Why have credit models broken down in the COVID-19 environment?
Credit models for the last 10 years have undergone significant scrutiny and governance, driven by regulatory expectations and a determination that they are deemed “fit for purpose” prior to their use. So why did they become “unfit for purpose” in a matter of days?
Most of the models were built on historical data from the last decade, which is not representative of the current environment. Also, credit models generally presume a gradual impact of the environment on losses, with lags ranging from one to six months. The estimated model parameters will exacerbate predictions due to any sudden macroeconomic movements. Current economic volatility is likely to generate unintuitive or counterintuitive estimates if one relies heavily on the models.
Additionally, the payment holiday and forbearance interventions, along with the closure of asset markets, have clouded typical indicators, such as current delinquencies, that are often used to project future losses. Credit models rely on inputs about the presumed macro-economic forecasts that typically use traditional economic theory concepts of general or partial equilibrium at their core to project the future. Such forecasts may be completely unreliable as the artificial shut-down of many consumer goods and services markets has pushed the economy into a state of disequilibrium.
To compound the economic forecasting problem, government interventions, such as temporary income replacement programs to mitigate unemployment, may not be fully factored into projected unemployment metrics challenging the credibility of the forecast. Clearly, there are sufficient limitations in the use of existing credit models in current environment.
We believe there are areas that model owners should be exploring in order to ensure that the output of their models remains valid and robust under the current circumstances.
New macroeconomic scenario models with forward-looking relationships to credit drivers
Defining a baseline macroeconomic projection is one of the main focus areas for credit risk modeling. The severity of the local lockdown seems to be the simplest key driver impacting economic expectations in the short- and mid-term, while duration of the local lockdown is driving the longer-term effects.
However, given the global nature of both today’s economies and the pandemic, we must understand how shocks caused by lockdowns in different parts of the world can propagate across economies through global value chains in order to develop a medium or long-term macroeconomic scenario. We can already observe this in China, where in spite of the withdrawal of a majority of the social distancing measures, the economic activity remains subdued due to the outbreaks of the pandemic at China’s main trading partners.
EY’s experience suggests that we can apply a combination of macroeconomic approaches (general equilibrium and input-output) and pandemic susceptible, infected and recovered (SIR) models, as well as bottom-up sector and geographic recovery perspectives, in order to generate scenarios accounting for lockdown risk, sectoral impacts, policy responses and international risk transmission. Given the abundance of daily published country-level pandemic statistics and the continuous flow of sectoral indicators from the exposure monitoring processes, such a modeling framework enables the necessary flexibility to update scenario outlooks daily.
Credit risk models will also need to be recalibrated to reflect a forward-looking impact of macroeconomic scenarios on structural credit factors, challenging where historical relationships hold – and applying new approaches where they don’t. Once applied to the COVID-19 pandemic, the approach can also be leveraged and extended to other use-cases related to an external shock impact on credit portfolios. One good example is the capture and the understating of the forward-looking implications of climate change risk.
Quest for new data and early warning indicators
Credit risk models attempt to effectively discriminate healthy and distressed exposures. The traditional data sources they typically use (financial and behavioural) struggle to capture the complexity and pace of the current economic environment. Economic indicators and borrower financial information are often observed on a lagged basis, and certain current indicators are distorted by the private and public relief programs offered in response to COVID-19. Greater emphasis is needed on augmenting traditional data with inferences from alternative data sources.
Transactional data offers a highly accessible, real-time indicator of financial health in both retail and non-retail portfolios that can enhance various components of the credit life cycle. Analytics around the nature of incoming and outgoing payments can provide deeper insights on credit capacity, quality and behavioral changes, particularly across retail and micro business.
Analysis of current transaction flow (level, frequency and volatility) against pre-COVID-19 levels can help track the performance (and risk) of SMEs and corporates during the recovery period and allow targeted intervention. In some cases, historical financial indicators could be supplemented or replaced with a transactional data based financial index. Going forward, banks should explore opportunities to gain better insights by using a range of other data sources including value chain linkage data, health/geolocation data, e-commerce and electronic tax filings.
This will raise questions around the suitability of current data management infrastructures. Institutions that, until now, were reluctant to invest in high-frequency big data platforms may now need to accelerate their technology spend as part of their next and beyond COVID-19-triggered change-the-bank initiatives. To gain access to untapped data sources, banks may need to expand their ecosystem and establish new relationships with external providers. The recent efforts to strengthen customer data protection and data integrity, as well as the broader third-party risk management agenda, should provide a necessary framework to facilitate this trend.
According to the EY/IIF global bank risk management survey, firms expected a significant increase in the application of these methods for credit decisioning over the next five years. Utilizing the broader range of accessible data, we believe the pandemic will accelerate this process and will act as a trigger to formulate complementary credit risk assessment frameworks that can also be used for new waves of challenges related to climate change, geopolitical risk or broader sustainability issues. It is critical to design approaches that do not follow the same over-reliance on historical trends that may not fit today’s crisis, while also not inappropriately amplifying short-term correlations in current data.
While it is important to enhance the efficiency of the methodology for today’s model risk management capabilities and approaches, the lack or distortion of data is fundamental.
