Last year, investment in real estate tech startups totaled more than $6 billion, which isn’t surprising given the implications of big data on the housing economy.
Everything from evaluating risk management to analyzing the next trend in housing depends on gathering “clean” information on mortgage borrowers and home sellers. But one group remains largely absent from this data collection, and this is the group that most intrigued a panel of big data experts assembled by Amherst InsightLabs at the SXSW conference and festivals on Sunday.
The panel spoke to a packed house braving the surprisingly cold Austin weather, attentive despite the temperatures outside Bungalow, a bar on popular Rainey Street. The session included an intro by Ben Stein (he opened with “Bueller, Bueller,” to loud applause) and featured Erin Glenn, CEO of Fintech Collective, as moderator, and Doug Fashenpour, CEO of Amherst, Joanne Gaskin, senior director, score and analytics at FICO, and Selma Hepp, chief economist and vice president of business intelligence at Pacific Union.
The panel explored how companies in the housing space are using big data today and what is quantifiable at the current time. The data foundation of the housing ecosystem, according to Fashenpour, includes data on economics/demographics, properties, transactions, mortgages and investments.
This massive data trove is necessary to generate meaningful analysis for real estate investment, credit scoring and underwriting decisions. And mining it for borrower behavior could yield the most valuable insights of all.
“To be able to access data is incredibly important,” FICO’s Gaskin said. “For example, going back to the economic downturn, we saw a change in the priority of pay — what people prioritized in bill paying. Paying the mortgage had always been No. 1, but in the downturn it shifted to auto payments. Why? People need a car to get to work, and long mortgage modification timelines meant homeowners got to stay in their homes for a long time after default. Consumer behavior changed as a result.”
Gathering consumer data is easy for the 190 million consumers — 92% of the country — who can get a FICO score based on the credit they have accumulated. It’s still not hard on the 28 million who are in the credit system but won’t generate a FICO score due to inactivity.
But then there’s a large segment of consumers — 25 million people — who have no traditional credit file at all. These people are “credit invisible” and their behaviors and motivations are harder to understand. They tend to be young and new to the country, Gaskin said, but those are just broad-stroke demographics.
All of the panelists had their own reasons for wanting to quantify this group, and their own methods of doing it.
In Fashenpour’s case, Amherst combines property-level information with more than 10 years of borrower behavior to measure credit and other risk. The company provides proprietary software, data and analytics to support its existing business lines and third parties of single family equity, asset management, investment banking and venture investments.
Pacific Union, a luxury real estate brand in North California, develops neighborhood-level market conditions for its real estate agents and partners. With that in mind, Hepp said the company wants to understand who consumers are, why they are buying or selling, and why they want to move to a particular location. As part of their efforts to discern and predict what buyers in different areas want, they collect quite a bit of data from their own agents, in addition to information from the National Association of Realtors and state-level sources.
At FICO, consumer data is the cornerstone of the company’s offering — measuring consumer credit risk. For Gaskin, the credit invisible aren’t invisible at all. These customers do generate credit data, it just isn’t regularly reported to the credit bureaus. Many will have payment records for smartphones and rent, but less than 4% of utility data is reported to credit unions, and less than 1% of rental data is in credit repositories, she said.
To help bridge the gap in data, if a consumer doesn’t have a credit score, FICO will generate a “second-chance score” by using public record data and phone and cable information, which can only be used for unsecured lending. After six months of payments on those loans, FICO will generally have enough data on those consumers to generate a traditional FICO score.
FICO will also consider transactional information, SMS data and psychometric data, which is defined as the psychological theory or technique of mental measurement, but makes more sense if you think about observing behavioral data. For instance, Gaskin said, if the company surveys consumers and asks whether they are married or not, the time that it takes someone to respond could mean something beyond whatever answer they give.
One thing all the panelists agreed on was that social media was not a good indicator of consumer behavior or risk, mainly because it could be manipulated so easily — by consumers themselves or by the social media outlet.
Hepp noted that predictive analytics is based on past behaviors, but that past consumer behaviors don’t always translate to accurate forecasts. “We are looking at the past, but patterns and behaviors have changed a lot — we don’t have a good grasp on what the current behavior patterns are like.”