Best practices for banks when implementing q-factors

We recommend developing a reasonable framework supported by quantitative information. Although there are many ways to create a framework that is reasonable and supportable, our recommendations consider these key concepts: 

• Max loss assumption 
• Risk scales and anchoring scales 
• Weightings 

The first step in quantifying the qualitative is determining the maximum loss scenario. Many financial institutions will look at the previous financial crisis to calculate this maximum loss amount as a “ceiling” for determining how much to allocate to q-factors. Then it can consider which q-factors are involved in its operations. Previously, most financial institutions were applying all nine regulatory q-factors, but with the introduction of forecast-based CECL models, as well as the requirement to incorporate “reasonable and supportable forecasts” into the model, many financial institutions have used their CECL implementations to reevaluate the nine factors and gauge if any are either currently being incorporated into the calculation or if any do not pertain to the institution any longer. 

A max loss assumption is a common assumption used in q-factor frameworks to define the goal posts or the high watermark of the reserve. Max loss assumptions can be developed at the segment level or in total, representing the highest possible reserve amount. This sets limits so institutions do not continue to add qualitative adjustments to the point where they represent a risk that is too high or has never been reached by the institution. Effectively, it is a quantifiable way to set limits on the overall reserve calculation.  

Institutions have various methods for developing the maximum loss assumption. One approach is to model a stressed economic scenario to determine the reserve required under the worst possible economic conditions. This may involve stressing various assumptions, such as loss rates reflecting those observed during the previous Financial Crisis and setting prepayment rates to 0%. Further, if your model uses regression analysis and predicted economic factors, such as unemployment, the user may set employment to historical highs to model a max loss environment. The conditions of a max loss scenario to use in the model are determined by the user. However, we recommend documenting what assumptions are made to model the max loss scenario. Alternatively, specific black box models can model different economic scenarios. The model may use various sources for “severe” economic conditions, which would be integrated into the model to generate a reserve calculation under these severe economic conditions. The reserve calculation for these severe scenarios is often used as the max loss assumption.  

If your model cannot model stressed economic scenarios, institutions often analyze their own or peer group loss history during previous stressed economic periods, such as the Financial Crisis. For example, institutions may use the worst 2-, 3-, or 4-year stretch in terms of losses by segment or total as your maximum loss assumption. The length of the period to use can depend on the typical average life of your portfolio. If the portfolio contains more long-term loans, such as mortgages, a longer period may be appropriate, whereas a shorter period may be suitable for short-term loans. 

Overall, developing a max loss assumption will help institutions quantify the high-water mark of their overall CECL reserve to avoid calculating an unreasonable reserve. 

Based on relevant internal or external data series, the selected q-factors will be assigned a risk level according to their potential impact on the financial institution at that time. Many financial institutions typically incorporate “low,” “moderate,” and “high” risk environments when developing their q-factor framework. There can be “improvement” risk environments if the institution’s historical experience underperformed expectations for the future. The use of “improvement” risk levels has been minimal due to the low loss environment experienced recently.  

“Low” risk environments would translate to a small percentage of q-factor adjustment, or many times none at all. As risk increases based on the q-factor, the financial institution would move from “low” to “moderate” to “high,” which would continue to increase the q-factor basis points added to its CECL estimate. We recommend that the financial institution look at its historical data when determining these risk levels. For instance, what was the highest level of delinquency that the financial institution sustained? This metric can then be used as the “high” risk level for the trends in the delinquency q-factor. On the other hand, where does the financial institution typically operate in an average environment for delinquencies (i.e., a five-year average may be 25 basis points of its total loan portfolio)? This can be considered its “low” risk environment. “Moderate” risk would be somewhere in the middle. 

This type of analysis should be completed for each q-factor that the financial institution will use, and it should be supported by internal or external data. Financial institutions will then evaluate their q-factors, and the metrics used for each will determine the financial institution's risk level. The financial institution's qualitative adjustment will automatically be calculated using a combination of the "max loss" scenario and the identified risk level. 

