How good is the Good Faith Estimate?
An analysis of new research suggests that, contrary to the views of some observers, the Good Faith Estimate disclosure has been an accurate predictor of actual mortgage closing costs.
J. Michael Collins
Published January 1, 2011 | January 2011 issue
Federal and state policies mandate that lenders disclose an array of timely and potentially useful information during the mortgage application process. The Good Faith Estimate (GFE) and the HUD-1 Settlement Statement are the primary disclosure documents lenders provide to mortgage applicants.1/ As its name implies, the GFE lists the lender or mortgage broker's best estimate, in "good faith," of closing costs. It must be provided within three business days after a borrower applies for a loan. The HUD-1, the companion document to the GFE, is provided one day prior to the loan closing or mortgage settlement and lists the actual costs that the borrower will pay.
Critics of the mortgage application process have long derided the GFE for being inaccurate. A 2002 U.S. Department of Housing and Urban Development report asserted that "three decades of experience has shown that too often the estimates appearing on GFEs are significantly lower than the amount ultimately charged at settlement, are not made in good faith (e.g., a range of $0–$10,000), and do not provide meaningful guidance on the costs borrowers ultimately pay at settlement."2/ Until last year, lenders faced no penalty for providing inaccurate information on the GFE. Critics have often speculated that in the absence of such a penalty, lenders have an incentive to underestimate closing costs in order to lure unsuspecting borrowers into high-cost loans using a bait-and-switch strategy. In other words, lenders might promise low or no closing costs on the GFE and then charge higher costs at the closing, leaving borrowers to notice the change and either abandon the loan application or go through with the loan and pay higher costs than anticipated.
In January 2010, federal regulators implemented reforms in an attempt to refine the mortgage disclosure process.3/ Among the reforms is a requirement that lenders reimburse borrowers for any costs that exceed GFE estimates by more than allowable margins of error.4/ The new requirement raises a question: Before the reforms took effect, how big a problem was the underestimating of costs? Perhaps not big at all, it turns out. A preliminary analysis based on new research into the accuracy and usefulness of the pre-2010 GFE suggests that the underestimating of costs was not widespread. Instead, the analysis suggests that the GFE, even in the form it took prior to the new regulations, has been an accurate predictor of actual closing costs.
Degrees of divergence
The accuracy of the GFE has not been analyzed extensively in the past, but the existing evidence suggests that estimates on the GFE may indeed diverge from actual closing costs. However, findings about the degree of divergence vary widely. For example, one analysis that was cited in a U.S. Senate hearing suggests that 83 percent of borrowers end up with higher closing costs than those estimated on their GFEs.5/ Another analysis found that estimates on the GFEs were actually larger than closing costs, on average.6/
The new analysis described here examines the extent to which pre-2010 mortgage disclosures accurately reflect information about loan costs. In the long history of mortgage disclosures, this is the first study to examine differences between estimated costs on GFE forms and actual costs on the HUD-1 while controlling for borrower and loan characteristics. This study tests some theoretical work that suggests how sellers will reveal information to buyers—or, in this case, how lenders will reveal information to borrowers. According to information-economics theory, lenders have an incentive to intentionally underestimate costs to make their loan offer appear more attractive to mortgage applicants. This theory implies that lenders' GFE values would be less than the HUD-1 values for the same loans. However, in theory, this incentive is balanced by the possibility that if lenders underestimate settlement costs by too great a margin, some loan applicants may view the lender negatively, to the point of walking away from a loan offer. In other words, lenders have an incentive to underestimate costs on the GFE, but not by such a large margin that borrowers will reject the loan. Of course, all of this behavior is rooted in the theory that borrowers pay attention to the GFE, which may or may not be an accurate description of consumer behavior.
Defining the data
To determine whether lenders underestimated costs and, if so, how much the underestimates diverged from actual costs, this analysis compares GFE and HUD-1 data from a sample of 619 loans in the National Mortgage Data Repository (NMDR), a data set that the National Community Law Center collected from community-based organizations from 1994 to 2007. The data set is unique in that it tracks all aspects of a loan application, from the initial forms and disclosures to the closing documents. The fact that the data were collected before the introduction of the new disclosure regulations is useful in that all loans in the database were made under a similar regulatory framework. The median loan amount in the NMDR sample is $83,990 and the median borrower income is $45,972. Approximately one-third of the sample—34 percent—is made up of minority-race borrowers. A comparison with Home Mortgage Disclosure Act (HMDA) data from roughly the same time period suggests that borrowers in the NMDR sample are fairly typical of mortgage borrowers overall. Given the long time period covered by the NMDR sample, a direct comparison is challenging, but relative to HMDA data from 2003 (an approximate midpoint of the 1994–2007 time frame), the NMDR sample is similar with respect to income and loan amounts, and it slightly overrepresents minority borrowers (34 percent in the NMDR sample and 25 percent in the HMDA data).
