April 19, 2009

The Guns of Singapore (Cont.)

As I've repeatedly noted, America has a huge data-collecting and data-interpreting infrastructure serving a political and media juggernaut designed to make sure that minorities get plenty of mortgages. In contrast, we have virtually no infrastructure to let us know whether minorities are getting too many mortgages to pay them back.

Our system for making sure minorities get enough lending is like driving a car rigged up with a complex safety system that warns you loudly and incessantly to make sure you don't drift too far in your lane to the left, and that makes it hard to turn to the steering wheel to the left, but totally ignores any drifting you do to the right. Not surprisingly, you are eventually going to wind up in a ditch on the right.

So, social scientists are still debating last century's question -- are minorities being discriminated against by lenders? -- while largely ignoring this century's big questions about minorities and mortgages.

Here's a new study from economists at the New York Fed on the question that the media has been much more up in arms over: did minorities get charged higher mortgage interest rates? You've read plenty of articles complaining that minorities who deserved prime mortgages were selectively discriminated against into getting subprime mortgages. Are they true?

A new study by economists at the New York Federal Reserve says "No."
Subprime Mortgage Pricing: The Impact of Race, Ethnicity, and Gender on the Cost of Borrowing
Andrew Haughwout, Christopher Mayer, and Joseph Tracy

Some observers have argued that minority borrowers and neighborhoods were targeted for expensive credit in 2004-06, the peak period for subprime lending. To investigate this claim, we take advantage of a new data set that merges demographic information on subprime borrowers with information on the mortgages they took out. In a sample of more than 75,000 adjustable-rate mortgages, we find no evidence of adverse pricing by race, ethnicity, or gender in either the initial rate or the reset margin. Indeed, if any pricing differential exists, minority borrowers appear to pay slightly lower rates, as do those borrowers in Zip codes with a larger percentage of black or Hispanic residents or a higher unemployment rate. Mortgage rates are also lower in locations that previously had higher rates of house price appreciation. These results suggest some economies of scale in subprime lending. Yet there are important caveats: we are unable to measure points and fees at loan origination, and the data do not indicate whether borrowers might have qualified for less expensive conforming mortgages.

They looked at 75,000 subprime loans with 2-year teaser rates handed across the country out in August 2005 for which they were able to match the race/ethnicity info in the federal government's Home Mortgage Disclosure Act database with the expensive creditworthiness data from LoanPerformance, which is considered the gold standard of mortgage creditworthiness. With this breakthrough in data, they found:
"In contrast to previous findings, our results show that if anything, minority borrowers get slightly favorable terms, although the size of these effects are quite small."

Bloomberg reports on the study:

The authors said they were not able to determine whether minority borrowers were forced to pay higher upfront costs in points or fees, or whether the subprime borrowers might have qualified for less expensive conforming loans.

From what we know about car-buying -- blacks tend to be poorer negotiators, perhaps because they don't want to look like cheapskates to the salesman, perhaps because they have less information -- I would have thought it's inevitable that blacks and perhaps Hispanics would, on average, be more likely to get stuck with loan terms less favorable than, say, Armenians or Koreans would manage to extract. I mean, if you were Angelo Mozilo of Countrywide or Roland Arnall of Ameriquest or Bill Cook of of ConquistAmerica, why wouldn't you hire a whole bunch of Spanish-speaking mortgage peddlers to pursue the burgeoning Uneducated and Innumerate Latino Market?

On the other hand, there are vast government and NGO resources devoted to fighting discrimination against legally-protected minorities. So, maybe they just tried even harder to market subprime loans to dumb whites to make their race numbers balance out. Who knows?

As Peter Brimelow pointed out in 1992 after the inane Boston Fed study about mortgage discrimination kicked off this whole catastrophic cycle, the researchers shouldn't be focusing on the inputs (e.g., FICO scores) but on the outputs (i.e., default rates).

The claim that blacks and Hispanics took out so many subprime loans because they were discriminated against when they were really highly creditworthy borrowers who should have been getting prime loans is, in effect, claiming that black and Hispanic subprime borrowers should have been, on average, more creditworthy than white borrowers because a lot of minorities of prime creditworthiness had been forced down into the subprime ranks by The Man.

This prevalent theory is hard to reconcile with the facts of higher default rates among blacks and Hispanics.

Obviously, the gigantic default losses in heavily minority regions such as Southern California's Inland Empire (where about 9% of all defaulted dollars lost in the country went down the drain) suggest that subprime borrowers weren't charged enough interest to make up for the risk. But, we're not supposed to talk about that.

