10-25-2016 03:51 PM
I built a credit card application score model and one of the variable doesnt make any sense, here are the details:
The variable I am having trouble with is "revolving Debt_to_credit percent", the data and common knowledge tell me that as the
revolving Debt_to_credit percent increases, the riskier you become and therefore a lower score should be assigned to you, here is a picture of my weights of evidence:
the picture above makes sense because as "revolving_Debt_to_credit percent" increases the riskier that population is.
Now my problem is with the coefficient of the logistic regression, this is the output:
I was expecting the cofficient to be negative so that as the bad rate increases, the scorcard ponts assigned to this particular variable would decrease as well and not increase as shown in the picture above. so basically the picture above is saying as the revolving_debt_to_credit_percent increases, the bad rate increases as well and the score points do too, which it doesnt make sense.
my question is, should I eliminatre that variable from the analysis, could the values in the varible are erronous or that's the way it is from the optimization? What should I do?
Thank you very much
10-26-2016 09:19 AM
Is it possible there is another input for your Scorecard that is highly correlated with revolving Debt_to_credit percent? That is typically the cause of scorecard points going in the opposite direction than you would expect.