The mortgages in the NMDR sample were collected from 27 states, although the majority (55 percent) of the loans are from Massachusetts. Most of the loans are refinance loans (83 percent). Furthermore, most of the loans are to owners of single-family properties, 90 percent of which the owners claimed to use as their primary residences. While the sample is not broadly representative, differences between GFE and HUD-1 values are nevertheless illustrative of the extent to which problems may have existed under the previous regulations. In turn, the results of the analysis facilitate conjecture about the effects of the new regulations.
The GFE and HUD-1 forms have the same cost categories but different layouts and line item numbers. For example, a lender might list a closing fee under code 803 on the GFE but under code 1202 on the HUD-1. The differences make a side-by-side comparison of the two forms challenging. In order to capture all the costs within each cost category, each line item in the analysis was coded based on key words that appear in the text fields describing the line on the respective form. Costs were coded into origination fees, broker fees, and then the categories of "fixed" and "variable" costs. Fixed costs were defined as items that should be well known and should not vary significantly across loans, such as credit report fees and underwriting fees. Variable costs were defined as items that lenders may not be able to estimate as readily and that are more likely to vary across loans, including appraisal, attorney, flood insurance, survey, title, and title insurance fees.
A good overall predictor
The table below summarizes the GFE and HUD-1 values across the cost categories. As shown in the table, average total fees were $9,046 on the GFE and $8,686 on the HUD-1. Overall, this result indicates the GFE is a relatively good predictor of final HUD-1 costs. Total fees on the HUD-1 exceeded the estimates on the GFE for 39 percent of loans, which is a considerably lower rate than critics of the GFE have previously indicated. Within individual cost categories, the results vary. Fees defined as variable are greater on the HUD-1 than on the GFE only 30 percent of the time. Fees defined as fixed are greater on the HUD-1 than on the GFE in 52 percent of cases, but the average difference is only $57. Actual lender origination fees on the HUD-1 exceed estimates on the GFE in just 11 percent of cases. Regarding broker fees, the HUD-1 costs exceeded the estimates on the GFE in only 30 percent of cases, but the estimates were off by a large enough margin that the difference between the average underestimated cost and the average actual cost is $535.
The 2010 regulatory reforms introduced a 10 percent tolerance margin, such that if lenders underestimate costs on the GFE by 10 or more percent of the actual costs on the HUD-1, they could be held liable for the difference. How often do mortgage borrowers see costs that are at least 10 percent larger than estimated? The graph below shows rates based on the NMDR data used in this analysis. Overall, less than 1 in 5 loans analyzed had total fees that were at least 10 percent higher than estimated on the GFE. The graph also shows that the tolerance margin was exceeded less than 10 percent of the time among broker fees and less than 15 percent of the time among origination fees. Among fixed fees, the rate was higher—just over 30 percent—but, as noted in the table, the absolute cost of these fixed fees is relatively small, on average. Variable fees exceeded the 10 percent tolerance margin in less than 1 in 5 cases. Overall, the evidence does not suggest widespread use of GFE disclosures to lure borrowers into paying unexpectedly large fees at closing.
A statistical analysis7/ of the factors affecting total costs shows that, after controlling for loan, lender, and borrower factors, GFE estimates predict most of the final HUD-1 costs, and few other factors seem to matter. Loan amount, loan type, origination date, and application method are statistically significant in many cases. This is to be expected, as these factors should affect closing costs and fees. But the influence of each of these loan-level factors is relatively small, and even controlling for these factors, the GFE remains a good predictor of the HUD-1. Further analysis indicates that the loan amount is the strongest predictor of HUD-1 costs, as would be expected, since larger loans have larger fees. Borrower characteristics—race, income, education, and age—are not significant in the statistical models.
