Method for estimating the value of real property

[0002] This is a continuation-in-part of application Ser. No. 08/519,266 filed Aug. 25, 1995, now abandoned.

[0003] This application is related in subject matter to application Ser. No. 09/118,103 filed Jul. 17, 1998, and application Ser. No. 09,118,188 filed Jul. 17, 1998.

[0004] This invention relates generally to real estate appraisals and more particularly to a method for estimating the value of real property.

[0005] Real estate appraisals are used to estimate the defined value of a real property interest in real estate. The real estate appraisals are quite useful in many types of real estate transactions. However, a problem with real estate appraisals is that they take a lot of time to perform and are expensive. Typically, an appraiser is sent to a subject property (i.e. a property to be appraised) to inspect and determine an estimate price at which the property would most likely sell in the market under certain specified conditions. In order to estimate the price, the real estate appraiser finds recent sales that are comparable to the subject property. Then the most relevant units of comparison for each sale are determined from the comparables. Next, the sales prices of the comparables are adjusted to reflect their differences from the subject property. The adjusted sales prices are then reconciled to the comparables in order to derive a single value estimate of the subject property, which is a reflection of the probable price that would be agreed upon between knowledgeable parties acting without duress in a competitive market. This real estate appraisal process typically takes about four days to finish, which may be too long in many of today's fast paced real estate transactions. In addition, this appraisal process does not provide much insight on how to select comparable properties, which often results in estimates that do not accurately reflect the best price. Therefore, there is a need for a method that speeds up the appraisal process of a subject property and that provides a better estimate.

[0006] Thus, in accordance with this invention, there is provided a computer-implemented method for estimating the price of a subject property. The method comprises retrieving a set of real estate properties comparable to the subject property from a case base. The comparable properties and the subject property are characterized by a plurality of common attributes each having a respective value. Each attribute value from the comparable properties are evaluated to the same attribute value of the subject property on a fuzzy preference scale indicating desirable and tolerable deviations from an ideal match with the subject property. Each evaluation generates a preference vector having a value between 0 and 1. A measurement of similarity between each comparable property and the subject property is then determined. Next, comparable properties are extracted according to a predetermined threshold. The price of the extracted comparable properties are then adjusted to the value of the subject property by evaluating additional characteristics between the comparable properties and the subject property. Comparable properties are then extracted according to predetermined selection criteria. The extracted comparable properties are then aggregated into an estimate price of the subject property.

[0007] While the present invention will hereinafter be described in connection with an illustrative embodiment and method of use, it will be understood that it is not intended to limit the invention to this embodiment. Instead, it is intended to cover all alternatives, modifications and equivalents as may be included within the spirit and scope of the present invention as defined by the appended claims.

[0008] FIG. 1 is a flow chart setting forth case retrieval and case adaptation operations performed in this invention;

[0009] FIG. 2 is a system for performing the case retrieval and case adaptation operations according to this invention;

[0010] FIG. 3 is a flow diagram illustrating the operations performed by the system shown in FIG. 2;

[0011] FIG. 4 is a graphical display illustrating the preference criteria used in this invention; and

[0012] FIG. 5 is an example of a similarity distribution for a set of comparable properties.

[0013] This invention estimates the value of a subject property by using case-based reasoning principles. In particular, case-based reasoning is used to automate the steps of finding recent sales comparable to the subject property, determining the most relevant units of comparison, comparing the subject property with the comparables, adjusting the sales price of the comparables to reflect the differences from the subject, and reconciling the adjusted sales prices of the comparables to derive an estimate of the subject. This invention includes two modules. The first module is case retrieval, where recent sales of properties most comparable to the subject property are retrieved. The second module is case adaptation, where the selection of comparable properties is further refined and prices of these properties are then adjusted to reflect the differences with the subject property. Both the case retrieval and case adaptation modules are explained below in further detail.

