Skip to main content

Capital stress testing for large banks in light of the COVID-19 pandemic

May 12, 2020

Capital stress testing for large banks in light of the COVID-19 pandemic

The COVID-19 pandemic has imposed tremendous human costs around the world. In response, households, firms, and governments have substantially reduced economic activity, which has, in turn, severely reduced and disrupted their incomes. Such massive economic dislocation can prevent households and firms from repaying their debts to banks on time or in full, which could impose costs on banks. At the same time, the value of securities held by banks could fall and their losses from operations could rise. In total, COVID-19 could ultimately lead banks to suffer very significant losses. However, the potential economic outcomes from COVID 19 are extremely uncertain.

Banks fund themselves, in part, with equity, which is the final backstop to absorb such losses. Losses that erode equity too much will lead banks to fail. In the past, governments have required taxpayers to absorb these losses to prevent such outcomes. In other words, these large banks have been deemed too big to fail.

Could potential bank losses from the economic fallout of COVID-19 materially erode banks’ equity cushion and, if yes, by how much? To answer that question, the Federal Reserve Bank of Minneapolis ran a series of “stress tests.” Stress tests, which typically look over a nine-quarter horizon, are based on two inputs. First, we need a financial model of how various economic and financial variables, such as the unemployment rate or the value of the stock market, and changes to those variables affect the losses that banks suffer. Second, we need scenarios that describe the values that these variables might take over time. Importantly, the scenarios are not forecasts of what is expected to happen in the future. Instead, they represent examples of potential bad outcomes that could occur; by definition, these scenarios must be worse than the expected economic and financial outcomes to make this a “stress” test.

The rest of this document describes large-bank stress test results produced by the Federal Reserve Bank of Minneapolis assuming several alternative COVID-19-driven economic scenarios in which banks continue paying dividends. We find that large banks would experience capital declines ranging from roughly $510 billion to $560 billion under a range of plausible alternative scenarios.

These declines are very large. On average, they cut banks’ equity levels roughly in half. The aggregate ratio of capital to risk-weighted assets for the group falls from 11.5 percent to between 5.4 percent and 5.9 percent depending on scenario. While this average ratio remains just above the regulatory minimum of 4.5 percent, it is at financial crisis levels. This figure was 5.4 percent as of the first quarter of 2009 even after the government injected equity into the banks. COVID-19 scenarios would wipe out nearly all the additional capital the firms have accumulated over the past decade.

Another key measure of capital adequacy, one that does not account for the risk of bank assets, actually falls below regulatory minimums. The average ratio of aggregate capital to total assets declines to a range of 3.4 percent to 3.7 percent. The required minimum amount used in the 2019 Federal Reserve stress tests was 4.0 percent for this measure. This makes clear that equity levels for banks would be so low post-stress that markets would almost certainly question their ability to lend during the downturn, and perhaps question their ability to continue as a going concern.

And the situation is worse if we consider the variation of equity capital levels for individual firms. Some firms in the Minneapolis stress test have equity funding falling below their regulatory minimum requirements under both measures of capital. In the 2008 crisis, ambiguity about which institutions had sufficient capital and which didn’t caused market participants to question the resiliency of the banking sector as a whole. As noted, the estimates from this modeling effort come with significant uncertainty, reflecting the uncertainty that exists due to COVID-19.

In the rest of this document, we describe our stress test modeling approach and the scenarios we consider.

Stress test model description

The modeling framework we use is a slightly modified version of the “CLASS model” developed by economists at the Federal Reserve Bank of New York.1 CLASS developers describe it as a top down model of the U.S. banking industry generating projections of bank income, losses, and capital under various macroeconomic scenarios.

Specifically, we relied on the software code and description of CLASS found here: https://www.newyorkfed.org/research/staff_reports/sr663.html.

We modified the CLASS model or the implementation of the model in eight ways. We did so either to make it easier for others to replicate our results or because a change in law, reporting, or policy required it, or simply to account for the change in dates between this analysis and the original CLASS ones. We now describe the changes.

