Sample data

The examples on these pages and the tasks in the Homework repository use the same small, made-up loan portfolio. Nothing in it comes from a real bank: the clients, the loans and all values are invented. Only the shape is typical: a table of clients, a table of loans that points to the clients, and a nomenclature (the rating scale).

File	Rows	What it is
	150	one row per client
	400	one row per loan; client_id points to clients.csv
	10	the rating scale: one row per rating grade, with its PD band

From session 7 on, more files of the same made-up portfolio - with the problems of real data on purpose: messy text, several formats, bad rows, history over time.

File	Rows	What it is	Sessions
	150	the clients with their text fields as people typed them	7, 8
	231	free-text comments of credit officers about loans	8
	280	the payments of September 2026, with some bad rows	9
	-	the settings of a monthly report	9
	3 sheets	daily exchange rates, one sheet per month	9, 14, 16
	7	the regions in Bulgarian and English - saved in cp1251	7, 9
	1800	the rating of each client at 12 month ends	12, 13, 16
	3992	the outstanding amount of each loan at 12 month ends	16
	406	loans.csv as a local export: ;, decimal comma, dd.mm.yyyy, missing values, duplicates	14, 15
	4 tables	an SQLite database: clients, loans, rating_scale, ratings_history	13, 16

The same files are in the data/ folder of the Homework repository. On these pages they are in data/ of every runnable example.

See the files

Click a file name in the tables to look inside the file. A CSV file, a sheet of the Excel file or a table of the database is shown as a table that you can filter; rows with a wrong number of fields and empty lines are marked. Text shows a CSV or JSON file as it is - with its separator, quotes and spaces - and SQL the tables of the database. Download saves the file. The preview finds the encoding of regions_bg.csv (cp1251) by itself; Python does not - see the file below.

clients.csv

Column	Example	Meaning
client_id	C0002	the id of the client
segment	sme	retail, sme or corporate
region	Ruse	the region of the client
industry	M	an industry code (a letter, like the sections of NACE); empty for retail clients
rating	5	the internal rating, 1 (best) to 10 (default) - see rating_scale.csv
client_since	2014-06-08	the start of the relationship (ISO date)

import pandas as pd

clients = pd.read_csv("data/clients.csv")
clients.head()

import pandas as pd

clients = pd.read_csv("data/clients.csv")
clients.head()

Edit the code and run it again

loans.csv

Column	Example	Meaning
loan_id	L000001	the id of the loan
client_id	C0080	the client (see clients.csv)
product	investment	mortgage, consumer, credit_card, overdraft, working_capital, investment, project_finance
currency	EUR	EUR or USD
start_date, maturity_date	2026-01-22	ISO dates
amount	1156900.00	the original amount
outstanding	1044231.30	the amount still to be paid on the report date (30.09.2026)
interest_rate	8.83	in percent per year
days_past_due	0	days the payments are late
stage	1	1, 2 or 3 - stage 3 means more than 90 days past due or rating 10

import pandas as pd

loans = pd.read_csv("data/loans.csv")
loans.groupby("product")["outstanding"].sum().round(2)

import pandas as pd

loans = pd.read_csv("data/loans.csv")
loans.groupby("product")["outstanding"].sum().round(2)

Edit the code and run it again

rating_scale.csv

A nomenclature: one row per rating, with its grade and the band of the probability of default (PD).

Column	Example	Meaning
rating	4	1 to 10
grade	BBB	the letter grade
pd_min, pd_max	0.0025, 0.006	the PD band of the grade (0.0025 = 0.25%)
description	moderate risk

import csv

with open("data/rating_scale.csv", encoding="utf-8") as f:
    for row in csv.DictReader(f):
        print(row["rating"], row["grade"], row["description"])

import csv

with open("data/rating_scale.csv", encoding="utf-8") as f:
    for row in csv.DictReader(f):
        print(row["rating"], row["grade"], row["description"])

Edit the code and run it again

The files of the later sessions

client_register.csv

The same 150 clients as clients.csv, but the text fields as they were typed into an old system: client_id in many forms (c0014, C 0009, C-0026, C17), names in any case with extra spaces and quotes, legal forms written as E.A.D., EIKs with spaces or a BG in front, IBANs in groups of four or in lower case, e-mails in upper case or with " at ", cities in Latin or Cyrillic, with gr. in front, registration dates in five formats, and the turnover of companies in Bulgarian and English number formats. Columns: client_id, name, eik, iban, email, city, registered, turnover. The IBANs use made-up bank codes (DEMO, TEST, XMPL, FAKE) and the e-mails the reserved example.* domains.

loan_comments.csv

loan_id, comment_date (ISO), comment: short notes like "Payment of 1 250,00 EUR received on 15.09.2026." or "Disbursement to IBAN BG71 DEMO ... confirmed" - amounts, dates, IBANs, EIKs, e-mails and loan ids inside free text.

payments_2026-09.csv

payment_id, loan_id, pay_date, amount, currency. Most rows are fine; some are not: a missing amount, 12O.50 (a letter O), n/a, a date that does not exist, an unknown loan, a negative amount, a field too few or too many, an empty line.

report_config.json, fx_rates.xlsx, regions_bg.csv

• report_config.json: the report date, the LGD, the DPD buckets, the segments and the output settings of a monthly report.
• fx_rates.xlsx: one sheet per month (2026-07, 2026-08, 2026-09); columns date, USD, GBP, CHF - how many units of the currency one euro buys.
• regions_bg.csv: code, name_bg, name_en - saved in the Windows Cyrillic encoding cp1251, like many old exports. Open it with encoding="cp1251".

ratings_history.csv and balances_monthly.csv

The history of 12 month ends, from 31.10.2025 to 30.09.2026; the last month agrees with clients.csv and loans.csv. A client or a loan appears only from the month it started.

• ratings_history.csv: client_id, month_end, rating.
• balances_monthly.csv: loan_id, month_end, outstanding, days_past_due.

loans_export.csv

The 400 loans as an export from a local system: the separator is ;, the decimals have a comma (651927,30), dates are dd.mm.yyyy, the column names have spaces (Loan ID), the file starts with a BOM (Excel's "CSV UTF-8"). Some values are missing - empty, or . like a SAS missing value - and 6 rows are duplicated.

portfolio.db

An SQLite database with four tables: clients, loans, rating_scale and ratings_history - the same data as the CSV files, with primary and foreign keys.

import sqlite3

con = sqlite3.connect("data/portfolio.db")
for name, sql in con.execute("SELECT name, sql FROM sqlite_master WHERE type = 'table'"):
    print(name)
print(con.execute("SELECT COUNT(*), ROUND(SUM(outstanding), 2) FROM loans").fetchone())
con.close()

import sqlite3

con = sqlite3.connect("data/portfolio.db")
for name, sql in con.execute("SELECT name, sql FROM sqlite_master WHERE type = 'table'"):
    print(name)
print(con.execute("SELECT COUNT(*), ROUND(SUM(outstanding), 2) FROM loans").fetchone())
con.close()

Edit the code and run it again

Without pandas

The first lectures read the files with the standard library (open, csv), and pandas comes in session 14. Both ways read the same files.