AI, Appraisers & Buggy Whips
Most of these items cannot be ‘seen’ by Artificial Intelligence models or Big Data…
The article below in Mortgage News Daily e-newsletter on July 3, 2018, got me to pound my noggin… for a couple of seconds at least. Actually, I think I need to do it again!
Artificial Intelligence (AI) and ‘Big Data’ have been buzz words (and activity) for several years. And based on evidence in this article, lots of research and development leading to incorporation of AI is being done.
While the article does not mention it directly, both Freddie Mac and Fannie Mae have instituted appraisal waivers on, currently, a low number of mortgage loan applications. Freddie just announced last week that they are going to allow waivers on certain Condo loans, for a type of housing where obtaining of correct and accurate ‘data’ from the association may not be readily available. Will waivers increase on other real estate? I don’t think the answer is ‘no.’
For real estate, what is stunning about this is the apparent acceptance that real estate is static as the models apparently presume. In other words, what exists ‘today’ must certainly exist many months from now. We all know that is Ludacris! Any time something is built out in the open, exposed to the weather, degradation and depreciation occurs. Secondly, as tastes and occupants change, internal components may be modified. Most of those items cannot be ‘seen’ by the AI models or Big Data, until someone actually gets eyeballs on the property and updates the data file for the property. This is why appraisers (or some version thereof) will be needed as time progresses for mortgage lending. Maybe ‘we’ won’t be the buggy whip makers after all – put out to pasture by newer technology.
I’ve always said that everything in the world is cyclic. Cycles repeat. Unfortunately, young people without extensive experience, and incorrectly designed algorithms, cannot predict the exactness that is presently required for accurate property valuations for changing real estate. At some point in the future, presumptions in current cycles and assumed data may prove to be inaccurate.
And you know what? Another recession is coming. Should it be shown that ‘waivers’ which allowed properties to be sold or refi’d at higher than market value evidence indicates, which later are foreclosed, the culprits in this nonsense will be exposed. And they won’t be appraisers, thankfully. We’ll be able to take the buggy whips made by our ancestors and flog the perps!
Overcoming Vocational Irony as AI Role Grows – by: Jann Swanson – Jul 3 2018
A sure sign that a topic has progressed beyond trending and is now “trended” is when its acronym has firmly replaced the actual name. Artificial intelligence, the theory and development of computer systems able to perform tasks that normally require human intelligence, such as speech recognition or decision-making, has been discussed by techies since at least the early 1970s. However, its journey from something few understood to wide recognition of the verbal shortcut “” has happened in only a few short years. Now that it is mainstream there is a lot of discussion about how best to manage it and get the most from it.
Loretta Ibanez, Mortgage Innovation Director Single-Family Strategic Delivery for Freddie Mac, focuses on how companies use big data in their internal operations in a recent article in the company’s Perspectives blog. She uses the concept of “vocational irony” as a starting point; the cobblers children has no shoes, the accountant who goes broke, and the technology company that provides cutting edge financial models for their customers maintains internal machinery firmly rooted in the 20th century.”
Management consulting firm McKinsey & Company recently published a study that looked into big data and the building influence of artificial intelligence and Ibanez said one point stood out to her: “While investments in analytics are booming, many companies aren’t seeing the ROI they expected. They struggle to move from employing analytics in a few successful use cases to scaling it across the enterprise, embedding it in organizational culture and everyday decision-making.” How, she asks, do companies ensure that big data and AI, which are transforming nearly every industry in the world, improve their own internal operational processes?
Freddie Mac’s venture into this brave new world started with the introduction of Loan Advisor Suite in 2016. This allowed the company’s customers, i.e. banks and mortgage lenders, access to Freddie’s risk assessment tools. Loan Product Advisor, the Suite’s automated underwriting system, handles hundreds of thousands of single-family loan files and appraisals each month. It can assess borrowers who lack credit scores and is expanding use of its automated collateral evaluation (ACE) system to eliminate the requirement of an appraisal for many loans. While these are ways to look at the mortgage experience differently and to make smarter decisions, she says, harkening back to the cobbler analogy, we have to make sure “that we aren’t walking around barefoot while we accomplish our mission.”
