Vectors
Predictive Modeling companies attempt to create projections for the future based on various economic scenarios. There are quite a few companies producing these vectors and they have a wide variety of uses for financial institutions.
Inputs to vectors.
Some of the standard inputs for producing vectors are:
*Home Price projections
*Unemployment projections
*Forward Interest Rates
*Property Valuations
*Borrower Credit Profiles
*Loan Characteristics and related historical data
Uncertainty Principles and Quantum Effects
Questions from various market participants arise as to how “granular” should these items be. For example, should we create projections only at the national level or drill down to state or county, even zip code projections? Similarly, with unemployment projections, should we do this at the state or county level or are national level statistics sufficient? Arguments for and against each point of view exist. One view is that doing it at too granular a level gives way too many inputs to take in, and by focusing so intently on the “micro” level, you’re losing sight of the forest. Another opinion is that too much of a macro view gives one not enough insight into what’s happening at the detail level, and you’re losing touch with reality.
Another way to think of this is with a “quantum world” view – where when you attempt to determine the exact position of a particle, you cannot determine its exact speed. If you determine the exact speed, you cannot find its exact position. By drilling down to zip code + 4 unemployment data – you might lose sight of more macro trends that would impact that area. Yet, by focusing on, say, too wide a range of home price indices, to use another example, you tend not to see the actuality of what’s happening with properties relevant to specific RMBS of interest. It’s perhaps a bit of a stretch to apply this to the same quantum effects, but maybe not. Certainly the question arises as to how much granularity is sufficient. When do you need more detail and when do you not?
One vendor I spoke with recently has even gone so far as to produce the property address of homes backing RMBS, but apparently you can only actually SEE the property address for yourself, if you sign a document stating that you will not then go looking at the borrower credit profile (such as from Equifax, Experian, or TransUnion). Because RMBS loans are “anonymous” (aka “de-identified) in the sense that you don’t know who the borrower is, nor do you know the property address, then solving the problem of “where is the property located exactly so that the most granular level of property valuation can be performed” can be highly valuable indeed. Here again, though, we have this almost “quantum” oddity of “being able to determine the exact location of a property, but not then being able to determine the credit profile of the borrower who owns the property.
Of course, if a property is now REO, then the current owner of the loan IS the owner of the property and no credit profile is needed particularly. In the case of RMBS, the Trust itself has become the “owner” of the property. With foreclosures and REO at such high percentages of deals, then the need lessens for credit profiles of the original borrowers of these particular loans, as these original borrowers have been evicted and no longer have any rights to the property itself.
Surely, I’m just imagining this “quantum effect.” It can’t possibly apply to finance… Enough with all this uncertainty!
So what is the way forward here? We maintain that thinking things through in the above manner leaves one without any really defensible viewpoints. How about we go forward and use a “results-driven” approach? In this approach we try a wide variety of approaches – trying each one of them under a wide-variety of levels of granularity. Don’t stick too much on any one approach, but then save these predictions for ALL of these variations. Then each month look at the actualities of what occurred in the real world and see which approaches most closely approximated what was found in the real world. Do this month after month and don’t develop any particular prejudices. In other words, constantly be on the alert as to which approaches produce the most practical real-world results. Perhaps then, a pattern may emerge as to which “solution” fits best.
Hopefully, then, we won’t have a situation where the “solution” itself only “resolves itself” when we observe it closely; but maybe the next time we observe it, it’ll be a different answer – just like, quantumly speaking, when a particle is observed, it’s location cannot be determined.
One thing is for sure, you want your predictive modeling company to be able to show you what their predictions were at various points in time (without the benefit of 20-20 hindsight) and have them show you how did their predictions do. Any predictive modeling company worth their salt for their crystal ball techniques should be able to show you how their predictions performed. We’re not saying they should be 100% perfect in all their predictions under all circumstances, but they should be able to show you exactly how they did – unless of course, they’re embarrassed to show you how badly they did.
In any case, if you want to get information on “the exact property address and home valuation”, check out a vendor which provides a very interesting solution as regards to home property valuations matched against the anonymous loan-level securitized data. See Lewtan’s ABSNet Home Val ™ solution here:
http://www.lewtan.com/products/ABSNEThomeval.html
Have a nice day. See you next observation – maybe.
Quark Out!
No comments:
Post a Comment