Score the unscored!

By Felix Salmon
June 17, 2013

I went to two conferences in the past couple of weeks: the Underbanked Financial Services Forum, in Miami, and the Clinton Global Initiative’s America conference, in Chicago. At the former, I was introduced to a company called Cognical, which pitched itself as a tool which will allow lenders to lend money to a much broader group of people than they currently accept.

Cognical was set up, at least in part, to address the Catch-22 built in to the US lending system: you can’t get credit if you don’t have a credit score, and you can’t get a credit score until you’ve been extended some credit. The result is a system where many borrowers, especially immigrants and the poor, find themselves forced to pay through the nose to loan sharks, pawn shops, payday lenders, installment lenders, and other institutions which range from the consumer-unfriendly to the downright predatory.

It’s hard to wade through the jargon on the Cognical website. Exempli gratia: “Leveraging our experience with machine learning algorithms and unstructured big data, we assess and transform application variables in specific ways to expand the data and its predictive value.” But the general idea is pretty simple: rather than just look at credit score, lenders can use Cognical to mine enormous quantities of data, in the hope and expectation that buried in there somewhere the firm will be able to find patterns which can predict whether or not any given individual is going to repay a loan.

Cognical claims spectacular results, but I’m not completely sold: with big data comes a massive rise in spurious correlations, and in any case it’s not like the borrowers in question are leaving a massive data trail behind them in the first place. On top of that, even if Cognical is doing amazing work, all of it will be for naught unless and until someone happens to apply for a loan from one of Cognical’s clients.

As a result, what you really want, it seems to me, is two things. First is a dataset which is obviously germane in terms of throwing light on an individual’s ability to pay her obligations in a timely manner. and second is a way to get that dataset into the hands of the institutions which really matter, when it comes to this particular game: the three big credit bureaus, as well as FICO.

Wonderfully, that is exactly what a bunch of us, including Sasha Orloff of LendUp, ended up talking about at the CGI conference the following week. At the moment, it’s only an idea, but I’m keeping my fingers crossed that the idea is so good, and so obvious, that it’s going to end up being put into effect at some point.

What’s the dataset? It’s not a bunch of Facebook likes and Twitter favorites, the kind of inchoate data only Max Levchin could love. Rather, it’s a big and simple obligation: rent payments. It’s pretty obvious that if you’re good at making your rent payments on time every month, you’re also more likely to pay other obligations in a timely manner. And so if rent payments were reported to the ratings agencies, that would give them valuable information about the group currently known as “thin files” — people about whom there’s too little data to make a determination as to creditworthiness.

The place to start is HUD, along with the enormous housing agencies in cities like New York and Chicago. As public agencies, they have every reason to want to improve the lives of the people they’re renting apartments to — and one easy way of doing that would be by improving those people’s access to affordable credit. (And there would be a financial benefit for the agencies, too: if tenants knew that their rent payments would be a part of their credit score, they might be more inclined to pay on time.)

The ratings bureaus are weirdly low-tech: they don’t exactly have convenient APIs which allow anybody to upload payment datasets. Instead, they use a clunky old thing called Metro 2, which was designed back when people were mainly worried about Y2K issues. Still, it’s possible to write a program which converts any structured data into Metro 2 and then uploads it to the ratings bureaus — Orloff has done it, and he says he would be happy to donate the code to any open service or housing agency which wanted to report rental information.

Of course, once the rental information was uploaded to Experian and Equifax and TransUnion, they would have to actually do something with it — as would FICO. That might take a while. But Experian says that it’s already incorporating rental data into its own scores, and certainly landlords would be very interested in a FICO score which included such data. The demand is out there, and if the information is just dropped into the companies’ laps, it would be hard for them to simply ignore it.

The goal here is, simply, to score the unscored. Such people might not have high FICO scores, at the beginning, especially if their history of rental payments is spotty. But score beats no score, and once you have one, you can start working to improve it. What’s more, people who have been diligently paying rent on time for years, and who have sensibly be avoiding debt, might actually end up with high FICO scores rather than none at all.

It doesn’t need to end with public housing agencies, of course. The government could set up a central system where all landlords could, if they wanted to, upload the rental information of their tenants. And once such a system was set up, it could conceivably be extended beyond rent payments, to include things like utility bills as well. The idea being that when a company like FICO attempts to ascertain creditworthiness, all information has to have some value. And the more information that can be obtained about people who haven’t formally borrowed much money in the past, the more likely it is that they’ll be able to get a score. Which is a very useful thing indeed, in today’s economy.

Comments
11 comments so far

“As public agencies, they have every reason to want to improve the lives of the people they’re renting apartments to”

sooooooo adorable . . .

Posted by billyjoerob | Report as abusive

“Leveraging our experience with machine learning algorithms and unstructured big data, we assess and transform application variables in specific ways to expand the data and its predictive value.”

Cohort data is used to predict how people without credit histories will perform.

Posted by GRRR | Report as abusive

I was expecting to see “utility bills” much higher up than the last paragraph; the number of relevant creditors is far smaller than the number of landlords, and are typically subject to heavy regulation, so it would be comparatively easy to persuade all of them to report their data.

Posted by dWj | Report as abusive

All great except it’s giving possibly unscrupulous landlords the ability to blow up your credit rating.

Credit scores seem like exhibit number one in individuals not having enough control over how their personal data is used.

