FICOs and AUS: We Will Add Your Distinctiveness to Our Collective
Submitted by
CalculatedRisk on Mon, 03/19/2007 - 9:13am (57 comments)
Shopping at Amazon?
Wait! If you click on the Amazon banner link at the top of the page first, HCN will get a percentage of anything you purchase for 24 hours!
Open predictions
Our local book authors
User login
Who's online
There are currently 8 members and 174 guests online.
Currently online
Recent Recommendations
Am I correct in assuming that "borrower-directed" documentation relief is often "broker-directed" documentation relief?
Thanks again, Tanta.
coffee and chocolate? this smells like redbull and no-doze to me !
FICO != financial literacy. That is the faulty assumption I see by lenders. The lenders should still stop people from committing financial suicide instead of letting them based on FICO.
No need to get all long winded about it
I worked on one of those mortgage lending systems for HomeFed back in the day, before the savings and loan crisis took them out. It's so fun to watch the game break down again today...
Our biggest problem was the constant changing of the underwriting guidelines made it very difficult to keep the system up to date - trying to get the updates out to the field was really complicated in the pre-wireless days. I think it would be much easier these days.
I've also worked on credit-scoring systems at other companies, and it's pretty much the same. I really don't think anything beats having an actual, experienced expert who knows what they are looking for to underwrite loans. The automation is really part of the problem, for me. If you're using the tool to help the expert, that's one thing, but when you decide to make the tool the expert, you are asking for trouble.
And we got it. What a surprise.
FICO seems to be one of those things destroyed by measuring it. A guy with a high FICO gets a stretch mortgage and then cannot pay his other debts. His FICO crashes.
Of course, the FICO on the mortgage does not change and so appears to be much better than it is.
Unverified yet but here is this from a mortgage broker:
Indymac Bank announces immediate changes to their 80/20 program, to take effect Tuesday, 3/20/07 (note todays date).
All 80/20s in their pipeline need to be locked no later than today for Indymac Bank to honor this product. This includes all subprime, Alt A, and prime loans that dont fall within the guidelines below.
These loans have top be locked through the Emits system not the Quick Pricer.
Here is a highlight of the changes:
The No Ratio 80/20 will be gone
Full doc will now require a min, credit score of 680 and 3 mos. PITI reserves
Stated will now require a min. score of 700 and 4 mos. of reserves
Maximum loan amount for a 1st mortgage will be $417,000
200% Payment shock requirement will be eliminated
2/6 Libor program will be eliminated
Alternative sources of credit will no longer be acceptable
The line I heard today i thought was funny, "95 is the new 100"! (95% CLTV is the new 100% CLTV)
albrt, you broke the code. And on the first comment, too. Quite impressive.
Donna, I'm sure I would have loved working with you. I was one of the "business experts" sitting at the other end of the conference table, with my head in my hands, while the CEO kept asking you why it took so long to update those systems with the guideline changes, dammit! I would raise my weary head and point out that we were, after all, in this rut of constant guideline changes in part because we rushed through the programming of the original guidelines, because we were in this huge competitive fever to get some of the "new" business from Countrywide (or whomever), and so then we had problems, which required guideline changes, and now you want to rush the system updates, which will mean another rush job in two months . . .
But there you go. Take something like neg am, which we discussed downblog a ways. People have a hell of a time handling that shit. You need a computer to do it. Then you get the computer to do it, and the people don't understand it any more. Plus, the customers aren't computers, so they're in the dark at all times. It's possible that the customers should all get better technology so that they will be worthy of us, but, you know, I wonder if maybe we should just design some products that people--us and them--can understand.
Expired
SEC broad based investigation of subprime.
Tanta sweetie, you give me a PMI lapdance, and I would be pleased to put some coin in your tip-cup.
Bankers issuing neg-am ARMS claim that requiring PMI for loans issued at >80% LTV makes the bullet-proof.
Please tell me the story about how the risk-bearing relationships between the lender, PMI issuer, piggyback/silent second issuer are likely to play out.
What happens in the event the original neg-am ARM is issued at 80LTV (or less), goes to 110% original loan amount, recasts, and the bank forecloses? Would the Bank have required PMI, and in either case, how much do they recover? How does the piggy back/silent second loan factor in all of this?
