Thu, 21 Feb 2008
Thomas Brown has an interesting article at bankstocks.com which indicates that US mortgage defaults might not be driven by low FICO scores at all. Brown compares the various mortgage pools underlying the widely tracked ABX index and shows that in many cases, the worst performing pools have higher FICO scores than the best performing pools. In some cases, the average FICO scores of the worst pools are around 640 so that they do not even count as subprime according to the standard definition. Until last year, subprime was generally defined as FICO scores below 620, but after the subprime turmoil began, some lenders have raised the cutoff to as high as 680.
In some sense, this is not surprising. FICO scores are not about creditworthiness as measured by income, assets or cash flows. They are simply about past credit histories. If a household with an impeccable credit history is given a housing loan that is well beyond what it can afford in terms of its income levels, the FICO score would be excellent, but the default risk would be high.
Brown’s data shows a clear pattern where pools originated by some lenders like Wells Fargo have the lowest default rates while those originated by other lenders like WMC have the highest default rates. Geography matters too – the worst pools have high exposure to some of the most frothy housing markets of 2006.
Brown interprets his data to mean that mortgage losses are not likely to be as high as feared because many subprime mortgages will not default. While I grant this, an equally valid conclusion from the data is that many non subprime mortgages will default because the trajectory of housing prices matters more than FICO scores.
All this has implications for India where many hopes are being pinned on the creation of credit registries and similar agencies that would make FICO-like scores possible in India as well. In a country where recovery is even harder than in the United States, excessive reliance on credit histories might not be such a good idea at all. The smartest crooks will build excellent credit histories with a series of small loans until they can take out a large loan. Long ago in the United States, the term Brazilian straddle was used to describe a huge market position which the trader intended to run away from if it proved unprofitable (the other leg of the straddle was supposed to be a one way air ticket to Brazil). What the equivalent straddle should be called in India is left to the imagination of the reader. I would confine myself to the observation that credit histories provide little protection against such straddles.
ANonReg wrote on Sat, 23 Feb 2008 11:08
Re: FICO scores and subprime defaults
FICO scores are due for an overhaul, see recent article in BW. While loans were made relying on FICO scores, the validity of FICO to predict crediworthiness is now coming into question. Goes to show that blindly transacting loans based on a secretive formula is a bad bet... whatever happened to good old underwriting, which was prevalent a few years ago?
George Thomas wrote on Tue, 08 Apr 2008 10:27
Re: FICO scores and subprime defaults
I agree with your point that FICO/credit history is not a substitute for having a rigorous & conservative underwriting discipline. But I would respectfully disagree with your perspective that this is evidence that the credit bureau infrastructure is therefore less valuable (esp for India/other countries, where recovery rates are low).
1) The mortgage tranche data cited by Tom Brown is evidence of the incremental value of "full doc" type assessment in addition to FICO. But I am willing to wager that the losses would still rank order based on FICO within each orginator or alternately controlling for the "incremental" underwriting variables.(eg Full Doc).
2) FICO/credit history is predominantly a measure of the "willingness to pay" or "personal behavior" rather than "ability to pay", which is more prospective based on asset values/cash flows. So it is a complementary piece to what is available to Indian lenders today. A philosophical level, risk should be evaluated holistically on both these dimensions.
3) Credit History represents a credible solution to the moral hazard problem of the "willful defaulters", as a poor score results in less access to credit and higher effective rates and conversely rewards good behavior over the longer term. Also since it builds up over time (weighted), there is a certain amount of "reputation value" that is created by a good history, which the individual would not want to lose, unless there is a material impairment in their "ability to pay".
4) To your point about the possibility of "gaming" the history/score to score the "hit", I agree that it is a cautionary tale but can suggest two ways to mitigate
a) Cap the weight in the scoring algorithm for credit history relative to asset/cash flow measures/criteria
b) Change a) to reduce the weight of credit history for large ticket loans. In general FICO/any statistical algorithm is best used for low and medium size loans, not for commercial or even bigger small business type loans.
Net-net, I think that creating the credit bureau infrastructure in India will expand the ability to extend retail credit (particularly unsecured) while effectively containing credit costs. I also think that a similar concept will work for car insurance, where the driving records drive insurance premiums. (used in US to price auto insurance)
Prof. Jayanth R. Varma wrote on Tue, 08 Apr 2008 10:34
Re: Re: FICO scores and subprime defaults
Sure, credit histories do contain incremental information and are therefore valuable. I am only cautioning against excessive reliance on them.
Also I think, willingness to pay is partly dependent on ability to pay and more importantly incentive to pay. For example, mortgage defaults depend on whether the home equity is positive or negative. Credit history is therefore only an imperfect measure of what traditional bankers called "character", it is partly contaminated by "collateral" and partly by "cash flows" both of which have cyclical biases.
Its deficiencies must be recognized while using it.