  1. We utilized data only from top-tier bank holding companies as reported on the FR Y-9C (consolidated financial statements for holding companies). The original CLASS model used bank holding company information as well as data on large independent banks that were not part of a holding company structure. We followed the CLASS methodology and included observations from the 200 largest institutions in a given quarter. All of the remaining firms are aggregated as a single entity.
  2. We measured loan losses and loan balances over 13 categories, as opposed to the 15 specified in the original CLASS model. Loans to depository institutions and loans to foreign governments were combined with “all other loans” (one item is no longer reported as a separate component on the FR Y-9C, and the other is quite small in overall size).
  3. Loan losses from first lien and junior lien mortgages were not reported as separate items until 2002 (prior to that they were combined into a single category). We imputed the losses on the two categories from 1996–2001 by simply allocating them on the basis of the individual loan balance in each category at a given firm. The CLASS model imputation was based on an industry-level ratio.
  4. We began measuring our data in 1996q1 and included information through 2019q4. The original CLASS model started with data in 1991q1 and ended in 2013q3.
  5. We updated the calculation of taxes and capital to reflect the current regulatory system. The primary measure of bank capital is now “common equity tier 1 capital” (CET1) and not tier 1 common capital. CET1 capital is primarily the common stock issued by an institution plus its retained earnings and accumulated other comprehensive income, minus the required regulatory deductions.
  6. To simplify our analysis and allow for easier replication, we removed the impact of any future deferred tax assets on our projections of bank capital. Now future CET1 capital is a simple function of starting capital and projected revenue, expenses, loan losses, taxes, dividends, and adjustments for minority interests. Dividends are assumed to be held constant over the projection horizon at the 2019q4 level.
  7. We assumed that all balance sheet positions, including risk-weighted assets, would remain constant over the projection horizon (so no growth or contraction in any assets or liabilities). This assumption is consistent with the instructions released for the 2020 CCAR and DFAST stress tests.
  8. The original CLASS model did not include estimates of losses from operational risk, the global market shock to large bank trading portfolios, or the counterparty default scenario component, all of which are part of the annual stress test exercises. We included those by assuming the losses would be equivalent to what was produced in the Fed’s 2019 stress tests (and adjusted them to reflect the set of 21 firms used in our analysis).2

We used our modified version of the CLASS model to estimate income, losses, and final CET1 capital of the 21 domestic U.S. banking organizations that are part of the 2020 CCAR and DFAST stress tests (we excluded all international holding company firms that are part of the tests).3

Scenario description

We measured bank capital under three separate economic scenarios designed to capture the impact of a prolonged COVID-19 crisis. The first scenario was constructed by the Minneapolis Fed and was intended to describe a lengthy U-shaped recession and recovery. The second scenario is a more severe version of the U-shaped recession. The third scenario is based on the most negative individual forecasts from the Blue Chip Consensus survey released on April 10 and describes a sharp but still relatively more V-shaped recession.

The “U Scenario” assumes that initial attempts to end the lockdown need to be reversed as a second wave of COVID-19 cases emerges. A strategy of test-and-trace at a national scale remains infeasible, and the prolonged shutdown leads to more firm exits and a slower rebound in economic activity. In this scenario, unemployment rises to nearly 18 percent in 2020q3 and then gradually declines over the remainder of the scenario. The unemployment rate falls below 9 percent at the end of 2021 and is just above 7 percent in 2023q1. Output contracts by 17 percent over 2020 before recovering by 12 percent in 2021; output grows a bit faster than a 4½ percent annual rate over the last five quarters of the scenario. Treasury yields are extremely low throughout the scenario; the 3-month yield is 0.1 percent throughout, and the 10-year yield quickly falls to 0.25 percent before rising gradually to just above 2 percent at the end. The spread between BBB corporates and the 10-year Treasury widens to nearly 6 percent at the end of 2020 before shrinking to 2 percent in 2023q1. The stock market falls 53 percent over 2021 and then gradually recovers, ending the scenario about 5 percent below its 2019q4 level. House prices fall 23 percent before nearly recovering as well. Commercial property prices decline 40 percent, but only make up about one-third of that loss by the end of the scenario.

