Contract language is holding mortgage lenders hostage

Register now

As reported in National Mortgage News in May, the Federal Housing Finance Agency announced a comprehensive Libor transition plan for Fannie Mae, Freddie Mac and the Federal Home Loan Banks for residential mortgages. And while the Libor deadline is December 2021, it was recommended that lenders stop pricing loans against the benchmark by September — leaving less than 350 days to transition.

However, it seems that not everyone is ahead of the game. Moody's Investors Service says that while many financial institutions are prepared, their clients are not — with borrowers being "very passive" about transition.

Libor is deeply rooted in many mortgage contracts, particularly in the specialist lending market, MBS and warehouse lines. Substituting one rate for another can mean changes in clauses and a whole new variation of terms to be approved by both the lender and customer.

According to a recent analysis, Libor serves as a reference in more than 100 million contracts, including mortgages, around the world, representing over $400 trillion. Even worse, two out of every five of these contracts contain no language dealing with cessation of the benchmark and will therefore require remediation prior to Libor's expiration. If financial institutions mismanage the transition — or fail to address the issue altogether — they risk incorrect payments and potential fraud charges as well as facing business disruption and loss of competitive advantage.

Overcoming the challenges

You may ask how such a seemingly little change in protocol can be turning the industry upside down, creating panic and tens of billions of dollars in reviewing contracts and looking for solutions for remediation. The answer is simpler than meets the eye — it is because contract language, the place where Libor is coded into the business of lending, has become hard-coded in all practices of setting lending rates.

Combing through this digital mountain of documents to look for specific contract language requires trained legal specialists, attorneys and regulators at a cost of time and money simply impossible to fathom, not to mention a high risk of human error.

In fact, Momenta Group has now warned that lenders are vastly underestimating the number of experts required to handle the complicated contract transition — with a shortfall of up to 250,000 skilled individuals. It's estimated that to remedy a simple individual contract will take at least one hour.

By applying artificial intelligence and machine learning to the complex Libor remediation process, digital intelligence enables lenders to implement a frictionless process for reviewing hundreds or thousands of documents, extracting Libor-related entities, and funneling identified documents to the legal team. With the tedious, time-consuming, error-prone task of sifting through mountains of documents being handled for them, legal specialists can focus 100% of their efforts on applying their expertise to the task of remediation.

If ever there was a use case for AI with ML

Let's talk about contract language and how it is holding lenders hostage. The billion-dollar question is, why are consultants needed and specifically, how can artificial intelligence with machine learning make a difference? The answer is simple, even if the solution is sophisticated. As I mentioned earlier, Libor is embedded in tens of thousands of contracts and agreements by which lenders do business with each other, client businesses and government regulators.

While AI with ML has been hyped in recent years, AI is building computer programs (also called computer vision) to help them find, interpret, make decisions and take action on complex language embedded in a corpus of data or documents as expert or highly trained users would. Far from being a replacement for high-paid skilled labor, such as attorneys, paralegals, analysts and consultants, AI with ML helps them scale to global challenges such as Libor transition.

A particularly useful application of AI for Libor transition is recent advancements in named entity extraction, whereby AI-based programs can learn how to recognize legal entities in the gigabytes of contracts and related documents. An entity is a specific person, place and/or action that is composed of multiple data fields that can be found anywhere in a contract. Often implied and never in a predictable place, named entities can be particularly vexing for legal experts to find as they plow through contracts.

Libor rates, clauses and affected parties are all entities in contracts that are rarely expressed in the same way within a single contract. But AI with machine learning can learn all of the permutations of these entities, including all of the stated and implied references to such entities. They can identify and understand them in seconds on a machine while a legal expert could take an hour or more to perform the same review. But AI does not get tired or distracted, like humans can, and can deliver results more consistently across thousands of documents.

With recent advancements in named entity recognition and extraction techniques, such as the ability of data scientists and legal experts to train the software with comprehensive taxonomies and variations (thesaurus, legal codes, aliases), modern AI solutions can be in production in a matter of days or weeks at a fraction of the cost of hiring an army of consultants and legal teams to do the work. Libor remediation experts can now train their "virtual legal team" of AI/ML software and supporting RPA bots to be their army of expert assistants to find, remediate or flag for additional review all the impacted contracts and legal business entities.

Here is where AI with ML can shine as an expert solution for Libor transition. By supplementing and learning how to work as an army of virtual assistants, a specialized named entity recognition solution, powered by modern AL with ML, can reduce the time of locating, understanding and remediating the sheer volume of Libor-related documents from tens of person years to weeks or months of computer time. The payback in risk abatement, time and money can speak for itself.

For reprint and licensing requests for this article, click here.
LIBOR SOFR Secondary market Artificial intelligence Machine learning Mortgage technology