8/6/13
`
`S&P | ARCHIVE | Criteria | Structured Finance | RMBS: Guidelines For The Use Of Automated Valuation Models For U.K. RMBS Transactions | Americas
`
`Feedback | Americas [Select Region] | Update Profile | Logout
`
`General site search...
`
`
`
`ARCHIVE | Criteria | Structured Finance | RMBS: Guidelines For The Use Of
`Automated Valuation Models For U.K. RMBS Transactions
`
`Publication date: 26-Sep-2005 05:51:33 EST
`
`View Analyst Contact Information
`
`Table of Contents
`
`Contact Client
`Services
`1-877-SPCLIENT
`1-877-772-5436
`Call Tree Options
`Contact Us
`
`(Editor's Note: This criteria article is no longer current. It has been superseded by the article "U.K. RMBS Methodology And
`Assumptions," published on Dec. 9, 2011.)
`
`Property valuation is a key driver in Standard & Poor's credit assessment of mortgage portfolios for RMBS transactions. This credit
`assessment estimates the likelihood of default (the foreclosure frequency) and the amount of loss on default (the loss severity) for each
`loan in the portfolio. The proportion of the loan to the property value (the LTV ratio) is a major factor in determining the probability of
`borrower default and the magnitude of loss severity. More specifically, a higher LTV ratio results in greater loss severity and higher
`probability of default.
`
`Automated Valuation Models (AVMs)
`A valuation carried out by a surveyor is based on a physical inspection of the property. In contrast, an AVM is a desktop model that
`excludes any physical check of the property, and relies on comparables and market information. A user inputs the target address and
`certain property characteristics and the model returns a value for the target property.
`
`In the U.K., AVMs have been available to the market since 1999. The two major providers at present are UKValuation Ltd. and Hometrack
`Data Systems. Both providers have produced hedonic-based valuation models. Hedonic-based AVMs rely on a large database of property
`valuations and sale prices, and conduct an automated search for comparable properties to the target property.
`
`Comparable properties are selected by their similarity and proximity to the target property, based around various property characteristics
`such as room counts, age, floor area, and property type.
`
`Each valuation estimated by these AVMs is also associated with a corresponding measure of confidence (e.g., 1 to 7 for Hometrack Data
`Systems), or a level of uncertainty (e.g., 8 to 20 for UKValuation), which is generated from the fit of the target property to the comparables.
`More specifically, the main ingredients that contribute to the measurement of confidence (or uncertainty) are physical similarity between
`the target property and the comparables, the proximity of the target property to the comparables, and the homogeneity of the neighborhood
`of the target property.
`
`Risks Of Using An AVM
`If any particular valuation technique results in consistently overvalued properties, the calculated LTV ratios are likely to be underestimated.
`For example, if a loan value is assumed to be 80 and the "true" value of the property is 100 (hence a "true" LTV of 80%), an overvaluation of
`110 will result in an LTV of approximately 73%. As lower LTVs are viewed as more favorable in terms of default and loss, the overvaluation
`of the property would result in an underestimation of both default frequency and loss severity. If this were to occur across a significant
`proportion of a mortgage portfolio, the potential loss on the portfolio could be underestimated.
`
`Accuracy Of An Automated Valuation And Confidence Levels/Uncertainty
`To test the accuracy of AVM-generated valuations, the estimated valuations from these models were compared to those reported by an
`actual full survey of a large sample of properties. The differences between the automated valuations and their corresponding surveyor values
`across a portfolio tended to follow a random normal distribution centered on a mean of 0 for both of the U.K. AVM providers.
`
`Generally for every "overvalued" property, there was also an equivalently "undervalued" property. There was, however, much variation in the
`magnitude and frequency of these under/overvaluations, the level of which was directly related to the confidence or uncertainty associated
`with the valuation. The higher the level of confidence (or conversely, the lower the uncertainty), the less variation there was in the accuracy
`of the valuation.
`
`Chart 1 shows indicative examples of this variation based on confidence or uncertainty. To standardize Standard & Poor's approach across
`providers, confidence levels and uncertainties have been grouped into five levels of accuracy, and are plotted in chart 1 as "variation levels"
`1 through to 5.
`
`Chart 1
`
`Show
`
`www.standardandpoors.com/prot/ratings/articles/en/us/?articleType=HTML&assetID=1245330509010
`
`1/4
`
` 1
`
`MICROSTRATEGY 1020
`Microstrategy, Inc. v. Zillow, Inc.
`IPR2013-00034
`
`

`
`8/6/13
`
`S&P | ARCHIVE | Criteria | Structured Finance | RMBS: Guidelines For The Use Of Automated Valuation Models For U.K. RMBS Transactions | Americas
`
`The horizontal axis in chart 1 represents the difference between an automated value and its corresponding observed surveyor value
`(expressed as a percentage of the surveyor value). Positive differences indicate overvaluation by the AVM provider, while negative
`differences indicate undervaluation by the AVM provider.
