Establishing a Benchmark for Your CECL Build

March 31, 2020 marked the first quarter-end¹ in which all banks that adopted the new Current Expected Credit Loss (CECL) standard as of January 2020 were to publish their estimates of allowance for credit loss. Unexpectedly, the March 31 time period also coincided with the COVID-19 crisis, which sent macroeconomic forecasts into turmoil and sparked volatility in the newly minted measure of allowance. The result of our study puts the peer group lower bound at 1.64% and the upper bound at 3.24%. Any bank in the peer group outside the range should be looked at carefully and consideration of riskiness indicators will inform whether a said bank is reserved appropriately.

Executive management needs to assess whether its bank is more or less well reserved than its peers, and our triangulation method provides a range of estimates for CECL based on a well-rounded set of measures.

Based on their own historical experience and current portfolio mix, every institution should understand where they stand within their peer group. A common set of triangulation metrics can help standardize how institutions look at their estimates in a given business cycle.
Below are five approaches to providing the benchmark metrics for an institution. The methodology is presented here at the bank level, but would also work in a portfolio-level analysis. We will use:

DFAST² results: We use the total loss ratio and the DFAST/expected credit loss (ECL) coverage ratio to understand how well-reserved banks are. The DFAST losses are based on the nine-quarter cumulative losses under the Severely Adverse scenarios. We also create a DFAST scaler for added differentiation across DFAST banks.
NCO averages for 2007-2010: Using net charge-offs (NCOs) observed during the Great Recession, we will estimate, based on quarterly maximum and average NCO during the Great Recession, how many quarters of coverage are provided for in the bank's current estimates.
NCO under economic conditions: Using NCOs observed at different levels of economic conditions based on unemployment levels (UERs), we estimate potential reserves based on a weighted average life assumption given the forecast UER levels as of March 31, 2020.
Historical CECL – full foresight: Using historical full foresight analysis, we compute the maximum, minimum, and average observed lifetime loss through³ the Great Recession.
CECL forecaster: Using the Moody’s Analytics CECL forecaster,⁴ we compute the product-level NCOs and, based on a set of CECL assumptions, compute the lifetime loss estimate as of March 31, 2020.

While reviewing each approach (Figure 1) of triangulation methodology (index⁵), we classify each into the two classes of metrics:

Riskiness indicators

- These metrics provide a useful way to think about portfolio riskiness at a specific point but offer limited ways to be incorporated as part of the index to determine an ECL range.

- These metrics can be used to apply an override on the range of ECL moving the lower or the upper bound, creating a tighter (looser) range for the estimate.

ECL range indicators

- These metrics directly provide a range for ECL estimates based on historical experience, forecast of future conditions, or a mix of both using heuristics measures. For each metric in our analysis, we have a Baseline and a Stressed version (S4) that can be used to model the lower and upper bound of the range of ECL estimates.

- These metrics can be used directly to create a lower- and upper-bound estimate for ECL, keeping in mind that metrics based on Baseline scenario data would be applied to the lower bound, while metrics based on the S4 scenario would be applied to the upper bound.

To construct our index’s lower and upper bound for the peer group,^​6 we first computed the weighted-average ECL for the sample banks in our dataset (1.98%) shown in Figure 2. Then, we computed the weighted average for every ECL metric used in the indexes per Figure 3. Our lower- and upper-bound index constructed from our set of metrics gives us a lower bound of 1.64% and an upper bound of 3.24%, which is intuitively commensurate with the values observed from our sample set. Four banks fall below the lower bound based on the peer group averages: Wells, BofA, Key, and M&T.

In the creation of the index, expert judgment plays a role in both the selection of the metrics as well as the weights assigned to each metric. In the case of CECL estimates, there are no historical CECL estimate benchmarks, so all we can do is use our best judgment, ensure results are intuitive, and directionally correct and repeat the analysis over the coming quarters^​7 to solidify our expert judgment approximations.

