Data Structure

Understanding how FCA structures its Call Report files helps you work effectively with the output of process_data().

Files in an archive

Each quarterly zip contains pairs of files sharing a root name:

File pattern	Role
D_<ROOT>.TXT	Metadata - column names, types, decimal positions, and definitions in a fixed-width file
<ROOT>_Q<YYYYMM>_G<YYYYMMDD>.TXT	Data - headerless CSV values

For example, the September 2025 archive contains D_RCB.TXT (metadata) and RCB_Q202509_G20251112.TXT (data).

Note

Some metadata files in the RCI2* family carry an extra _<YEAR> suffix (D_RCI2B_2016.TXT, for instance) that is not present in the data filename. fcall handles this automatically.

The three scenarios

process_metadata_file() classifies each dataset into one of three scenarios based on the structure of the metadata file.

"single"

Plain CSV with no repeating column groups. Example: INST (institution information).

UNINUM  NAME  STATE  ...
12345   Acme  CA     ...
23456   Beta  TX     ...

"single_multiple"

Each row in the raw file contains data for \(N\) codes (e.g. investment types). The raw wide format is automatically pivoted by process_data() into one row per entity per code.

Before pivot (wide):

UNINUM  INV_CODE__1  AMOUNT__1  INV_CODE__2  AMOUNT__2  ...
12345   10           100.00     15           200.00     ...

After pivot (tidy):

UNINUM  INV_CODE  AMOUNT
12345   10        100.00
12345   15        200.00

Datasets in this category (based on March 2026 data): RCB, RCF, RCO, and others.

"single_multiple_single"

The raw data file wraps each record across n_codes + 2 physical lines: leading single columns on line 1, one line per code in the middle, and trailing single columns on the last line. After parsing, the same pivot as "single_multiple" is applied.

Datasets in this category (based on March 2026 data): RCR7.

Code dictionaries

Datasets with repeating code groups use a code dictionary to map integer codes to human-readable labels. fcall ships 13 such dictionaries (ported from the R package’s internal data):

import fcall

# Look up the code dictionary for a root file name
result = fcall.get_codes_dict("RCB")
print(result["codes_dict"])

Metadata schema

process_metadata_file() returns a dict with two keys:

Key	Type	Description
"scenario"	str	"single", "single_multiple", or "single_multiple_single"
"vars_info"	polars.DataFrame	One row per column; see columns below

vars_info columns:

Column	Type	Description
ColumnName	Utf8	Variable name (cleaned of **)
ColumnType	Utf8	"Numeric" or "Alphanum."
DecimalPosition	Int64	Decimal places (0 = integer)
Definition	Utf8	Free-text description from FCA
MultipleOccurrenceColumn	Boolean	True for repeating columns
CodeColumn	Boolean	True for the first repeating column (the code key)
ColumnTypeSQL	Utf8	"text", "integer", or "float"