> Posted by Andrew Fixler, Freelance Journalist
On August 4, Facebook received approval on a patent it had purchased in a bundle from the defunct social network Friendster. It primarily describes a mechanism to weed out content depending on whether it travels via trusted nodes in a user’s social network. This might not have caused much of a stir, had it not been for entrepreneur and blogger Mikhail Avady’s revelation that the patent also includes the following application:
“In a fourth embodiment of the invention, the service provider is a lender. When an individual applies for a loan, the lender examines the credit ratings of members of the individual’s social network who are connected to the individual through authorized nodes. If the average credit rating of these members is at least a minimum credit score, the lender continues to process the loan application. Otherwise, the loan application is rejected.”
Many commentators and journalists reacted with alarm, while Facebook has not offered comment on the story. It is unclear whether or not a product will be developed out of this particular embodiment of the invention. A Daily KOS headline proclaims that “Facebook Gets Patent to Discriminate Against You Based on Your Social Network”, and a Popular Science writer notes that “It’s totally not something straight out of a cyberpunk dystopia”. This MSN article warns readers to purge their less trustworthy friends, though it also notes that the technology could relegate some consumers to riskier lenders. In the non-financial press, less attention is given to the potential upshots for thin-file loan applicants. The list of concerned news outlets stretches well beyond the first page of search results I examined after Googling the patent’s text.
Using “big data” from social media isn’t exactly new news in credit underwriting. For instance, Hamburg based Kreditech requests access to their clients’ social network profiles in order to gather socioeconomic data on the circles they belong to. Hong Kong’s Lenddo asks your closest friends to vouch for your trustworthiness, and penalizes their credit scores should your payments go delinquent. In my opinion, the most interesting of Lenddo’s many features is its capacity to distinguish your “Facebook friends” from your real friends that you are also connected to on Facebook. This distinction is crucial to accurately assessing credit-worthiness using social media data. Data science is making this distinction possible, and its importance goes unmentioned by commentators who have highlighted the absurdity of lenders making judgments based on our more dubious online connections. The social enforcement mechanism that Lenddo uses recalls the social collateral that underpins group lending in microfinance.
Social media closely tracks many potentially prejudical elements of our lives, so it is natural to be concerned that loan underwriting based on virtual networks could become a kind of online “redlining”, the historical (though still not entirely abolished) practice of discriminating against mortgage clients along racial and geographical lines, a practice that helped precipitate the Equal Credit Opportunity Act (ECOA) in the United States. Although many have leapt to the conclusion that this Facebook patent will induce discriminatory lending, it is important to weigh both the potential upsides and downsides of a new means of vetting loan applicants.
Redlining imposes unfair barriers to finance. The term originated with the creation of color-coded residential security maps by the U.S. Federal Housing Administration during the 1930s; this is the same agency that created many first-time American homeowners through the creation of a mortgage guarantee program, but which, according to The Atlantic, “explicitly refused to back loans to black people or even other people who lived near black people”. It’s a hugely problematic and irrational way to allocate capital, like any decision calculus grounded in prejudice. According to its legal definition, redlining also includes underwriting metrics, which, from the point of view of a given lender, correlate with risk and are therefore efficient, but that are nevertheless unacceptable because they systematically deny the poor financial access. The ECOA protects against or significantly limits discrimination based on sex, race, age, nation of origin, and a number of other factors, in order to mitigate redlining.
Will social media underwriting end up being a net financial democratizer or divider? Jeff Stewart, co-founder and CEO of Lenddo in Hong Kong, has something to say about this: “Artificial intelligence is simply better at administering credit in a fair way.”
It is possible for underwriting algorithms to explicitly or implicitly consider data points deemed discriminatory by ECOA while scouring for things that correlate with loan payment performance. This problem aligns with a decline in the “verifiability of discrimination” in the social media age, when those in charge of hiring, admissions or setting the terms of financial products can access your personal information without explicitly asking for it. Managing conduct related to discrimination and privacy is therefore important for those pioneering new approaches to underwriting, a point well made in this PWC report.