The “Severe U Scenario” also assumes that initial attempts to end the lockdown are reversed and remain in place until a vaccine is deployed at the end of 2021q1. This results in the most severe recession on record. In this scenario, the unemployment rate climbs to nearly 29 percent in 2020q3 and output falls by 25 percent over the year. After a couple of essentially flat quarters, output recovers sharply over the rest of 2021 and ends the scenario at about its 2019q4 level. Unemployment ends the scenario at just above 7 percent. This scenario inherits most of its financial variables from the U Scenario baseline, but assumes equities fall an additional 5 percentage points before recovering.

The “Negative Blue Chip Scenario” starts with the average of the lowest 10 GDP growth rates and the average of the top 10 unemployment rates for each of the first eight quarters, with continued recovery in subsequent quarters. Relative to the U Scenario, the decline is not quite as bad, and the recovery begins earlier, although it is quite gradual. It uses the same paths for financial data as the U Scenario.

Stress test results

The table below reports the aggregate amount of CET1 capital funding of the 21 banks as of 2019q4 and the projected amounts over the next nine quarters based on the stress test model of the three scenarios described above. We find that the firms would lose over $510 billion in capital in the U Scenario, over $530 billion in the Negative Blue Chip Scenario, and roughly $560 billion in the Severe U Scenario. CET1 expressed as a percentage of risk-weighted assets would fall from 11.5 percent in 2019q4 to levels at or below 5.9 percent and to levels at or below 3.7 percent when measured as a percentage of average assets (depending on the scenario).4

For questions about this article, please email mpls.src.stress.test.estimates@mpls.frb.org.

Aggregate Bank Capital under Potential Coronavirus Scenarios
(billions of $ of CET1 capital for the 21 firms)
Date U-Shaped Scenario Negative Blue Chip Severe U-Shaped Scenario

2019q4

1050.4 1050.4 1050.4

2020q1

864.7 870.2 801.5

2020q2

716.2 729.1 579.1

2020q3

661.1 656.7 631.0

2020q4

660.2 634.2 604.3

2021q1

629.8 604.6 576.9

2021q2

594.1 569.0 547.9

2021q3

562.6 540.1 519.2

2021q4

539.5 517.9 492.0

2022q1

539.8 518.7 502.6
Minimum level 539.5 517.9 492.0
Change from 2019q4 -510.9 -532.5 -558.4

Additional metrics
Capital and Leverage Ratios U-Shaped Scenario Negative Blue Chip Severe U-Shaped Scenario
CET1 / RWA [2019q4] 11.5% 11.5% 11.5%
CET1 / RWA [min quarter] 5.9% 5.7% 5.4%
Change in the ratio -5.6% -5.8% -6.1%
CET1 / Average Assets [2019q4] 7.3% 7.3% 7.3%
CET1 / Average Assets [min quarter] 3.7% 3.6% 3.4%
Change in the ratio -3.5% -3.7% -3.9%

Endnotes

1 This research was subsequently published in “Assessing Financial Stability: The Capital and Loss Assessment under Stress Scenarios (CLASS) Model,” Journal of Banking and Finance, Volume 69, Supplement 1, August 2016.

2 Twelve of the 21 firms included in our analysis were also part of the 2019 DFAST exercise. These 12 firms accounted for 91 percent of the total assets of the firms in the 2019 exercise. The Fed disclosed in the 2019 Supervisory Stress Test Results document that total operation risk losses were $123 billion. We assumed that 91 percent of these losses (or $112 billion) were due to domestic firms and attributed that among the 21 firms based on their respective asset sizes. We also assumed that the domestic firms subject to the global market shock and the counterparty default scenario would have the same magnitude of losses as they did in the 2019 DFAST exercise ($80 billion). These losses, along with the operational risk losses, were then incorporated into the capital projection for each firm.

3 The following firms were included in the analysis: ALLY, AXP, BAC, BK, C, COF, DFS, FITB, GS, HBAN, JPM, KEY, MS, MTB, NTRS, PNC, RF, STT, TFC, USB, and WFC.

4 Our analysis does not include any impact from reduced income related to mortgage loan forbearance or additional losses from greater utilization of loan commitments by businesses (which would further increase the decline in capital).