The index lower and upper bound include only one riskiness indicator and all of the ECL range indicator metrics. We left out most of the riskiness indicators to make bank-level adjustments in the comparison of the results section for individual banks. Figure 3 outlines the metrics used for the lower and upper bound, as well as the weights derived based on expert judgment.

For the lower bound, our logic was to assign equal weights to the first three estimates, given their similar level of precision and directionality, whereas the last two—which were based on a three-year average—would tend to drive the actual rate lower based on the March 31, 2020 economic environment. We decided to assign equal weighting of 30% to the first three and a weight of 5% to the 3-year averages. The lower-bound estimate now gave us a baseline estimate. To make it a lower-bound estimate, we then took 70% of the final value. The 70% is based on ensuring that our lower and upper bound are grounded in credit risk observed behavior (asymmetry of credit risk).

For the upper bound, we focused on 10 stressed metrics that gave us a view of worsening economic conditions as of March 31, 2020. All 10 metrics provided a similar range of worst-case scenarios ranging from a low of 3.5% to a high of 6.3% for the entire peer group. We assigned equal weights of 10% to each metric, which were all based on either Moody’s Analytics S4, the Great Recession period, or the DFAST Severely Adverse scenario. Just as with the lower bound, to make the metric a realistic upper bound and not a worst-case upper bound, we applied a 70% factor to the index.

Credit asymmetry behavior depicts that downside credit risk is much larger than upside credit risk. Therefore, our lower- and upper-bound metrics adhere to the principle. Our average for the sample peer dataset average is 2.35%, with a lower bound at 1.64% and an upper bound at 3.24%, simply reflecting that the downside difference (3.24%-2.35%) is much larger than the upside difference (2.35%-1.64%).

Comparison of results – averages⁸

Next, we applied these weights for each bank to understand where, based on each bank's results, they fell within our lower- and upper-bound index. Figure 4 presents each bank-specific upper- and lower-bound index as well as the peer group upper- and lower-bound average estimate depicted as the red lines on the graph.

Based on the chosen methodology and weights assigned, we produced a range of estimates for each bank, which seems reasonable given what we know as of March 31, 2020. The banks that aggressively provisioned seem to be over the lower-bound threshold for the peer group (JPM, Truist, Citi, USB, Ally, Fifth Third, and so on), whereas the banks that were seen as being less conservative fall at or just below the lower bound for the peer group (Wells, Key, M&T). Figure 5 shows the individual banks’ upper- and lower-bound thresholds.

We discuss below a sample of the individual banks’ results, with added commentary based on the riskiness indicators. The riskiness dashboard in Figure 6 outlines the banks that have better-than-average riskiness metrics. This could help add to our interpretation of whether any given bank is more or less reserved as of March 31, 2020.

For example, Bank of America sits just below the peer group average lower-bound index, but quite a fair bit above the BofA bank-specific index. BofA performed on average much better in a crisis during the last stress test, having the lowest DFAST riskiness scaler for the entire peer group. Therefore, we believe that it is better reserved on average than most of its peers, and indicates that BofA is probably being much more aggressive than its peers in reserving for a lower risk portfolio.

In another example, M&T Bank sits below the peer group lower-bound index and yet above its bank-specific upper-bound index due to very low net charge-off experience on which our top-down models are conditioned. However, the DFAST riskiness index tells a different story, showing that their portfolio is one of the riskiest in the peer group. We believe that even with the added scaler adjustment, M&T should be sufficiently reserved compared to its peer group.

Figure 6 provides a view of the different riskiness indicators. All cells in green are metrics that are above average for the peer group.

Comparison of results – specific institutions⁹

Here, we present the sample results of two banks within the peer group, showing the contribution of each of our metrics to the upper- and lower-bound index. We add commentary based on 1) our observations and 2) DFAST riskiness scaler, and whether we believe the banks were aggressively reserving as of Q1 2020. There is little doubt that given economic deterioration, banks will be increasing their reserves in Q2. We believe our upper-bound estimate for Q1 2020 to be a good benchmark for what to expect in Q2 2020.