Statistical analyses show that the GFE is not as strong a predictor of actual HUD-1 costs for variable fees as it is for fixed fees. This finding suggests that these fees are indeed more variable than the other fee categories, as might be expected. Notably, the difference between the variable fee estimate on the GFE and actual variable fees on the HUD-1 declines as income increases. This relationship perhaps suggests that borrowers with higher incomes are more attentive to variable costs and that their attentiveness restrains lenders from inflating variable costs before closing.
While lenders may have an incentive to underestimate settlement costs on the GFE and then charge unexpected fees at closing, there is little evidence of this practice in the data, with the exception of mortgage broker fees. There is no conclusive evidence that borrowers who may be perceived as less financially sophisticated are more likely to encounter cost increases from the GFE to the HUD-1, although income had statistically significant yet relatively small effects on variable costs, as discussed above. The latter finding may be consistent with lower-income loan applicants receiving low initial estimates of variable costs on the GFE, only to see costs rise on the HUD-1 at the closing. This may be due either to lenders sizing up lower-income borrowers' sensitivity to closing costs and then manipulating the GFE estimates accordingly, or to genuine differences in loan applicants and loan application processes that result in unpredictable costs. The use of the GFE and HUD-1 by lower-income borrowers may be worthy of continued observation and analysis.
On reforms and timing
Overall, the analysis reveals that a majority of borrowers (61 percent) paid closing costs that were equal to or lower than estimated on the GFE. The findings indicate the GFE is a useful predictor of actual HUD-1 costs, and few loan- or borrower-level factors are strong predictors of the differences between costs on the GFE and the HUD-1. These findings suggest that the reforms adopted in 2010, including liability for underestimates, may not have much effect on the accuracy of the GFE. The reforms refine the format of disclosure documents and expand the GFE significantly, but it is unclear whether the new formatting will facilitate comparison shopping. However, based on the NMDR data and controlling for loan-level factors, it appears the new regulations may help borrowers obtain more accurate statements regarding broker fees. There also is the potential that less predictable variable costs will be estimated more carefully on the GFE under the new provisions.
The average total closing costs in this analysis ($8,686) represent about 8 percent of the mean loan amount. Obviously, closing costs are a significant expense for borrowers and are worthy of attention and scrutiny. But another important issue—one that was beyond the scope of this analysis—is whether the disclosure process provides information in a timely way so borrowers can actually use it. A borrower who finds out on the day of the loan closing that settlement costs are greater than estimated may be in a difficult position. If the loan is necessary to purchase a home or, in the case of a cash-out refinance, to gain access to needed funds, the borrower may be unwilling to back out of a standing loan offer. In the rush to settle the loan, borrowers may not even recognize cost increases. In future analyses of the disclosure process, the matter of timing warrants careful assessment.
J. Michael Collins is an assistant professor of consumer science in the School of Human Ecology at the University of Wisconsin–Madison. He also serves as Faculty Director of the university's Center for Financial Security and is a specialist with University of Wisconsin Cooperative Extension.
1/ The HUD-1 is generally used for a mortgage closing combined with a home purchase; the HUD-1A is for a mortgage refinance. For simplicity, "HUD-1" is used here to refer to both forms.
2/ "Real Estate Settlement Procedures Act (RESPA): Simplifying and Improving the Process of Obtaining Mortgages to Reduce Settlement Costs to Consumers," Federal Register, Department of Housing and Urban Development, 67 (145): 49133–49174, 2002.
4/ Shopping for Your Home Loan: HUD's Settlement Cost Booklet is available at www.hud.gov/offices/hsg/rmra/res/settlementaug17english.pdf.
5/ Lauren E. Willis, "Decision Making and the Limits of Disclosure: The Problem of Predatory Lending," Maryland Law Review, 65(3), 2006, p. 707–840.
6/ Mark D. Shroder, "The Value of the Sunshine Cure: The Efficacy of the Real Estate Settlement Procedures Act Disclosure Strategy," Cityscape, 9 (1), 2007, p. 73–91.
7/ The analysis performed was an ordinary least squares (OLS) regression, which is a commonly used statistical method for estimating the degree to which fluctuations of one variable are proportional to movements in another variable (or group of other variables). Further information on OLS regression is available in numerous texts on statistics or econometrics. For example, see A Guide to Econometrics by Peter Kennedy (5th Edition, Blackwell Publishing, 2003).