BookmarkSubscribeRSS Feed
AnshulS
Obsidian | Level 7

Hi,

 

I am building a PD model on SAS EM using credit score node.

 

I can see in results as Scorecard window, and this has a score for all the variables displayed here. 

 

Point is i want to calculate the probability of default for a particular customer i.e. lets say 0.6 is the probability of default.

 

How should I do this in SAS EM??

7 REPLIES 7
Reeza
Super User

Once you have a model built you can usually connect a SCORE node to score new data. Or export your model as score code to be used in several languages. 

 

If you want the first option you need to import your data into SAS and use a Score node to score the output. The input data needs to have the same structure as your original data as well as the same range of values. 

 

For the second method, you can export the score code and then you can use that in Base SAS or EG or a different language  if you choose that option, and run it against the data in a new program. Note that 'skeleton' code is developed, you need to wrap the logic in a data step and provide the input data. 

 


@AnshulS wrote:

Hi,

 

I am building a PD model on SAS EM using credit score node.

 

I can see in results as Scorecard window, and this has a score for all the variables displayed here. 

 

Point is i want to calculate the probability of default for a particular customer i.e. lets say 0.6 is the probability of default.

 

How should I do this in SAS EM??


 

AnshulS
Obsidian | Level 7
Hi Reeza,
Thanks for your reply. I actually wanted to see probability of default for each observation in my input data. Is there a way to see estimated default probabilities for each input observation in SAS EM??

Thanks,
Anshul.
SASKiwi
PROC Star

As I understand it, you have behaviour scores for a number of variables. Do you have an overall score for each row - I'm assuming here that each row represents a loan account?

 

If you do then the next step would be to compare your actual account defaults with the overall behaviour score. You could group your scores into bands. Typically a behaviour score is a number between 0 and 1,000 (low score means poor behaviour, high score means good behaviour). That means you could split your loans into behaviour scores bands of say 100. You would expect the lower the score the more defaults you are likely to get. By getting the proportion of accounts going into default over total accounts in each band you can start to estimate the probability of default for each score band.This is pretty much a manual coding process. There is no "magic button" in Enterprise Miner to do this. Also it helps to have big volumes of data covering many years so you can calibrate your scorecard accurately

 

I am being very simplistic here to illustrate how you could go about it, as the path to calculating PDs is tricky and complicated. This is why scorecard experts earn so much money and building scorecards is so expensive! Unfortunately I'm not one of those people. I just know a bit about them... 

Ksharp
Super User

Use Regression Node. If you have a target variable which has binary value, then EM will invoke PROC LOGISTIC to model data, the predicted value of Y is the PD probability .

AnshulS
Obsidian | Level 7

Hi,

Thanks for your reply, conceptually i understand this, but when i use logistic regression where in output i will see probability of defaults? As i only see fit statistics and output windows, but nowhere the probabilities....

 

I have attached the output for your reference here please.

 

 

Ksharp
Super User

At Left panel , there should have INPUT DATASET and OUTPUT DATASET items, click small button ( ... )  check output dataset. 

Reeza
Super User

If I understand correctly, you already have a model. You can score your input data, you don’t have to have new data. I would try that first. My instinct is to say it’s actually calculated in a table already, you just need to know how to access it, but I don't have SAS EM with Credit Score node to test it out at the moment. 

sas-innovate-2024.png

Available on demand!

Missed SAS Innovate Las Vegas? Watch all the action for free! View the keynotes, general sessions and 22 breakouts on demand.

 

Register now!

How to choose a machine learning algorithm

Use this tutorial as a handy guide to weigh the pros and cons of these commonly used machine learning algorithms.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 7 replies
  • 1934 views
  • 5 likes
  • 4 in conversation