01
Oct
Posted by Tasso Argyros in Analytics on October 1, 2008

Perhaps if you’ve got a hammer, everything looks like a nail, but it does strike us here at Aster that one of the underlying causes of the liquidity crisis is a problem with data and analytics. After all, the reason that banks don’t want to lend to each other anymore is that they don’t trust that the other banks really know the value of their mortgage-backed securities (MBS) – mainly because they themselves don’t trust the value of their own.

So why is this? Why is it hard for banks to understand the value of a particular mortgage backed security? Simple – a bank holding those MBS’s doesn’t have access to the granular data on the underlying individual assets (the mortgages, with their underlying properties and payees) that make up the pooled asset. Simple questions such as “In what zip codes are the properties backing these mortgages, and by what percentage have median property prices changed since the mortgage was issued?� can’t easily be answered. In other words, unlike stocks, for which you have easy access to a company’s full financial statements to value, the underlying value of an MBS is difficult to figure out because the data needed to price the underlying asset simply is unavailable.

The reason for this is that traditional MBS modeling has focused much more on the behavior of those securities under varying interest rate scenarios because, historically, the ratio of pre-payment vs. held-to-term of those mortgages has been the big swing factor in how much those securities were worth. When interest rates fall, people are more likely to pay off or refinance their existing mortgage, and when interest rates rise, people are more likely to hold onto their existing (now lower interest rate) mortgage. The default rate on the underlying mortgage was traditionally taken as a given, and not predictively modeled at all.

People sometimes ask us – what is the price for not implementing a world-class analytics capability. I think now we can answer, “$700 billion, and counting.” What has emerged to be crucial in this crisis are two pieces of data that are not tracked by standard models: 1) Is the current property value above or below the value of the mortgage(s) on the property, and 2) Can a mortgage holder afford his or her mortgage payments?

This has to change. If mortgages are going to continue to be securitized in the future, it will be necessary to have valuation models that track details on the underlying real assets (specifically, the property values, and the owners’ ability to pay) on at least a monthly, if not more frequent basis.  We would propose a national centralized registry for the mortgage securitization industry that matches property locations to mortgage pools, and provides a number of valuation models for those properties (analogous to Zillow’s property valuation model). It would also provide ongoing tracking of either a FICO score or more granular credit quality information for those mortgage holders.  In addition – and this is crucial - a centralized repository of the detailed data disclosed by the mortgage applicant in the original approved mortgage application needs to be retained in this registry. On top of this shared database, financial services companies would be free to create valuation and pricing models based on whatever predictive drivers they belief influence the default and payment rates.

Obviously such a centralized repository would require bullet-proof security and privacy handling, but such a registry needs to be established if we’re not going to fall into this mess all over again. Similar registries would be appropriate for student loan pools and other securitized loan categories such as auto-loans. Is anyone else thinking that this is all a matter of data?

Bookmark and Share

Comments:
Jeff on October 2nd, 2008 at 10:11 am #

Hey Tasso,

Having priced mortgage-backed securities at a once-venerable Wall Street institution, I can tell you that your statement is not true. There have been per-mortgage prepayment models running on the Street for nearly ten years now. While they certainly could be augmented with additional data per household, as you mention above, things like FICO scores are a standard feature in many prepayment predictive models that exist today.

Regards,
Jeff

Tasso Argyros on October 2nd, 2008 at 4:48 pm #

Hello Jeff,

You mention that you were used to modeling the prepayment rate based on fundamental inputs — that wasn’t really our point, but it’s good to know at least some fundamentals have been used on these securities. My observation was based on a conversation with an MBS modeler who’s currently working on valuation (not pricing) for his bank’s MBS portfolio. His bank’s approach is to use an arbitrage-free pricing model where the underlying mortgage pool (usually 5-10k individual mortgages) is treated as roughly homogeneous with an average default rate, pre-payment rate and correlation. The average default rate is an exogenous variable in the model — our point, was maybe that default rate could be endogenously modeled if the data were available to do so.

Jeff on October 2nd, 2008 at 7:54 pm #

Yo Tasso,

Complex words for complex times. I guess I should have been more clear: the prepayment modeling team also predicts defaults, as that’s just the other side of prepayment. I’m surprised that the modeler you were speaking with did not have a model to predict both the prepayment and the default rate, though I am sure they were using copulas to model the correlation. Most mortgage desks have a prepayment team and a valuation (pricing?) team. The prepayment team worries about prepayment and credit risk, while the valuation team adds interest rate risk to the pricing model. Note that I am using “pricing” and “valuation” interchangeably; I’m not clear on what distinction you are drawing between the two.

I’m also not clear on your use of “endogenous” versus “exogenous”: if you’re predicting the prepayment and default rate from aggregate metrics of the mortgage pool and its underlying collateral, and those rates then serve as parameters in the valuation model, does that not make the prepayment and default rate endogenous to the model?

In addition, often loan-level detail is available. For nonagency loans, Bear Stearns was looking at loan-level prepayment modeling for nonagency mortgage pools in 1996: http://findarticles.com/p/articles/mi_m0EIN/is_1996_Feb_29/ai_18037602. I was under the impression that the agencies were starting to release more loan-level detail for many of their securitized products (http://www.freddiemac.com/news/archives/mbs/2006/20060630_loanlevel.html), though I could be wrong, as I haven’t been on the street for a few years.

Most financial services firms keep large databases of loan-level prepayment and default information; they don’t put this into a centralized repository, as they view this data as a competitive advantage for model building. It’s the same scenario as exists for behavioral targeting companies today; perhaps a BlueKai for MBS would instantiate the centralized database you propose!

Either way, thanks for highlighting the esoteric yet suddenly critical corner of the world that is MBS pricing.

Later,
Jeff

[...] Big Data and the Financial Meltdown Growing Your Business with Frontline Data Warehouses » Aster nCluster 3.0: Aligning [...]

Philip on October 17th, 2008 at 9:28 am #

Tasso, Your comment is correct that the valuation models of loan pools, and securities backed by loan pools, were based on too thin a data set.

Our company has created a solution that is based upon the rigorous automated underwriting processes that should have occurred at origination (but which did not in the case of many of the “toxic loans”). Send me a note if you would like to discuss.

with kind regards,

Philip

BusinessRx Reading List : The Software Cure For A Financial Meltdown on January 21st, 2009 at 11:10 pm #

[...] “…the reason that banks don’t want to lend to each other anymore is that they don’t trust that the other banks really know the value of their mortgage-backed securities … because they themselves don’t trust the value of their own.� – Aster Data [...]

Jason on March 2nd, 2009 at 11:30 am #

One stupid thing about these securities is the fact that they could make it and sell it to others without holding any responsibility on them. I think if they were forced to keep these securities on their balance sheets, they would’ve been more responsible and conservative.

Post a comment
Name: 
Email: 
URL: 
Comments: