Credit Scoring modelling for retail banking sector.

Problem raised by Accenture

Coordinating teachers of the problem:

Ignacio Villanueva (Universidad Complutense de Madrid)

Estela Luna (Accenture)

 

Exposition of the problem

Credit scoring is a systematic approach to the admission process of the retail banking activity. It substitutes the traditional way, based on personal impression, of admitting commercial operations involving loans, mortgages, credit cards etc. It allows the bank to quantify the potential risk of default of every single customer and use it to forecast the credit behaviour of incoming clients

 

Scheme of the work to be done:

 

1) Data analysis .

 

2) Single factor analysis

 

Descriptive Statistics of the main variables of the model. We will obtain mean, median, standard deviation, kurtosis, skewness, maximum, minimum, 95 percentile, 5 percentile.

 

3) Multi factor analysis

 

We will obtain correlation between each variable and the default variable. Highly correlated variables with default will be rejected because of lacking of explanatory capacity.

Chi-square hypothesis testing to know independence between each variable and default.  Test statistic is:

 

4) Modeling

Probability linear models and non linear models are analyzed and we select logit model for our case. Logit model is:

 

 

5) Model performance analysis

We obtained percentage of defaulted forecasted and non forecasted.   

 

6) Validation of the model

 

We obtain Powerstat and RAR statistics with their confidence interval

 

7) Calibration of the model

 

Defaulted clients are counted, obtaining a yearly defaulted rate (YDR). The main prupose of calibration is to determine the variables A, B and C of the formula:

A, B and C must be obtained so to minimize root mean square error.

We perform Hosmer- Lemeshow test to know if the obtained formula adjusts to the actual YDR. The statistic is:

, where

Oi=# of defaulted

Ni=# of clients in block i

= average deafult probability in block i

 

 

8) Business analysis:

this part links the quantitative analysis performed in steps 1 through 7 to the general credit policies required by the bank to cover regulatory and business goals. This point wraps up the overall activity.