US Stress Test Capital Requirements Are Excessively Volatile and Overestimate Losses

Identifying the Problem and How to Solve It

  • Excessive year-over-year volatility in the stress test capital requirements constrains banks’ ability to facilitate the growth of the real economy and capital markets via lending and market-making activities.
  • In the U.S., this volatility is driven heavily by the Global Market Shock (GMS) component of the stress tests. A key reason for the volatility is that the GMS loss estimates are not correlated with the actual riskiness of banks’ trading book positions. The result is a GMS loss estimate that is both volatile and that significantly overestimates stress losses.
  • To address these problems, the Federal Reserve should design GMS scenarios that are appropriately stressful yet plausible, and address the double count between the GMS and market risk capital requirements as they implement the Fundamental Review of the Trading Book (FRTB) reforms over the coming years.

Background: Large U.S. Banks’ CET1 Capital Levels Have Grown Significantly Since 2009

Since their initiation in late 2010, the CCAR and DFAST processes have become the cornerstones of the regulatory capital framework for large U.S. banks.  To comply with CCAR/DFAST requirements, U.S. banks have built-up robust capital adequacy processes and have steadfastly increased their CET1 capital levels.[1]  Large U.S. banks’ common equity tier 1 (CET1) capital ratios and levels have grown significantly since 2009 as shown in Figure 1.

Figure 1.  CET1 capital ratios and levels of all CCAR firms since 2009.



Source: SIFMA Research Quarterly [2]

Policymakers broadly agree that the quantity and quality of capital in the system have improved dramatically over the past decade.  For example, Federal Reserve Board Chairman Jerome Powell observed in a February 2019 statement to House members that capital levels are “just right”.[3]  Similarly, Acting Comptroller of the Currency Michael Hsu said, in May 2021, that banks’ “capital and liquidity ratios are strong”[4] and he’s generally comfortable with big banks’ capital levels.[5]

As a result of these strong capital levels and with the unprecedented support from regulators, e.g., temporary exemption of U.S. Treasury securities and central bank reserves from SLR,[6] the U.S. banking system weathered the COVID-19 event and its associated severe market stresses without any bank failures, while also continuing to support the capital market and the broader economy.

One example of this was the support lent to the U.S. Treasury markets by the largest banks.  Figure 2, Panel A, plots the U.S. Treasury securities held for trading by eight U.S. G-SIBs collectively between December 2017 and December 2021.  The banks collectively held $345bn for trading in December 2019 and peaked at $452 in June 2020 during the height of Treasury market stress.  This suggests that the largest banks were sufficiently well capitalized to provide significant support to a crucial funding market during stressed conditions.

Figure 2.  Treasury security held for trading by 8 U.S. G-SIBs during Dec. 2017 – Dec. 2021.

Treasury security held for trading by 8 U.S. G-SIBs during Dec. 2017 - Dec. 2021.

Data Source: FR Y9-C and FFIEC 102[7]

Volatile or Excessively Conservative Capital Requirements May Have Negative Economic Impacts

Excessive volatility or over-estimation of stress test related capital requirements constrain banks’ ability to facilitate the growth of the real economy and capital markets via lending and market-making activities.  For example, investigating the connection between loan rates and capital requirements, Glancy and Kurtzman (2022) find that “a one-percentage-point increase in capital requirements raises loan rates by 8.5 basis points for high volatility commercial real estate.[8]  Over the last decade, the role of banks in the global financial system has diminished while the non-bank financial intermediation (NBFI) sector has substantially grown.  According to the Financial Stability Board (FSB), at the end of 2020, banks accounted for $180 trillion of the $469 trillion total global financial assets and the size of the NBFI sector was at $227 trillion; that marks a significant shift since the end of 2008, when their respective sizes were $120 trillion and $100 trillion.[9] That shift has occurred as stringent capital requirements, including stress test capital requirements, were implemented in the U.S. and in other major jurisdictions.  Thus, while it is crucial that banks’ capital requirements appropriately reflect risks, they ought not to be excessively volatile, since doing so restricts the ability of banks to support the economy.

The U.S. Stress Capital Requirements are Excessively Volatile

The CCAR stress capital requirement is calculated by determining the largest depletion of CET1 ratio, i.e., the CET1 ratio as of the final quarter of the previous capital plan cycle minus the lowest projected CET1 ratio in any quarter of the planning horizon under a supervisory stress scenario (or start-to-trough CET1 ratio drawdown).[10] Figure 3 (the red dashed line) plots the weighted average of start-to-trough CET1 ratio drawdowns, and the weighted average CET1 ratio of 8 U.S. G-SIBs (the black solid line) since 2012.[11] It is clear that the CET1 ratio drawdown is quite volatile, with volatility of 1.06% more than doubling the volatility of actual CET1 ratios (or 0.47%).  Excess volatility of the CET1 ratio drawdowns directly translates into excess volatility of capital requirements for large U.S. banks, thereby constraining banks’ ability to provide financing and market-making to their full potential.

