SAS Support Communities

The issue you are describing illustrates the problem known as 'temporal infidelity'.  This problem occurs when the relationships modeled for the time periods you had available during modeling had shifted by the time the model was applied to new data.   In general, models will not perform on newer data as well as they performed on your historical data.  You need to monitor the amount of the change and the nature of the change to assess when a model needs to be refit.   Using out-of-time samples to validate your model is a reasonable practice and gives you a more realistic assessment of how your model will perform, but do not be surprised when it does not perform as well.  Simply including all of the training data will make some of your metrics look better but will be misleading as they also mask the temporal infidelity which you seem to have identified.  Tools such as SAS Model Manager allow you to monitor the performance of a model over time so that you can refit the model when the performance has degraded too much.     I hope this helps! Doug  ... View more

Well I don't think there are any hard and fast rules on this. It also depends how your data behaves and whether you want to take into account both good and bad economic conditions. For example if you only use data from 12 months, say 2015, when economic conditions are good, would your behaviour score model also work when times are bad, like 2006/2007 (GFC)?   If you want your model to work over a variety of economic conditions then you need to use observational data from those periods, so I'd say you need at least 5 years data and probably more for long-term mortgage products. What you may also find is that getting enough overdue events to build a highly predictive model may be hard when times are good but definitely not so hard in adverse conditions.   The outcome period should match what you are trying to predict. For example where I work we need to predict the probability of going into default (90 days past due) in the next 12 months. That means we need to look forward 13 months from the observation point and see if the loan went into default in any of those months.     ... View more

1. Clarify the definition with the business - VERY IMPORTANT 2. If you're analyzing credit, I would assume the amount paid was actually more significant than the amount intended. You can also test them in the model one at a time and see which provides a better fit. ... View more

You have several parts in your proces. 1/ Eguide that is your front end interface is supporting utf8 as it is build in .NET. 2/ your sas system base/foundation is the one that executes the code can run in single byte mostly Latin1.  It can also run in utf8 mode but in that case it should use other binaries. When it is on a server the SAS APP setting scripting os is the one doing that switch. 3/ there is an Access to oracle client software installed that must support the utf8 or the Latin1 encoding. This one is communicating with 4/ the oracle server and that database having defined some encoding fir the database and tables Having issues with thai you are most likely into some wrong configuration in the 2,3,4 aligning. Not an eguide issue but things that are used behind those. ... View more

Hi Ros, Not a statistician here, but I am a big fan of the Interactive Grouping and Scorecard nodes in Enterprise Miner. I used to do everything by hand back in the day, and these nodes do a lot of the hard work behind the scenes. Two things that I thing you could improve: You are using data from 2007 to 2012. This time window is too large. The standard is to have 12 month periods, but it could vary depending on whether it is new applications vs behavior, the specific credit practice, or country regulations. But I would rather do yearly periods e.g. use June 2010 through May 2011 to predict the next year. I would think that this way the variable TERMS would be more stable. Instead of using the moving average to determine how to bin or group the variable TERMS, I would use the Interactive Grouping node in the credit scoring tab directly. This node bins the variables for you, and most importantly calculates and graphs the weight of evidence of each of the bins, it also calculates the Information Value and only passes the variables that meet a threshold to the next nodes in your diagram (think of a Scorecard node). I would much rather use weight of evidence and information value than moving average to assess a variable. You can learn more about these nodes in the reference help. Press F1 if you are on Enterprise Miner. This book is also a must read for credit scoring professionals: Credit Risk Scorecards: Developing and Implementing Intelligent Credit Scoring Naeem Siddiqi I hope this helps, Miguel ... View more

What is C ? and SD means STD ? data have; input Contract     Contract_Date : date11.     Contract_Amount     ; format      Contract_Date date9.; cards; 1     01-Jan-13     10000 2     02-Jan-13     20000 3     01-Feb-14     30000 4     02-Feb-14     25000 5     10-Jun-14     22000 6     11-Jun-14     22000 ; run; proc sql; create table want as select *,(select avg(Contract_Amount) from have where Contract_Date le a.Contract_Date) as AVG_Contract_AMT,           (select std(Contract_Amount) from have where Contract_Date le a.Contract_Date) as SD   from have as a ; quit; Xia Keshan ... View more

Thanks, but what the different between Variable Selection and Regression? I think both method are able to eliminate correlation. I have never used Variable Selection before. How does it work? ... View more

Thanks so much. I am so new for this field. You answer is so helpful. ... View more

I have already got the answer. Thanks ... View more