Recently you noted that a lender had an insurance claim the was rejected. What's the potential hit to the bank's investment portfolio.
PMI doesn't cover 100% - with defects maybe nothing. How shakey are lender's balance sheets?
Does RMBS/CDO pricing provide a good proxy to value Bank loan portfolios?
If these loans haven't been corrctly originally priced, who takes it in the shorts?
Is any of this a concern to you?
So many questions, so little time.
From Roubini :
"
Also, did the originators care or had any interest or incentive to warn the borrower that he/she could not afford such predatory and deceptive mortgages? No way as the originator would get very fat originations fees/commissions. then package this toxic waste into mortgage backed securities, get a nice rating on that garbage from oligopolistic credit rating agencies whose income derived from giving a high rating to this junk under the pretense that tens of thousands of piles of toxic waste would turn by miracle into gourmet food and then sell this securitized toxic junk as if it was fresh roses to even more clueless savvy investors desperately searching for yield and being dumbly ignorant of the risks that they were taking. These investors were not innocent victims; they were blinded by their own search-for-yield greed and did not bother to figure out non-transparent and totally opaque new financial instruments may be toxic waste."
The issue I see as a potential major turning point ahead is the "oligopolistic Credit agencies" and the ratings racket they are running on the CDOs. It appears they were doing some fishy business already with some of their recent upgrades, but was has me the most worried is that they will prolong their risk ratings too long in unrealistically healthy territory, only to eventually be forced to mark the ratings toward an unavoidable reality, which will then massively shift the demand for these instruments and set off the real financial volatility.
Any comments on how this fraud can be maintained if the likely credit tightening cuts demand, pushes up supply, and keeps prices on a declining trend? Id think at this point we are guaranteed a hit to the Alt A chunk of junk which is apparently the other $400b to the $600b of subprime.
Assuming the financial crisis is avoided, is there any way we avoid a recession if housing keeps heading down the tubes? Im trying hard to write a believable non-hard landing scenario for clients and am having an awful lot of trouble thinking one up. Rate cuts dont make much difference to payment shock issues, so that doesnt help. There is not enough upside to falling oil prices. Job growth looks like it will have yet another slice removed from it in coming months. Is there any upside? Given the price to rent and price to income ratios, we are still far away from the bottom in terms of current price reductions stimulating much housing demand.
Hapsburger, I've got PMI on my list for the next Tantawonk post, 'Kay?
I've enjoyed looking at this cartoon every now and then over the last couple of years.
Nicholson Cartoons :
about 95% being the new 100% CLTV.
all it takes is a bit fudjing of the appraised value:
Mortgage Grapevine: CA, $500,000, 731 FICO, 90-100% LTV, Refinance, TTF, stated, OO
I hear you, Tanta. Hey, the food was great, though, huh? All those reps bringing in new goodies every day, and I was pregnant, so I ate whatever I wanted! Actually, I was seven months pregnant before my boss noticed - we all thought was hilarious, that he was so oblivious... I laughed so hard when he suggested we ought to maybe have a "talk"! But then he pissed me off when I came back after maternity leave, complaining I wasn't at the eight o'clock meeting he hadn't bothered to tell me about that he didn't make either, and why wasn't I there in time, in spite of figuring out emergency day care for a sick kid - an hour of yelling at me, and he probably never did figure out why I quit, the clueless jerk.
Ah, those were the days...
I did a lot of work in Artificial Intelligence and expert systems. I know the aim was to "replace" the "expensive" experts with the cheaper, faster computers - but you're right, the problem then is the people using them don't understand what's going on. I have many friends with advanced math and physics degrees doing financial programming these days since the money is good and the math is complicated - I really dread to think what's going to happen when these fancy systems crater. It won't be pretty at all, and will probably make the subprime meltdown look tame in comparison.
And meanwhile the brokers take home huge bonuses that could have been used to keep the real experts in charge and let everyone know when stupidity was taking place. Or maybe that's the idea, get the controls out of the way so everything can be obscured.
But I can't help thinking it will end badly...
Take a read of Freemont's Cease and Desist Order posted on the RE blog at OC Register. Doesn't sound like a credit scoring problem. Are they the only ones?
Give big daddy some sugah.
I'm tingling.
