topic forecasting delinquency in Statistical Procedures

forecasting delinquency

Steelers_In_DC — Thu, 23 Mar 2017 17:41:56 GMT

Good Afternoon All,

I have current data, along with the history for credit card balances. I have grouped these by month and delinquency buckets(1 cycle = 30 days past due, 2 cycle = 60 days past due etc.)

There is an old process that I've inherited to forecast delinquencies, it is an markov chain that I don't like how it's written so I'm going to try to do it over using IML. I'm also going to do a moving average.

Why I'm reaching out to the group is to find out if someone has experience with an exercise like this, if they have a certain model or procedure that they thought fit this kind of data best.

Thank You,

Mark

Re: forecasting delinquency

Rick_SAS — Thu, 23 Mar 2017 17:52:04 GMT

I cannot speak to "the exercise" personally, but it sounds similar to some SGF papers that Gongwei Chen wrote. If you decide to rewrite the analysis in IML, there are a few articles about Markov chains and moving averages in IML that you might find helpful for writing efficient code:

Re: forecasting delinquency

Steelers_In_DC — Thu, 23 Mar 2017 18:00:13 GMT

That's great, thanks Rick!

Re: forecasting delinquency

Reeza — Thu, 23 Mar 2017 18:05:15 GMT

No idea of the accuracy compared to a Markov model, but 2 stage regression are models are what I've seen. First stage calculates the probability of default and the second calculates the amount of the default assuming they're going to default.