Building risk models after great recession.

[Anonymous]

I am curious what FICO's approach is on using data before and after 2008 to build risk models.

Build something based on pre '08 data, post '08, combination?

Obviously, we are talking about two significant macro economic environments and all I've read seems to indicate that behaviors have changed in the Great Recession.

Any ideas on how to deal with this?

[RobertEG]

I would imagine, since FICO is primarily a behavioral model, that it is constantly updated based on changes in credit statistics.

I would also presume that the segmentation portion of the modeling, which separates consumers into risk categories, or "buckets," would be the first to be effected.

I doubt that creditors in the current market want '08 statistics as basis for their '12 decision making.

[MattH]

I think it is way too soon to determine how scoring models need to be changed to reflect the Great Recession, because although the recession is technically over its impacts are still playing out. The economy may not be sinking, but it also is not rising nearly fast enough to create sufficient jobs. One can argue about whether this or that policy would help more, but -- without getting into partisan arguments -- I doubt anybody thinks either party could bring about robust recovery any time soon. This recession was not just a temporary drop in confidence, from which one can expect a quick recovery. This recession was a global deleveraging that made trilions of dollats of paper wealth vanish into thin air. When Nobel-prize wining economists loudly disagree on the causes and consequence of the Great Crash, I'm sure not able to predict where all this will end up. Nor I think, can anybody else make reliable predictions, "especially about the future" as Yogi Berra said.
 
Also, remember FICO scores are mainly intended to rank-order individuals by credit risk. If the main effect of the crash was to make everybody equally more at risk, then lenders need only raise their cutoff scores based on their loss experience. In data-mining geek lingo, if everybody has become a tad higher risk that is a "main effect." The scoring model still works if there is mostly the same overall main effect of the crash on everybody. But it may be case that some groups of cinsumers were hit harder than others by the crash, in ways not already captured by the scoring models. In data-mining geek lingo, an effect that is bigger for some subsets of your population is known as the "interaction term" in models. Now we come to the crucial point: fitting the interaction term is vastly more sensitive to statistical noise than is fitting the main effect term. Unless great care is taken, interaction terms tend to find signal that ain't really there in the noise. In my day job as a phramaceutical researcher, we call finding signals in tbe noise "overfitting." It may take years to learn how the crash has hit different FICO buckets versus others.

[jamesdwi]

---

@Anonymous wrote:

I am curious what FICO's approach is on using data before and after 2008 to build risk models.

Build something based on pre '08 data, post '08, combination?

Obviously, we are talking about two significant macro economic environments and all I've read seems to indicate that behaviors have changed in the Great Recession.

Any ideas on how to deal with this?

---

I don't think they are going to change anything, thankfully FICO scores are point in time and baddies fall off and have less impact as time goes on. get back to paying on time, don't pay late and reduce your UTL and write GW lettters and PFD the worst things on your reports. 

Creditors want there customers to be able to handle the unexpected without missing payments and going down hill and basicly make smart credit decisions.  Buit if you don't have a BK or a lot of baddies its possible to get back into prime land in 3-4 years. and there is FP and CapOne and others that will help you get back into the credit world.