We know from the Boston Fed's recent study of subprime foreclosures in Massachusetts that foreclosure rates have long been significantly higher for blacks and Hispanics, at least in that state:

Similarly, FHA loans, which served, more conservatively, the least creditworthy, saw much higher default rates among blacks and Hispanics in the 1990s, when there wasn't a housing bubble, suggesting that we are dealing with a deeper pattern extending back before the recent housing follies.

So, either high quality black and Hispanic borrowers were being forced into the subprime category, as the conventional wisdom holds, which implies that blacks and Hispanics should have had lower default rates. Or, the NY Fed economists were right, and the system worked in a colorblind fashion, which should give us equal default rates among Non-Asian Minorities.

And yet, we still see higher default rates among NAMs.

What all this default data suggests is the opposite of the conventional wisdom: that colorblind systems of assessing creditworthiness tend to overestimate the creditworthiness of black and Hispanic applicants. Apparently, if a lender wants equal default rates among his white, black, and Hispanic borrowers, he needs to hold the blacks and Hispanics to higher standards -- i.e., to attain equality of outputs, he needs to discriminate against NAMs in terms of inputs.

Of course, he's not allowed to do that.

Another insight is the wisdom of anti-usury laws that set some sort of upper limit on interest rates. These can serve to prevent loans likely to go bad, which tend to impose widespread economic liabilities. Of course, in the giant push to get more lending to minorities from 1968 onward, these laws were in bad odor because they have a "disparate impact" -- the tighter the caps on interest rates, the fewer NAMs will get mortgages.

My published articles are archived at iSteve.com -- Steve Sailer


Anonymous said...

What all this default data suggests is the opposite of the conventional wisdom: that colorblind systems of assessing creditworthiness tend to overestimate the creditworthiness of blacks and Hispanics.
You can't say that. It's racist!

Van Cleef said...

iSteve, you need to brace yourself for a lot more batsh*t USGov sponsored mortgage lending going forward, and no 'lessons' will have been learned.

For Obama, and the rest of the left, the grievous injury to the economy [via 'social justice' lending practices gone shady] is not a concern. Leftists want the economy to tank permanently, and they want to discredit capitalism. 'Give the bankers enough rope to hang themselves' is their motto.

Every other day, underneath his suit & tie, Barack Obama [seventeen years in a Black Liberation Theology church] wears a 'What's Not To Like?' t-shirt. On the other days he switches to a 'It's All Good' t-shirt. And on special days, like when he fired the CEO of GM, Obama wears a 'Suckas!' t-shirt.

Because that's the way he rolls.

Anonymous said...

Has anybody other than me noticed (since it is April 15) that we do not check a box on our tax returns indicating our race / ethnicity? While the liberals taunt conservatives as "tea-baggers" and call them racist nobody is bothering to ask why there are only whites protesting.

eh said...

"...although the size of these effects are quite small."I think this should be is quite small, because size is the subject of this clause, not effects, and size is singular.

iSteve, you need to brace yourself for a lot more batsh*t USGov sponsored mortgage lending going forward, and no 'lessons' will have been learned.Maybe. But private investors will be skittish about mortgage paper for some time. So if it happens it'll be some sort of arrangement like big financial firms get now -- they can issue debt guaranteed by the FDIC, which means low interest rates and willing buyers. Still allowing (lasting) lower mortgage rates to non-whites in the name of raising home ownership among them would be hard to defend politically, and probably also in court.

By now I think it's clear that when all the needed data is available, if it ever is, it will show that non-whites benefited from and defaulted on subprime loans disproportionately -- it's only a matter of to what degree.

lån said...

blog is presented in a nice way!!

Frank said...

Perhaps I'm misusing the supplied data from this blog but the total amount of dollars lost on the subprime crisis seems modest. If we extrapolate the MA foreclosure for the the entire US and use the data in post on 17 April:


we're talking about 15 billion lost in 2005 (5 billion on whites, 5.9 billion on Hispanics and 4.1 billion on blacks). Granted that was one year but it doesn't seem to add up the trillions lost. Is this is a domino effect where foreclosures force down real estate values, tax revenues, etc?

Anonymous said...

Brilliant article.

Hatshepsut said...

As La Griffe du Lion once explained very clearly: when the mean score for any group is lower than your merit threshold, you really have to set the "passing score" for members of that group higher (on any given test) in order to select the truly qualified. Why? Because each candidate's "measured" score is really just an approximation to her "true" score. To be sure of selecting only the truly qualified, you have to give yourself a cushion. Otherwise, if you set the passing score precisely at your minimum threshold you will often select unqualified candidates; those having true scores lower than their measured scores.

