How much is a property worth?

Property tax is the most widely used source of municipal tax in the developing world ( William Dillinger , 1988) but remains relatively untapped ( Riël Franzsen and William McCluskey , 2017). Since Governments are challenged by a lack of revenue, essential in addressing a multitude of civic challenges, levying a tax on immovable property offers a means to raise revenue but to do this they need efficient and equitable means to assess property values. Automated Valuation Models and PLACE data offer a solution.

Regression-based Automated Valuation Models (AVMs) are algorithms used property characteristics like size, bathroom count, location, etc., to estimate value. They are used by Valuation Offices to more efficiently scale valuation accuracy and assessment uniformity especially when dealing with thousands of properties thereby reducing the cost involved in building property rolls.

Computer Aided Mass Appraisal (CAMA) systems help manage sophisticated AVM’s capable of providing valuation estimates for thousands or millions of properties, cost-effectively (RICS   AVM RoadMap   2021).

PLACE high resolution imagery can be used to further improve AVM accuracy and defensibility, especially when data quality is lacking. The ability of AVMs to yield reasonable estimates of value are mostly determined by the quality and availability of data coupled with the activity/efficiency of the market. PLACE imagery helps identify and create new variables to feed into AVMs for more accurate valuation estimates.

We have been working with the Center for Appraisal Research and Technology (CART) who are using PLACE imagery to develop an AVM for the island of Providenciales (Turks and Caicos Islands) in conjunction the country's  Valuation Office . Using publicly available sales records from the  Turks and Caicos Real Estate Association , initial models have been developed to test attributes that best explain the variation in property price (e.g. square footage, number of baths and year built). A regression model was created to test whether property prices could be reasonable estimated/predicted given these features. If the model found a reliable and consistent pattern between a property feature and price, it was deemed to be “statistically significant” and included as a model variable. The resulting model yielded a mathematical formula that may be used to arrive at an estimated sale price for a property based on the public records on which it was trained.

PLACE imagery showed great promise for not only improving valuation models, but also populating core physical attributes in a cadaster, including building footprints and other property attributes that are commonly associated with value, including waterfront status, pool status, solar panels, and more. The image quality, including color, shading, and resolution allowed for better identification of buildings and amenities as compared to Google Maps and for identification of new properties, construction, and finished roads given its currency.

After adding in variables derived from PLACE imagery, regression models could explain 70-78% percent of the variation in sale price meeting International Association of Assessing Officers ( IAAO ) standards for valuation accuracy and uniformity. PLACE imagery also helped models better understand market behavior of all variables, as evidenced by more statistically significant variables with results of the study showing:

• Pools, on average, are associated with a price increase of 30%
• Oceanfront properties, on average, sell for 56% more than non-oceanfront ones
• Properties located on a channel/canal, on average, sell for 45% more than those that are not