An Automated Valuation Model (AVM) provides an estimate of the market value of a subject property at a specific time. To determine a value for the property in question, the most common AVM method is to use the prices of comparable properties recently sold. This is typically done by identifying all properties within a certain distance of the subject, then applying guidelines to find those that are most similar to the subject in terms of property characteristics and location factors.
This is a methodology similar to social networks, such as Facebook and LinkedIn: people are connected based on the similarity of relationships. For example, a person may be connected to their colleagues or someone they met at a conference, with the strength of a connection depending on how many connections they have in common.
Greater similarity… stronger connection
By applying this concept to the housing market, it is possible to build a network of properties – a network graph algorithm. The nodes in the graph represent individual properties, and the strength of each link indicates the similarity of the connected properties. A link can also be made on factors beyond the characteristics of the property. Take the example of a property used as a comparable sale by an appraiser for another property; this indicates that the two are considered similar to each other, so the bond between the pair would be very strong.
CoreLogic data sources can be leveraged to create a network of all properties in the country, simplifying the selection of comparable properties, with the best being those most closely related to the topic. The attached illustration of a network graph shows a connection between a subject (red node) and its “comparables” (blue nodes); all connections are linked directly or indirectly.
For a better conceptualization of a network graph, consider placing these interconnected data points over a selected area in Google Maps. In this example, the red pushpin represents the property in question, the pink circles indicate comparable properties previously used by an appraiser, and the blue pins are ideal comparables identified by the CoreLogic network graph algorithm.
When analyzing the map, the first point to remember is that the CoreLogic algorithm retrieves all but one of the appraiser’s comparables. It’s a comparable indicator of screening success, because no one knows a neighborhood better than a local assessor. The second finding is that comparables tend to be in the same neighborhood as the subject, or in the same general neighborhood as those previously selected by the appraiser.
The single model methodology is the best single approach
Leveraging a network graph algorithm to select comparables in the same way as a trained local expert is just one of the innovations driving the new Total home valueX MY V.
This recently introduced assessment solution leverages artificial intelligence and cloud-based machine learning capabilities. Using a single model methodology to support diverse use cases and markets, users only need to validate a single model, which is tuned per use case to deliver success rates, accuracy and consistency.
With access to a property database of over 5.5 billion records (updated daily) that captures 99.9% of US properties and spans over 50 years, THVX produces automated appraisals that can be used wherever a property’s current value is relevant.