JP Morgan
JPM reserved 2.5% of loans for Q1 2020 (Figure 7). Our lower- and upper-bound estimates were 1.65% and 3.32%, which represents quite a range in their case. We emphasize that the upper bound is based on a worst-case scenario as of March 31, 2020, but as of April/May became the baseline levels—which means that the reserve level required by JPM could be substantial. In fact, JPM already mentioned at its investor conference that the company expects to reserve another approximately $7 billion for Q2 2020¹⁰. P JPM’s reserve ratio would in that case rise to about 3.4%, which would put it above the upper bound of 3.32% computed for JPM. Given that the Q2 scenario is very close to our upper-bound metrics, this is right in line with where JPM plans to be. Based on the latest July earnings release, and as expected by our analysis, JPM now sits at the upper bound of the March triangulation index with a reserve of 3.32%.

U.S. Bank
USB reserved 2.07% of loans for Q1 2020 (Figure 8). Our lower- and upper-bound estimates were 1.45% and 2.43%. The Q1 reserve sits above its midpoint of the upper- and lower-bound estimates. The provision for loan losses for Q1 was low compared to peers but the overall allowance is similar or even higher than peers. On a recent analyst call, USB disclosed.¹¹ “Our expectation is that unemployment is going to be pretty significant in the second quarter and then starting to dissipate or moderate as we go through the rest of the year and into 2021. So, our expectation is that the reserve builds in the second quarter by the industry as well as USB is going to be pretty significant.” USB’s reserve is above that of the peer group lower-bound threshold and the riskiness of its portfolio is just slightly better than the peer group average. So, we view it as well reserved for the March 31, 2020 reporting period. Based on the latest July earnings release, and as expected by our analysis, USB now sits at the upper bound of the March triangulation index with a reserve of 2.54%.

Key takeaways

The goal of this study was to create a bounded index that could be used as a proxy for the level of available reserves across a set of peer banks. We found evidence that combining top-down methodologies with risk indicators can help clarify where each bank finds itself among its peer group. In the current macroeconomic environment, we expect most banks in the peer group analyzed to move much closer to the upper bound in Q2 2020. This finding has anecdotal support in statements from JPM and USB.

As of March 31, 2020, the lower- and upper-bound indexes for the peer group sat at 1.64% and 3.24%, respectively. Our expectation, based on the banks’ current portfolios and evolution of the economic environment, is that the lower-bound index will move well above 2%, while the upper bound will shift into the low 4% range. Results suggest that a few banks in the peer group may ramp up reserve in Q2 more aggressively than the rest—namely Wells Fargo, Key Bank, Fifth Third Bank, Citizens Financial Group, and Regions Bank. The latter three have a riskier loss profile than the peer group average, and Wells relies on its expected recoveries forecast in Q1.

Further results will be released after the Q2 2020 call report data becomes available. It should help refine the weights of our triangulation index.
Repeating our proposed methodology on portfolio-level call reports may help to further narrow the index range. In times of extreme volatility, a combination of heuristics can offer much-needed reassurance about the range of possible outcomes. The tools used in this type of benchmarking are available from Moody’s Analytics. When internal models seem inadequate, these tools can provide executive management help for future planning.

For more information, read the full paper: CECL Build – Is it Enough?

Additional resources from Moody’s and Moody’s Analytics

Moody’s Topic Page on COVID-19
Forecasting the Impact of the COVID-19 Recession on Consumer Credit Losses
Moody’s Analytics – CECL And IFRS 9 Recommendations: Handling Shifting Economic Scenarios
Moody’s Analytics – Severe Pandemic, a Protracted Economic Slump, and Commercial Real Estate Forecast Scenarios
Moody’s Investors Service - Worldwide, Coronavirus Aid Will Benefit Financial Institutions, But Could Raise Long-Term Risks