topic Re: GInin difference between train data and out of time data in SAS Data Science

GInin difference between train data and out of time data

Ronein — Mon, 23 Dec 2024 09:55:38 GMT

Hello
I build a credit risk model on 100,000 customers.

I split the data into train 70% and test 30% and built the model on train data.

Then ,the results are Gini 79% on train data.

I calculated PD based on model coefficients on out of time data.

Gini of Out of time data is 80%

My question- is this difference of 1% between out of time and train is okay or mention a problem ?

What other tests would you do in such case?

I checked PSI for all predictors and it show high stability for each of predictors

Re: GInin difference between train data and out of time data

PaigeMiller — Mon, 23 Dec 2024 10:37:29 GMT

You are asking the same question as https://communities.sas.com/t5/SAS-Data-Science/Gini-credit-risk-model/m-p/954433#M10965

Please do not double post questions.

PS: There is an answer at your other thread.

Re: GInin difference between train data and out of time data

Ronein — Mon, 23 Dec 2024 16:00:39 GMT

No. Here is out of time vs train data and other question was train data vs test data

Re: GInin difference between train data and out of time data

Ronein — Mon, 23 Dec 2024 20:03:27 GMT

In this question I ask about train data vs out of time data ( model develop on data from June 2021 and out of time data is for example June 2023). In previous post my question was about train data ( 70%) vs test data (30%) ...

Re: GInin difference between train data and out of time data

PaigeMiller — Mon, 23 Dec 2024 21:41:44 GMT

@Ronein wrote:
No. Here is out of time vs train data and other question was train data vs test data

Changing a few of the words doesn't change the meaning of the question. Your two posts have the same meaning, and likely the same answer.