Carlos Montenegro

Sébastien Pavot
Edward Vrijghem

Group Assignment Financial Programming

2019
Prof. Dr. Minh Phan
Contents
Report ..................................................................................................................................................... 1
General data overview......................................................................................................................... 1
Insights and opportunities ................................................................................................................... 2
Trends ................................................................................................................................................. 3
Technical Aspects .................................................................................................................................... 4
General Structure of the code.............................................................................................................. 4
Libraries .............................................................................................................................................. 4
Functions created ................................................................................................................................ 5
Variable information ............................................................................................................................... 5

Report

General data overview

We found out that in general the total number of transactions and the total balance per district were
highly correlated, this came as no surprise. The possession of a high number of clients results in a high
number of transactions, as well as a high total balance. More clients result into more accounts, which
results in more transactions as well as a higher total balance.

When comparing the average salary per district size we concluded that Prague, the capital of
Czechoslovakia, scored best in both categories. Furthermore we observed a slight correlation between
district size and average salary. Our clients were earning more in the capital district, compared to the in
the smaller, lesser populated districts.

When digging deeper into loan statuses we start to notice some age-group related differences. When
observing contracts where the loan was not payed, we see and increase for the age group of 50 year olds.
This could indicate an increased risk when borrowing to this group, the risk assessment division should
take this into consideration. Next to this, the high level of indebted clients (indicated in green) who take
a loan in their 20’s to 50’s should also be considered.
Carlos Montenegro
Sébastien Pavot
Edward Vrijghem
Furthermore, every trend goes as expected. Younger and older age-groups tend to borrow less in general.
This is firstly because the group of 60-69 tends to spend less money on so called investments (houses, cars
and others) than the younger generation (10-19yo.) which doesn’t have a lot of loans. This is most likely
because the legal minimum age to borrow money is 18 in Czechoslovakia.

Insights and opportunities

As of the relationship between seniority and number of transactions our findings were just as expected.
Senior clients tended to have a higher total number of transactions compared to the newer clients. This
means the newer clients possess a high potential and should thoroughly be targeted. More transactions
mean more transaction fees which translates into more revenue for the bank.

As for the cards we can see that there are huge opportunities, the grey part of the bar charts represent
clients which are not in possession of a card yet. The clients situated in “the grey zone” could form a target
group for later advertising actions.

As per average loan payments per district (a direct derivative of average loan), we included the top
regions. Surprisingly the capital district Praha didn’t came out on top. This information could be useful to
analyze in terms of marketing opportunities. Maybe this could be a result of the high population which
results in a better indication of the average loan. Another reason could be that there is more competition
in the capital to get a decent loan, resulting in banks being more reluctant in giving out big loans. In both
cases the bank could take this into account when doing credit scores/assigning loans.
Carlos Montenegro
Sébastien Pavot
Edward Vrijghem
As per the relation between the number of transactions and the balance of your clients you can see a
strong positive correlation. Wealthier clients tend to do more transactions, as expected.

Trends
As for trends we see a steady decrease in the average yearly loan payment as well as the interests credited.
These obviously go together. This could be an industry trend or could a company specific trend, in both
cases this is a very worrying indicator and should thoroughly be examined.

Yearly transactions and total clients increased from 1993 to 1994, remained steady until 1997. After 1997
we notice a decrease, this is equally worrying as the decreasing average loan payments and average
interests credited. All of these factors need a solid analysis and should be reviewed.
Carlos Montenegro
Sébastien Pavot
Edward Vrijghem

Technical Aspects

General Structure of the code

Explore the data
We explore the data using the following main functions:

1. nunique(): which gave us the number of unique values for specific variables
2. head(): which in most cases was used to return the first 5 rows of each dataset
3. info(): which gave us a summary for each variable, containing the number of non-missing values
and its type.
4. Describe(): which gave the maximum and minimum numbers, which was specially important to
check in which ranges our variables fluctuated.

Clean the data

1. We changed the name of the categories of some variables to more meaningful ones.
2. We left the missing values untouched and created new dummy variables for them.
3. We read the dates and calculated some new variables from them (ex. birthday or recency)
4. We decided which variables where not important to maintain and we dropped them.

Merge the tables

1. We found out that there was a problem with the information in the data set of transactions
(“trans”). While most of the other tables were account based, the “trans” data set was
transaction based, which mean that each row of data was uniquely identifying each transaction
made in each account. We modified this data set and made it account based.
2. Once the previous step was done, we created to temporal big data sets:
a. One containing the ‘loan’, ‘order’, ‘trans’, ‘district’ and ‘account’, in an account-based
dataset.
b. And the other containing the ‘card’, ‘disp’, and ‘client’, in an account-based dataset.
3. Finally, we were able to combine this temporal datasets, and group them by client Id to be able
to get a client-based dataset and have a DataMart that uniquely identifies each client.

Create new variables

1. We created more variables which were mostly client oriented with the exception of the
demographic variables which were aggregated figures.
2. We explored the categories and values of our variables and created new variables based on
certain condition that we found meaningful.

