COVID-19: A Positive Disruptor to Expected Credit Loss Models

By Amnon Levy and Tim Daly

With COVID-19 continuing to batter the global economy, many banks are struggling to model projected credit losses that enter third quarter reports under the Current Expected Credit Loss (CECL) model and, for large and regional banks, in their Comprehensive Capital Analysis and Review submissions.

It is clear that COVID-19 is affecting credit risk across industries in ways that differ from past recessions, and that macroeconomic relationships are not holding to their historical patterns. Referencing the relationship between unemployment and inflation, Federal Reserve Vice Chairman Richard Clarida recently noted that economic models “can be and have been wrong.” In a similar vein, highly sophisticated quantitative credit loss models are often requiring significant qualitative adjustments that are based on cumbersome supplemental analyses.

Credit loss models: challenges in the current environment

The current challenges of quantitative credit loss modeling relate primarily to insufficient data that accurately describes the current credit environment used in bank estimates along with rigidity in model oversight. For example, CECL models may rely on macroeconomic scenarios defined by broad-brush variables, such as unemployment, that are insufficiently differentiated across industry segments.

Moreover, the models may be calibrated to historical data that include recessions that are themselves unique.

Models that calibrate the sensitivity of credit losses using the last 20 years of pre-COVID data simply will not pick up on the varying degrees to which different industries are affected by COVID-19.

Macroeconomic scenarios described only by an increase in national unemployment, for example, cannot differentiate among the 2001 dot-com bust’s effects on technology and telecom, the 2008-09 crisis for financial institutions, and the current COVID-19 effects on hospitality, entertainment and leisure. On top of all this, each recession brings different fiscal and monetary stimulus programs, which further distances the current environment from being properly described by the historical relationships between credit quality across industry segments and macroeconomic variables.

Plenty of uncertainty in credit market signals

Millions of payment deferrals, combined with various regulatory and legislative relief, present another challenge to understanding credit risk, as the magnitude of a looming wave of defaults and evictions is unclear. While funding needed to bolster corporate loans is available, especially to those with access to securities markets, much uncertainty remains about the effectiveness of its form. Programs to further support individuals remain in congressional gridlock. Uncertainty in mortgage, auto, and credit card defaults will, in turn, result in uncertainty in losses on ABS and RMBS. Fiscal and monetary programs designed to bolster financial markets and the firms that rely on securities in ways that may be distorting market signals which are often used in credit loss models, such as bond spreads and equity market performance.

Meanwhile, the epidemiological progression of the pandemic and the sociological response have resulted in further breakdown of historical relationships. For example, unlike in previous recessions, the auto industry has this time been affected by both a demand shock and a supply shock (the latter caused by both breakdowns in the supply chain as well as virus-related labor shortages) that have sometimes resulted in seemingly puzzling mid-recession price increases. At the same time, data from home office and outdoor furniture manufacturers highlight even more “un-recession”-like patterns, with shifting workplace and outdoor sporting cultures driving increased demand.

More granular data are essential

It is ever more apparent that the segmentation and classifications initially being used by financial institutions are often insufficient.

More current and granular data are needed. A classification of “restaurant,” for example, is not nearly as relevant as whether it is a takeout restaurant in Utah or a fine-dining restaurant in midtown Manhattan. Roadside hotels and suburban office space are exhibiting resiliency and not experiencing the same impact of cultural shifts to remote work and away from downtown office space and hotels, along with the wide range of downstream cross-industry implications and credit consequences. COVID-19 has also unveiled hidden concentration risks, such as credit to the airline and cruise industries which, like restaurants, rely on business models in which customers are within close physical proximity to each other—inherently problematic in the current environment.

The murky picture offered by traditional data, segmentation and modeling has forced a reckoning of sorts and the incorporation of nontraditional data in credit modeling. Cross-state epidemiological and Google mobility data, for example, can provide guidance to address shortcomings in a model’s ability to discriminate across portfolio segments, as well as the extent to which credit quality may be impacted by possible permanent localized cultural shifts. Granular data on government programs, disposable income and consumer spending patterns may further comprise more informative and robust economic indicators than, say, unemployment alone.

Changing conditions require model agility

Financial institutions need to expand the data they monitor—and how they use it in credit modeling. But they also must recognize the need for quantitative models that can adapt to quickly changing environments. The extensive and cumbersome two-to-four-month review by model risk management renders many analyses stale, given the speed of COVID-19 developments. This limits the relevance of quantitative models in loan origination and business strategy, often relegating the models merely to regulatory compliance and financial reporting. This rigidity can often be largely overcome through the availability of granular data that banks can adopt, depending on the circumstances, their individual portfolios, and economic conditions.

While third quarter reporting and CCAR submissions used models that are ostensibly validated, this situation calls for a more dynamic monitoring by which financial institutions can quickly leverage benchmarks and analyses recognizing immediate and practical considerations. This is critical today as organizations prepare for fourth quarter 2020, for 2021, and for the next inevitable economic crisis down the road. After all, by their very nature, crises will reveal behavior incongruent to historic patterns.

Amnon Levy is managing director and head of portfolio and balance sheet research at Moody’s Analytics, where Tim Daly is senior director and head of America strategic relationships.