`
`The vertical axis represents the probability of these differences occurring. The five curves are associated with different variation levels.
`Lower variation levels produce taller, narrower distributions. Here, the difference between an automated value and its corresponding
`surveyor value tends to zero more often than for higher variation levels; that is, lower variation levels tend to be more accurate than high
`variation levels.
`
`Accuracy Of A Portfolio Of Automated Valuations
`From an RMBS rating perspective, it is more relevant to examine the accuracy of an entire pool of automated valuations rather than the
`accuracy of an individual valuation, given that portfolios of mortgages are securitized.
`
`Having established some patterns for the valuation error in an individual valuation, it is possible to estimate the overall under/overestimation
`of a portfolio of automated valuations. For example, a portfolio of 100 automated valuations may have one valuation that is overvalued by
`20%, but is unlikely to have all 100 valuations overvalued by 20%.
`
`It would be even less likely for a portfolio of 500 valuations to be overvalued by 20% overall. Some properties would be undervalued to
`varying degrees while some would be overvalued to varying degrees. Hence, the overall under/overvaluation of a portfolio of automated
`valuations has a similar distribution to that of a single AVM generated valuation; that is, it follows a random normal distribution centered on
`zero, but with a comparatively smaller variation than the distribution described in the preceding section.
`
`The distribution for portfolio under/overvaluation is illustrated in chart 2.
`
`Chart 2
`
`Here the horizontal axis represents the difference between the total automated values of a portfolio and the "true" value of the portfolio.
`
`www.standardandpoors.com/prot/ratings/articles/en/us/?articleType=HTML&assetID=1245330509010
`
`2/4
`
` 2
`
`

`
`S&P | ARCHIVE | Criteria | Structured Finance | RMBS: Guidelines For The Use Of Automated Valuation Models For U.K. RMBS Transactions | Americas
`8/6/13
`Positive differences indicate overvaluation of the entire portfolio by the AVM provider, while negative differences indicate an undervaluation of
`the entire portfolio by the AVM provider.
`
`Quantifying Risks By Variation Level
`A theoretical overestimation of the value of a portfolio of AVM-generated valuations can be determined by the distribution of portfolio level
`under/overestimation given by each variation level. High overvaluation lies on the extreme right of the curve, while lower overvaluation lies on
`the center-right of the curve. Standard & Poor's has assumed a rating dependent amount of overvaluation per variation level. The higher the
`rating level, the higher the amount of assumed overvaluation and the further to the right of the distribution the overvaluation amount will lie
`(as illustrated by chart 3).
`
`Chart 3
`
`To reflect the risk of overvaluation across a portfolio of mortgages, AVM-generated valuations will be reduced by the amount of assumed
`overvaluation per rating level for each variation level. This will have the effect of producing higher LTVs for the portfolio. The extent to which
`this impacts the probability of default and degree of loss severity will depend on both the initial LTV of the pool and the average portfolio
`variation level. The table below summarizes the reductions that will be applied to portfolios with AVM-generated valuations.
`
`Reductions To AVM-Generated Valuations
`BB
`BBB
`Variation level
`AAA
`AA
`A
`1
`1.07% 0.72% 0.63% 0.54% 0.27%
`2
`1.74% 1.16% 1.02% 0.87% 0.44%
`3
`2.95% 1.97% 1.72% 1.48% 0.74%
`4
`4.16% 2.77% 2.43% 2.08% 1.04%
`5
`5.37% 3.58% 3.13% 2.68% 1.34%
`
`For example, in a 'AAA' rating scenario, a portfolio (with variation level 4 for all valuations) is assumed to be overvalued by 4.16%, whereas
`in a 'BBB' scenario, the same portfolio with variation level 4 is assumed to be overvalued by 2.08%. A portfolio with an AVM valuation of
`£100,000,000 with variation level 4 would receive a corresponding reduction of 4.16% in a 'AAA' rating scenario. This would bring down the
`portfolio value to its "true" value of £96,006,144 (see formula below). If the original weighted-average LTV for the portfolio was 75%, this
`reduction would raise the weighted-average LTV to 78%.
`
`The Future Of AVMs
`AVM use in the U.K. is growing rapidly. There are various companies exploring the idea of introducing their own AVMs; for example,
`Rightmove.co.uk Ltd. is set to launch its own AVM in early 2006. The key drivers for the success of AVMs are their relative cost and time
`efficiencies. Future regulatory issues may further drive their use, with the mandatory Home Condition Report coming into effect in 2007.
`Here, a property is inspected (but not officially valued) prior to the property going on the market. This report could be used in conjunction
`with an AVM to underwrite a mortgage speedily, or as a quality control spot check. Given that AVMs can generate valuations in moments
`and are relatively inexpensive, AVM use in property purchasing and origination of mortgage loans will likely continue to grow in the future.
