June 11, 2015 / 5:01 AM / 4 years ago

INSIGHT-Automated lenders threaten to eat banks' lunch

NEW YORK, June 11 (Reuters) - When Kevin Pereira needed a loan last year for his shaving gear business, his bank, Wells Fargo, considered his application and turned him down.

After a Google search, Pereira, 26, found Kabbage, an on-line lender that used high tech tools to evaluate his credit, including analyzing his company’s Facebook page and looking for quirks in how he navigated the Kabbage Web site. The lender sifted through his company’s credit card data and its checking account information.

In the end, Kabbage approved Pereira’s company, Wet Shave Club, for a $19,000 line of credit.

“I can make an underwriting decision based on Facebook data that is as effective as using credit score data,” said Kabbage Chief Operating Officer Kathryn Petralia, 44, noting that personal credit ratings known as FICO scores represent a single data point among many that the company considers in its underwriting process.

Kabbage, Lending Club and On Deck are among the lenders making small business and consumer loans that more established banks are less eager to approve. The new lenders say their edge is technology - they largely use automated algorithms to approve borrowers, while a bank uses more costly humans.

That may be good news for borrowers and bad news for banks, analysts said. Non-traditional lenders could siphon away as much as 7 percent of annual U.S. bank profits, amounting to more than $11 billion, in five years or more, according to Goldman Sachs research analysts, mostly from loans to consumers and small businesses.

Kabbage expects to make more than $1 billion of loans this year, up from about $600 million in 2014, with 95 percent of those loans processed without any human involvement, many within as little as seven minutes, according to company officials.

To be sure, at a time when interest rates are at historic lows, Kabbage’s rates can be high — fees for Wet Shave Club’s six month line of credit are about 10 percent of the amount loaned out, Pereira said. That’s equivalent to an annual percentage rate of more than 30 percent. But even though the first loan ended up skimming nearly all of the company’s profits while he was paying it back, Pereira is happy with the process.

“When you have a young business, it’s really important to get as much traction as you can,” he said. “We were able to buy a month ahead of inventory without worrying.”

By comparison, at Wells Fargo, an unsecured line of credit for small businesses has annual rates that range from 5 percent to 13 percent, excluding fees, according to a spokesman.

The on-line lenders are small in the scheme of the broader market, accounting for about 3 percent of the roughly $1 trillion of personal and small business loans outstanding, but the sector has caught the attention of traditional banks who are increasingly looking at automated lending programs and unconventional data sets as ways to lend more efficiently, that is, using fewer people.

Citigroup has an arrangement which may help Lending Club make loans to impoverished and minority borrowers. In May, Goldman Sachs hired an executive to build an online lending platform focused on consumer and small business loans. BancAlliance, a group of more than 200 community banks, partnered with Lending Club in February to originate consumer loans. Kabbage said it is in talks to license its platform to two top 20 banks.

Leaving lending decisions up to computers has proved dangerous in the past. Countrywide Financial Corp collapsed after using automated loan underwriting technology before the financial crisis. Bank of America - which bought Countrywide in 2008 - paid a fine of almost $1.3 billion because of defective loans Countrywide made using a mostly automated process. Loan underwriting standards could deteriorate if automated lending becomes widespread, critics said

Big data “creates complacency,” said William Black, a former bank regulator who is now a professor at the University of Missouri Kansas City. “It is easily manipulated. It doesn’t give you objective answers — it gives you whatever was programmed.”

It’s easy to tweak algorithms to make more loans and boost near term profits while taking on more credit risk, Black said. Moreover, he said the models are often too complex for executives or regulators to question.

The data that underwriters use for automated loans is often incomplete or flat out wrong, said the National Consumer Law Center, a watchdog for low-income Americans, in a recent study. The group criticized the “astounding” lack of transparency in their underwriting practices.

“The black box is getting bigger and bigger and more mysterious,” said Persis Yu, a lawyer for the NCLC.

Some traditional banks are skeptical as well. JPMorgan Chase & Co CEO Jamie Dimon recently told a conference that he feared many of the new lenders companies would not survive a credit downturn.

ROBOT, RUN

Automated underwriting has been successful over time in other arenas, such as consumer credit card lending, where banks have been using algorithms to lend since at least the 1990s. Small business loans are often personally guaranteed by the proprietor, and are in some ways similar to credit card loans.

Lenders also say they have discovered sources of information that are useful for lending and were not available in the 1990s.

For example, Kabbage’s Petralia said the company has found that a borrower seeking a small business loan whose company is active on Facebook is 20 percent less likely to default on a loan than a borrower who doesn’t use Facebook. Being active on the site reflects a business that knows its customers well, Petralia said.

If applicants agree to allow Kabbage to vet their data from package delivery company United Parcel Service, the lender looks at the size and weight of the packages a company sends out, how many customers the business ships to, and how often the business ships to the same clients.

These data points give a sense of how big customer orders are, how many clients the business has, and whether satisfied customers are ordering repeatedly. Kabbage can access UPS data because UPS is an investor in the lender— both are based in Atlanta.

None of the lenders would give explicit details on their algorithms, which are proprietary, but many discussed a few of the data points they look at. Kabbage and other online lenders also look at borrowers’ navigation habits on their web sites for hidden clues to their likelihood to repay.

Subprime lender ZestFinance - run by former Google Chief Information Officer Douglas Merrill - uses data points such as whether an applicant uses all capital letters on their application as part of their underwriting process. Those borrowers are higher risk, ZestFinance data show, compared with those who capitalize conventionally.

Upstart, another online consumer lender, uses a borrower’s college grades and SAT scores in its algorithms to determine credit for borrowers who lack a credit history.

Wall Street has begun to package the loans in transactions like BlackRock’s securitization of more than $300 million of consumer loans originated by P2P lending company Prosper. The securitization was given a credit rating by Moody’s in January.

Former Citibank Citigroup Chief Executive Vikram Pandit has invested in Orchard, which helps institutional investors buy loans originated by marketplace lenders. Pandit said algorithms can yield the same results as a traditional person-to-person process of underwriting for many loan types.

“I can’t imagine banks aren’t looking at that and saying, ‘Oh my god, I can do this cheaper and better,’” he said. (Reporting by Michael Erman in New York, Editing by Charles Levinson, Dan Wilchins and John Pickering)

0 : 0
  • narrow-browser-and-phone
  • medium-browser-and-portrait-tablet
  • landscape-tablet
  • medium-wide-browser
  • wide-browser-and-larger
  • medium-browser-and-landscape-tablet
  • medium-wide-browser-and-larger
  • above-phone
  • portrait-tablet-and-above
  • above-portrait-tablet
  • landscape-tablet-and-above
  • landscape-tablet-and-medium-wide-browser
  • portrait-tablet-and-below
  • landscape-tablet-and-below