by Tanta on 10/11/2007 10:18:00 AM

This is a follow up to CR's post last night on the WSJ article using HMDA data to make some observations about "subprime" loans.

Trust the Wall Street Journal to fail to understand the point of reporting regulation. They are not the only media outfit to have made a major logical error using HMDA data, but they make a nice poster child for the problem. The WSJ's basic point of departure:

The analysis of loan data by The Wall Street Journal indicates that from 2004 to 2006, when home prices peaked in many parts of the country, more than 2,500 banks, thrifts, credit unions and mortgage companies made a combined $1.5 trillion in high-interest-rate loans. Most subprime loans, which are extended to borrowers with sketchy credit or stretched finances, fall into this basket.

Of course, if you simply assume that borrowers get a high interest rate for a good reason--they are subprime credits--then the fact of "high interest rates" needs no explaining, and the fact that such loans may be concentrated in low-income or minority borrower groups can be explained as something other than discrimination (if you simply assume that low-income or minority borrowers are more likely to have subprime credit profiles than other borrowers).

I guess we're going to have to take a walk down memory lane here. There was a time --around 1975 to be precise--when Congress was a bit concerned about anecdotal evidence that banks and thrifts were engaging in "redlining," or refusal to lend money at all in certain (mostly minority or low-income) neighborhoods. (The idea is that a lender drew a red line on a map around areas in the "wrong side of town," and loans would not be made for properties in the red areas.) In a rare recognition by Congress that anecdotal evidence is unsatisfactory, the Home Mortgage Disclosure Act (HMDA) was born. Its original incarnation required lenders to report data on the geographical location (down to census tract level) of loans originated. This data could be analyzed to see if apparent redlined areas appeared.

In 1989, HMDA got an overhaul which required lenders to report on denied applications, as well as originated loans. Further, the borrower's race, sex, and income level had to be reported on all loans. This change grew out of more anecdotal evidence--some of which was being probed in courts of law--that minority and female applicants were being turned down at disproportionate rates. Congress also gave the Federal Reserve the authority to augment HMDA reporting requirements.

In 2002, the Fed used that authority to require price data on loans originated. You see, there was anecdotal evidence that some classes of borrowers were getting higher rates on loans in ways that couldn't be explained by the loan or borrower characteristics.

This whole dynamic may be hard for the WSJ and its fellows in the Big Paid Media, so let me explain this very clearly. In 1975, some folks accused lenders of redlining, which means not granting credit at all to some people. The lenders said they weren't doing that. Congress passed HMDA, and then there was actual data about geographic lending patterns to analyze instead of anecdotes. Once we got some HMDA data under our belts, the Community Reinvestment Act came into being (in 1977) precisely because it was clear that redlining had been going on. CRA in essence forces lenders to show that they are willing to make loans in neighborhoods in which they are willing to take deposits (i.e., those deposits need to be "reinvested" in the neighborhood they came from in the form of loans, not just mortgage loans, to that neighborhood. You can't extract deposits from poor people and use them exclusively to fund loans to rich people.) CRA does not mandate price levels, or even address the question of price levels.

You may be surprised to hear this, but over time accusations of discriminatory lending practices did not go away. In a number of cases, "mystery shopper" tests were performed, in which a white applicant and a black applicant each applied for credit at the same instutition with identical credentials (employment, income, credit history, loan terms), and the results showed that black applicants were more likely to be turned down. This cast some doubt on the lenders' claims that loan rates in minority neighborhoods were a function of the lower credit quality of those borrowers. That became a hypothesis in need of some testing, you see, not an accepted explanation.

So the 1989 revision to HMDA forced collection of demographic data, for the precise purpose of testing the assumption that poor and minority people are just always bad credit risks. This resulted, as you might expect, in conjunction with CRA and other fair lending laws, in much higher rates of home mortgage lending in those areas that were once redlined.

But were these poor and minority people happy, at last? Why no, they weren't. Turns out, anecdotal evidence began to emerge that while these good people were finally getting loans, they were getting them at much higher interest rates than higher-income folks and whites generally got, and that this could not be accounted for by the difference in creditworthiness of the borrowers or the quality of the collateral (the latter proxied by census tract).

So the 2002 change to HMDA, to collect data on mortgage loan pricing, was an attempt to collect empirical data on pricing patterns to test claims about what might explain higher loan pricing, not to accept them without further probing. The whole point of the HMDA dataset is to ask if wide disparities in loan pricing exist, in the same geographical area at the same time. If they do, the data can be analyzed controlling for income level, race, and sex, to see if any of those things correlate with loan pricing. If they do--and they certainly do--there is still the question of why this happens.

The lending industry will tell you without fail that this correlation exists because low-income and minority borrowers have lousy credit histories. But the HMDA data does not support (or disprove) that claim. The HMDA data shows that these borrowers get higher rates on average, but since the HMDA dataset does not include FICO or any other reliable measure of credit history, it cannot be used to conclude that these higher rates are explained by the subprime credit of the borrowers.

And you cannot use subprime mortgage lending patterns to prove or disprove this claim, either. You are trying to test whether "subprime mortgages" are being given only to truly "subprime borrowers." Your test results will look funny if you assume your conclusion.