Geoff,
Enron was investment grade until a day or two before it filed. In general, if there is a credit crisis, you can be sure the rating agencies will be the next to last people to discover it, right before the regulatory agencies. The rating agency business model is flawed because they are paid by the issuer and nothing can fix it. My advice is watch CDS prices. That is a market based assessment of risk by people that have skin in the game. They have proven to be much better leading indicators of credit risk - not perfect, but generally very accurate and far more timely than paid to rate crowd.
Donna,
I've also worked on credit-scoring systems at other companies, and it's pretty much the same. I really don't think anything beats having an actual, experienced expert who knows what they are looking for to underwrite loans. The automation is really part of the problem, for me. If you're using the tool to help the expert, that's one thing, but when you decide to make the tool the expert, you are asking for trouble.
Bah, humbug! Looking at it from a non-mortgage, non-prime POV, most of the underwriters I run into are, to put it kindly, thimblewits. I don't believe there's an underwriter alive that can't be outperformed by a simple logistic regression + a decent portfolio history that spans a credit/business cycle. (That said, FICO ain't so hot either. The big flaw is, of course, that it doesn't take loan structure into account at all. But that's easily enough fixed with in-house design.)
Tanta,
Methinks I'm going to have to go into consulting. Granted that applicant FICO by itself is a weak measure of a credit's quality, but that can be worked around easily enough by implementing a matrix. Set FICO as the Y axis and let it determine tiers of pricing, given that it is the closest thing available to a retail credit "lingua franca". Then ration credit and/or adjust pricing based on the (doubtless) more accurate internal models that capture loan structure, etc. Plain and simple, it works.
Brian,
Enron was investment grade until a day or two before it filed. In general, if there is a credit crisis, you can be sure the rating agencies will be the next to last people to discover it, right before the regulatory agencies.
Mostly true, though you need to replace "discover" with "report". Analytic models like the KMV EDF/"distance to default" model were screaming "trouble" a good 3 quarters before the BK. Which makes the current claims in favor of human underwriting for retail credits all the more inexplicable to me. John Henry was a Myth; the steam-drill wins every time, assuming you don't have an idiot for a machine tender....
BG, you ought to go into consulting. Your kind of oversimplistic bravado is lapped up in certain parts of the mortgage world. They'd love you. Me, they get annoyed with, since I tend to try to point out that no, it's not that simple.
Your theory--put the "lingua franca" on the matrix, price that puppy, and presto! on to the next deal--is exactly the kind of thinking that got us into this mess of mortgage mills churning out crap that fit some simple-minded matrix, overinvested in FICO, and thought pricing was just the application of a simple adjustment arithmetic. Plain and simple, it ain't workin'.
But thanks for the stark display of that mindset; I'm afraid sometimes people think I make people like you up.
Not that I'd dare wholly defend BG, but poor originating and underwriting is part of what keeps me employed. As an REO asset manager, I'm left to pick up the pieces of failed M/F loans and assess why those 5-7% rent increases couldn't last forever and why we didn't foresee sewer rates doubling or real estate taxes skyrocketing. I can generally pick out which deals were 4Q deals done to clinch the bonus, which rarely match the write down we take on a deal.
Not that it rests wholly with the underwriter. Quite often the developer with stars in his eyes will provide eight great amenities, all of which are expensive to maintain and so structurally unique that there's no alternate use for the space. The developer who slyly notes that 'this is easy', forgetting that other developers have that very same 'insight'. In the tax credit world, they are ably abetted by the local housing authorities who require everything to be so close to class A (for NIMBY, for developer-as-Santa, for all sorts of reasons) to receive a credit allocation, that, why not build class A?
Needless to say, more and more conversations I have begin like this:
"Hi. I am your new general partner."
REObserver, of course I've met underwriters who are smarter than a box of rocks only by the thinnest of margins. In any profession you have your 50% who are inexplicably below the median.
But, you also have professions that are busy dumbing themselves down, and I thought that was Donna's original point. The only thing more annoying than some Luddite old fart is some silly young puppy who doesn't know what to do when the blue screen comes up. Cheap shots are, well, cheap.
