Current State of LIBOR contracts

Utilizing Machine Learning to Identify and Categorize LIBOR Fallback Terms

My nine-year-old daughter recently asked me what I do for a living and I found the above graphic handy to explain the concept of unstructured data and how my company, Pendo Systems, is helping global financial institutions and our partner professional services, law and technology firms understand and derive value from their unstructured documents.  It was a brief but effective conversation as my daughter described the before “data” picture above as a “hot mess”.   I now use “hot mess” in meetings to describe the current state of unstructured data and this conversation served as the inspiration for this LIBOR transition article.

LIBOR Fallback Terms & DEFCON 1

In his February 26, 2019 speech at the SIFMA C&L Society Luncheon, Michael Held, General Counsel at The Federal Reserve Bank of New York stated, “When you looked at the underlying contracts that used LIBOR, they didn’t provide very well for LIBOR simply disappearing. This is a DEFCON 1 litigation event if I’ve ever seen one.”   He outlined two key tasks for market participants with LIBOR exposures: “stop writing new contracts on LIBOR and start using SOFR or at least another robust alternative” and address“the trillions of dollars of existing contracts that extend past 2021 and don’t have effective fallbacks.”  Held emphasized that“inaction at this point in the LIBOR transition is short-sighted and futile and only extends the uncertainty. We need decisive action by everyone in the market to avoid damage to individual firms and the financial system.”

Historically, LIBOR for a particular interest period was determined by averaging quotes of several reference banks as of approximately 11:00 a.m. London time. Today, LIBOR is determined by reference to a screen quote from a customary market quotation service, such as Reuters or Bloomberg, with reference bank quotes used only as a fallback if the screen quote is unavailable. 

Seems straightforward enough but fast forward to July 27, 2017, when Andrew Bailey, the FCA’s Chief Executive, announced that, after 2021, the FCA would no longer exercise its authority to compel panel banks to submit quotes used to determine LIBOR. In his speech, Mr. Bailey emphasized that there are insufficient underlying interbank transactions to continue to rely on LIBOR as a benchmark, noting that it is unsustainable “for market participants to rely indefinitely on reference rates that do not have active underlying markets to support them”.

Currently, LIBOR serves as the reference rate for an estimated $400 trillion of financial contracts, including: commercial/retail/securitized/syndicated loans, floating rate notes, derivatives, and swap/hedge agreements.  The outstanding financial products referencing just the USD LIBOR are estimated to be approximately $240 trillion in notional value.  For financial institutions and corporates around the world, the transition is massive and fraught with the potential for legal and operational risk…and the clock is ticking.

For the past several years, regulators and industry leaders around the world have been collaborating on policy initiatives to transition to alternative, nearly risk-free reference rates (RFRs).  Given LIBOR’s lost relevance as the “world’s most important number” these initiatives have evolved from policy development to pragmatic execution as the conversion of agreements that reference LIBOR with maturities beyond 2021 to an alternative reference rate will be critical for market stability.

The Underlying Complexities of the LIBOR Transition

The related operational transition is enormous as it includes managing outstanding positions, transitioning to new non-LIBOR products, and reworking significant industry infrastructure.  Client and bank economics are impacted; certain businesses such as derivates, corporate trust and mortgages are in effect “ground zero”.  Further, significant conduct risk exists for any LIBOR-based products sold that extend beyond 2021.

Fallbacks for legacy positions entail high risk as these terms are buried in documents (i.e. unstructured data) not captured by automated systems. For financial institutions the challenge will be balancing the economic impact against client needs, conduct, legal and reputational concerns while ensuring readiness.  Imagine the overwhelming manual task of reviewing thousands of contracts across lines of businesses and products to organize and prioritize the needed transition work.  A handful of examples include the following:

Corporate Trust Services

Corporate Trusts provide trustee, agency and fiduciary services for bondholders, investors and lenders for asset- and mortgage-backed securitizations, municipals, and warehouse/conduits.  Documentation issues abound as fallback language is largely not captured in systems and the documentation likely exists in multiple areas or by product.

Syndicated Loans

Legacy credit agreements in the U.S. syndicated loan market typically have fallback mechanisms in the event that LIBOR is temporarily unavailable. Although variations exist, the LIBOR definition may provide that if the screen rate is unavailable, the rate is determined by interpolating between LIBOR rates of other specified durations, or by the rate offered to the agent bank by major banks for U.S. dollar deposits in the London interbank market for delivery on the first day of the interest period in the approximate amount of the loans, however, these fallbacks are inadequate if LIBOR no longer exists.

  • Market disruption clauses addressing a temporary unavailability of LIBOR or other trigger event by establishing fallback pricing at an alternative base rate are not a solution as borrowers would be dissatisfied with more expensive base rate pricing on a permanent basis if LIBOR disappears.
  • Some provisions allow the agent alone to select a comparable or successor rate and apply it in a manner consistent with market practice.
  • Others provisions allow the agent and the borrower to choose a successor rate, with many, but not all, of these provisions giving the majority lenders affirmative or negative consent rights to the successor rate amendment. 
  • Some replacement rate provisions are drafted narrowly to refer to a successor reference rate or index rate, suggesting that only a new benchmark can be selected without making other changes.

