Portfolio Risk & CECL Allowance Calculation CECL Accounting CECL Models Incurred Loss Stress Testing

Making Qualitative Adjustments and Stress Testing in Uncertain Economic Times

June 19, 2020

Read Time: 0 min

Key Takeaways

The unpredictability of the economy has created questions around reporting losses.
Without historical data to rely on, institutions can explain why it determined a forecast and provide support for that determination.
A financial institution should reflect its belief in the degree of the severity of the economy expected in the next few years.

The coronavirus pandemic has upended so many things and created uncertainty within the financial services industry. Credit risk operations, such as the allowance and stress testing, are not exempt.

When making allowance for loan and lease loss (ALLL) or allowance for credit loss (ACL) calculations, financial institutions must consider the uncertainty presented during our current economic and societal times. The 2008 financial crisis exposed significant weaknesses of relying on incurred losses. In response, the FASB replaced the standard with the current expected credit loss (CECL) model to allow for more timely adjustment of reserve levels. After more than ten years of economic expansion, financial institutions are now grappling with how the pandemic may impact reporting losses. One thing is clear: “directional consistency” with forecasts is not the problem. Financial institutions know if they need to make adjustments up or down, but the problem lies in determining the actual adjustment needed.

Change in reasonable and supportable forecasts

The incurred loss model and CECL require reasonable and supportable forecasts; however, there are little to no historical factors to base forecasts on due to the unprecedented nature of the current economy. By the end of Q1, institutions were planning forecasts around 8.5 – 10% unemployment, and within days, the forecasts jumped to 12%. Institutions were suddenly unable to forecast for three days confidently. Now, “reasonable and supportable” has shifted. An institution can explain why it determined a particular forecast and provide support for that determination without it necessarily being correct.

“Recently, we haven’t seen forecasts longer than a year,” said Neekis Hammond, Managing Director of Advisory Services at Abrigo, in a recent Abrigo webinar, Qualitative Adjustment in Times of Crisis and Stress (Testing). “Any guess is better than modeling the average in the next 9-12 months. On average, we saw one-year forecasts with a relatively quick reversion between one to two years. I don’t expect forecasts to get much longer, but I do expect quicker reversions than what we’ve seen in the past in coming out of recession.”

Save time on CECL calculation and documentation

learn more

Using a three-phased approach, a financial institution should reflect its belief in the degree of the severity of the economy expected in the next few years when reporting on current end of quarter financial statements.

Phase 1 - The changes to risk are very general. Because of the unprecedented nature of the current situation, many financial institutions will lack strong historical data to pull from. Therefore, institutions will estimate the impact of the current economic conditions in a general sense. Apply reasonable first principle estimates to make a claim that something is likely to be impacted. “A two-year, multi-family project in development, for example, might be a separate risk profile than a restaurant, bar, or other establishment that is fundamentally dependent on discretionary human gathering,” explained Garver Moore, Managing Director for Advisory Services at Abrigo.

Historical analysis will not reflect future risk or impact, Hammond added. “Taking a slice of the portfolio and using a basic model as evidence of support for a qualitative factor is important, especially as you’re speaking to the magnitude of that adjustment.”

Phase 2 – Institutions have specifically identified risks and the “error bars” are smaller in the projections due to a better understanding of the actual best- and worst-case scenarios with more data available. With more specific analysis to pull from, financial institutions have a specific evaluation of risks.

Phase 3 - Improvement is general in this third phase. The macro headwinds expected on performing loans are dying down or even reversing, and the watchlist is in a runoff mode because the institution is not repopulating problem loans at the same rate. Businesses and borrowers who have survived the economic downturn are seeing new opportunities and are looking to take out new loans.

Ongoing Considerations (9-30 and Beyond)

An incurred loss model requires a lookback at the data to adjust the results, which is especially difficult given the impact of the coronavirus pandemic. The data will be irrelevant on both sides of the recession’s impacts, as the periods heading into recession are now insignificant, and the periods coming out of recession will overstate the pain.

The transition to the CECL, on the other hand, uses forward-looking forecasts allowing data to be responsive as data changes occur – especially important in uncertain times.

CECL vs. ILM

Forward-looking, reasonable and supportable forecasts should include unemployment, GDP changes, and vacancies or delinquencies. Consider recent trends. How fast have things been changing? Is the acceleration or deceleration of that change in line with your forward-looking estimates?

If the foreseeable future is unlike recent data, your CECL calculations most likely need an adjustment. Financial institutions can leverage the following models to adjust for the lack of historical data regarding the current economy:

“Peg” to model training range max and adjust post hoc. The highest input will feed it as the top of what it was trained on and then reserve the right to make another adjustment afterward. Those that use this model will be assuming the data to an extent and then guessing – but will be explicit about guessing.
“Trust” model and underlying assumptions – Run the model, trusting the math and the underlying assumptions that 20% unemployment is twice as bad as 10% unemployment levels. Back-testing is imperative if an institution is running a model out of its calibration range.
Run “in-model” for a longer timeframe – Map a short-term, extreme unemployment number, and run the number in the model for a longer period.

Take account of the current economy and continue to calibrate models to observe trends and projections.