Examiners are expecting bankers to take new approaches to stress testing loans, especially commercial real estate ones, which could have a profound impact on loan-loss reserves.
Ron Piccinini of First National of Nebraska Inc.'s treasury department has developed an out-of-the-box model.
Take a bank that has a simple risk scorecard for commercial real estate loans. It assigns each loan a score of 1 to 10 on the basis of its debt service coverage ratio, loan-to-value ratio, occupancy rate, and market quality — a subjective yet important variable determined jointly with the commercial banker for each and every transaction.
First, identify and collect the data necessary to compute each component of the score. To compute the debt service coverage ratio, you may need the current loan amount, interest rate, and number of amortization periods, as well as the property's projected income, expenses, vacancy rate, and cap rate. For the LTV computation, the same inputs are needed. The occupancy rate is one minus the vacancy rate, and market quality can be tied to the cap rate.
Second, create stress/what-if scenarios. How would the portfolio look if expenses rose 15% across the board and vacancies rose 5 percentage points? What would happen if cap rates rose 2 percentage points?
Third, given the stress scenarios, recalculate the score for each loan. Use the resulting scores to make a qualitative assessment of the portfolio's health.
What about probabilities? Some regulators insist that stress tests be conducted at a given confidence level (e.g., 95% or 99%). But there are no regulatory tables where the physical probability of all future events is readily available, so the idea is to model the joint probability distribution of the variables. For example, it makes sense to force cap rates and vacancy rates to move together in the opposite direction of market quality.
The question of how to calibrate these models is beyond the scope of this article, but a simple, pessimistic way is to model each variable independently and use the 95th-worst percentile of each one for the stress scenario. The results can be used as an estimate of the 95% level of confidence and may satisfy regulatory demands.