B-Score is the continuous scoring done month on month basis of complete portfolio to rate them or segment the borrowers into different segment according to there riskiness or goodness. Bank uses this B-score to offer new top-up loan over your current loan without any documentation or other different product like credit card , Personal Loan , etc all product are pre approved without any documentation to make user experience much better . Credit Card companies use this B-score to increase the Monthly limit of card without asking for any new document of proof and sometimes also decrease the limit of you card. ECL is one of the applications of B-Score which subsequently help the organization to future project and minimize the Expected Credit Loss for their organization. Also this help them in decreasing the projected expected credit loss depend on how well your B-Score is performing for complete portfolio.
For calculation of B-Score we take consideration of Month on Month performance of the borrowers on-book like Average Delay in EMI payment , Maximum Delay in EMI payment, Average Outstanding, Maximum Outstanding, etc as well as we also take consideration of OFF-Book Performance because it might be possible that borrower is performing well on book but off-book performance of the user has become very bad and suspicious so to have a complete view and to make more stable model using the OFF-Book performance is must . To get OFF-Book performance we pull the credit bureau report of the borrowers and take features like Information About the Delinquency , Outstanding, Enquires etc and use this features and on-Book performance feature in combination to build the final model for the B-Score.
To select data we take snapshot at point of time or to make B-score more stable we can take snapshot from 2 to 4 different quarters. Lets consider we are building the model on the data from just 1 snapshot of the portfolio so lets say the organization disperse the loan for tenure ranges from 1 to 10 year so we have to make it sure the data selected should have representation of loan from all tenure like 10% from loan at there 1 year ,10% from loan 2 year and same till 10 year. If we are considering snapshot from more than 1 quarter then there should not be the duplicate entry of the loan as the same loan are present in every quarter so we will be selecting different loan from different quarters.
| Start Date | End Date | |
|---|---|---|
| Model Development | 1-Sep-2018 | 31-Dec-2018 |
| Observation Start | Observation End | Performance Start | Performance End | Sample Size | Bad Count | Bad Definition | |
|---|---|---|---|---|---|---|---|
| Train | 01-Jun-2017 | 01-Jun-2017 | 01-July-2017 | 30-June-2018 | 50000 | 2500 | 90 DPD 12 MOB EVER |
| Test | 01-Jun-2017 | 01-Jun-2017 | 01-July-2017 | 30-June-2018 | 20000 | 1000 | 90 DPD 12 MOB EVER |
| OOT | 01-Aug-2017 | 01-Aug-2017 | 01-sep-2017 | 31-Aug-2018 | 10000 | 500 | 90 DPD 12 MOB EVER |
Independent Variable : Borrower On-Book Performance Data + Borrower OFF-Book Performance Data
Target Variable : Borrower 90 DPD 12 MOB EVER
| On-Book Features | OFF-Book Features | Target |
|---|
| Average Delay | Average Outstanding | 30 DPD Count | Enquires 3 Months | Enquires 12 Month | Delinquency 3 Months | Outstanding 3 months | 90 DPD EVER |
|---|---|---|---|---|---|---|---|
| 0 | 30000 | 0 | 2 | 4 | 0 | 20000 | 0 |
| 48 | 45000 | 4 | 27 | 12 | 8 | 70000 | 1 |
| 15 | 23000 | 1 | 1 | 2 | 1 | 25000 | 0 |
| 143 | 70000 | 12 | 134 | 45 | 24 | 83000 | 1 |
Training:-
After all data Pre-processing we train a classification model which returns the output in terms of probabilities . We bucket or divide all borrowers in terms of the probability into 5 different segments A to E where A being the least Probability of default and E with highest Probability of Default
| # Borrowers | Probability Range | # Defaulter | # Non Defaulter | Percentage Defaulter | |
|---|---|---|---|---|---|
| E | 2000 | 0.81 to 1.00 | 600 | 1400 | 37.5% |
| D | 2000 | 0.61 to 0.80 | 500 | 1500 | 31.25% |
| C | 2000 | 0.41 to 0.60 | 300 | 1700 | 18.75% |
| B | 2000 | 0.21 to 0.40 | 150 | 1850 | 9.3% |
| A | 2000 | 0.00 to 0.20 | 50 | 1950 | 3.12% |
| Total | 10000 | 1600 | 8400 | 100% |
The Garde Should Follow the rank Order i.e the percentage of Default in E should be greater than D to A
Result / Action: –
If the user is in A we can offer them Different product like pre-Approved Loan Top Up, Pre-approved Credit Card, Pre Approved Credit Limit Increase
If The user is In E we have take action like decreasing the credit limit or blocking the credit card, Blocking the 6 months of EMI in his account or frequent reminding them to pay the loan.
How B-Score Can Lead Organization to More Profitability?
Any organization into the Lending business have some cost involved for its operation like: –
Cost Of Fund: – The term cost of funds refers to how much banks and financial institutions spend to acquire money to lend to their customers like 6% interest are given to saving account holder whose saving are used to lend to other customers so cost to fund is 6% in this case.
Cost Of Default: – The term cost of funds refers to how much Lender loses on money lent if the default occurs e.g. Bank loses Rs 8 on every Rs 100 lent in market so the cost of default is 8 out of every 100 or 8% of the total money exposed to market.
CapEx :- Cost involved in setting up infrastructure for the operation example development of buildings, vehicles, land, or data center etc. let’s say it is 2%.
OpEx :- Cost involved in monthly operation like salaries to employees , office rent, marketing etc. let’s say it is 2%.
So here the point is that we cant do anything much in Cost to fund ,CapEx , OpEx as they are pretty much fix but we have fair chance to decrease the Cost Of Default.
We can either decrease the cost of default by making our A-Score much more sophisticated and strong or we can Decreases it by Using B-Score.
B-score is the process or scoring model to score the existing / On-Book Borrower to their respective Goodness or Riskiness Which is done on monthly basis.
Borrower which has been graded the best i.e. A grade Bank can offer them Cross Sell, Limit Increase and Loan Top-up without asking for any new document just on the basis of their On-Book and Off-Book payment history behavior.
Let take an example of Loan Top-up to existing A grade Borrower. So as this customer are already very good in terms of their repayment and financial stability status the default rate of this Borrower are very less let say 2% so overall profit of 4% are increased if we give loan to this customer as the Cost of default for A-Score was 6% and using B-Score we are giving loan new loan to customer base whose default rate is 2%. Also as we are lending more money we are also increasing the profit from the interest rate from all lent money. Finally, we can say that bank can increase their profitability to 4x by using Just B-Score through the Cross Selling, Renewability and Limit Increase.
An online data science course aimed at helping you launch a career. Students from 33 countries and counting, meet the 100+ alumni’s who cracked career in data science, Credit Risk. Contact at +91 8806930487 or drop a mail at scientistexpress@gmail.com for a free demo class.
