On Monday, the Fed, OCC, FDIC, CFPB and NCUA (the Agencies) released a request for information (RFI) on financial institutions’ use of artificial intelligence and machine learning (AI/ML). The RFI explains that the Agencies support the responsible development of innovation and highlights various ways financial institutions have used AI/ML to improve their services, including flagging suspicious activity, personalizing customer services, enhancing credit decisioning, and augmenting risk management practices. It also provides an analysis of the risks associated with AI/ML and notes that the Agencies expect financial institutions to have processes in place to address such risks. In particular, it notes that the lack of explainability associated with certain uses of AI/ML makes it difficult to review decisions and evaluate how the system will function in unforeseen circumstances. It also observes that the ability for AI/ML to learn and update without human interaction can result in errors, especially when inputs vary from training data due to a change in circumstances such as an economic downturn. Other risks highlighted by the RFI include data quality, fair lending, cybersecurity and third party risk.
Questions about whether the risks stemming from the use of AI/ML techniques warrant new regulatory action have been swirling around for several years now and have amplified in the wake of high-profile incidents revealing algorithmic bias in credit decisioning. Given this RFI, the recent issuance of AI/ML regulatory guidance in other countries, and guidance from the EU regulators expected at the end of this month, it appears highly likely that the Agencies will take regulatory action in the near future. Regardless of whether it takes the form of an update to 2011 joint Fed and OCC guidance on model risk management, new stand-alone guidance or a formal rulemaking, we expect that the Agencies will release a comprehensive framework that addresses the risks of AI/ML-based models at each stage of the model lifecycle and across their full range of business uses. As firms have expanded their use of AI/ML to include areas not considered by existing guidance such as marketing, customer support and human resources, they would be wise to ensure that they fully understand the uses of AI/ML throughout the organization and update their risk tiering accordingly. We also expect that the RFI will expand the scope of the Fed and OCC joint guidance to include fintechs and other financial institutions that are currently out of scope. While it will likely contain new expectations for bias detection and mitigation as well as explainability of AI/ML systems, it remains to be seen whether it will follow its European counterparts in driving toward formalized risk tiering and certification standards. Such standards may be challenging for US implementation considering the existing regulatory principles and industry practices for model risk management -- especially regarding vendor models -- that require banks to perform their own model testing rather than rely on a third-party certification.
As this area represents an overlap of several related risk areas and existing and potential new regulations, including data privacy, alternative data, fair lending, fair banking, and cybersecurity, designing a new regulatory framework that dovetails with other existing regulations and guidelines will inevitably be challenging. The fact that five regulatory agencies in various stages of transition came together to jointly issue this RFI represents the implicit recognition of these challenges and the necessity to take careful bipartisan action. As such, we view this release as a strong signal that the Agencies are committed to working together to create a regulatory framework that is effective, efficient and not a hindrance to innovation.
On Monday, the Alternative Reference Rates Committee (ARRC) published a white paper recommending the use of 30-day average SOFR set in advance in asset-backed securities (ABS). SOFR is the recommended alternative to USD LIBOR. An “in advance setting” applies a reset of the rate at the beginning of the interest period, which differs from the committee’s recommendation for the use of SOFR in arrears for business loans and other products, which would involve a reset of the rate toward the end of the accrual period.
The ARRC concedes that future product development may lead to different applications of SOFR in securitized products, including the use of a forward-looking term SOFR should it become available. However, only last week the ARRC had announced that it would not be in a position to recommend a forward-looking SOFR term rate by its original target date of Q2 2021. At this time, insufficient liquidity in SOFR derivatives would prevent the construction of such a term rate. In light of regulatory expectation to end the use of USD LIBOR in new products as soon as practicable, the committee suggests that the use of SOFR set in advance today represents the most direct way for firms to meet that expectation with respect to securitized products.
The ARRC’s recommendation represents a practical solution to life without a term rate and is the result of three years of discussion within the subgroup on securitizations. While an in arrears solution would have certain economic advantages (i.e., aligning coupon payments to incurred funding costs in the same time period), its use for securitizations would likely require at least a seven business day lookback. Given the prevalence of monthly paying instruments in the market, such an extended lookback period could reflect 35% or more of the payment period, negating much of the economic advantage of alignment.
The use of SOFR in advance illustrates the compromise that needed to be struck between economic accuracy and the practical realities of operating in the market. Especially at a time when regulatory pressure to end the use of LIBOR in new products is becoming more and more pronounced, solutions that can be implemented more quickly provide an advantage. There is little doubt that additional variations in conventions will continue to be developed, but for now the ARRC’s proposal appears to provide a practical solution adhering to an age-old principle – don’t let perfection be the enemy of good.
Subscribe to PwC’s LIBOR Transition Market Update here to read more about these and other developments.