[0014] FIG. 1 is a flow chart setting forth the case retrieval and case adaptation operations performed according to this invention. Referring now to FIG. 1, an initial set of real properties that are comparable to the subject property are retrieved at 10. The initial set of comparable real properties and the subject property are both characterized by a plurality of common attributes each having a respective value. The attributes include transaction characteristics such as location of the property, date of sale of the property, the living area, the lot size, the number of bedrooms, and the number of bathrooms. Although the illustrative embodiment is described with reference to these attributes, it is within the scope of this invention to use other attributes such as type of construction of the property, parking spaces, location influence of the property, foundation of the property, the roof type and roof cover of the property, garage or carport, style of the property, etc. Each attribute value from the comparable properties is then evaluated to the same attribute value of the subject property and a measure of similarity is then generated at 12. The price of each of the comparable properties is then adjusted at 14 to reflect the value of the subject property by applying a set of modification rules. The modification rules evaluate additional characteristics between the comparable properties and the subject property and adjust the price of the property accordingly. In the illustrative embodiment, the additional characteristics are the number of fireplaces, pools, the effective age of the properties, the quality of the properties, and the condition of the properties. After price adjustment, a smaller set of more closely related comparable properties is then extracted at 16. The extracted comparable properties are then aggregated at 18 into an estimate price of the subject property. After the estimate price has been generated, a measurement of confidence indicating the reliability of the estimated value is generated 20.

[0015] FIG. 2 shows a system 22 for performing the case retrieval and case adaptation operations according to this invention. The system 22 includes a comparable property database 24, which is preferably a case base having a plurality of properties. The system 22 also includes an adjustment rules database 26 containing rules for adjusting the prices of the comparable properties in the comparable property case base 24. A personal computer or work station 28 having a hard drive 30, an input device such as a keyboard 32 and a mouse 34, and an output device such as a display 36 and a printer 38, is connected to the case base 24 and the database 26. The computer 28 uses SQL queries to retrieve the comparable properties from the case base 24 and performs a series of case adaptation operations to the properties according to the adjustment rules.

[0016] The case retrieval and case adaptation operations are shown in further detail in the flow diagram of FIG. 3. After information about the subject property has been acquired, similar comparable properties are selected at 40 from the case base 24. In the illustrative embodiment, the case base 24 contains about 600,000 real estate properties, with each property being characterized by about 166 attributes. Typically, the selection at 40 returns anywhere from four to one hundred comparable properties. The selection is performed by comparing specific attributes (i.e., location, date of sale, living area, lot size, number of bedrooms, and number of bathrooms) of the subject property with the same attributes of each comparable properties. All of the retrieved potential comparable properties have values that are within allowable deviations for the subject property. If the size of the comparable set of properties is too small (e.g., less than 10), then the allowable deviations can be adjusted so that more comparables of lesser quality can be obtained.

[0017] After the initial set of comparable properties has been retrieved, each of the values for the attributes (i.e., location, date of sale, living area, lot size, number of bedrooms, and number of bathrooms) are evaluated against the same attributes of the subject property on a fuzzy preference scale that indicates desirable and tolerable deviations from an ideal match. For example, in the illustrative embodiment, the maximum allowable deviations for the attributes are within one mile for the location attribute, 12 months for the date of sale attribute, plus or minus 25% for the living area attribute, plus 100% or minus 50% for the lot size attribute, plus or minus two percent for the number of bedrooms attribute, and plus or minus two percent for the number of bathrooms attribute. After each attribute is evaluated, a preference vector having a value between 0 and 1 inclusive, is generated, with 1 being the best value. These values represent the partial degree of membership of each attribute value in the fuzzy sets and fuzzy relations representing the preference criteria provided in the fuzzy preference scales.

[0018] An understanding of the above preference criteria for the first four attributes (i.e., the date of sale, the distance from subject property, the living area, and the lot size) is graphically described in FIG. 4. Each of the attributes in FIG. 4 have a trapezoidal shape representing its criteria evaluation functions. For each attribute, the broader base of the trapezoidal shape represents the range of tolerable values and corresponds to the interval-value used in the preliminary retrieval query. The smaller top or core of the trapezoidal shape represents the most desirable range of values. An attribute value falling inside the most desirable region will receive a preference value of 1. As the feature value moves away from the most desirable range, its associated preference value will decrease from 1 to 0. At the end of this evaluation, each comparable will have a preference vector, with each element taking values in the [0,1] interval. These values represent the partial degree of membership of each attribute value in the fuzzy sets and fuzzy relations representing the preference criteria provided in the fuzzy preference scales. Typically, comparable properties selected in the preliminary retrieval that have attribute values falling outside the tolerable value range will not be evaluated.

