The reasonable and supportable forecast might be the most critical part of your CECL
forecast but is quite possibly the part that has the most uncertainty. The forecast should
be an objective exercise that relies on data and minimizes the use of subjective
components. With the convergence of CECL and the COVID-19 pandemic, banks are
looking for trends and correlations in their internal and external data to provide any
insight into their forecast. This is proving to be true with the commercial real estate
market, especially with critical components such as office, retail, and hospitality.

Delinquency rates for commercial real estate have, in fact, doubled but it should be
noted that this continues a pre-pandemic trend that extends back to 2019. The graph
also indicates that delinquencies for community banks are declining and the larger
banks are seeing delinquencies that now exceed community banks, possibly reflecting
differences in borrower profile and risk appetite. Context is important.

Note the timeframe of the graph: 2018 to 2020. When building a forecast model, is it
reasonable to increase the reserve when your delinquency rate, or some other rate,
doubled? What if the timeframe was extended?

Going back twelve years to the beginning of the last recession, the graph now shows that while recent delinquency rates are moving, they are still significantly below past recession numbers (and incredibly below figures from the S&L crisis of the early 80s (not shown)). It also shows that delinquency rates for community banks during the last recession peaked one year before larger banks, both generally after the recession was over. Again: more context.

When building a forecast model, or even a qualitative matrix, this graph, along with other data, indicates the importance of using historical data that reflects your bank and also extends through an economic cycle. The incurred allowance model, based on its parameters, appropriately focuses on near-term data. CECL requires a longer-term data driven view that will require banks to look to their past to help predict the future.

When using historical data for a forecast, it is important to go back and calculate the accuracy of that forecast. Data points are never perfectly correlated, and understanding why a forecast did or did not reasonably predict losses is the basic concept of backtesting. Using a backtesting program to constantly refine your Allowance for Credit Loss (ACL) model will ultimately lead to a more accurate ACL model. To learn more about backtesting, join us on March 25 for the Q1 Technical Series: Monitoring and Backtesting CECL.