`
`Group E-Mail Address
`StructuredFinanceEurope@standardandpoors.com
`
`Primary Credit Analysts: Victoria Johnstone, London (44) 20-7176-3864;
`
`www.standardandpoors.com/prot/ratings/articles/en/us/?articleType=HTML&assetID=1245330509010
`
`3/4
`
` 3
`
`

`
`8/6/13
`
`S&P | ARCHIVE | Criteria | Structured Finance | RMBS: Guidelines For The Use Of Automated Valuation Models For U.K. RMBS Transactions | Americas
`victoria_johnstone@standardandpoors.com
`Nadia Bahjat-Abbas, London (44) 20-7176-3655;
`nadia_bahjat-abbas@standardandpoors.com
`Alain Carron, Paris (33) 1-4420-7337;
`alain_carron@standardandpoors.com
`
`No content (including ratings, credit-related analyses and data, model, softw are or other application or output
`therefrom) or any part thereof (Content) may be modified, reverse engineered, reproduced or distributed in
`any form by any means, or stored in a database or retrieval system, w ithout the prior w ritten permission of
`S&P. The Content shall not be used for any unlaw ful or unauthorized purposes. S&P, its affiliates, and any
`third-party providers, as w ell as their directors, officers, shareholders, employees or agents (collectively
`S&P Parties) do not guarantee the accuracy, completeness, timeliness or availability of the Content. S&P
`Parties are not responsible for any errors or omissions, regardless of the cause, for the results obtained
`from the use of the Content, or for the security or maintenance of any data input by the user. The Content is
`provided on an “as is” basis. S&P PARTIES DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES,
`INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A
`PARTICULAR PURPOSE OR USE, FREEDOM FROM BUGS, SOFTWARE ERRORS OR DEFECTS, THAT THE
`CONTENT’S FUNCTIONING WILL BE UNINTERRUPTED OR THAT THE CONTENT WILL OPERATE WITH ANY
`SOFTWARE OR HARDWARE CONFIGURATION. In no event shall S&P Parties be liable to any party for any
`direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs,
`expenses, legal fees, or losses (including, w ithout limitation, lost income or lost profits and opportunity costs)
`in connection w ith any use of the Content even if advised of the possibility of such damages.
`
`Credit-related analyses, including ratings, and statements in the Content are statements of opinion as of the
`date they are expressed and not statements of fact or recommendations to purchase, hold, or sell any
`securities or to make any investment decisions. S&P assumes no obligation to update the Content follow ing
`publication in any form or format. The Content should not be relied on and is not a substitute for the skill,
`judgment and experience of the user, its management, employees, advisors and/or clients w hen making
`investment and other business decisions. S&P’s opinions and analyses do not address the suitability of any
`security. S&P does not act as a fiduciary or an investment advisor. While S&P has obtained information from
`sources it believes to be reliable, S&P does not perform an audit and undertakes no duty of due diligence or
`independent verification of any information it receives.
`
`S&P keeps certain activities of its business units separate from each other in order to preserve the
`independence and objectivity of their respective activities. As a result, certain business units of S&P may
`have information that is not available to other S&P business units. S&P has established policies and
`procedures to maintain the confidentiality of certain non-public information received in connection w ith each
`analytical process.
`
`S&P may receive compensation for its ratings and certain credit-related analyses, normally from issuers or
`underw riters of securities or from obligors. S&P reserves the right to disseminate its opinions and analyses.
`S&P's public ratings and analyses are made available on its Web sites, www.standardandpoors.com
`(free of charge), and www.ratingsdirect.com and www.globalcreditportal.com (subscription), and
`may be distributed through other means, including via S&P publications and third-party redistributors.
`Additional information about our ratings fees is available at
`www.standardandpoors.com/usratingsfees.
`
`Any Passw ords/user IDs issued by S&P to users are single user-dedicated and may ONLY be used by the
`individual to w hom they have been assigned. No sharing of passw ords/user IDs and no simultaneous
`access via the same passw ord/user ID is permitted. To reprint, translate, or use the data or information other
`than as provided herein, contact Client Services, 55 Water Street, New York, NY 10041; (1) 212-438-7280
`or by e-mail to: research_request@standardandpoors.com.
`
`C ontac t U s
`P rivac y and C ookie N otic e
`T erms of U s e
`Regulatory A ffairs and Dis c laimers
`C opyright © 2 0 1 3 Standard & P oor’s Financ ial Servic es LLC , a part of Mc Graw H ill Financ ial. A ll rights res erved.
`
`www.standardandpoors.com/prot/ratings/articles/en/us/?articleType=HTML&assetID=1245330509010
`
`4/4
`
` 4