[0019] In the illustrative embodiment, the preference distribution for the date-of-sale attribute for a comparable property that was sold within three months of the present date is 1. If the date of sale of the comparable property was 6 months ago, its preference value will be 0.67. If the date of sale of the comparable property was 9 months ago, its preference value will be 0.33. Any comparable property with a date of sale of more than 12 months is given a preference value of zero. For the distance from the subject property attribute, comparable properties located within 0.25 miles from the subject property have a preference value of 1. If the comparable property is located a half of a mile from the subject property, then its preference value will be 0.67. If the comparable property is located three quarters of a mile from the subject property, then its preference value will be 0.33. Any comparable property located more than one mile from the subject property is given a preference value of zero. For the living area attribute, a comparable property having a living area that is within 94% to 106% of the living area of the subject property will have a preference value of 1. If the comparable property has a living area that is within 75% to 94% or within 106% to 125% of the living area of the subject property, then the comparable property is given a preference value between zero and 1. If the comparable property has a living area that is less than 75% or greater than 125% of the living area of the subject property, then the comparable property is given a preference value of zero. For the lot size attribute, a comparable property having a lot size that is within 87.5% to 112.5% of the lot size of the subject property will have a preference value of 1. If the comparable property has a lot size that is within 50% to 87.5% or within 112.5% to 150% of the lot size of the subject property, then the comparable property is given a preference value between zero and 1. If the comparable property has a lot size that is less than 50% or greater than 150% of the lot size of the subject property, then the comparable property is given a preference value of zero. The tolerable and desirable ranges of values for the five attributes are illustrative of possible values for use with the present invention and are not intended to be limiting.

[0020] The remaining two attributes not shown in FIG. 4, the number of bedrooms and the number of bathrooms, have preference functions which are illustrated in Tables 1 and 2, respectively. For example, if the subject property has five bedrooms, then Table 1 will provide a preference value of 1 for comparable properties having five bedrooms. However, if the comparable property has six or more bedrooms, then the comparable will be given a preference value of 0.80. Also, Table 1 indicates that a comparable property having four bedrooms will have a preference value of 0.60, three bedrooms will have a preference value of 0.50, and two or less bedrooms will receive a preference value of zero.

[0021] Table 2 can be used in a similar manner to generate preference functions for the number of bathrooms attribute. For example, if the subject property has 2 bathrooms, then Table 2 will provide a preference value of 1 for comparable properties having two bathrooms. However, if the comparable property has two and a half bathrooms, then the comparable will be given a preference value of 0.70. Also, Table 2 indicates that a comparable property having three bathrooms will have a preference value of 0.25, three and half bathrooms will have a preference value of 0.05, four or more bathrooms will have a preference value of zero. In addition, Table 2 indicates that a comparable property having one and a half bathrooms will have a preference value of 0.70 and one bathroom will have a preference of 0.1.

[0022] After each attribute of the comparable properties has been evaluated against the subject property and a preference vector has been generated, the measurement of similarity between each comparable and the subject property is determined. The measurement of similarity is a function of the preference vector computed above and of the priorities of the attributes, which are reflected by a set of predetermined weights. The predetermined weights for the illustrative embodiment are shown in Table 3 under the weight column. In the illustrative embodiment, the living area attribute has a weight of 0.3, the date of sale and distance attributes both have a weight of 0.2, the lot size attributes have a weight of 0.1, while the number of bedrooms and bathrooms attributes have a weight of 0.05.

[0023] The measurement of similarity for a comparable property is determined by multiplying the predetermined weight by the preference vector generated for each attribute. This product results in a weighted preference value. After all of the weighted preference values have been determined, the weighted preferences are summed together to generate the measurement of similarity. An example of a measurement of similarity computation between a comparable property and a subject property is provided in Table 3. In the example provided in Table 3, the subject property has a living area of 2000 square feet, a lot size of 20,000 square feet, three bedrooms, and two and a half bathrooms. The comparable property was sold six months ago, is located 0.2 miles from the subject property, has a living area of 1800 square feet, a lot size of 35,000 square feet, three bedrooms and two bathrooms. A comparison between the subject property and the comparable property is provided in the fourth column for each attribute. In Table 3, the living area comparison is 90%, the lot size comparison is 175%, and the number of bedroom comparison is 0%. As described above, each comparison results in a preference which is multiplied by the predetermined weight. The weighted preferences for each attribute for the comparable property are listed in the weighted preference column and the measurement of similarity is the sum of the weighted preferences. In Table 3, the measurement of similarity for this particular comparable property is 0.7915.