Making things worse, the farther away (in SD units) the measured score of any candidate from her group's mean score may be, the more likely her measured score is a fluke (and the more likely that her true score is some value closer to her group's mean). So if you want to reliably select people over a threshold when the mean for their group is below it, you have to make your cushion even larger than the cushion for members of groups whose mean scores are above or at least closer to your threshold.

What does this all mean with respect to loan applications and default rates?

It explains why NAM's with "similar qualifications" have higher default rates!

Suppose you are a lender. Think of the loan application as a test. Based on experience with loans to millions of applicants, you figure out what threshold score (on the credit application) predicts the (final) default rate you want (the default rate won't be zero, that's impossible. It'll be, say, 1%). (And yes, changed circumstances like, say, a recession will invalidate your ex-ante estimate.)

Whenever an applicant's true score is above that threshold you should go ahead and make him/her a loan. However, you can't be sure of an applicant's true score from one credit application (test)-- confusion, recent but undocumented changes in the applicant's circumstances (a bonus or raise, a theft loss or impending divorce), unmeasured personal qualities, etc-- may cause the applicant's "measured score" to deviate from her "true score."

So you will set a "passing score" a bit above your calculated threshold score. How far above? Well, that depends. Most likely your threshold is close to the mean score for whites and east asians. For white/east-asian applicants you only need a modest cushion between your threshold and the passing score, since it is not improbable a white/east-asian applicant will score above the threshold.

But what happens when an NAM applies for a loan? The mean score for her group is well below the mean score for whites and east asians, and even farther below the passing score for white/asian-applicants (which is slightly above the merit threshold, since you need a bit of a cushion to maintain quality). Fewer NAM's will turn in high scores, and when one does, her score is more likely to be a fluke (and her true score is probably lower rather than higher).

Mathematically you know that the passing score for NAM's should be set higher than the passing score for whites and east-asians. Again, only NAM candidates with scores well above the mean for their group exceed your threshold. And the farther away from her group's mean any candidate's score is, the more likely it's wrong (or in your particular business, a fraud). Mathematically, you need a bigger cushion between your merit threshold and passing score for NAM's to maintain quality.

This is no idle theory--it is validated by experience. If you look at just experience with loans to millions of NAM's to (re-)calculate the credit-application threshold score which predicts the desired default rate, you will discover that the threshold for NAM's is higher than for all-borrowers-together (or whites and east-asians separately). The difference will match the extra cushion you ought to set between the NAM passing score and your original threshold number.

Nevertheless, if you set a passing score for NAM's higher than that for whites and east asians, the government will call that "invidious discrimination" and punish you severely. Innumerate (and innumerable!) government and "public interest" lawyers won't care that both theory and experience show that NAM's have higher default rates for any given credit-score level. They will yawn when you call your experts in statistics, in quality control, in loan performance. They will only get excited when they declaim to the court, or to the newspapers "those crooks lend to whites with lower scores than the blacks they turn away!"

So you grit your teeth and obey the reigning taboos. You calculate the desired-default-rate credit-score threshold for all borrowers. You set the passing (credit) score to make sure whites and east asians have true scores over that threshold. You know full well that NAM's with scores near the passing score frequently have true scores below the threshold. You make loans to them anyway, on which they default with depressing regularity. You charge white and asian borrowers higher interest rates to make up for the losses on loans to NAM's.

The USA staggers on, enfeebled by a thousand little bleeding wounds, of which forced loans to unqualified NAM's is just one.

Glaivester said...

A quick point here - the statement that equal default rates amongst whites and blacks should occur in a fair system is not strictly true, unless the goal of the banks is to make as many loans as possible without making a negative profit rather than making loans until the next loan makes no profit.

Put another way, if we assume that on average, a default costs $95,000 and the profit of a successful loan is $5000, banks are not going to (at least as a goal) make loans until they have a 5% default rate (no profit). Rather, they will make loans until they have a 5% marginal default rate (that is, the next loan would have a 5% chance of default). Once they start losing money on the next loan, the bank will try not to make more loans; it isnt goingto keep losing money on loans until it reduces its profit to exactly zero.

To further complicate things, the cost of default varies, the profit from a successful loan varies, and these variations are presumably not random, so the model becomes even more complicated than the one I presented (which assumes that the average cost of default is the same at all levels of creditworthiness).

In any case, while similar default rates for minorities and whites might suggest that they are getting fair terms, it does not necessarily prove it; to prove it you would need a model that would predict that the next marginal white and black buyer would give the same expected return to the banker when taking into account the risk of default (at different points on amortization), the amount of profit the loan would generate if successful, and the probabilities of all of the various outcomes.

The only way that default rates would prove the fairness of treatment would be if the default rate vs marginal profit graphs were the same (or at least the marginal profit hit zero at the same default rate for both).