German credit dataset : interpretation of checking_status feature

Question

I am struggling to understand the meaning of some features of the german credit dataset.

I am particularly interested in, the categorical feature checking_status (Status of existing checking account) which has 4 possible values :

• '< 0'
• '0<X<200'
• '>=200'
• 'no checking'

I am not a banking expert but my intuition would be than the higher the checking status the higher the probability to be classified as good. TO check my intuition I computed the proportion of applicant classified as good in the dataset depending on their checking_status values. The results were surprising :

• no checking : 88 %
• <0 : 35 %
• 0<=X<200 : 42 %
• '>=200' : 12 %

Can be read as : 88 % of applicants with 'no checking' are classified as good.

So according to german credit dataset it is harder to have a credit with more money '>=200' < '<0' and it is much easier when the applicant does not have a bank account !

Any idea/links on the interpretation of this feature?

Answer 1

The dataset description was vague, so it's a guess - but I suspect there is a bit of a Base Rate Fallacy/Survivorship Bias at play.

It's easier to have less money in your account, and therefore there are more people with little or no money than people with a lot of money. On the other extreme, people with private swimming pools full of cash don't go applying for credit - they don't need to.

Therefore, the classifier mostly sees a subset of those people who either need more money than your average applicant (and so, if they default, you lose more) or burn through the money they do have faster (and therefore have a higher risk of defaulting).

There might be some more sociological factors at play, too - you're not seeing all the people who would have had a high-value checking account - if they had one at all, but they chose not to for some reason. I don't feel qualified to speculate on the exact factors at play in 90s-era Germany, though.