[0024] After the measurement of similarities have been computed for all of the comparable properties, the comparables are then sorted in decreasing order of similarity. After sorting, the comparables are arranged in a preference distribution as shown in FIG. 5, with the comparable property having the highest measurement of similarity placed at one end of the distribution and the comparable property having the lowest measurement of similarity placed at the opposite end of the distribution. The comparable properties are then compared against a predetermined threshold that reflects desirable and tolerable deviations of an ideal match with the subject property. More specifically, the comparable properties that have a measurement of similarity above the predetermined threshold will be extracted for further review, while the comparable properties below the threshold are removed from further consideration. FIG. 5 shows two possible similarity distributions for two different retrievals. In these distributions, a value of 0.5 is used as the predetermined threshold. Therefore, comparable properties having a measurement of similarity above 0.5 are extracted for further review, while the comparables with measurements of similarities less than 0.5 are removed and no longer considered. In FIG. 5, retrieval number one has 11 comparable properties having a measurement of similarity above 0.5, while retrieval number two has five comparable properties with a measurement of similarity above 0.5.

[0025] Instead of using a predetermined threshold to determine which retrieval provides the best results, an alternative approach is to take the average of the similarity values of the retrieved comparables. This corresponds to the area under the curve of the distributions and is determined by taking the average measurement of similarity. For example, the average similarity measure for retrievals one and two in FIG. 5 would be determined as follows:

[0026] Average Similarity Measure Subject 1 (from best 8 comps):

[0027] (1+1+0.85+0.8+0.7+0.7+0.7+0.5)/8=0.78125

[0028] Average Similarity Measure Subject 2 (from best 8 comps):

[0029] (1+0.9+0.8+0.7+0.7+0.4+0.035+0.25)/8=0.6375

[0030] Referring again to FIG. 3, the comparable properties that have been selected for further review at 40 are then adjusted to reflect the value of the subject property at 42. In particular, any difference between the subject property and the comparable properties that would cause the comparables to be more or less valuable than the subject property will require an adjustment. Thus, if a comparable property is superior to the subject property, then an adjustment is needed to decrease the price of the comparable. However, if the comparable property is inferior to the subject property, then an adjustment is needed to increase the price of the comparable.

[0031] The adjustments to the price of the comparable properties are performed by using the plurality of adjustment rules stored in the adjustment rule database 26. The adjustment rules are generated from the plurality of attributes stored in the case base 24 for all of the comparable properties. As mentioned earlier, there are approximately 166 attributes available for the subject property and the comparable properties in the illustrative embodiment. A illustrative listing of the attributes are presented below. The attributes described with a # are numeric and the remaining attributes are textual. The numeric attributes are described with a number and the textual attributes are described with text. For example, the attribute total room is described with a number such as three, four, or the like, and the pool attribute is described with a text format such as indoor, spa, etc.

[0032] Based on these attributes, the following adjustment rules are generated in the case base 24 and stored in the adjustment rule database 26.

[0033] In order to accommodate for even more or less bathrooms, Table 4 takes the difference between the subject property and the comparative property (i.e., @) and multiplies the difference by five. For example, if the subject property has seven bathrooms and the comparable has three, then the adjustment would be 20 ([7-3]*5). If the subject property has three bathrooms and the comparable has seven, then the adjustment would be −20 ([7-3]*-5).

[0034] In order to accommodate for even more or less bedrooms, Table 5 takes the difference between the subject property and the comparable property (i.e., @) and subtracts the difference by one and multiplies the difference by 3.5. For example, if the subject property has six bedrooms and the comparable has four, then the adjustment would be 3.5 [[(6-4)-1]*3.5]. If the subject property has four bedrooms and the comparable has six, then the adjustment would be −3.5 [[(6-4)-1]*−3.5].

[0035] These adjustment rules are then applied to the comparable properties selected at 40 in order to adjust for the value of the subject property.

[0036] An example of an adjustment for a comparable property is provided in Table 6. In the example provided in Table 6, the comparable property has a sale price of $175,000 dollars. However, the comparable property has a building area of 1800 square feet, while the subject property has a building area of 2000. Using the adjustment rules for the attribute building area, the price of the comparable is adjusted by $5450 (i.e., 22+(175000*0.00003)=$27.25 per square foot which is (200*$27.25=$5450)). Also, the price of the comparable is adjusted for the lot area since the comparable has a larger lot size. In Table 6, the lot area attribute is adjusted by $1/sq ft for a total of −$5000. Since the comparable has two bathrooms and the subject property has two and a half bathrooms, the price needs to be adjusted by using the rules provided in Table 4, which turns out to be $2000. There are no adjustments necessary for the bedroom attribute because both the subject property and the comparable property have the same number of bedrooms. Since the comparable does not have a fireplace and the subject property has one, the price needs to be adjusted accordingly. Using the adjustment rule for fireplaces, the price is adjusted $2000. If the adjustment rules are used for the effective year, quality, condition, and pool attributes for the subject and comparable property, the rules will generate an adjustment of $2800, $3500, $0, and $10,000, respectively. All of the adjustments are then summed with the sale price of the comparable property to arrive at the adjusted price. In Table 6, the adjusted price of the comparable property is $195,750.

