bulk dataset distributed by the FDIC through its FOIA page.

What is RIS?

The FDIC Research Information System is a relational database of bank structure and financial data covering all FDIC-insured institutions. It is the primary source for:

• Bank structure data — institution name, charter type, location, RSSD ID
• Call report data — financial statements from regulatory filings
• Financial ratios — performance metrics derived from call reports
• Failure/assistance data — banks that failed or received FDIC assistance
• Structural changes — mergers, acquisitions, charter conversions

The FDIC publishes RIS as quarterly CSV snapshots covering 2006-Q1 through the present.

Quickstart

from datasets import load_dataset

# Load the RAT table (financial ratios, all quarters)
ds = load_dataset("kairusama/ris-one", "RAT", split="train")

# Stream a large table without downloading everything
ds = load_dataset("kairusama/ris-one", "RAT", split="train", streaming=True)

# Filter to a specific quarter
ds = load_dataset("kairusama/ris-one", "RAT", split="train")
q1_2024 = ds.filter(lambda x: x["report_date"] == "2024-03-31")

# Use with pandas directly
import pandas as pd
df = pd.read_parquet("hf://datasets/kairusama/ris-one/data/RAT/RAT2403.parquet")

Dataset Layout

data/
  RAT/                    # Financial ratios (viewer-safe, <1000 cols)
    RAT0603.parquet       # 2006-Q1
    RAT0606.parquet       # 2006-Q2
    ...
    RAT2512.parquet       # 2025-Q4
  STRU/                   # Bank structure data
    STRU0603.parquet
    ...
  MERG/                   # Mergers and structural changes
    MERG0603.parquet
    ...
  CDI/                    # Consolidated financial data (>1000 cols)
    CDI0603.parquet
    ...
  FTS/                    # Financial time series (>1000 cols)
    FTS0603.parquet
    ...
metadata/
  ris_periods.json        # List of available report dates
  ris_run_log_*.json      # ETL execution records (append-only)

Each table is a named config (e.g., "RAT", "STRU"). Each parquet file contains one quarter's data with a report_date column (ISO date string like "2024-03-31"). Files are named <TABLE><YYMM>.parquet.

Tables (Configs)

Config	Description	Viewer
RAT	Financial ratios derived from call reports	OK (<1000 cols)
STRU	Bank structure data (name, charter, location, RSSD)	OK
MERG	Mergers and structural changes	OK
CDI	Consolidated financial data	TooManyColumns (>1000)
FTS	Financial time series / call report metrics	TooManyColumns (>1000)

Coverage

Date Range	Source ZIP
2006-Q1 – 2015-Q4	ris0603-ris1512-csv.zip
2016-Q1 – 2023-Q4	ris1603-ris2312-csv.zip
2024-Q1 – 2025-Q4	ris2403-ris2512-csv.zip

Source

• FDIC FOIA RIS page: https://www.fdic.gov/foia/ris.html
• Data downloads: Direct HTTP (no authentication required)
• Update frequency: Quarterly (the FDIC publishes new data ~45 days after quarter-end)

Reproducibility

Each ETL run produces a metadata/ris_run_log_<timestamp>.json recording:

• Source ZIP URLs
• Tables and report dates processed
• Row/column counts per file
• Any conversion failures
• Commit SHA in the dataset repo

Known Limitations

• No incremental updates: Each ETL run re-processes all available data
• CSV quirks: FDIC CSVs use comma-separated thousands (handled automatically)
• Schema drift: FDIC may add/remove columns across quarters
• Wide tables: Some tables (e.g., CDI, FTS) exceed HF Data Studio's 1000-column viewer cap

License

CC0-1.0 — see LICENSE.

The underlying FDIC RIS data is a U.S. government work and is not subject to copyright protection.