Enter your e-mail address and password below.


Forgot your password? New User? Register Now.
   ARTICLE   |   From Scotsman Guide Residential Edition   |   November 2016

Machine Learning Offers a Way Forward

Computer technology promises to spur a rebirth in mortgage margins

Machine Learning Offers a Way ForwardPendulums swing. Regulation increases. Margins decrease. That’s not entirely bad, however. The Russian composer Igor Stravinsky said: “The more constraints one imposes, the more one frees one’s self.” Based on Stravinsky’s logic, the mortgage industry is as free as a bird.

A plethora of regulatory bodies govern the typical mortgage lender: the Consumer Financial Protection Bureau (CFPB), the U.S. Department of Housing and Urban Development, the Federal Housing Administration, numerous state agencies and more. It’s no surprise that one result of all that regulation is increased cost.

The cost to originate a mortgage increased by more than 70 percent between 2008 and year-end 2015, to more than $7,700 per loan. A survey conducted by the Stratmor group revealed that the TRID consumer-protection rules rolled out last year by the CFPB have increased mortgage-origination costs by $210 per loan.

One way to think about the changes in origination costs in relation to regulation is as follows: More regulation creates an increased focus on compliance; more compliance leads to slower, more cautious processing and more paperwork; more paperwork lowers throughput for underwriters; that translates into fewer loan approvals per underwriter.

In such an environment, how can mortgage lenders stay compliant while decreasing costs and increasing profit margins? For the answer, we can look to the efficiencies created by economies of scale as well as emerging technology.

Achieving scale

Many lenders have reacted to higher loan-production costs by originating fewer mortgages. In his 2016 letter to JP Morgan Chase shareholders, the lender’s outspoken CEO Jamie Dimon included a section titled, “Why are you still in the mortgage business?” Dimon’s answer, paraphrased: It sure isn’t because of the margins.

Beyond increased loan-production costs, the complexity of many regulations also can lead to costly errors. Since its inception in 2011, the CFPB has levied over $5 billion in penalties against companies it oversees, with many of those fines meted out to mortgage lenders.

Another side effect of increased regulation is more paperwork. In recent years, the average mortgage file has mushroomed to more than 500 pages, according to the Mortgage Bankers Association (MBA). Larger files lead to lower underwriter productivity (fewer files approved per day), which, in turn, leads to higher costs per loan. According to the MBA, the number of loans reviewed per underwriter per month decreased from about 165 to 33 over the last 10 years.

When examining regulation’s impact and the cost of loan production at the individual underwriter level, every borrower is like a snowflake — a unique creation unlike any other. In that sense, every loan handled by an underwriter is unique and being processed for the first time. In the finite universe of one mortgage originator, that assumption may hold. But when extrapolated to the scale of some 8 million home loans financed annually, that “snowflake” framing starts to break down at the same time new possibilities for controlling costs emerge.

By way of example, there is Flatiron Health, a company that is developing one of the most powerful cloud-software programs for oncology research. Like mortgages, oncology research is a vast and complex field, with many researchers studying all aspects of cancer — yet only 4 percent of adult cancer patients are enrolled in clinical trials.

Machine learning will become a requirement in
the near future to compete in the mortgage industry.

Flatiron’s thesis, however, is that if data from the remaining 96 percent of patients can be captured in some way, and provided to researchers globally via technological tools, that effort can leverage the collective power of the entire oncology profession to seek out patterns, insights and, eventually, cures. Even with small gains, such a project could affect many people. Even a 5 percent improvement in patient-survival rates would result in tens of thousands of lives saved every year.

This power of scale can be applied to the mortgage industry as well and lead to the rebirth of profit margins for the industry. How can lenders take advantage of the data that surrounds them, however?

Tapping technology

Machine learning, also referred to in its advanced form as artificial intelligence, is simply the act of teaching a computer to learn from experience. If a computer can improve its performance on a task by taking into account its past experience, it has engaged in what can be best described as machine learning.

The classic computer game Chessmaster 2000 used brute-force programming to beat its human opponents. The computer would search through millions of possible moves and choose the best one for the given situation, with each move being like a new snowflake and requiring a new search for the best move.

With machine learning, a programmer can provide the computer with just the rules of the game and, over time, it would learn and adapt to its opponent’s moves and eventually become better at chess than most of the best players in the world. In fact, such a deep-learning machine taught itself to play chess at an International Master level in just three days. Said another way, a computer was able to learn a set of complex rules and perform better than 98 percent of professional chess players worldwide in a short period of time.

Machine learning will become a requirement in the near future to compete in the mortgage industry, given the expanding regulatory and other cost constraints lenders face. Machine learning has already begun to impact lending in the area of underwriting. Just like a computer learned to play chess in three days, machine learning in the mortgage sector is making it possible for computers to begin to make sophisticated underwriting decisions.

Startup lenders like LendingHome, Lenda and Clara are investing heavily in this type of technology. Ten years ago, it would have made sense for some mortgage companies to have 100 underwriters on staff. Machine learning makes it possible for a handful of underwriters to perform at the same level as a much larger team. The most effective way to increase margins is to expand productivity, which machine learning enables.

•  •  •

Technology, and more specifically machine learning, holds the key to advancing productivity in the mortgage industry. As regulatory and other costs continue to balloon, the pressure to find alternatives for maintaining profit margins is enhanced. These cost constraints help to define, and change, the rules of the game.

The evolution of technology and machine learning, combined with the power of human productivity, promises to propel future profits upward in a mortgage industry that is now suffering through a bout of margin compression.


Fins A Lender Post a Loan
Residential Find a Lender Commercial Find a Lender
Scotsman Guide Digital Magazine

Related Articles



© 2019 Scotsman Guide Media. All Rights Reserved.  Terms of Use  |  Privacy Policy