Moving Targets: Using Scorecards in Volatile Times

Maintaining a flow of originations that accurately reflects the risk appetite of the business is always challenging, and even more so in today’s volatile world.

Currently, there is much concern amongst lenders around the impact of COVID-19 on scorecard performance. However, businesses can and should be able to make an informed response to these difficulties with the right approach and tools.

Short-Term Safety Measures

Scorecard deterioration is inevitable in periods of change, but until enough time has passed to yield sufficient performance data in the new world, redevelopment or realignment can be difficult. In the light of COVID-19 impacts, lenders are looking at how to successfully manage risk and bolster potentially under-performing scorecards so they can continue to operate effectively.

When the landscape changes it can be prudent to mitigate weaknesses in current application tools through additional conservatism such as extra policy rules or adjusted score cut-offs. A clear understanding of consumer stresses and additional sources of insight such as Open Banking data can allow well-selected rules and overrides to significantly strengthen the origination process.

Understanding Predictability Changes in the New World

When functioning correctly, scorecards are powerful tools, allowing automated, objective decisions combining a large number of factors. However, in the current environment, not only will scorecard performance change, but so will the predictive patterns of the underlying variables. For example, Payment Holidays may mask or delay arrears cases, possibly weakening or misaligning the arrears variables returned from the Credit Bureaus.

A key mitigant here is scorecard monitoring. Existing scorecards will retain significant power (assuming they were functioning effectively in the first place!) but it is critical to know where weaknesses have appeared. In stable times, monitoring is often an ad-hoc, low-priority exercise to check model performance is broadly in line with expectation. However, this analysis must move to the fore, be regularly executed, and evolve to cover earlier outcomes and a wider array of variables. This will enable management to react quickly to the changing predictive landscape, validate and refine stop-gap policy rules, and grow their understanding of their model performance.

Closing The Gap: From Model Rebuilds to Machine Learning

An ideal outcome would be to replace ailing models with new versions perfectly tailored to the new world – but in the absence of empirical data this in not plausible right away. Additional policy rules and monitoring will help plug the gaps and mitigate weaknesses, but lenders want to regain the power of true scorecards as quickly as possible.

One option is to leverage dynamic, short-term machine learning models after the initial scoring decision. This would serve as a ‘safety net’ to cover gaps in the current model and capture the latest customer behaviours based on shorter term outcomes. Material deltas between the initial and ML scores could trigger manual underwriting, auto-reject, or price increases to help counter any deterioration in core scorecard predictability.

As performance data emerges, the most favourable results will come from fully rebuilding and redeploying models, enabling the emerging risk trends in the new world to be captured. In the meantime, it is vital to identify where the dangers lie and steer the business around them.

For more information on how we can help you strengthen your decision framework, check out our case study.

If you are interested in discussing the challenges and solutions outlined above, please get in touch.