How to create bins when doing credit score validation using decile analysis?

[Jennifer]

Up

0

Down

When doing credit score validation using decile analysis, how to create bins ?

1. Divide credit score into 10 bins. Or
2. Sort data by credit score first, and then divide all samples into 10 bins?

With (1), each bin may have different number of samples

With (2), samples with the same credit score may go to adjacent bins.

BTW, in reality, is it possible to have credit score of ZERO?

Thanks!

• This discussion was modified 4 years ago by  Jennifer.

[Datura]

Up

2

Down

Credit score of zero or negative is theoretically possible because we can choose PDO or offset to scale a predicted probability to any wanted score range

But actually, it’s always more convenient for us to set score range between 100 —1000. In this case, zero or a negative score are not real scores they are just special value codes, meaning no score or sth else. We need exclude these rows from our analysis

To create the decile we divide it into approximately equal size bins based on the score. In Python we can use qcut() function to do it. In SAS use Proc Rank

[Jennifer]

Up

0

Down

Very clear! Thx!

[norahx]

Up

0

Down

If the data is very skewed, for example, sometimes there are many same value credit scores for different customers, should we avoid to put same score to different category?

In this way we might have not very even group.

Thanks!

[Jennifer]

Up

0

Down

A blog which explains usage of qcut and cut very well

[Datura]

Up

0

Down

???

[Datura]

Up

0

Down

That’s ok. In this case we may only get 8 or 9 bins rather than 10 bins. A bin may have about big chunk, say 20% of records rather than 10 %. It happens in real work. It is fine