Posted by albertsun | Report as abusive

I share Albert Sun’s concern; I’ve had at least one really nasty landlord who I wouldn’t have trusted with power to affect my credit rating. If you were going to let random private landlords have access to this system, you’d want to do it by having them register with a payment processor, so the tenant sends payment (maybe with a direct debit option of some kind, as well as sending checks) to the processor, and the processor forwards it to the landlord, and the date of the transaction gets recorded for scoring.

Posted by Auros | Report as abusive

Disclaimer – I’m with Cognical, so that’s my frame, but I think there’s a pretty significant misunderstanding: Cognical does not mine data. On the contrary , we use deep learning technology to help lenders better understand the data they are already using. Again: We do not add data. (Yes, some of our funding is going to a website overhaul).

The conversation we are perpetuating is that everyone is looking to “big data” as the path to enlightenment, which implicitly implies we’ve exhausted all of the predictive signals in smaller data sets. We are proving the opposite. There is a ton of predictive left to extract in lenders’ smaller data sets, but existing modeling solutions can’t find them. In the payday space, where LendUp plays, we’re growing lender profit 25% – 75% not by mining data, but by giving applicants credit for compensating factors unrecognized by less sophisticated underwriting models. It’s not a black box, just a more accurate solution.

But to address the greater point, we agree that rental data is awesome. And to the degree lenders use it then we’ll help them objectively value that data. But Cognical doesn’t procure data in any way. We just make the best sense of it.

Posted by Btdubya | Report as abusive

Disclaimer – I’m with Cognical, so that’s my frame, but I think there’s a pretty significant misunderstanding: Cognical does not mine data. On the contrary , we use deep learning technology to help lenders better understand the data they are already using. Again: We do not add data. (Yes, some of our funding is going to a website overhaul).

The conversation we are perpetuating is that everyone is looking to “big data” as the path to enlightenment, which implicitly implies we’ve exhausted all of the predictive signals in smaller data sets. We are proving the opposite. There is a ton of predictive left to extract in lenders’ smaller data sets, but existing modeling solutions can’t find them. In the payday space, where LendUp plays, we’re growing lender profit 25% – 75% not by mining data, but by giving applicants credit for compensating factors unrecognized by less sophisticated underwriting models. It’s not a black box, just a more accurate solution.

But to address the greater point, we agree that rental data is awesome. And to the degree lenders use it then we’ll help them objectively value that data. But Cognical doesn’t procure data in any way. We just make the best sense of it.

Posted by Btdubya | Report as abusive

Felix: HUD and local housing authorities do not collect rent from tenants, except in rare cases. What happens is this: The tenant is part of a program called Section 8, the rules of which state that they (the tenant) will have to pay no more than 35 percent of their income (give or take–the rules differ place to place) for rent. HUD–i.e. federal taxpayers–will pick up the rest.

Most such tenants have no reportable income. They pay 35 percent of that figure–0–towards rent. Many have small under the table gigs that generate cash income but not a lot.

The way to make your idea work is to get electric companies and other utilities, including the big cable & dish TV outfits, to report. Those things are not so heavily subsidized by the government and so those are the things that poor people actually have to pay.

Posted by Eericsonjr | Report as abusive

“As a result, what you really want, it seems to me, is two things. First is a dataset which is obviously germane in terms of throwing light on an individual’s ability to pay her obligations in a timely manner. and second is a way to get that dataset into the hands of the institutions which really matter, when it comes to this particular game: the three big credit bureaus, as well as FICO.”

ERRRRR

The institutions that really matter are BANKS! the ones who lend the money and face the risk! givint it to credit bureaus means giving it to an intermediary; who will they finally give it to? BANKS. So giving it to credit bureaus is just adding an innecesary rise in the cost by giving a mark-up to these intermediaries. The reason for Credit Bureaus to exist is the lack of information other than credit history; if we finally find alternative information that can replace or even better, outperform credit history, then Credit Bureaus have their days numbered as the de facto providers of information to assess credit risk of new applicants for banks.

Posted by carlos2401 | Report as abusive

“As a result, what you really want, it seems to me, is two things. First is a dataset which is obviously germane in terms of throwing light on an individual’s ability to pay her obligations in a timely manner. and second is a way to get that dataset into the hands of the institutions which really matter, when it comes to this particular game: the three big credit bureaus, as well as FICO.”

ERRRRR

The institutions that really matter are BANKS! The ones who lend the money and face the risk! Giving it to credit bureaus means giving it to an intermediary; who will they finally give it to? BANKS. So giving it to credit bureaus is just adding an unnecessary rise in the cost by giving a mark-up to these intermediaries. The reason for Credit Bureaus to exist is the lack of information other than credit history; if we finally find alternative information that can replace or even better, outperform credit history, then Credit Bureaus have their days numbered as the de facto providers of information to assess credit risk of new applicants for banks.

Posted by Bethany12 | Report as abusive

Do you mind if I quote a couple of your articles as long as I provide credit and sources back to your blog? My website is in the exact same niche as yours and my users would certainly benefit from some of the information you provide here. Please let me know if this okay with you. Cheers!

Posted by traduceri daneza romana | Report as abusive
Post Your Comment

We welcome comments that advance the story through relevant opinion, anecdotes, links and data. If you see a comment that you believe is irrelevant or inappropriate, you can flag it to our editors by using the report abuse links. Views expressed in the comments do not represent those of Reuters. For more information on our comment policy, see http://blogs.reuters.com/fulldisclosure/2010/09/27/toward-a-more-thoughtful-conversation-on-stories/