I didn't think my post involved cheap shots. I happen to like credit analysis technology. God knows I don't miss the drudgery they've in large measure eliminated. But I challenge anyone to prove to me that what we are currently seeing in the mortgage world is a "validation" of these models. What, did we intend this shit to fall apart?
Someone has to do the validating; the models don't validate themselves. The Borg part of my post was a joke, but it is true that borrowers, like your developers, do adapt. And investors who told themselves that it must all be OK because S&P's software said so are, I suspect, not looking back on that and laughing right now.
I think that the models will get better, and that many of them will then show what a bad idea some of these mortgage instruments were.
When somebody actually models the process of self-selection into stated income loans, we'll find a huge moral hazard problem that these loans attract people who have no intention of paying these loans if things go south. Why would you refuse to have your income verified and take a higher interest rate (absent some very odd tax issues), unless it was to get as much of the bank's money as you can to make a heads-I-win or tails-bank-loses investment in real estate? (Hey, maybe we should incorporate recent appreciation rates into that model.)
I'm guessing that's a big hole in the existing alt-a mortgage models: No modeling of unobservables that drive both the tendency to default and the tendency to self-select into a stated income loan.
In the end, though, it'd be all too easy to dismiss the financial innovations that have and will enable lenders to better evaluate risk and extend credit to borrowers for a home.
Let's remember that, despite the big junk bond crash of 1987, junk bonds still delivered higher returns over the long run.
Financial innovations often cause short-term speculative bubbles but make long-term contributions.
In the long run, maybe there won't be very many 100% 80-20 loans, but there'll be fixed 80-10 or 80-15's that carry low risks and require small risk premiums. And that'll be good.
Thank you Tanta,I knew you did not make people like this one up.I have worked for and with people who are not able to deal with complex,fluid situations.they have to cook by recipe as well.they are very good at some things,but i sure as shit do not want them pruning my japanese maple.
Keith has it exactly right. It's hard to model Alt-A when the composition of the Alt-A borrower pool keeps changing. To model self-selection into Alt-A you need a lot of information about people, observed before they make their choice. I don't know of a dataset that would explicitly let you do that (they don't call them unobservables for nothing). People will eventually figure out ways to make inferences, but too late for today's hapless CDO owner.
I think the strong desire here was to commoditize things that can't be entirely commoditized. I think that happened for several reasons:
I recognize I'm mixing a few trends here and that this most obviously applies on the CRE side, where I work. The loans I used to get (on behalf of ownership) were a lot more document intensive than some of these no-doc loans. I couldn't begin to count the amount of ridiculous paperwork required to secure a loan, or how many things were just 'check list' items. At the same time, I didn't ask too much about the outstanding LTV/LTC we got, because hey, the bank is taking the risk and I won't be working here in two years anyway. In those too rare circumstances where either an appraiser or a credit risk officer raised a few questions, well, just call up the account rep and have them tell those folks in Charlotte just who they're dealing with.
I deal with M/F assets on a daily basis. I know the submarkets of Dallas and Fort Worth better than I know the aisles of my DC supermarket. I know who's buying in the northwest better than I know my NCAA bracket. I know when to worry about delinquency, how to cut carpet costs, how to goose NOI for the 6 months prior to a refi.
There are plenty of professions and professionals who will take a beating this year. Some more deserved than others, I can promise. There are an awful lot of new loan products coming from Fannie and Freddie, and I know that most of the people pitching them and receiving them don't know how it all works. Just because your macro script says 'Safe' doesn't make it so. Lots of noise will be made about tightening standards and we'll all get down to fundamentals until the next time we're yelled at for not taking more risks.
If developers were a little more realistic, if appraisers weren't so eager to maintain the relationship, if underwriters took a little more time to understand the project, if loan committee asked a few more questions, if investors didn't demand you place all their money...
well, I'd have a pony.
re: Freemont's Cease and Desist Order
It appears that New Century's problems with the Ohio AG's office started when they sent a letters to various state regulators listing the loans that they had made but could not fund. Ohio obviously took exception. However, there is an argument that suing nearly bankrupt lenders ranges somewhere between problematic to counter-productive.
It does highlight the risk to the lenders of even a momentary liquidity problems.
"But I challenge anyone to prove to me that what we are currently seeing in the mortgage world is a "validation" of these models. What, did we intend this shit to fall apart?"
