AVM's (Automated Valuation Models), or APV's, (Automated Property Valuations), hereafter referred to as 'AVM's', have been in use since the early 1990's by Banks, Financial Institutions, Appraisal Districts, Appraisers, and Investors.

[AVM's deal only with data; they do not and cannot 'know' what interior or exterior remodeling has been completed, what landscaping is in place, etc. as an Appraiser would, unless that information is already on the Tax Assessor rolls.]

AVM's are commercially judged on two major criteria: 'HIT RATES', wherein the initial 'hit' is classed as locating the Subject Property, and the second 'hit' is classed as valuing the Subject Property, and secondly, the reliability and variance of the valuations. [The best commercial AVM's find the Subject Property approximately 75% of the time, and value 80% of those so found. In National Tests, the RDI AVM located the Subject Property 98% of the time and valued same 95% of the time.]

RDI's expertise in approximate string matching, its Intellectual Property Algorithms and mathematical sophistication account for the tremendous uplift in RDI's AVM locating and valuing properties over the best commercial AVM's on the market today. Major Financial Institutions, such as First American Real Estate Tax Services and Data Quick, Inc. have licensed RDI's underlying Matching Technology and Products.

AVM's utilize sales data, property data characteristics,( such as bedrooms, baths, half-baths, year built, condition, grade, stories, topography), influence factors, lot size, gross living area, land and improved assessed values, land use codes, and various other secondary vectors relative to the physical property to generate valuations. Secondary vectors would include factors such as latitude, longitude, census blocks, groups and tracts and zip codes, as well as GIS (Geographical Information Systems) data, typically using a property address to produce the GIS characteristics. [If a closed polygon of the property, using latitude-longitude coordinates is available, even when a property address is not present in the assessor files, it is possible to calculate a centroid, or mid point of the property polygon in a X-Y coordinate system, thereby determining the property location.]

AVM's typically use HEDONIC models of the Subject Property and selected comparables based upon both the aforementioned feature vectors and SIMILARITY between the Subject Property and the selected comparables. NON-HEDONIC models use REPEAT SALES of a single property over the course of many years to develop indexes of value at the appropriate distance from the subject, namely as street range, census block and block group, census tract and Zip code. AVM's accept a Subject Property as input, using either TaxID or address, and then attempt to locate same in its data bases. If successful, the AVM locates comparable properties, similarly located by range of distance from the subject, and time of sale-said constraints are generally specified by the user, Some AVM's, like RDI's, permit the user to specify additional criteria such as 'SAME SUBDIVISION' or 'SAME SUBDIVISION - SAME SECTION', Date Range of sales, Upper - Lower bounds of Sale Price, Gross Living Area, etc. Once the AVM finds comparables based upon the specified constraints, it begins the task of determining 'SIMILARITY'.

Determining 'similarity' requires utilization of mathematical methods and models, along with statistical measures indicating the reliability of the model's results. AVM models range from a SINGLE MODEL, such OLS (Ordinary Least Squares), Regression, and Multiple Regression to COMPOUND MODELS using several model methods. Model terms, such as 'Predictive Modeling', 'Neural Nets', 'HMM' [Hidden Markov Model], Non-Linear Models, Replacement Cost Models, (e.g. Marshall & Swift,) etc. all refer to specific mathematical approaches which take the property characteristic vectors, minimize differences between the subject and the comparables, and then automatically select comparables that are most 'similar' to the Subject Property vectors.

The similarity is typically determined in two mathematical phases:

1. First, find the MAX (Min( SV1, SV2…SVN)) ~= (MAX(MIN(CV1,CV2…CVN )) for all candidate comparables.

2. Second, utilizing these values as DATA POINTS, apply the relevant mathematical model(s) in terms of which data points have the 'best fit, exercising statistical measure and special mathematical models to select the 'best comparables,( perhaps arbitrarily, X < 10 and X > 3, from the initial comparables set selected.)

RDI's AVM has 12 Mathematical Models, a Replacement Cost Model, and NON-HEDONIC indexing, all of which are fed into the RDI-AVM mathematical engine. As a final step, the RDI-AVM uses an EXPERT SYSTEM, a form of Artificial Intelligence, to determine the optimal 3-4 comparables, using statistical measures of reliability and confidence; the selected comparables are then displayed on the RDI-AVM report, These same measures are utilized to determine the 'Highest and Lowest Reasonable Value' values, which also on the RDI-AVM Report.

RDI's AVM was the first to have what is termed, 'AAVM (Appraiser-Assisted AVM') capability, which allows almost every subject or comparable vector to be adjusted, over-ridden, or added to by the user. As a result, the RDI-AVM report has an appearance very similar to an Appraisal Report. Superscripts identify user-modified values as appropriate and serve as fraud reduction tools-lenders are alerted of the user modification to the RDI-AVM report. An additional method of fraud reduction permits only EXISTING properties on an Assessor roll to be used as additional comparables, thereby eliminating any fictional properties in the RDI-AVM value.

Finally, RDI's AVM is a complete Modeling System that cannot only accurately value BACKWARD IN TIME, it can value a home assessed at $125,000 against comparables of no less than $500,000. This allows accurate determination of value of the $125,000 home placed in Tanglewilde, at $329,450. The RDI-AVM has a 13-year proven accuracy of +/- 1% across all property value ranges between $20,000 and $10,000,000 homes.

For more information regarding RDI's AVM, please contact Dr. Frank Kautzmann III at Real Data Inc. 16420 Park Ten Place, Suite 200 Houston, Texas 77084 (281) 398-0880 ext 101, Dr. Kautzmann may also be contacted via email at drfnk@realdata.net.