A Quantitative Perspective on the Next Economic Downturn
Key performance indicators are showing signs of borrower stress in the consumer, commercial, and commercial real estate lending arenas as a result of prolonged higher interest rates and inflation. This, in turn, may prompt concern over the sufficiency and reliability of models commonly used to evaluate potential credit and capital exposures.
How can institutions ensure their models are robust and reliable during changing and challenging environments – exactly when they rely on them the most? In DCG’s work with banking clients, we see five principal credit modeling challenges. The good news is that there are a number of actions that model owners and management teams may consider in response.
Understanding Credit Terms
At a high level, credit risk metrics are trending up. Consumer loans and Commercial Real Estate (CRE) loans have experienced rising delinquency rates over the past few years; consumer loans, CRE loans, and commercial loans have all recorded rising charge-off rates over the same period.
The changes in consumer loans have been more pronounced than other asset classes, which is expected given their nature and historical experience compounded by COVID aid (which artificially reduced consumer defaults and losses). Even though trends are rising, the overall levels of delinquencies, defaults, and losses are not significantly higher than their basal (non-recession) levels.
These trends are driven in part by the macroeconomic climate. As a direct result of COVID aid and initially sparked by disruptions to global trade, US inflation reached and has remained at levels not experienced since the late 1980s and early 1990s. The higher cost of goods and reduced purchasing power has strained consumers and businesses. This has been exacerbated by the Federal Reserve’s (Fed’s) policy actions designed to slow the rate of inflation. The more than 500 bps increase in the Fed’s target rate has both slowed the rate of inflation (though not completely) and increased the cost of borrowing. This combination of higher prices and higher cost of debt is reminiscent of its last occurrence 35 years ago during and after the Savings and Loans Crisis.
The economic story does have some positive notes: unemployment enjoys levels as low as the 1960s, Gross Domestic Product (GDP) growth has remained average, and disposable personal income growth has been marginally higher for many workers.
5 Credit Modeling Challenges
This mix of economic signals may pose meaningful challenges for developing and using credit models. A common scenario is that management expects models to capture recent trends, but instead the models ‘miss’ this increase in risk and produce forecasts that may be considered too optimistic. Five key modeling challenges that may lead to this include:
1) Missing Factors: Models may not be developed using many independent variables. Thus, when credit stress is caused by a factor not included in the model, the forecasts may be less reliable. This is a common challenge for single-factor regressions in credit reserve models that rely solely on unemployment.
2) Sensitivity: Models may include many independent variables, but some variables are more important than others. Relative sensitivity can be easy to see when reviewing standardized coefficients. But when the credit stress is caused by a factor with a smaller standardized coefficient, the relative impact on the forecast may be muted – directionally consistent with expectations, but not as strong as perhaps it should be.
3) Interactions: A model may not specify that some effects may change depending on the behavior of other effects. For example, if inflation were the sole factor that increased, there may be a different response to overall credit risk than if there were a change in both inflation and short-term rates.
4) Time Series Length: Continuing the example of interaction effects, models may only be predictive on data that have been observed and included in an estimation. For example, if the last time high inflation and high short-term rates occurred was around 1990, then these models would require a time series significantly longer than what is often possible in practice.
5) Exogenous Forces: Some events may influence the underlying data in a model in unintuitive ways, breaking down expected relationships and weakening predictive power. A perfect example is the impact of COVID era direct transfer payments from the government to consumers. Individuals were losing jobs and the economy was under significant strain, but losses on credit cards and other higher-risk asset classes were lower than they have ever been.
3 Actions to Consider
In the face of these challenges, there are a variety of alternatives for model developers, model owners, and management to consider. While there is no perfect solution, implementing some combination of these actions may further bolster credit model usability.
1) Develop a framework for ongoing testing and performance monitoring.
Conducting scenario testing before those scenarios actually occur can illustrate how and when models may become unreliable. This gifts management with valuable time to deliberately address limitations in advance, rather than be forced to respond hastily.
Monitoring processes provides real-time feedback. If trends indicate more risk but forecasts predict less risk, this misalignment of expectations could alert management sooner than waiting until results are unusable.
Including well-defined triggers not only alerts management, but also indicates severity, next steps, and ownership for completing necessary remediation steps.
Creating institutional memory through robust documentation of testing performed over time allows developers to understand what has been tried before or where limitations exist and to eliminate key person risk.
2) Consider changes to the model development and governance process.
Develop a Champion/Challenger Framework: If a current model has a well-defined limitation, develop a challenger model that overcomes this limitation. It does not have to be used but can nonetheless help management interpret and better use the results from the champion. This may be helpful in addressing champion models that use a single factor.
Use Benchmarks: There are times where limitations are not easily overcome, but other contextual information may allow results to be interpreted. This may be helpful for incorporating peer data into a process in a way that does not directly influence predictions.
Adjust Historical Data: Particularly when exogenous forces are in play, there are instances when adjusting historical data may be an appropriate way to reinforce expected relationships and lengthen the historical time series. For example, a CRE construction model may include the behavior of land, lots, and development in addition to vertical construction. If the current portfolio makeup is almost exclusively vertical construction but historical credit stress was predominantly the result of poor performing land loans, model developers may reasonably adjust the historical data.
Qualitative Adjustments: While ubiquitous in credit reserve models, qualitative overlays and adjustments may be powerful tools for any credit model. Because the challenges itemized above can present insurmountable difficulties, it may be appropriate for management to develop a rules-based and consistent approach for making qualitative adjustments.
3) Give Model Risk Management (MRM) a pivotal role.
An MRM program establishes standards for the entire organization (for example, a defined ongoing monitoring framework), so that model owners have consistent guidance.
MRM also provides assurance to management by maintaining a relationship with all models in the inventory; model owners are required to demonstrate compliance with MRM’s standards regularly. In the unlikely event that a model owner is not meeting their responsibilities, MRM provides a mechanism for escalation, independent of the line of business.
Finally, MRM defines requirements for model validation, which can serve as an excellent resource for exploring these challenges and learning from the horizontal experience of others.
For more information about how to help ensure credit models are robust and reliable during changing and challenging environments, Contact DCG.
ABOUT THE AUTHOR
J. Chase Ogden is a Quantitative Consultant with Darling Consulting Group's Quantitative Risk Analysis and Strategy team. As a practitioner at large and mid-sized financial institutions, Chase has experience in a wide array of modeling approaches, applications, and techniques, including asset/liability models, pricing and profitability, capital models, credit risk and reserve models, operational risk models, deposit studies, prepayment models, branch site analytics, associate goals and incentives, customer attrition models, householding algorithms, and next-most-likely product association.
Chase is a graduate of the University of Mississippi and holds master’s degrees in international commerce policy and applied statistics from George Mason University and the University of Alabama, respectively. A teacher at heart, Chase frequents as an adjunct instructor of mathematics and statistics.
© 2024 Darling Consulting Group, Inc.
Kommentare