I don't think what we are seeing validates AU anymore than it validates raspy-voiced, old coffee smelling underwriters with nicotine stained fingers.
BG said:
Bah, humbug! Looking at it from a non-mortgage, non-prime POV, most of the underwriters I run into are, to put it kindly, thimblewits. I don't believe there's an underwriter alive that can't be outperformed by a simple logistic regression + a decent portfolio history that spans a credit/business cycle. (That said, FICO ain't so hot either. The big flaw is, of course, that it doesn't take loan structure into account at all. But that's easily enough fixed with in-house design.)
For God's sake, what do think the major investment houses have been doing for the last five years?
Simple (and logistic) regressions were so 1980s. When I left the business over a year ago, it was about high-dimensional adaptive clustering algorithms, compute farms estimating derivatives against S&P Levels and Intex code, and reversible models that could use emergent models plus bond salespeoples' input to back out the daily (realtime) rate sheets.
Lovely business, but naive statisticians don't have much to add.
I remember when an AVM (automated valuation model) was the flavor of the month. That lasted a NY minute.
For all the sophisticated multi linear regression analysis and 'high-dimensional adaptive clustering algorithms', it boils down to ...
garbage in - garbage out
On high value loans that are high risk, lenders will often order two field appraisals AND an AVM. It's not uncommon for there to be a 30% variance between the three. Its hard to tell what the collateral value is in many cases.
Only in the more conforming data rich markets do you see consistent valuations.
I see quite a few review appraisals on a daily basis that vary significantly from the origination appraisal. (and these are mostly audit reviews with no pretext of potential fraud) ???
This conversation about automated underwriting, predictive modeling, analytics, data mining is interesting;
I do predictive modelling - not in mortgages - the closest I came to that was P&C insurance - The same issues bedevil that field - the model has to be understandable, people said ( try explaining Support Vector Machines ( SVM) ) to people; also the data was noisy( garbage in garbage out).
Oddly enough, simple binning algorithms worked very well and people understand the concept of bucketing easily so underwriter acceptance wasn't a problem.
But, it always left me rather deflated - I did histogram analysis when I was 11 years old - things hadn't progressed much further it seemed.
-K
The exchange between Tanta and BG makes the important point that predictive models, no matter how they're built, tend to perform significantly worse, if not downright badly, when applied to data significantly different from the development sample. Folks who make their living building prepayment and default models (like me) would agree wholeheartedly with this point.
Most models at this point attempt to address the adverse selection and "unobserved heterogeneity" noted by Tanta, but of course the technique are just proxying for the detailed credit report data and underwriting information that may be unavailable.
However, I feel like the Tanta/BG exchange misses the biggest issue - unrealistic expectations about house prices. All major mortgage modeling tools either implicitly or explicitly make assumptions about future house prices. Any good underwriter must do the same thing. To Admn Revwr's point, GIGO applies to the estimated current house price, AND to estimated future prices.
I think the real problem is that 10-15 years of high house price appreciation created very unrealistic expections on the part of underwriters, bankers, investors, etc. Slowly but surely, "traditional" underwriting limitations have been relaxed, under the assumption that rising house prices would hide these sins. The logical end-game is the outright fraud that we're starting to see.
In that sense, I'd argue that FICO is not "over-weighted". Along with the loan-to-value ratio, FICO remains one of the best predictors of mortgage default. Regardless of how these and other variables are weighted, however, the projections will be terrible if folks assume 4+% appreciation in what's turning out to be a 0% appreciation world (at least in the near term).
Keep up the good work CR/Tanta! It's a pleasure to read such well-informed and thoughtful commentary.
AR -
It's not garbage in, garbage out, it's garbage in, high-quality multi-tranche asset backed securities out. Haven't you been paying attention?
The models aren't trying to predict anything other than your ability to move the risk to the bond buyers for the best price. That's what all these articles are about.
If you want to know about individual default risk, move to a small town, open an S&L, go to church every week.
I have the attention span of a gnat or a goldfish.
This article was too long.
Couldn't read it. Never will.
In that sense, I'd argue that FICO is not "over-weighted". Along with the loan-to-value ratio, FICO remains one of the best predictors of mortgage default.