Figure 3.  Risk-Weighted Assets (RWA) weighted average actual CET1 capital ratio, start-to-trough CET1 ratio drawdown, and their variability, i.e., standard deviation, in bracket of 8 U.S. G-SIBs during Sept. 2012 – Dec. 2021.

Risk-Weighted Assets

Data Source: Federal Reserve Board[12]

In his remarks at the Brookings Institution on September 7, 2022, Federal Reserve Board Vice Chairman Michael Barr announced that “[w]e are looking holistically at our capital tools [and] work to minimize unintended consequences [and] consider adjustments, if any, to the supplementary leverage ratio, countercyclical capital buffer, and stress testing”.[13] On stress testing, former Vice Chairman Randal Quarles, in his farewell speech in December 2021, stated that “the volatility of these requirements from year-to-year indicates that we still do not have those requirements quite right … excessive volatility in a firm’s capital requirements limits the firm’s ability to manage its capital effectively … and we must guard against excess volatility when it has no particular relationship to changing risks at firms”.[14] In order to determine how to mitigate the excess volatility of stress test capital requirements for large U.S. banks, we must first understand what causes the excess volatility of the CET1 ratio drawdowns.

The Global Market Shock is a Key Source of Volatility in the Stress Test Capital Requirements

Of the $612bn aggregate losses in the 2022 CCAR/DFAST exercise, trading and counterparty losses under the GMS scenario (designed to stress trading positions held by banks with significant trading operations) accounted for $100bn of which $92.2bn were borne by eight U.S. G-SIBs because of their substantial trading or custodial operations.  The aggregate losses in 2021 was $474bn and eight U.S. G-SIBs reported $80.7bn of the $86.5bn trading and counterparty losses.  Naturally for these eight banks trading and counterparty loss is a critical component of pre-tax net income estimate during the capital planning horizon.  And so is the corresponding volatility.

Through a panel regression approach, Table 1 investigates the relationship between start-to-trough CET1 ratio drawdown, GMS loss estimate, and riskiness of 8 U.S. G-SIBs’ trading positions.  Two models were considered, i.e., (1) start-to-trough CET1 ratio drawdown as a function of GMS loss estimate, market risk RWA density, and bank-specific effect (see column 1), and (2) GMS loss estimate as a function of market risk RWA density and bank-specific effect (see column 2).  For cross-bank comparison, a bank’s GMS loss estimate is normalized by its total trading assets and liabilities.  Riskiness of trading positions is measured by market risk RWA (risk-weighted asset) density which is defined as the ratio of advanced approach market risk RWA and fair value of gross trading assets and liabilities.

Table 1: The relationship between start-to-trough CET1 ratio drawdown, GMS loss estimate, and riskiness of 8 U.S. G-SIBs’ trading positions.

Start-to-trough CET1 ratio drawdown is defined as the ratio of a bank’s CET1 ratio as of the final quarter of the previous capital plan cycle minus the lowest projected CET1 ratio in any quarter of the planning horizon under a supervisory stress test; GMS Loss Estimate is normalized by total trading assets/liabilities; Market Risk RWA Density is the ratio of advanced approach market risk RWA and total trading assets/liabilities. Data Source: FR Y9-C and FFIEC 102

As expected, the GMS loss estimate has material impacts on the bank’s start-to-trough CET1 ratio drawdown (i.e., the correlation between GMS loss estimate and start-to-trough CET1 ratio drawdown is statistically significant at 0.1% level).  The result indicates that on average for every 1% increase in the GMS loss estimate, the bank’s CET1 ratio drawdown (and the resulting capital requirement) increases 0.22%.  For the eight U.S. G-SIBs, between 2021 and 2022 CCAR/DFAST exercise, GMS loss estimate increased 14.25% (from $80.7bn to $92.2bn).  As the result of 2022 stress testing, the stress capital buffer requirement for three of the 8 U.S. G-SIBs will each rise about 1% point which translates into an increase in minimum CET1 capital requirement about $55bn for these three banks alone effective on October 1, 2022.[15]

However, both the start-to-trough CET1 ratio drawdown and GMS loss estimate appear to have little connection with the riskiness of the banks’ trading positions, as shown in Column 2 of Table 1 (i.e., the correlation between market risk RWA density and GMS loss estimate or start-to-trough CET1 ratio drawdown is not statistically significant).  In fact, whereas the eight U.S. G-SIBs observed a 14.25% increase in GMS loss estimate from $80.7bn in 2021 to $92.2bn in 2022 CCAR/DFAST exercise, the weighted average riskiness (Figure 2 Panel B) of their trading positions decreased from 20.13% to 17.16% during this same one-year period.

It is not surprising that the GMS loss estimate is not perfectly correlated with market risk RWA as market risk RWA is calibrated using historical stress scenarios, while GMS loss is estimated using supervisory specified GMS scenario.  And the two sets of scenarios differ.  What is surprising is that loss estimates under the two scenarios are not correlated at all, which indicates that GMS scenario bears no resemblance with historical stress scenario in any meaningful way.

