Comparing the Results of Stress Tests: CCAR 2016 Versus NYU Stern V-Lab Model

On June 29, the Federal Reserve published the results of the 2016 Comprehensive Capital Analysis and Review (CCAR) for the 33 bank holding companies that were included in the exercise.  All banks passed the quantitative assessment of CCAR by maintaining projected post-stress regulatory capital ratios under their own planned capital actions that were well above the minimum required regulatory capital ratios in an excess of about $275 billion.  Notwithstanding the substantial amounts of capital held by U.S. banks, Bloomberg View published a recent column indicating that the largest six U.S. bank holding companies have a $350 billion capital shortfall based on NYU Stern’s Volatility Laboratory (V-Lab) model.

In this blog post we show why the comparison between CCAR and the V-Lab results is an “apples-to-oranges” comparison and unfair for the following two main reasons: (i) the capital requirement used to derive the capital shortfall under the V-Lab methodology is twice as large as the minimum tier 1 leverage ratio requirement in CCAR; and (ii) the capital shortfall obtained using the V-Lab methodology is very procyclical, for instance the “capital shortfall” of the six largest banks at the end of 2015 was 40 percent lower relative to the amount cited in the Bloomberg column.

In general, it is useful to benchmark the results of CCAR against alternative approaches.  For instance, the supervisory scenarios included in CCAR don’t span all types of risk banks face, thus assessing banks’ resilience under different stress scenarios is a worthwhile exercise.  A group of researchers at NYU Stern School of Business have developed a simple methodology to benchmark macroprudential stress tests that relies on publicly available market data.  In addition, the results of the so called “mark-to-market” stress tests are readily accessible on the Volatility Laboratory (V-Lab) website maintained by these researchers, allowing anyone to compare the results of supervisory stress tests to this alternative benchmark.

The two key differences between the CCAR and V-Lab methodologies concern the level of data granularity required to conduct the stress tests and the specification of the stress scenarios.  On the data side, CCAR relies mainly on confidential supervisory data whereas the V-Lab stress tests use publicly available market data.  Another key difference is the specification of the scenarios under these two approaches.  Under the Dodd-Frank Act stress test rules, the more severe stress scenario in CCAR assumes an increase in the unemployment rate between 3 to 5 percentage points, but at a minimum, an increase sufficient to result in a projected unemployment rate that reaches at least 10 percent over the nine-quarter stress test horizon.  Moreover, the paths of the remaining macroeconomic and financial variables provided in the supervisory scenarios are based on the underlying structure of the scenario as defined by the Federal Reserve.  In contrast, the V-Lab stress test scenario is defined by a 40 percent fall in the stock market over a six-month period, making it a considerably simpler stress scenario.

To compare the results between the U.S. stress tests and V-Lab we consider both the projected losses over the stress scenario under each approach and the post-stress capital ratios.  Losses under the V-Lab methodology are calculated as the stock return of a bank during the stress scenario times the market value of equity at the start of the exercise.  A roughly equivalent measure of losses in CCAR is the difference between net income before taxes under the severely adverse scenario and net income before taxes under the baseline scenario.  For comparison, we also report a gross measure of CCAR losses that exclude pre-provision net revenue.  In terms of post-stress capital ratios, the V-Lab market leverage ratio under stress is defined as the ratio between the post-stress market capitalization to the quasi-market value of a bank’s assets.  The market leverage ratio under stress is not directly available on the V-Lab website, but it is straightforward to calculate since the information on the stock market return under stress is reported.1   The most relevant post-stress capital ratio under CCAR is the common equity tier 1 ratio.

image of graph

As shown at the top of Table 1, market losses under V-Lab are about 20 percent higher than the difference between net income before taxes under stress in CCAR and pre-tax net income under the baseline scenario, so it appears that the V-Lab stress scenario is slightly more severe than the CCAR stress scenario.  Moreover, V-Lab market losses are just slightly higher than the sum of losses from loans held on portfolio, trading, counterparty and operational risk events, and other additional losses obtained under CCAR’s severely adverse scenario.  The bottom panel of Table 1 shows the correlation between V-Lab market losses and CCAR losses across all banks that are included in both samples.2  Overall, the correlations are somewhat elevated suggesting both methods are fairly consistent in terms of which banks are more likely to incur large losses.
image of graph
Figure 1 compares the stressed tier 1 leverage ratio under CCAR against the V-Lab market leverage ratio under stress.  Under CCAR, all bank holding companies report a minimum tier 1 leverage ratio above 4 percent.  The V-Lab methodology uses a default market leverage ratio minimum requirement of 8 percent, a much more stringent level.3   Moreover, the V-Lab measure is very procyclical; between the end of 2015 (the start of CCAR 2016 exercise) and the end of June of this year the capital shortfall under the V-Lab methodology (using a 8 percent market leverage ratio minimum requirement) more than doubled for all banks included in Figure 1 because of the marked decline in bank equity values that occurred in the first half of this year.  For the six largest banks, the capital shortfall using the default 8 percent minimum requirement rises from $243 billion at the end of 2015 to $365 billion at the end of June of this year.  If one uses a market leverage ratio minimum requirement of 4 percent – a number much more comparable to the CCAR post-stress minimum requirement, the capital shortfall for the six largest banks drops very substantially, to just $4 billion at 2015 year-end and $69 billion at the end of June 2016.

The CCAR requirements for post-stress capital ratios are derived from the calibration of regulatory minimum capital requirements under Basel III; it is not clear where the V-Lab website’s default 8 percent market leverage requirement comes from; it may represent the minimum requirement for total capital under Basel I.  However, the capital elements eligible to be included in total capital are much broader than common equity, the numerator of the V-Lab’s market leverage ratio under stress.  Another issue is the procyclicality of V-Lab’s market leverage ratio under stress.  In particular, a procyclical capital requirement forces banks to excessively deleverage in response to negative shocks to the equity market, which leads to unnecessary swings in lending and real economic activity.

In summary, the results from the two stress test methodologies indicate that U.S. banks are extremely well capitalized to sustain the impact of very large aggregate shocks.  In this blog post we demonstrate that it is misleading to compare the capital shortfall estimated using the V-Lab model to CCAR 2016 for at least two main reasons.  First, the default minimum capital requirement used on the V-Lab website is twice the minimum tier 1 leverage ratio requirement under CCAR.  Second, the V-Lab estimate is procyclical and the size of the capital shortfall is quite sensitive to changes in stock prices, therefore it makes it a poor measure to assess whether banks have sufficient capital to absorb losses during stressful conditions.

1Following Acharya et al (2014), the book value of debt is assumed to be unchanged over the six-month stress period.

2To control for bank size, V-Lab market losses are normalized by the market capitalization of the bank and CCAR losses are divided by average assets over the 9-quarter stress horizon.

3The V-Lab website is actually agnostic on the appropriate level, and allows a user to choose any market leverage requirement; the default, however, is 8 percent, and this is the level from which the Bloomberg View editorial derived its numbers.