Kyle, it's possible that you and I are saying the same thing, but that I'm using the wrong words for it. It wouldn't be the first time. I'm trying to write for a general audience of non-experts, so I do oversimplify. And I am not a mathematician, so I can be as egregious about certain terms as the next policy wonk.
Certainly you can consider FICO and LTV to be the two best predictors. But no model I know of treats them either 1) independently or 2) only as a threshold. What I mean is that the model allows high FICO to offset high LTV, and low FICO to be offset by low LTV, and so on; it works similarly with other characteristics. Also, there is a threshold level somewhere, usually, but it's very low: a model might have an absolute FICO minimum of 620 if it's dealing with prime loans. So having the "minimum required FICO" for the "maximum allowable LTV" just gets you into the ring, it doesn't mean you win the fight (BG just thinks it does). As the actual FICO on a given loan rises, the "thresholds" on other required characteristics relax. That is both perfectly commonsensical and occasionally quite problematic. When I say "overinvested," what I guess I really mean is that FICO is allowed to offset too much other stuff, particularly DTI, but also LTV, on an increment-per-increment basis (that is, the point increase in FICO translates into too much acceptable increase in LTV). I do not think the problem is adequately dealt with by calculating a default probability for a given FICO score, a default probability for a given LTV, and then accepting the "average" on a given loan. That, in fact, is one of those "cracks" where adverse self-selection can flourish. That gets to Keith's point that borrowers with a certain FICO can "elect" to overleverage, because the way the system weights FICOs lets them do it.
LP and DU do take into account HPA, insofar as they have internal AVMs. Admn Revwr is an excellent example of an expert human being "validating" those AVMs and coming up concerned about them. It's actually a lot easier to "game" an appraised value than it is a FICO, so people get back to thinking of FICO as the more reliable value. But a given FICO borrower will "perform" like a lower FICO borrower when HPA gets into negative territory, other things being equal.
I just got a UBS report showing 60+ DLQ in Alt-A pools. The low-doc purchase loans were running a DLQ rate of 2.16 at an average FICO of 718. The full doc purchase loans were running at 0.91, with average FICO of 721. The DTIs and LTVs were nearly identical, but slightly higher for the better performing loans. (The big difference was the presence of a second lien, but not total CLTV, interestingly.) This is only one data point, but put together with quite a few other ones, it suggests that "doc level" is more predictive than we thought, and what it predicts is adverse self-selection, which in turn doesn't correlate with FICO in the way we thought.
Back to how FICOs are, actually predictive of default: they are, in mortgages, to the extent that entities with huge databases spanning the required time periods--i.e., prime outfits like the GSEs and the MIs--were able to "calibrate" them to loan performance. Fair, Isaacs didn't come out with the claim that 620 is a good cutoff for a mortgage loan, the GSEs did. They did so by looking at the relationship of their loan porfolio performance to the reported FICOs over time. So far, so good, until you introduce "non-GSE" lending practices. So we're starting to get some data on performance of "Alt" loans, and I think it means we have to "re-calibrate" FICO to the Alt performance book specifically. Then they can "predict" as well for Alt as they have for prime conforming.
OK, that last sentence above didn't say what I think I wanted it to say. It's early.
What I'm suggesting is that FICO might not predict for Alt in the way it does for prime, not just that it might require a new set of buckets. With too much other risk layering, you may have signal, not noise, in such characteristics as the presence of subordinate financing, for instance. I fail to see why, a priori, we have to keep insisting that the higher the FICO the more prudent the borrower. In the past, when FICOs were being statistically measured, the lender was prudent for the borrower. I just didn't let you do anything really stupid, regardless of your FICO. We've come to some place where we've decided that past performance guarantees future results, by allowing the borrower with the past performance (high FICO) to define his or her own risk tolerance. "Rational agent" theory falls apart here; I'm getting borrowers who want high-LTV Option ARMs because they watch too much television. In other words, the information set underlying the past behavior patterns that produced the current FICO snapshot is changing. I will not assume, a priori, that that can't change the predictive value of FICOs.
If that didn't make any sense, I'll just excuse myself and make another pot of coffee.
Tanta,
I'm loving your posts. I don't have a house or mortgage (yet), and so you've been a fascinating introduction into the workings of these critters. (Your recent Option ARM tutorial was truly frightening.)