Regulation YY intends the prescribed GMS scenario to be “hypothetical but plausible” for the loss estimate to be meaningful.[16]  However, a prior SIFMA study concluded that the prescribed GMS scenario is not reasonably plausible (i.e., extremely low likelihood of occurrence based on empirical data) in terms of shock sizes and correlation assumptions.[17]  Additionally, year-over-year shock volatility doesn’t reflect gradual adjustments to emerging views of risks that are linked to feedbacks from the prior year’s analyses.  As a result, the GMS loss estimate is volatile, overly conservative, and exaggerates stress loss estimates and its volatility.  This is one of the key drivers of the excess volatility of banks’ start-to-trough CET1 ratio drawdown and, in turn, their stress test capital requirements.

How Can Excessive Volatility in the Stress Test Capital Requirement Be Mitigated?

There are at least three options that the Federal Reserve Board could adopt to mitigate the excess volatility of stress test capital requirement – (1) ensure the GMS scenario is reasonably plausible, (2) address the interplay between GMS and market risk capital by excluding market risk loss used to calibrate RWA from GMS loss estimate, and (3) institute a three-year average of GMS loss estimates.

To design a reasonably plausible GMS scenario, the Federal Reserve Board should tailor the shock sizes to reflect the liquidity and price stability characteristics of particular asset classes and validate the correlation assumption based on empirical analysis and historical observations.  A plausible GMS scenario would largely remove the spurious volatility of GMS loss estimate.

The GMS loss estimate captures the mark-to-market stress losses of trading positions banks held on the as-of date of the GMS scenario.  Market risk RWA, a component of the total RWA – the denominator of capital ratios – capitalizes the mark-to-market losses of trading positions banks held as of the final quarter of the previous capital plan cycle.  Therefore, the mark-to-market losses get counted twice to a large degree.  Addressing the interplay between GMS loss estimate and market risk RWA becomes more critical as the Fed replaces the current market risk capital framework with the Fundamental Review of the Trading Book (FRTB) as the result of Basel III end game implementation – a commitment the agencies recently reconfirmed.[18]

Similar to GMS, the FRTB is inherently a stress test framework that requires market risk RWA to be calibrated using a 1-year stress period going back to 2007 assuming conservative market risk factors shock sizes (via liquidity horizon, or LH) and correlation.[19]  As a result, excluding market risk losses used to calibrate market risk RWA (especially under FRTB) from the GMS loss estimate would mitigate the exaggeration of stress loss estimate and associated volatility.

Lastly, Former Federal Reserve Board Vice Chairman Randal Quarles proposed to “average the results of the current year’s stress test with the corresponding stress test results from the previous two years”.[20]  He concluded that “[t]his would not affect the overall stringency of the tests but would mean that firms that hold the risk profile of their balance sheets relatively constant would not see a large spike or plummet in their required capital from year-to-year based on changes in the Fed’s scenario choices.” Averaging results in this way would also be an important way of reducing the year-on-year volatility we see in the stress test capital requirement.


Banks play a key role in the global financial system.  But that role has diminished significantly over the last decade or two while non-bank financial intermediation sector has grown substantially during the same period.  Post-2008 financial crisis regulatory reforms forced large banks to cut back on trading and maintain a large portion of their reserves in the safest assets.  As a result, deposits at U.S. banks exceeded loans by over $7 trillion in 2021 up from about $250 billion in 2008.[21] And, many of the risks once held by banks have been transferred outside of banks resulting in many key funding markets becoming more volatile.[22]  Excessive volatility and over-estimation of losses under the stress test capital requirement is a key constraint on banks’ capacity to make markets for their clients and lend to consumers and businesses.

To ensure the smooth functioning of the U.S. capital markets and the broader financial system, the Federal Reserve Board should set capital requirements that are appropriately reflective of risks but not unduly conservative nor excessively volatile. Designing a GMS scenario that is stressful yet plausible (that is, based on real-world risks banks face) and reassessing the interaction of the GMS with other capital requirements would be a good start.

Dr. Guowei Zhang is Managing Director and Head of Capital Policy SIFMA.

Dr. Peter Ryan is Managing Director and Head of International Capital Markets and Strategic Initiatives SIFMA.










[8] David Glancy and Robert Kurtzman, “How Do Capital Requirements Affect Loan Rates? Evidence from High Volatility Commercial Real Estate”, The Review of Corporate Finance Studies, vol 11, February 2022.

[9] See Graph 1-1 of FSB’s Global Monitoring Report on Non-Bank Financial Intermediation 2021.

[10] Starting April 5, 2021, the start-to-trough CET1 ratio drawdown serves as the basis for Stress Capital Buffer requirement (SCB) under the SCB final rule, i.e.,






[16] 12 C.F.R. §252 Appendix A Section 5.2.3(c).

[17] For example, based on historical and crisis data, the probability of shock sizes of the 2019 GMS spreads for A/BBB/B- corporate bonds were less than or equal to 0.001% (or approximately an once every 400 calendar years event). See


[19] A risk factor LH is how long a bank is assumed to be exposed to the risk with no ability to either hedge or exit the trade.  And under FRTB, LH for 12 out of the 26 risk factors is 3-month or longer.

[20] See Endnote 14.


[22] See Endnote 21.