Automated valuation models (AVMs) like the Xome Value® are useful tools when deciding to buy or sell a home. However, some AVMs tend to be more accurate than others when estimating a property’s value due to a range of factors.
An AVM is only as reliable as its data. AVMs that follow the best practices for data verification, data analysis, market analysis, and ongoing quality assurance deliver the most reliable property valuation estimates, according to the Standard on Automated Valuation Models, a criteria set by the International Association of Assessing Officers that measures the fairness, quality, equity, and accuracy of AVMs.
If a model works with incomplete, incorrect, or outdated data, its property valuation estimates will naturally be off. It would end up omitting critical factors that influence the market value and as a result, it may be unable to capture relevant market dynamics that provide the necessary context for an accurate valuation.
There are important elements that exist in every well-designed AVM that can prevent this from happening. Data, validation, and testing are foundational in any AVM, but there are four key elements that our technical teams at Xome® keep in mind when building and maintaining the Xome Value.
Data source quality
Automated valuation models need several types of property, neighborhood, and market data to accurately estimate property values. Without high-quality data sources that encompass this data, an AVM’s performance can be significantly impaired, leading to misleading valuations.
For example, outdated data may not reflect recent renovations, comparable properties, or market trends, which can affect the final estimate. This is why it is important to have high-quality data sources, such as the Home Price Index and public records (like tax assessment records), to improve AVM results.
Data must also be current, and it is recommended that engineering teams regularly perform data quality reviews as a part of the machine learning lifecycle to check the completeness and consistency of data, along with identifying and removing outliers. Engineers and analysts should also use statistical analysis to confirm the quality of data will support the modeling process.
Comparable properties
Automated valuation models are unable to estimate a property value accurately without a fair amount of similar property data for comparison. Comparable properties are recently sold properties that have similar characteristics to the subject property, such as location, size, and condition. This helps to better reflect the market conditions with real-world transaction data and trends. Without comparable property data for the AVM to use, the generated valuation estimates would be incomplete and skew results.
Once that data is available, adjustments must be made between the subject property and comparable properties. These include value adjustments for discrepancies in key attributes such as size, age, condition, and location. Adjustments are calculated by quantifying how each difference affects the property’s value and are then applied to align the comparable properties more closely with the subject property.
Time series component
The time series component is an essential element for an automated valuation model to identify temporal changes. It refers to the individual factors within the time series data set that tracks property values and market trends over time.
Here’s how it works: the time series data set is broken down into trend, seasonality, and cycles components to track and forecast changes in market conditions. But the AVM cannot understand these components, adjust for variations over time, nor account for periodic fluctuations without performing a market trends analysis.
For example, a rising trend in local employment opportunities can impact property values. Significant job growth in an area may lead to an increased demand for housing as more people relocate to that area looking for employment, and where there is an increased demand there are typically higher property values as buyers are willing to pay more to live in an area with more jobs. If an AVM does not account for that market trend, its results may be misleading and could potentially misinform its users.
Xome shows some of this market trend analysis on each property’s page and is also included in the Xome Value.
Property characteristics statistics
Property characteristics are the data attributes of individual properties, such as the building and lot size, property type, and number of bedrooms and bathrooms, that are used to estimate their value.
Different property characteristics influence property value in a few ways. For instance, a property with a large amount of square footage may have a higher market value than comparable properties with a smaller amount because it offers more living space. This is also true for a property with more bedrooms and bathrooms.
This data is then analyzed to provide various insights, such as average square footage and median number of bedrooms within a dataset and is referred to as the property characteristics statistics. These both help in making comparisons and adjustments when performing property valuations, and battle regressivity in AVMs that may lead to systematic bias.
Engineers can use regression analysis and other statistical methods to quantify the relationship between property characteristics and value by modeling how changes in features such as size and location may affect the property’s price.
Maintaining high-performing and reliable real estate AVMs
It is a challenge to build a high-quality and accurate automated valuation model and will take several iterations over time to get there. The Xome Value has been years in the making and the work on it never truly ends.
Our engineering teams are continuously working on improvements, reviewing data updates, and evaluating its results to ensure we provide the most accurate property valuation estimates possible.