Anyway, there was a news article that was brought back to mind by your post. Apparently, those in-car navigation systems have led to some accidents, when the nav system tells the driver to do something and they do it, with no heed of the actual surroundings. One person actually managed to drive into a river this way.
It's not unlike the originators you spoke of, whose instincts almost certainly are telling them that the loan is problematic, but the computer says it's good! And who are they to argue with the infallible computer, right? Next stop: the bottom of the river.
Tanta, this whole FICO scoring thing reminds me a lot about SAT scores.
The SAT is an aptitude test; a standardized test designed to measure the ability of a person to develop skills or acquire knowledge. Students are admitted to university in large part because of their aptitude scores. With the premise being that students with higher scores will develop more skills and acquire more knowledge.
FICO is a 'test' also. I would argue that FICO scores do not measure aptitude. It is more a measure of your achievements. ie. Do you have multiple tradelines? Do you make payments in a timely fashion? Do you maintain low monthly balances? If you have all those 'achievements' you'll score high. If not you score low.
Were we agree to disagree, is the predictive value of FICO scores. Its my opinion that aptitude tests have greater predictive power than achievement test. But how do you determine someone's aptitude to borrow? Absent a reliable method of determining borrowing aptitude, borrower 'achievement' seems like a reasonable substitute.
BrooklynInDaHouse, that's an interesting comparison. You're suggesting, I think, that FICO is more like high school grades (a measure of "achievement") than like SATs (a measure of "aptitude").
I still have some fairly clear memories of being a new college student, and then again a new grad student, and witnessing some of my fellows in major melt-down mode over the "size of fish relative to size of pond" problem. Students who got As in high school automatically expected to get As in college, forgetting that colleges take overrrepresented samples of A students, and then maintain their own curve, so someobody who got As in high school has to get a C in college (unless you want to unleash rampant grade inflation, which some of us think has happened). Same thing in grad school. Each time we take the biggest fish out of the smaller pond and put it into a bigger pond with a bunch of big fish from other small ponds, the curve changes.
You could consider a certain class of Alt-A borrowers to be rather like those students who got As in high school--they did fine when they took lower-balance conforming mortgages with conservative terms--but who got Cs (or worse) when they got to college (big loans, high-risk terms).
You're right that it's hard, then, to come up with an equivalent of the SAT to replace grades as the predictive factor, if what you want is a predictive factor as easily quantified as an SAT score. I'm suggesting, I guess, that the totality of borrower behavior, including the request for no-down, no-doc loan, is a datum that can measure aptitude. Shorthand: ask me to give you a very high-risk loan, while asking me not to verify your ability to carry it, and I judge your risk-analysis skills accordingly, and therefore hold you to a much higher standard (FICO or anything else) than I would someone else.
I keep thinking of an alternative medical analogy: you and I both have a headache. You go to the doctor, submit to a full exam and usual tests, and then you get a prescription for painkillers. I call the nurse and try to get her to call in an Rx for Vicodin without having to come in to the office. I might (I hope) get told to go buy some Advil. I might be just as "deserving" of narcotics as you are, but my behavior doesn't suggest that.
Tanta:
I love your UberNerd segments. As a MBS investor, I've learned a lot about how to analyze the securities I buy. Could you someday do an UberNerd piece on how to interpret Fannie/Freddie's commitment rates? I've heard different stories from different people and thought you'd be the one to give a real answer.
Thanks in advance and keep up the great work.
Thanks, Tom. I'll see what I can do. I have to warn you that I the last time I had to price loans for an agency MBS, Mr. Clinton was president, Windows 98 was just a rumor, and I could still outrun my interns as well as out-think them. I've spent more time in the last several years on credit policy. But after I do what I promised with MI, I'll try to address your question.
Is it fair to say that it is difficult to create software that will tell you when to make a loan that shouldn't be made?
Tanta.
Love your posts.
If nothing more important comes up, I think a lot of people would enjoy a little overview of the definition of ALT-A, especially since ALT-A is being characterized as "next". (or perhaps you have done one before and CR can post it up?)
