Automated valuation models often come under special scrutiny as mortgage experts seek to increase technology in the industry while warding off the possibility of another housing crisis, but one expert explained why AVMs could work in today’s mortgage lending environment.
“Initially used as a way to decrease costs, this new resurgence of AVM use is proving, through man vs machine testing, that AVMs are a viable option for much of the current origination volume – this is where the game changing moment resides,” a spokesperson at HouseCanary, a 2019 HousingWire Tech100 winner, explained in an interview.
HousingWire’s Tech100 recognizes the most innovative and impactful companies in the housing and mortgage industries. Now, HousingWire is launching an all-new award program – the Tech Trendsetters, which recognizes the experts behind these innovative companies.
We want to recognize the experts behind the technology. The people who drive the innovation. The executives making an impact. This inaugural program will award 50 of the technology leaders driving innovation in the housing economy. Nominate your tech driver today, the nomination period closes July 26, 2019.
How do you become a leader in the tech space? One of our previous Tech100 winners says it like this:
“HouseCanary has become a leader in valuations through a strong focus on data quality management, development of image recognition methods to assess home conditions, innovation with the latest machine learning algorithms, and more,” the company told HousingWire.
HousingWire sat down with a spokesperson for HouseCanary to see how technology is affecting the housing and mortgage industries.
HousingWire: How can machine learning change the housing industry?
HouseCanary: Objectivity and measurability are the key benefits to changing the housing industry through machine learning. A single algorithm repeatedly exposed to the same dataset will give the same consistent and objective output each time. This output is not influenced by any party's hidden interests and/or subjective opinions. Furthermore, these algorithms are able to measure and quantify the error/uncertainty of their outputs informing the user of how accurate or inaccurate its estimate is. This allows users of such outputs to make better informed decisions by weighing the model's level of uncertainty against their particular use case and tolerance for risk.
HW: How can automated property valuations change the game for the valuation space?
HC: AVMs have had a checkered past in the housing industry. At their onset they were heralded for their ability to quickly and accurately determine property values at scale. During the housing crisis it was exposed that AVMs were not good at modeling value for all properties (which should have been known). This led to a period from 2008 to 2013 where the valuation space went almost entirely to human-based valuations. Now, however, there has been a renaissance to AVMs as the industry has learned that there is a sizable use-case where AVMs can and should be used responsibly. Initially used as a way to decrease costs, this new resurgence of AVM use is proving, through man vs machine testing, that AVMs are a viable option for much of the current origination volume – this is where the game changing moment resides. Once the major originators transition to data-driven decision-making workflows, AVMs will become a standard in lenders’ valuation playbook and eventually grow to represent the valuation method of choice for upwards of 70% of all originations.
HW: How do artificial intelligence valuations compare to traditional valuation methods?
HC: If you define traditional valuation methods to be the methods used by appraisers, primarily the comparative based approach which looks at three to five “comps,” then the biggest difference would be the number of properties considered for a valuation. For example, any algorithm considers and processes many more than just three to five properties in determining a value. In particular, the machine learning based methods are specifically designed to accommodate and utilize the ever-growing datasets available today.