So. There is no measure of borrower creditworthiness, specifically, in the HMDA data. The "high priced lending" data is an attempt to quantify the number of loans made at a threshold which is usually going to be unexplainable except in terms of either the risk factors of the loan or discrimination. (That is, the threshold is set to "weed out" spikes in market rates during which everybody gets high rates. If risk factors cannot explain the difference, the presumption must be discrimination).

It also uses APR, not note rate, as its measure. APR is calculated by taking into account fees and points over the stated term of the loan, and so using an APR measure lets you pick up loans that appear to have a low note rate, but that still involved unusually high charges to the borrower. APR on an ARM is calculated by assuming that the original index value is unchanged over the life of the loan, and then by using an interest rate in the APR calculation that takes into account scheduled rate increases up to the "fully indexed" value (that is, it's kind of a "blended rate" of the initial discounted rate and subsequent rates on the loan). So you can't evade high-cost loan reporting by putting people into teaser-rate ARMs, or by offsetting low note rates with outrageous fees, because the APR measure cannot be fooled like that.

The trouble, of course, is that an absolute level of APRs isn't very helpful: we all know that what counted as a "high rate" in 2003 is not the same as what counts as a high rate today, because market rates change. We also know that second liens get higher rates than first liens, and that the term of the loan affects rate. So the threshold was set based on the price of comparable-maturity Treasury securities at the approximate time the loan application was made, with a spread of 3.00% for first liens and 5.00% for second liens. If the APR on the loan exceeds these thresholds, it is reported as "high rate."

However, since the 2002 change to HMDA did not force lenders to collect other data that could account for pricing differences, such as LTV, doc type, FICO, or DTI, you have a set of pricing data but you're back to square one in terms of using it to decide whether this pricing is fair or predatory or discriminatory. (And yes, the Fed initially proposed collecting more loan level data, and yes, the industry lobbied long and hard over the "reporting burden" this would create.)

You also find that the data itself can be weirdly skewed when a lot of loans are ARMs and the yield curve is flat or inverted. Without the other data on borrower credit quality and loan terms, it's very hard to sort out the noise.

So we have data from HMDA showing that high-priced lending is not necessarily limited to low-income and minority neighborhoods. The WSJ takes this to mean that subprime lending is not necessarily limited to low-income and minority neighborhoods.

Subprime mortgages were initially aimed at lower-income consumers with spotty credit. But the data contradict the conventional wisdom that subprime borrowers are overwhelmingly low-income residents of inner cities. Although the concentration of high-rate loans is higher in poorer communities, the numbers show that high-rate lending also rose sharply in middle-class and wealthier communities.

First of all, I take major exception to the claim that "subprime mortgages were initially aimed at lower-income consumers with spotty credit." Near-prime programs like FHA were until quite recently the backbone of lending to lower-income consumers with "spotty" credit. Subprime mortgages were aimed at people with terrible credit, and if you think that problem is limited to those with low income, you'll have to explain to me how these low income folks racked up enough credit to have major problems with it in the first place. If people won't lend to you, you don't have a credit report full of late payments and charge-offs. I bring up the whole history o' HMDA to point out that the concern has always been as much about whether credit was granted at all to some folks as it has been about the terms on which that credit is granted. I will also note that FHA loans, because of HUD and Ginnie Mae rules, can rarely end up in the "high rate" category in HMDA. You can write FHA loans all day and not worry that your HMDA report will make you look like a predator. Yes, FHA loans have higher rates than conforming conventional loans, to account for their riskier nature, but they're comfortably within that "spread" used in the HMDA definitions, because they are near-prime or "spotty" credit, not terrible credit.

Further, the WSJ "contradicts" the "conventional wisdom" that subprime is mostly a matter of urban poor folk by showing that some suburban middle-and-upper income folk get high rate loans, too. However, because they accept without question that "subrime falls into the basket" of high-rate loans, they manage to assume that increasing numbers of high-rate loans to the suburban middle class means that these loans are "subprime." As CR notes, the whole phenomenon of "Alt-A" seems to have escaped them.

The possibility also seems to have escaped them that maybe subprime is, at the end of the day, just "high rate lending." If that's the only constant we can find in that category called "subprime"--if income, credit history, property location and price in that bucket is apparently rather random--then you begin to suspect that "subprime" is "loans to naive or desperate borrowers," not this ballyhooed "risk based priced" stuff of recent legend.

If you don't find that idea perfectly convincing, then you will indeed have to collect and analyze the data on LTV, FICO, doc type, etc. to establish that the category "subprime" means risk-based price, not just high-price.

The bottom line is, as CR notes, that "high-risk" lending was everywhere in the boom years. Of course there is a desire to collapse it all into the easy category of "subprime." And there has for a long time been a lot of political pressure to keep the association of "subprime" and "urban minorities" in place, because it has functioned as a good excuse for the subprime lenders (they "help" the poor and minorities, remember?). My view is that a whole lot of parties are very interested in maintaining rather than seriously analyzing a lot of faulty assumptions about risk, rates, and borrower credit characteristics. If this ain't "just a subprime problem," then an entire debt-based economy in which even the middle and upper middle class cannot afford homes given RE inflation and wage stagnation is suddenly in question. The last thing certain vested interests want to hear is that, basically, "we are all subprime now."