As one who came from the school that 'knew' ALT-A meant "prime credit & difficult to prove income" I find myself newly confused mainly because everyone has changed the meaning. So now I'm having to not only understand what ALT-A is, but to understand how each market participant defines ALT-A, so as to decipher all the PRs and reports. Some reporters think it's a credit quality between prime and subprime. Some people refer to option ARM portfolios as a sub-segment of ALT-A... and then there's ALT-B... and furthermore I am still confused by the relationship between ALT-A and doc types. Some people simply equate ALT-A with stated. For example: say a prime credit borrower gets a stated income 2/28 from a subprime lender, is that a subprime loan or ALT-A loan? And also, when they say "stated loans" or "stated income loans", do they usually mean SISA or could it be SINA/SIVA too?
Thank you again.
Regards.
Journeyman,
I was involved as a technologist in AUS (artificial intelligence and expert systems) at both Fannie and Freddie in the early days when they were creating LP and DU.
We would have had no problem then developing systems flagging loans not to buy (assuming rich enough data sets) and I am sure that remains the case today. The bigger issue is that for the last decade or so, most orgs fielding AUs (or more commonly today, automated pricing) almost assume that ALL loans should be approved; it is just a matter of finding the right price.
So again, garbage in/garbage out. But in most cases today, the garbage going in is the set of policies adopted by the lending institution.
that for the last decade or so, most orgs fielding AUs (or more commonly today, automated pricing) almost assume that ALL loans should be approved; it is just a matter of finding the right price.
Right. But that can only make sense if it is assumed that all potential loan scenarios can be given a reliable default prediction, since you have to have that to find the "right price." You cannot price what you cannot underwrite, as much as you cannot underwrite what you cannot price. At least in my part of the universe, underwriting was never intended to eliminate default risk; it was meant to keep it affordable. That is, it was supposed to try to match default frequency control (creditworthiness of the borrower) with loss severity control (liquidation value of the collateral) such that whatever defaults you ended up with did not consume all your profits off of mortgage lending. So "pricing the risk" was always there, even in the old days. As far as I'm concerned, what has changed is not our willingness to price risk, but our willingness to "pay up" for loss frequency control (FICO) while not "paying down" for the lack of loss severity control (low LTVs and honest appraisals).
Go to CreditCRM - Credit Repair Business Software for a hoot! The website has a video describing how it can improve your customers' credit scores, and has a New Century banner in the background!
Fannie Mae, by the way, has it's own internal mortgage credit score, and FICO is only one of many variables.
Thanks for your replies, Tanta!
Geeks like us gotta stick together.
Ficus, you might think that you don't have a house (yet) but you are definitely buying one (or several) with your tax dollars!
PZjYB8 qkprsjhwpvgq, [url=http://zsybfvxzddrv.com/]zsybfvxzddrv[/url], [link=http://oooysretnoby.com/]oooysretnoby[/link], http://gbrbomgmjzus.com/
Look it. , hairy sexy milfs + cumshot, [url="http://prettymilf.hottubecam.com/hairy-sexy-milfs-+-cumshot.html"]hairy sexy milfs + cumshot[/url], http://prettymilf.hottubecam.com/hairy-sexy-milfs-+-cumshot.html hairy sexy milfs + cumshot, milf hunter trixie, [url="http://prettymilf.hottubecam.com/busty-milfs-fucling-hard.html"]milf hunter trixie[/url], http://prettymilf.hottubecam.com/busty-milfs-fucling-hard.html milf hunter trixie, busty mature moms, [url="http://prettymilf.hottubecam.com/busty-mature-moms.html"]busty mature moms[/url], http://prettymilf.hottubecam.com/busty-mature-moms.html busty mature moms,
My submissions: , coumadin foods, aldactone vitamin d deficiency, caduet, cardizem interaction with cefixime,
He rised. , atacand generic, avodart pde5 flaccid erections, aleve taken with metformin symbestin, albuterol atrovent, augmentin 875 mg tablet, blood pressure medication cartia, losing weight on alesse, caduet discount, what time of day should i take clomid, aldactone and hair loss, casodex oral tablet 50 mg, avelox, asacol and weight gain,
Good time. And , how much would 5 aleve raise potassium levels,
Good time. And , levoxyl no prescription, keflex dosage, imuran reviews what people say,
Welcome friends! , cupons for doryx,
Must have: , omnicef susp, alternative to pentasa, procardia for raynaud's disease,