For example, a financial institution establishes the allowance ceiling for its maximum loss scenario as 3% based on its performance during the last financial crisis. The model calculation estimates a lifetime loss of around 1% in current economic times. The financial institution would have a q-factor range of 2%, meaning that if all q-factors were in a “high” risk level, the q-factor would equal 2% plus the 1% for the model, totaling 3%, which is consistent to how the financial institution performed in the last high-risk scenario. Putting one factor into practice, consider a hypothetical financial institution that lost its team of lenders who had been there for years. Since one of the q-factors assesses the experience, ability and depth of the relevant management and staff, that q-factor will be assigned a much higher risk rating than before. The financial institution would continue this exercise across all q-factors. 

Financial institutions must ensure accuracy and prevent double-counting risks by identifying if elements of q-factors are already integrated into their models. For instance, if a financial institution is forecasting metrics such as the unemployment rate, it may inadvertently double count that risk if it separately incorporates this factor in its q-factors. 

A common component of a qualitative framework is factor weightings. These are percentages that weigh the adjustments based on the perceived impact each factor has on the current risk of the portfolio. For example, if an institution feels the levels of delinquency have a greater impact on the current credit risk of the portfolio than changes in lending policy, the levels of delinquency will be given a higher weight, resulting in larger adjustments given the same risk scores. These weights may be evaluated periodically to confirm or review their validity. Management selects the weightings qualitatively, purely based on their discretion and judgment. It is recommended to document and justify the appropriateness of the weightings.  

Now that we have developed our risk factor scales and determined our risk scores, max loss assumptions, and weightings, this section will explain the most common approach to calculating the qualitative adjustments.  

The picture below illustrates the adjustment calculations for each segment, the risk score, and the final adjustment amount. First, the major risk adjustment (F) is calculated as the max loss rate (Y) minus the quantitative model-generated historical loss percentage (X) and multiplied by the factor weights (A). Then the moderate risk (E) is two-thirds of the major risk adjustment and minor risk (D) is one-third of the major risk adjustment. The improvement adjustment (B) is calculated as the historical loss percentage minus the minimum loss rate (typically set to 0%) and multiplied by the factor weightings. Lastly, as illustrated, the final adjustment is determined based on the risk scores. The sum of the adjustments is the total qualitative adjustment percentage to be applied to the loan balances to arrive at the total qualitative reserve for that specific segment. 

This is a common example of calculation, but it is not the only method for qualitative calculation. The methodology could be applied to the entire portfolio rather than each segment. The calculated adjustments could be ranges from which management makes selections. For instance, if the major risk adjustment range is between seven and 11 basis points, management could choose the high end, low end or midpoint of the range. Ultimately, the q-factor framework should be reasonable, supportable and well-documented. 

That indicates your model provider likely already considers certain typical q-factors in the quantitative portion of the model. However, if you do not have any q-factor framework or any documentation showing current conditions were considered in the CECL model, it is recommended that documentation is still created to support this decision. If you evaluated q-factors under Allowance for Loan and Lease Losses (ALLL), your auditors and/or regulators may look for documentation that shows that you evaluated those factors under CECL, or documentation that supports those factors are already considered in the quantitative portion of your model.  

For example, if a model considers economic factors to predict assumptions, it may be documented that no additional qualitative adjustments will be made related to economic factors to avoid double counting these factors. Furthermore, if the quantitative model segments and applies assumptions based on levels of delinquency, it should be documented that no additional q-factors will be applied for delinquency since it is already accounted for in the model, and higher levels of delinquency will receive higher loss rate assumptions.  

Alternatively, for factors that are not considered in the quantitative portion of the model, such as changes in lending policy, institutions may document that current conditions for these factors have not changed and thus do not require qualitative adjustment. However, if that is the case, we would still recommend developing a framework under which adjustments would be made had the current conditions of the factor been different. For example, if you made significant changes to lending policies that increase the risk of loss in the future, how will you quantify a qualitative adjustment to consider this? Although putting CECL in place has given financial institutions some additional work, it has also given them clearer forecasts supported by well-documented and defensible data.