Using Cluster Analysis to Improve Housing Appraisals
Location
CBM 149
Start Date
19-4-2019 1:40 PM
Department
Economics
Abstract
Residential housing appraisals are critical to well-functioning housing markets. They are used to inform listing prices, support mortgage applications, evaluate property taxes, support homeowners’ insurance estimates, and provide a guide to buyers and homeowners about the value of their form. We examine a sample of 20,000 homes in the Milwaukee, Wisconsin metropolitan area to examine the role of geography and proximity in constructing housing appraisal estimates. The process of buying a home nearly always comes with a standard comparison-based appraisal using the URAR (Uniform Real Estate Appraisal Report) form, where professional appraisers choose three properties comparable to the property being appraised and go through a formal but partially subjective process of adjustments to the comparisons to arrive at an estimated market value of the subject. The comparison homes are drawn from homes of similar style and similar geography to the subject. However, homes are unique and markets can be highly localized, making it difficult to find suitable comparison homes. An alternative method of appraisals uses large datasets of thousands of homes to estimate using regression techniques a value based on the characteristics of a particular home. This is the approach used by popular websites such as Zillow and Trulia. This approach is much less subjective, relying not on the appraiser’s judgment but instead on data and algorithms to provide appraisal values. However, good local appraisers with good local knowledge are able This paper uses clustering techniques to identify homes of similar characteristics, and asks the question whether appraisals can in fact gain information by including properties outside their immediate geographic area. Step one of this process is creating statistical clusters of homes using the k-means method, where homes of similar styles, ages, and sizes among other characteristics are included into a small number of clusters. Step two involves creating regression-based appraisal values using within-cluster homes from a narrow geography similar to what might form the selection area for a comparison-based residential appraisal. Goodness-of-fit measures are constructed from this regression. Step three involves expanding the pool of homes to include the full cluster, including homes that are of similar style but potentially from very diverse geography. Results are presented that evaluate the impact on goodness-of-fit of choosing nearby homes versus distant homes.
Using Cluster Analysis to Improve Housing Appraisals
CBM 149
Residential housing appraisals are critical to well-functioning housing markets. They are used to inform listing prices, support mortgage applications, evaluate property taxes, support homeowners’ insurance estimates, and provide a guide to buyers and homeowners about the value of their form. We examine a sample of 20,000 homes in the Milwaukee, Wisconsin metropolitan area to examine the role of geography and proximity in constructing housing appraisal estimates. The process of buying a home nearly always comes with a standard comparison-based appraisal using the URAR (Uniform Real Estate Appraisal Report) form, where professional appraisers choose three properties comparable to the property being appraised and go through a formal but partially subjective process of adjustments to the comparisons to arrive at an estimated market value of the subject. The comparison homes are drawn from homes of similar style and similar geography to the subject. However, homes are unique and markets can be highly localized, making it difficult to find suitable comparison homes. An alternative method of appraisals uses large datasets of thousands of homes to estimate using regression techniques a value based on the characteristics of a particular home. This is the approach used by popular websites such as Zillow and Trulia. This approach is much less subjective, relying not on the appraiser’s judgment but instead on data and algorithms to provide appraisal values. However, good local appraisers with good local knowledge are able This paper uses clustering techniques to identify homes of similar characteristics, and asks the question whether appraisals can in fact gain information by including properties outside their immediate geographic area. Step one of this process is creating statistical clusters of homes using the k-means method, where homes of similar styles, ages, and sizes among other characteristics are included into a small number of clusters. Step two involves creating regression-based appraisal values using within-cluster homes from a narrow geography similar to what might form the selection area for a comparison-based residential appraisal. Goodness-of-fit measures are constructed from this regression. Step three involves expanding the pool of homes to include the full cluster, including homes that are of similar style but potentially from very diverse geography. Results are presented that evaluate the impact on goodness-of-fit of choosing nearby homes versus distant homes.
Comments
Michael G. Wenz and Scott W. Hegerty are the faculty sponsors of this project.