[0037] Referring again to FIG. 3, after all of the adjustments are applied to the sales price of the comparable properties, another set of comparable properties that more closely match the subject property are extracted at 44. In the illustrative embodiment, 4-8 comparables are selected at 44. If less than four comparables are selected, then the comparables may not correctly reflect the market and if more than eight comparables are used, then some of the comparables may not be similar enough to the subject property. If it is not possible to find four comparables similar to the subject property, then no value estimate may be calculated for the subject property. However, if there are many comparables (i.e. about 100 hundred), then it is necessary to filter out the poorer comparables. In the illustrative embodiment, the best results are attained by keeping comparable properties that have no single adjustment larger than 10% of the sale price, a net adjustment that does not exceed 15% of the sale price, a gross adjustment that does not exceed 25% of the sale price, and a dollar per square foot that does not vary more than 15%. Basically, all of the adjusted comparable properties are excluded from further consideration if any comparable has a single adjustment larger than 10%, a net adjustment larger than 15% or a gross adjustment larger than 25%.

[0038] The best (i.e. four to eight) of the remaining adjusted comparable properties are selected by sorting and ranking each of the comparables' measurement of similarity, the net adjustment, and the gross adjustment, in the manner as shown in Table 7. In particular, the comparables having the highest measurement of similarity score are placed at the top of the list and ranked in descending order. In Table 7, comparable property 113-012 has the highest measurement of similarity score and is ranked one, while comparable property 331-018 has the lowest measurement of similarity score and is ranked nine. Next to the measurement of similarity score and rank are the net and gross adjustment and respective rankings for the comparable properties. The rankings for the net and gross adjustment are attained in the same manner as the measurement of similarity. The rankings for the measurement of similarity, net adjustment, and gross adjustment, for each comparable are then summed across the board to produce a total ranking. The comparables with the lowest total rank are considered the best. In Table 7, comparables 113-012, 306-008, and 334-010, are the three best comparables.

[0039] After the comparables have been ordered, it is necessary to determine how many of these comparables are to be used. Generally, the sales prices of the comparables should bound the sales price that will be estimated for the subject property. Therefore, it would be favorable to select comparables with both a negative and positive net adjustment. A comparable property with a negative net adjustment is likely to have an unadjusted price over the final estimate and a comparable with a positive net adjust is likely to have an unadjusted price under the final estimate. So, in order to do this, a temporary set of comparables is created by repeatedly adding the comparable with the best similarity score to the set until there are at least four comparables in the set and there is at least one comparable of each sign (negative and positive) net adjustment. In Table 7, the comparables with the top six similarity scores would be included in the set. All other comparables are discarded. Of the comparables in the set only four of each sign net adjustments are retained. The four retained are the four comparables with the lowest total rank. In the example, comparable 305-006 would be discarded since there are four comparables with a positive net adjustment and lower total rank. The five comparables selected form the final set of comparables.

[0040] Referring again to FIG. 3, after the best of the adjusted comparables have been selected, the adjusted prices of the selected comparables are aggregated into an estimate price of the subject property at 46. The aggregated estimated price is determined by multiplying the adjusted price of the comparable properties to their respective measurement of similarity and summed together to generate a total weighted price. Next, the total weighted price is divided by the total of the similarity measurements for the comparable properties. The result is an estimate price of the subject property. An example of the aggregation for the comparables provided in Table 7 is shown in Table 8. In this example, the total weighted price is $757,640 and the total similarity score is 3.83. Thus, dividing $757,640 by 3.83 results in an estimate price of 199,900 for the subject property.

[0041] After producing the final estimate of the value of the subject property, a measurement of confidence indicating the reliability is generated. In particular, the confidence measurement in the estimate can be obtained by averaging the similarity scores of the comparables in the final selection, or by averaging the number of comparables over a threshold in the primary retrieval. The estimate is justified by displaying the comparables in enough detail so that they can be shown to be similar to the subject.

[0042] It is therefore apparent that there has been provided in accordance with this invention, a method for estimating the price of a real property that fully satisfy the aims and advantages and objectives hereinbefore set forth. The invention has been described with reference to several embodiments, however, it will be appreciated that variations and modifications can be effected by a person of ordinary skill in the art without departing from the scope of the invention.