Appraisals, Automated Valuation
Models, and Mortgage Default
Austin Kelly
Associate Director, Division of Enterprise Regulation
Austin.kelly@fhfa.gov
Subtitle: “the results demonstrate the usefulness of appraisal estimates in the
prediction of claim propensities, over and above the information contained in AVMs.”
“Abstract http://www.fhfa.gov/webfiles/15047/Kelly.PDF
. . . Previous research has suggested the possibility that professional appraisals or
econometric estimates of collateral value may be indicative of credit risk. This paper
examines the issue by estimating the probability of a mortgage default (defined both as 90
day delinquency and as a claim on mortgage insurance) as a function of the difference
between sales price of a home and the estimated value of the home at the time of the
purchase, produced by both an appraisal and by an Automated Valuation Model (AVM).
Logistic regression is used to estimate the quarterly hazard of a serious delinquency, or. . .
1. Introduction and Literature Review
. . . Appraisers provide the estimate of value used in determining initial equity. A handful of
papers have examined the role of appraisers in the underwriting process. Horne and
Rosenblatt (1996) examine the distribution of the differences between appraised values
and purchase prices. They find that differences between appraised values and sale prices
are almost always less than one percent, and appraisals for less than the purchase price are
extremely rare. LaCour-Little and Malpezzi (2001) estimates a model similar to the one
in this paper.. . .
Shiller and Weiss (1999) lay out a framework for evaluating the profitability of AVM
deployment. This paper takes a step towards filling the data requirements of their
framework, estimating the correlation between appraisal/selling price and AVM
valuation, and demonstrating the effectiveness of AVM systems in predicting default,
foreclosure, and loss severity.
Section 6 offers concluding
remarks and some observations concerning the relative predictive power of appraisals and
AVMs.
2. Model
The focus of this paper is the effect of appraisal and AVM quality on the credit risk in
mortgages. An appraisal is a measure of the “market value” of a property. In a highly
liquid market with large numbers of identical commodities traded, this is a simple
concept. In the housing market, with infrequently traded heterogeneous properties,
market value is a more tenuous concept. To some extent, the fact that a buyer is willing
to pay $X for a house sets $X as the market value, rendering an appraisal somewhat
superfluous. From the perspective of the entity holding the credit risk on the mortgage
(lender or, in this case, insurer), the most relevant concept might be the value that the
second highest bidder is willing to spend on the property, a notion that mixes the concepts
of “market value” and “liquidity.” This is because the holder of the credit risk cares
about the price at which the buyer could later sell the property, which determines the
buyer’s choice of prepayment or default in the face of trigger events, and determines the
amount of recovery in case of default. This may be expressed as
Market Value = Transaction Price + Idiosyncrasies (1)
where Market Value refers to the expected selling price if a property were immediately
resold. Transaction Price is the price agreed upon by the buyer and seller. Idiosyncrasies
represent any unique characteristics attached to the transaction, such as a buyer uniquely
attracted to a particular property characteristic, or a seller motivated to sell exceptionally
quickly, or, for that matter, fraud.
The appraisal process can provide an estimate of property value independent of the
idiosyncratic circumstances that might cause a buyer to be the highest bidder. Single
family appraisals are generally based upon the sale prices of comparable properties, with
adjustments made for differences in characteristics between the property in question and
the comparables, and with adjustments made for area-wide trends in price. An appraisal
constitutes an estimate of the market value. Such an estimate may be biased or unbiased.
Sources of bias to the high side are pressure from buyers, sellers, brokers, etc. who need
an appraisal for at least the agreed upon price so that the transaction can take place. The
holder of the credit risk on the transaction, for example, the insurer, would presumably
wish to pressure appraisers for an accurate estimation, but in many cases the appraiser is
hired by the lender, although the risk is borne primarily by the insurer.3
Appraisal Value = Market Value + Bias1 + _1 (2)
where Appraisal Value is the value assigned by an appraiser, Bias1 represents any
possible tendency to assign a value other than the expectation of Market Value, and _1 is
the inherent noise in any estimation process.
(3FHA does maintain a list of approved appraisers, and can remove an appraiser from the list for fraud or
unethical behavior, but it is not clear how effective this might be in the case of modest upward bias of the
sort considered here. See US GAO (2004) for a discussion of FHA’s role in monitoring appraisers.
5)
An AVM produces a second estimate of the market value (wrong) of the property. An AVM
estimate may be less subject to bias (wrong), as AVM services are sold to a wide variety of
parties, such as lenders, insurers, GSEs, or MBS investors, with no clear incentive to
produce “high” or “low” estimates (wrong). On the other hand, AVMs constitute a mass-appraisal
approach, rely upon generally available characteristics, and do not involve visits to
properties to ascertain condition or incorporate local knowledge (the announcement of a
factory closing or plans for a new transit stop), so that their variances may be much higher
than the variances of appraisals.
AVM Value = Market Value + Bias2 + _
_ (3)
where AVM value is the value assigned by an AVM, Bias2 is the tendency (if any) for an
AVM to produce a value other than the expectation of market value, and _
_ is the inherent
noise in the AVM estimation process.
