A good, customized credit risk scorecard can significantly reduce an organizations charge-offs and is critical to running a successful lending organization. However, the success of a credit scoring program not only depends on the quality of the scorecard and the associated models, but also on how well the scorecard is being monitored and maintained.
The market and the economy often go through periods of turbulence. Changes to macroeconomic factors such as inflation, unemployment and GDP can impact the profitability of borrowers and alter their risk profile. Then, there are never before seen events like pandemics which significantly disrupt established norms for predicting credit risk. In addition, the condition of the market that the lender operates in can fluctuate. For instance, a variation in the mix of borrowers being booked by the lender can lead to changes in the key drivers of charge-off. Any of these situations (and many others) can make even the best crafted scorecard ineffective, and result in significant losses for the lender. It is therefore very important for an organization to closely monitor its scorecards. Here are some recommendations for putting together a good monitoring program.
(a) Monitor the distribution of the model inputs: A change in the distribution of model input variables might be driven by modifications in the process the data provider uses to collect and compute the data and that should be investigated. However, once that is examined and ruled out, it could signify a fundamental change in the underlying applicant pool. In either case, the models will need to be tested to make sure they are still effective and predictive. In the case where the change in distribution is caused by an alteration in the applicant pool, there is a strong possibility that the drivers of default are different for the newer population which could impact charge-offs, unless the models and the scorecards are modified. So, while it is possible that these changes are driven by data origination reasons, these shifts could be a leading indicator of charge-off problems yet to come and monitoring them can give the lender time for remedial action.
(b) Study the Defaulters: It is important to regularly study the accounts that have charged-off. Comparing the key statistics (firm type, size, income, industry etc.) of charged-off accounts over time, helps a lender understand and track the profile of their riskier accounts. Changes to this profile might occur as a result of some market disruption, even when the underlying applicant pool is still the same. For example, introduction of new cheaper technology might make firms relying on an older technology less profitable. Or a big player might decide to enter a niche industry taking away market share from existing smaller firms. Or else, a new regulatory requirement or discovery of health hazards, might shrink the market size for a certain industry. Whatever the reason for these changes, the models driving the scorecard need to be examined to confirm their continued effectiveness and if needed, the models should be adjusted or rebuilt.
(c) Check the relationship between charge-offs and predictors: It is possible that the scorecard still becomes less effective, due to a change in drivers (and associated thresholds) for default. This might be driven by reasons like economic turbulence. For example, a general worsening of economy might render the models less predictive. Credit attributes that were once considered safe (for example a certain level of credit score), might now lead to higher charge-offs. This is something that has been observed by most lenders over the pandemic but can creep in more quietly in other situations as well. A regular study of the predictors’ relationship with charge-offs can help detect this problem soon and allow the lender to take remedial measures which could vary anywhere between minor adjustments to the model (for smaller scale or temporary disturbances) to a full-scale model rebuild.
A systematic model monitoring program as outlined above, while essential, can also very taxing on an organization's resources. It is therefore important to design an automated system to minimize the time and effort needed to perform the tracking. Jigyasa Analytics has years of experience guiding its clients in model maintenance and developing custom automation around it to make the process cost effective and less intensive on resources.