Fusion’s Felix Salmon plays an excellent devil’s advocate in favor of Facebook’s patent on “The Dirty Filthy Loans Edition” of the Slate Money podcast. He notes that FICO scores don’t pick up on the creditworthiness of thin-file clients, including immigrants and young borrowers, and that “the obvious way to [underwrite these borrowers] is to use all of the information we have about you, including your social graph.” According to Salmon, all underwriting is inherently discriminatory to a degree, and regardless of whether loan terms are based on FICO scores or metrics gathered from our online presence, underwriting in a properly functioning credit market will generally favor the well-off.
Among young professionals, rising student debt burdens, which disproportionately weigh on students whose parents were unable to bankroll their higher education, have a significant impact on FICO scores (debt burden comprises 30 percent of the FICO score). In addition, student loan debtors over 90 days in arrears increased from roughly 8 to 16 percent between 2004 and 2012. This is among the reasons that 66 percent of generation Y is classified as subprime. Millennials are becoming more economically potent, and banks are increasingly trying to find ways to develop a competitive edge in this burgeoning, though difficult to bank, demographic. They aren’t the only demographic like this, though. Plenty of older people are still reeling from the acute, but ultimately short term financial hardship caused by the 2008 financial crisis and resulting recession.
The three biggest credit bureaus in the U.S., Experian, TransUnion and Equifax, have developed VantageScore, an alternative to FICO that they believed in a 2013 estimate to be able to score between 30-35 million of the estimated 64 million “unscorable” borrowers in the country. VantageScore discovered that roughly one-sixth of the newly scoreable market is not, in fact, as high risk as some assumed. Like other alternatives to FICO designed to penetrate un-bankable segments of the lending market, the key to VantageScore is more data. However, the data VantageScore uses does not drastically depart from that used in FICO other than a few alternative data sources including rent and utility payments. There are others who offer a proactive solution to thin-file customers without resorting to social media data. In the case of a company called LendUp, this is done through “education, gamification and a transparent fee structure” tied to loan products. While social media data clearly isn’t necessary to prove that a large number of borrowers are not unscoreable or subprime, as their credit scores would suggest, there’s no reason to think that social media data couldn’t more accurately price their risk now that they are in the safer bracket.
The practice of feeding algorithms with social media data to underwrite loans may or may not become the norm in the United States in the future, at least among large financial institutions. However, it’s worth noting that six of the 10 largest banks in the U.S. were among the users of the 1 billion VantageScore credit scores used last year. Ed Mierzwinski, Consumer Program Director at USPIRG, opined to Yahoo that credit underwriting is among the many ways Facebook is exploring monetizing data, but likely one that will remain untenable in the U.S. due to privacy concerns. Frank Eliason, head of social media for Citibank, remarked to The Economist that “using [social media] data to assess loan applicants is a ‘dangerous game’ that big banks are ducking for now”, in particular out of concern for the maintenance of public trust. In the same article, Jack Vonder Heide of Technology Briefing Centers notes that it is already common for underwriters at smaller financial institutions to peruse loan applicants’ online profiles, although “if that process was automated and industrialized, it could turn a big bank into ‘very juicy fodder’ for the press”, not to mention potentially landing banks in a legal quandary over violations of privacy laws – and potentially redlining.
Plenty of observers have met these underwriting innovations with optimism. Venture capital funding and lots of customers ratify underwriting startups. They also get media validation. Lenddo, for instance, had 350,000 customers two years after it was founded in 2011, and the World Economic Forum named it a 2014 Technology Pioneer. Forbes named Kreditech in its list of “The Next Billion Dollar Startups”. Given the circumspection around this particular story, it may just be the case that concern is grounded in a perennial mistrust of social networks’ handling and monetization of our personal data, rather than the usefulness and fairness of big data analytics in underwriting loans.
Have you read?