Libraries
 pandas: this library is used to convert the variables to date time format, to merge tables, to
create dummies for categorical variables, to get dummies, among others.
 numpy: this library is used to calculate a time difference, to find NaN values, to load a data file,
to fix a random state, to obtain absolute values, to arrange bins, among others.
Carlos Montenegro
Sébastien Pavot
Edward Vrijghem
Functions created
 explore:
o Logic: this function makes the data exploration simpler by generating the main
information about the data set
o Input: data set to be analyzed
o Output: it returns a print of the following functions ouput: describe(), info(), nunique()
 to_month_gender
o Logic: this function returns the gender of a text that contains the birthday and gender of
a person (50 + MonthNumber if the person is a woman)
o Input: data set value
o Output: the person’s gender

Variable information

Variable name Explanation From table

disp_id Record identifier of disposition Disp
client_id Record identifier for each client Client
account_id Record identifier for each account Account
Owner / Disp Is the client owner or disponent of the account Disp
Is_shared? Is the account shared? 1 for yes and 0 for no. Disp
card_id Record identifier of credit card Card
Card type Type of card (‘Classic’, ‘gold’, ‘junior’) Card
Date card issued Date the card was issued to the client Card
Time in days since the card was issued based on
Time since card issued Card
1999/01/01 as the today date
Flag column to identify client with card. Yes,
Has card means that the client has a card and no if he has Card
not.
district_id Record identifier of district District
age Age of the client in years Client
gender Gender of the client, F for female and M for male. Client
Type of issuance of statements (Monthly, weekly,
Issuance type Account
immediately)
Date account opened Date the account was opened Account
Seniority Time in years since the account was opened Account
Loan amount Amount of money the loan is valued Loan
Loan duration Time in months of the loan Loan
Loan payment by month Amount of money due by month Loan
Status of paying of the loan. A equal loan finished
and paid, B equal loan finished but not payed, C
Loan status Loan
for contract still running and OK so far, D for
contract running but client in debt.
Carlos Montenegro
Sébastien Pavot
Edward Vrijghem
Flag column to identify people who contracted a
contracted a loan loan. Yes for people who contracted a loan and Loan
No else.
date loan issued Date the loan was issued Loan
Yes if the loan is finished regarding time, no else.
Loan_finished Based on difference between the 1999/01/01 Loan
and the date the loan was issued.
If the loan isn’t finished, difference in month
from 1999/01/01 since the loan was issued.
Loan months remaining Loan
Means the number of months remaining in the
loan contract.
Amount of money remaining to pay if the client is
Amount loan remaining paying each month based on number of months Loan
remaining by the money due by month.
Total_order Total of amount of order. Order
Leasing Total of amount of order characterize as leasing. Order
Total of credit that the account has by doing the
Credit Trans
sum of all transactions characterize as Credit.
Total of debit that the account has by doing the
Debit Trans
sum of all transactions characterize as debit.
Total of amount characterize as operation type
Cash deposit Trans
cash deposit.
Total of amount characterize as operation type
Cash withdraw Trans
cash withdraw.
Money transfer to other Total of amount characterize as operation type
Trans
bank money transfer to other bank.
Total of amount characterize as operation type
Recovering other bank Trans
recovering other bank.
Total of amount characterize as operation type
Debit card Trans
debit card.
Total of amount characterize as operation type
Other operation Trans
other.
Total of amount characterize as transaction type
Insurance payment Trans
insurance payment.
Total of amount characterize as transaction type
Statement payment Trans
statement payment.
Total of amount characterize as transaction type
Interest credited Trans
interest credited.
Total of amount characterize as transaction type
Sanction interest negative Trans
sanction interest negative.
Total of amount characterize as transaction type
Household Trans
household.
Total of amount characterize as transaction type
Age pension Trans
age pension.
Total of amount characterize as transaction type
Loan payment Trans
loan payment.
Carlos Montenegro
Sébastien Pavot
Edward Vrijghem
Total of amount characterize as transaction type
Other transaction Trans
other.
Number transactions Count the number of transactions Trans
Difference between credit and debit columns to
Total balance Trans
know the actual solde.
district_name Name of the district District
Region Region of the district District
inhabitants Number of inhabitants per district District
ratio_urban Ratio of inhabitants District
avg_salary Average salary per district District
unempl_95 Unemployment rate in 1995 District
unempl_96 Unemployment rate in 1996 District
entrepren Percentage of entrepreneurs per district District
crime_95 Crime rate in 1995 District
crime_96 Crime rate in 1996 District
number_urban Number of urban inhabitants per district District
Number of inhabitants minus number of number
number_country District
of urban inhabitants
Increase or decrease of unemployment rate
unemployment_trend District
between 1995 and 1996
crime_per Number of crime per inhabitants District
Increase or decrease of crime rate between 1995
crime_rate District
and 1996
Flag column to identify district that have a crime
has crime rate District
rate available.
Flag column to identify district that have a
has unempl rate District
unemployment rate available.