The relevant questions for a holder of mortgage credit risk are, 1) “does an appraisal
contain any information helpful to the assessment of default propensities, and 2) “does an
AVM estimate contain any information beyond that contained in an appraisal?” The
latter will be the case if the mean square error of the AVM is not too large, relative to the
mean square error of the appraisal, and if the correlation between the two errors is not too
high. One way to test this proposition is to estimate equations such as
Prob(Default) = fn(Appraisal, AVM Estimate, other risk variables) (4)
Loss Given Default = fn(Appraisal, AVM Estimate, other risk variables) (5)
and test the coefficients on the Appraisal and AVM Estimate values.
Underwriters, and FHA guidelines in particular, generally take the minimum of the sale
price or the appraised value as the denominator when calculating the loan-to-value ratio,
used as a key indicator of default probability. Thus, the extent to which an appraisal
exceeds the transaction price has no effect on the underwriting decision, or perceived
degree of risk attached to the loan by the underwriter. An appraisal less than the
transaction price has serious consequences, however, generally requiring an increase in
the cash that the buyer has to bring to the table, or a decrease in the price received by the
seller, or the failure of the transaction to go through. Thus appraisals may produce
benefits in ways not captured by transaction data, either by preventing transactions on
overpriced properties, or by triggering renegotiated prices.
AVMs are generally not used in FHA underwriting4. However, AVM estimates may
provide an additional source of information on the value of the collateral; therefore on the
level of credit risk for a given mortgage. The extra predictive power could be useful for
risk monitoring on the part of FHA or other insurers, risk accounting, and for investor
evaluations of portfolios of mortgages.
6. Conclusions
AVM estimates are predictive of both claim and delinquency propensities. Appraisal
ratios also have predictive power for claims. Examined separately, each is useful as a
predictor of the claim propensity of a mortgage. Entered together, the correlation
between the two estimates is weak enough that each serves as a useful indicator of credit
18In the interest of space, only the full model for the GAOrisk specification is included for the Atlanta
results. Other results were similar, with AVM values predictive of risk and appraisal ratios insignificant,
and with the GAOrisk specification slightly outperforming the TOTAL scorecard specification.
27
risk, although the significance levels are higher on the AVM estimate when both are in
the regression.
The confidence measure attached to the AVM estimate also serves as a predictor of credit
risk. Properties that are easier to value have lower credit risk, even after conditioning on
a host of standard underwriting variables. Additionally, AVM estimates are a significant
predictor of loss given default, an important but often ignored dimension of credit risk.
Much of the value of an appraisal presumably comes prior to origination, in preventing
transactions at prices far above market value, or contributing to the renegotiation of price
prior to closing. The results here should not be taken to imply that appraisals have less
value than AVMs, only that appraisal values have less post-origination predictive power
than do AVMs.
These results confirm the value of econometric estimates of property value first found by
LaCour-Little and Malpezzi, using a more recent, larger, and nationally representative
sample, and focusing on claims and losses, not just delinquency. Additionally, this work
demonstrates the utility of commercial, off-the-shelf, AVM estimates for predicting credit
risk. Finally, the results demonstrate the usefulness of appraisal estimates in the
prediction of claim propensities, over and above the information contained in AVMs.”
Thanks!
Curtis D. Harris, BS, CGREA, REB
Bachelor of Science in Real Estate, CSULA
State Certified General Appraiser
Real Estate Broker
ASTM E-2018 Commercial Real Estate Inspector
HUD 203k Consultant
HUD/FHA Real Estate Appraiser/Reviewer
FannieMae REO Consultant
CTAC LEED Certification
The Harris Company, Forensic Appraisers and Real Estate Consultants
*PIRS/Harris Company and the Science of Real Estate-Partners*
1910 East Mariposa Avenue, Suite 115
El Segundo, CA. 90245
310-337-1973 Office
310-251-3959 Cell
WebSite: http://www.harriscompanyrec.com
Resume: http://www.harriscompanyrec.com/CURRICULUMVITAENAME2011a.pdf
Commercial Appraiser Blog: http://harriscompanyrec.com/blog/
We Make a Simple Pledge to
Communicate, in a timely fashion, each appraisal, analysis, and opinion without bias or partiality
Abstain from behavior that is deleterious to our clients, the appraisal profession, and the public
Hold paramount the confidential nature of the appraiser/consultant – client relationship
and
Comply with the requirements of the Uniform Standards of Professional Appraisal Practice and the
Code of Professional Ethics of the National Society of Real Estate Appraisers
IT’S THE LAW-Designation Discrimination is Illegal [FIRREA, Sec. 564.6]: Professional Association Membership http://www.orea.ca.gov/html/fed_regs.shtml#Statement7 Membership in an appraisal organization: A State Certified General Appraiser may not be excluded from consideration for an assignment for a federally related transaction by virtue of membership or lack of membership in any particular appraisal organization, including the appraisal institute.
CONFIDENTIALITY/PRIVILEGE NOTICE: This transmission and any attachments are intended solely for the addressee. The information contained in this transmission is confidential in nature and protected from further use or disclosure under U.S. Pub. L. 106-102, 113 U.S. Stat. 1338 (1999), and may be subject to consultant/appraiser-client or other legal privilege. Your use or disclosure of this information for any purpose other than that intended by its transmittal is strictly prohibited and may subject you to fines and/or penalties under federal and state law. If you are not the intended recipient of this transmission, please destroy all copies received and confirm destruction to the sender via return transmittal