Collateralized Loan Obligations (CLOs)

Indentures in the U.S. CLO market often contain a fallback provision if the LIBOR screen rate is unavailable on the interest determination date. In that case, the calculation agent would request quotes from reference banks in the London interbank market and determine the rate based on the mean of the quotes provided. If an insufficient number of quotes are obtained, LIBOR for the subject interest period will be LIBOR as calculated on the prior interest determination date. Since reference banks no longer will be providing quotes if LIBOR becomes permanently unavailable, this mechanism effectively turns floating rate obligations into those of a fixed rate instrument, which is not what investors bargained for.

Examples could go on and we haven’t touched on Floating Rate Notes, Derivatives or Swaps… 



All of the Answers are Held Captive by Unstructured Documents

Suffice it to say, the core challenge faced across the financial services industry is the ability to effectively and efficiently identify and categorize by common types terms and conditions the relevant corpus of contracts that must be transitioned from LIBOR to an alternative reference rate.  Add to this the need for auditable, evidence-based data lineage for contract remediation efforts and certain, future litigation. The reality is that systems of record that support LIBOR-based businesses typically contain single-purpose, transactional data as there were no fields envisioned to capture this extraneous context (i.e. LIBOR fall back terms). 

It is estimated that 80-90% of the information needed for the LIBOR transition is trapped in unstructured documents which are unsearchable for metadata identification and extraction.  As such, the current state approach to mining the corpus of LIBOR contacts would be manual, mind-numbing and expensive task that will not create enduring value for an organization as it is viewed through the lens of a one and done event.  I believe that this one-off view to LIBOR remediation is an opportunity.

According to Bill Woodley, Pendo’s President and COO and former CEO of Deutsche Bank Americas, “the burden of LIBOR transition is one of the most under-estimated challenges approaching the financial services industry today. The operational challenge is as intrinsic and of a scale similar to other recent major issues like Stress Testing and Living Wills – but most financial institutions are deeply under-prepared. Aside from uncertainty as to the final outcome of what will replace LIBOR and its derivations, the biggest problem is scale and transparency. Most Financial Institutions simply do not know what LIBOR replacement language exists in their hundreds of thousands of relevant documents.”

In speaking with professional services and law firms, the consensus is that a manual approach is untenable given the projected fees, the looming deadline, the need for accurate classification of contract types across millions of contracts. Quite frankly, it is not the best use of this top talent.  In our discussions with the market, we are hearing is that a traditional approach to identify the contracts that require repapering will not scale as the LIBOR transition clearly requires a digital solution and time is the enemy.



Forging a Human & Machine Learning Partnership to Structure Contractual Data

Through Natural Language Processing (NLP) and Machine Learning (ML), it is possible to create a single data layer, combining contractual data and deal attributes sourced from unstructured and structured data sources.  The nexus of structured contractual data (e.g. contract ID, fallback rate / terms / triggers, consent requirements, etc.) and product data (e.g. customer / contract ID, reference rate, principal, maturity, spread, etc.) enables linkage capability and completeness checks.  Above all, this hybrid structured data set allows for legal analysis to create the waterfall of logic of actions for each category/type of contract language.  

Pendo’s Machine Learning Platform can digitize unstructured data at enterprise scale to create data sets primed for downstream processing, insights and analytics.  By utilizing Pendo for the LIBOR transition, our clients and partners are transforming the corpus of unstructured contracts into machine-readable format to accelerate the identification and extraction of critical data elements needed to advance this work in an effective, efficient, economically sound and evidence-based manner.

As per Bill Woodley,“This is what Pendo Systems has been built to do. For three years, Pendo’s Machine Learning Platform has been extracting precise language and data from large volumes of unstructured documents – quickly, accurately and with auditable lineage back to source. This simply cannot be done manually. The Platform uses custom scripts to help clients extract, organize and evaluate complex financial documents, converting unstructured data into usable, structured data sets. Pendo is building a leading practice in LIBOR replacement and is an essential tool for those in early stages of preparation for this major change.”

Pendo’s approach to wrangling the unstructured data underpinning LIBOR transition programs allows our customers to define the critical data elements needed to determine the scope and scale of the remediation work. For example, Pendo can search through millions of contracts and index the ones that extend beyond the 2021 deadline and extract the 30 to 40+ data elements needed to categorize the fallback terms, legal triggers (i.e. reasonable, discretion, etc.) and provisions.  The output generated by Pendo is essentially a matrixed view of an enterprise’s LIBOR contracts which enables our clients and partners to manage the legal and operational transition while minimizing costs and risk.

Included below are two screen shots of our LIBOR use case; additional material can be found through the following links:   and



Creating Competitive Advantage through Exploiting Unstructured Data

Beyond the work at hand for the LIBOR transition, digitizing the corpus of these foundational business documents positions our clients to “Pendo once and reuse many times” as these contracts have been transformed into digital assets for the enterprise and the Pendo machine learning models developed for the LIBOR transition can be modified for future needs.

Pendo’s Founder and CEO, Pam Cytron sums up Pendo’s capability as “Pendo doesn’t help our customers get better at finding the needle in the haystack; Pendo simply removes the hay.”  Pendo transforms the unstructured data “hot mess” into just “hot” enabling our clients to build a competitive advantage through understanding and monetizing their unstructured data.