In the days before companies could embed AI and machine learning into their internal operation processes, they had to make tradeoff decisions among time, money, workforce and – importantly – their computing power. Inevitably they put improving the customer experience at the top of the list. But now, with cheaper, scalable cloud computing available, it makes sense to apply it to internal operations in order to get more done, faster, and at lower cost. She says, “Not only can the cobbler’s children have shoes, they can have the best shoes in town.”
To demonstrate how far artificial intelligence has come, Ibanez demonstrates how AI can be used for AI. Her company recently partnered with machine-learning company DataRobot to help its own data scientists find better methods for modeling historical data and are already finding ways to use machine learning algorithms in fast and powerful ways.
The old method of improving a predictive model would involve:
- Hand-coding programs to analyze the data.
- Using human experience to narrow the field of possible predictive model designs
- Manually run and re-run simulations and algorithms.
- Continuously adjust until a strong, predictive model was achieved.
This cycle typically took many, many months as humans waded through the thousands of available algorithms and data sets that have been introduced, using their best judgement to pick the correct data set and the correct algorithm.
AI provides the ability to rapidly analyze and experiment with any structured data. It doesn’t even need to know anything about the data to select the most likely algorithms to obtain the predictive information needed from it. Using unsupervised machine learning it selects dozens of algorithms to run on the data, scanning and profiling the results and creating a “leaderboard” that identifies the best algorithms to use. It can even evaluate combinations of algorithms.
That, the author says, is using artificial intelligence (algorithm) for artificial intelligence (algorithm selection). It helps Freddie Mac improve its models faster and experiment with non-traditional data and will ultimately help Fannie Mae’s customers improve the loan origination process.
She concludes that, at a time when we’re all drowning in email and data, we need to filter out the noise, and decide and focus on what is important. There are many instances of data-centric activities that require a person to apply judgment and make a decision; approving an invoice, flagging a suspicious transaction or identifying suspicious activity on an internal network.
"AI and machine-learning can help our employees make faster, better decisions on what might be an outlier, what we need to pay attention to, and what might be our next best step.” She urges that this is the time to do it, making it a part of Freddie Mac’s transformation into a better company helping to build a better housing finance system.
- Speed Regardless of Accuracy Under the Banner of Modernization - March 8, 2023
- Marin City Discrimination Case Settled - March 7, 2023
- Exactly How Are Property Data Collectors Professionally Trained? - March 3, 2023
I have encoded this comment for FNMA & the Appraisal Foundation so that only Big Data can decode it:
Well said Retired.
Let’s start adding “Value” to our service. Think about a DISCOUNT on a borrowers rate if they have a full inspection/report vs another short cut. Let the BORROWER decide, it’s their money. So they have to fork over $ 400-600. for an appraisal vs $ 200. for another type of analysis, they have more piece of mind and get a discount at the same time. Lender get’s a “better” valuation product and is willing to discount a fee.
I don’t give an RAT’S A** about longer turnaround, that is simply fiction. After 30+ years I can turnaround cookie cutter stuff in 1-2 days. Just as fast as it takes an AMC to get an order, shop for a vendor, QC the report and deliver to lender.
The paper would now be worth more for the investor, an “A” class group of loans vs a “B”.
Just my HO
If in a magical world AFTER completing 100 full appraisals, some 50, 60, or 70% could some how be shown to be a perfect match to the best AVM, then on the surface there would seem to be some validity for their use. The catch is, ALL 100 need to be done in order to find the few or many that may be in line with the computer. Giving me a percentage, 50, 60 or 70% of properties that might be in line, does not tell me which 50, 60, or 70% are in line.
Seek the truth.
Related. The first white paper on the secretive nature of online censorship. This is a very amazing read and highly recommended.
I think this has merit and is related to the above article. Why won’t these companies publish their algorithms, methods, and results. Automatic valuation takes a similar approach to automatic online censorship, shrouded in mystery and secrecy. Both use similar deployment tools, have devastating consequences to many participators, resulting in a deplatforming and demonetization of our provided services. Similar to news monopolies the emerging data management monopolies seek total control and dominance over their respective industries.
The Censorship Master Plan Decoded