Are Machines to Blame for the Sub-Prime Crisis?
Technology and the Collapse of Mortgage Markets
When Rob Lux, EMTM ‘01, attended the Mortgage Bankers Association’s technology conference in
Dallas in March, one of the key discussions was the role of technology in the sub-prime mortgage crisis.
Was the whole system just moving too fast, with near-instant loan approvals, tough competition, and record
volumes? Was the infrastructure up to the task? Were the machines to blame? What could companies do to make
Lux became Chief Technology Officer for GMAC ResCap, one of the nation’s largest mortgage lenders, at
the end of 2007. He has been running ever since. He noted that you could probably tell from the Wall Street
Journal why he has been so busy – and that is never a good sign. The financial sector has been in
free fall, with Bear Stearns as one of the latest casualties. And one of the first large stones that set this
global avalanche rolling was the collapse of sub-prime mortgages in the United States.
This naturally leads to the next question: What caused sub-prime mortgages to roll off the cliff? In talking
to his colleagues in Dallas, Lux said the consensus was that you could not lay the blame primarily at the
feet of the silicon chips. But issues such as data management and automation may have played a role. The
crisis may offer some important lessons for any organization.
Automatic Weapons: Making Bad Decisions Faster
Automation has been criticized for contributing to the sub-prime mess. It won’t be the first time that
automated systems have taken the fall. Back in the late 1980s, when Lux was working at the Philadelphia Stock
Exchange, observers blamed automated trading for the stock market meltdown. Later studies concluded that, in
fact, technology probably wasn’t the cause.
"The consensus is that technology didn’t really create the crisis per se,” Lux said in an interview
from the Dallas conference. “But it certainly may have played a role in accelerating it. As one gentleman
at the conference said ‘Technology does not make bad decisions, it just helps humans make bad decisions
People create the models and write the programs that run on the machines. To paraphrase the National Rifle Association,
Technology doesn’t kill markets; people kill markets. But like an automatic weapon, automated technology certainly
can make the end come a great deal more quickly.
Automated underwriting revolutionized mortgage making. In the early 1990’s, borrowers brought a wheelbarrow
full of forms to a lender. If they were lucky, they might receive a response in 30 days. If they were even luckier,
the response would be “yes.” This was a critical bottleneck in the complex process of buying and selling homes.
Enter automated underwriting. Customers could go onto a website or pick up a phone, answer a few questions, and
receive a response in minutes. There were systems that could go out to multiple lenders and give borrowers several
offers. For good loan prospects, automated systems such as FannienMae’s Desktop Underwriter could offer rapid
decisions with their guarantees. A loan officer with a laptop could give a decision almost instantly.
This worked well for the best borrowers, the “conforming” business. As lenders looked for new opportunities, they
began to focus on borrowers outside the conforming limits. These might be entrepreneurs without salary or income verification.
They also included jumbo loans above the conforming limit of $417,000 and loans to borrowers with less than stellar credit
scores, which Freddie or Fannie would not cover. The larger lenders with scale developed their own models for the risks of
lending to these non-conforming segments. They created automated systems, modeling the risks and offering automated decisions.
Thanks to automated systems and record low interest rates that fueled refinancing, mortgage volumes skyrocketed.
Over the last five years, about 90 percent of homeowners refinanced their mortgages to take advantage of lower rates;
some refinanced two or three times. As the process speeded up, mortgage makers sometimes cut corners, such as not asking
for income verification. “When everyone is refinancing, it becomes a process where you need to accelerate a lot
of the steps,” Lux said. “In the exuberance to accelerate, people sometimes took short cuts.” He points
out, however, that manual systems using the same rules would have had similar, if slower, outcomes.
Automation led to a heady time when money for real estate was just a few clicks away. But it all depended on
having the right models the right data. And there’s the rub.
Automation Replaces Data
In this fast-paced environment, the industry began increasing its reliance on automated models in other areas.
For example, instead of conducting a visual appraisal of a property before writing the loan, mortgage banks
used valuation models to estimate property values. These models looked at comparable sales in the neighborhood
and automatically generated a valuation. Automation replaced data. This saved a lot of time, but may have missed
some variations in homes. It was less accurate in areas that were sparsely populated or had fewer home sales –
or when prices were changing rapidly.
The irony is that the mortgage industry is one of the most data-rich industries in the world. To obtain a
mortgage, homeowners are willing to share almost any detail of their life down to their pets’ names and
financial statements. Even when lenders collected the information, the trick was to capture and use it –
particularly as the roller coaster of refinancing headed down its fastest hills. “As an industry we could
do a better job of managing data as a strategic asset,” Lux said. Yet unlike Wal-Mart, which has to move
a lot of widgets in addition to data, for mortgage lenders, data is the business.
No History: Limits of Models
Models look at history to predict the future. But looking back 15 years in mortgage markets did not give a hint
about what actually would transpire. All the trends were going up and up.
“The big mistake was in the way they valued some of the things that they were doing,” said Wharton Adjunct
Finance Professor John Percival. “They used models that were based upon history. In their history, they didn’t
have any experience that told them there was any possibility that what has happened was going to happen. Nothing in their
models said the U.S. real estate market could go down by 20 to 40 percent. Nothing in history said that was going to happen.”
When the market shifted, all the answers were wrong. “The models were built on the premise that prices are appreciating,”
Lux aid. “When the market changes, you may not be assessing all your risk properly. The data may be saying everything
looks great, but you may be taking on more risk than you anticipated.”
Soul of a New Mortgage Machine
At the conference in Dallas, Lux and his colleagues were focused most intently on what they could do to prevent future disasters.
“We are discussing what we should be doing as an industry, to improve our ability to perceive these types of things.
This downturn has been a mess for consumers, homeowners and companies. No one ever wants to see this happen again. Our industry
is committed to making these changes.”
Even as mortgage companies address the meltdown, new technology challenges are emerging. Regulators are implementing
new rules that have raised conforming loan limits in certain geographic areas, effective in June. This means lenders
are scrambling to rebuild their systems for a policy that will only be in effect for this year. “It is a huge
technology initiative,” Lux said. “At the same time, technology people are in crisis mode to respond
quickly to market conditions and make sure the industry is stable.” To add to the workload, the Fed’s
interest rate cuts have made this a very attractive time for borrowing, so loan volumes have been growing.
“It is a perfect storm,” he said.
Next issue: How broader integration and cultural issues contributed to the sub-prime crisis.