Full text of Working Papers (Federal Reserve Bank of Richmond) : On the Measurement of Large Financial Firm Resolvability, Working Paper 18-06R : Working Paper

View original document

The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.

On the Measurement of Large Financial
Firm Resolvability

WP 18-06R

Arantxa Jarque
Federal Reserve Bank of Richmond
John R. Walter
Federal Reserve Bank of Richmond
Jackson Evert
Federal Reserve Bank of Richmond

On the Measurement of Large Financial Firm Resolvability∗
Arantxa Jarque

John Walter
Richmond Fed

Jackson Evert

March 2018
Working Paper No. 18-06R

Abstract
We say that a large financial institution is “resolvable” if policymakers would allow it to
go through unassisted bankruptcy in the event of failure. The choice between bankruptcy or
bailout trades oﬀ the higher loss imposed on the economy in a potentially disruptive bankruptcy
resolution against the incentive for excessive risk—taking created by an assisted resolution or a
bailout. The resolution plans (“living wills”) of large financial institutions contain information
needed to evaluate this trade—oﬀ. In this paper, we propose a tool to complement the living
will review process: an impact score that compares expected losses in the economy stemming
from a resolution in bankruptcy with those expected under an assisted resolution or a bailout,
based solely on objective characteristics of a bank holding company. We provide a framework
that allows us to discuss the data needed and the concepts that underlie the construction of
such a score. Importantly, the same firm characteristics may be ascribed diﬀerent impacts under
diﬀerent resolution methods or crisis scenarios, and these impacts can depend on policymakers’
assessments. We say that a firm’s structure is acceptable if its impact score under bankruptcy
is lower than that of any other resolution method. We study the current score used to designate
firms as GSIBs and propose a modified version that we view as a starting point for an impact
score.
Keywords: Resolution, Bankruptcy, Financial Regulation, Safety Net.
JEL codes: G01, G21, G28, G33.

Introduction

Financial troubles experienced by large, critical financial institutions were major contributors to the
2007-08 financial crisis and garnered significant public attention.1 The U.S. decision to intervene
in the financial system to prevent the collapse of troubled firms is commonly attributed to the fear
∗

We would like to thank Keith Goodwin, Justin Kirschner, Hoossam Malek, Tim Pudner and John Weinberg for
valuable input during the preparation of this paper and Huberto Ennis for early encouragement with the project and
for many helpful conversations. The views expressed here do not necessarily reflect those of the Federal Reserve Bank
of Richmond or the Federal Reserve System. Correspondence: Arantxa.Jarque@rich.frb.org.
1
White and Yorulmazer (2014) documents the largest firms to receive financial aid from public sources during the
2007-08 crisis, and the types of interventions used to help them.

that the failure of such firms could be very damaging to financial stability and the real economy.
A typical concern with these types of interventions is that they provide incentives for firms to
engage in excessive risk—taking, given that shareholders and creditors expect public support in the
event of financial distress (i.e., the implicit public support expected by creditors, or “safety net,”
creates a moral hazard problem). In the aftermath of the 2007-08 crisis, reforms have emphasized
macro—prudential regulations in an attempt to ensure that critical firms behave in a way that
promotes financial stability, and that the system is suﬃciently protected against the collapse of
these institutions.
These new regulations, following mandates in the Dodd—Frank Act (DFA), require that large,
critical financial firms be subject to enhanced prudential regulation and require that those with
assets larger than $50 billion submit a resolution plan, or “living will.”2 A living will (LW) is
meant to give details on the “structure” of the firm, as well as describe in detail how, if the firm
fails, it would be resolved through bankruptcy, without government support (its “plan” for orderly
wind—down). The idea is for policymakers to be able to anticipate the consequences of unassisted
failures and to make sure such consequences are acceptable. Regulators may conclude that the LW
has deficiencies either because the “plan” fails to outline a feasible means of unwinding the firm
without government aid, or because the firm is structured in a way that makes it hard to resolve
through bankruptcy in a pre—specified series of crises scenarios, even under the best plan. If the
LW has “deficiencies” it is deemed “non—credible” and the firm is required to revise it and resubmit
it. If regulators are not satisfied with subsequent revisions of the LW the firm can potentially face
pressure to restructure. Over the last four years, starting in 2012, the Board of Governors (Board)
and the FDIC have been jointly reviewing the LW submissions for the largest of these firms. In
2015, several of the firms’ LWs were found to contain deficiencies so that the LWs were considered
non—credible, either because of problems with their plans or because of inadequate structures, and
regulators engaged with the firms in an iterative process to remedy these deficiencies.3
LWs should help deter moral hazard if, when no deficiencies are found in a LW, market participants revise down their priors on whether policymakers will be tempted to bail out the firm
in the event of financial trouble. If expectations of a bailout are reduced, creditors will monitor
risk—taking and charge a premium for it, eﬀectively controlling excessive risk—taking.
In practice, however, LWs are lengthy and hard to digest, since they include significant amounts
of qualitative information and nonstandardized content across firms. In addition, only a small
portion of the information contained in each firm’s LW is made public. These factors have created
opacity and uncertainty about the review process; indeed there have been calls to make public
more information regarding either the evaluation framework or the results of the LW process.4
2

For information on the living will process, see https://www.federalreserve.gov/supervisionreg/resolutionplans.htm.
3
The Federal Reserve and the FDIC do no typically refer to an LW that is acceptable — requiring no significant
modifications — as "credible." Instead such a LW is referred to, in press releases and letters to the firm, as containing
no deficiencies that would "trigger a resubmission" of the LW or even more ""stringent reqirements" (Board of
Governors and Federal Deposit Insurance Corporation, 2017). For simplicity, in our paper, we will call a plan that
meets supervisory standards "acceptable."
4
For example, see Hoenig (2015) or GAO (2016).

In the absence of this increased transparency, however, it is diﬃcult to know how the market is
interpreting LW evaluation outcomes, and hence the role that resolution plans may be able to play
in ending the too—big—to—fail problem.56
Our objective in this article is to clarify the factors that influence a bailout decision and, in
particular, the way in which the structure of a firm may influence it. Given the opacity of the LW
review process, we propose a new tool to complement the process: an impact score that provides a
quantitative evaluation of how fit a firm’s structure is for unassisted resolution. We consider a score,
based on publicly—available data on firm characteristics, which compares the impact on the economy
if the firm files for bankruptcy with the impact of a resolution involving various levels of public
support. Our framework highlights that the decision to aid a failing firm trades oﬀ the loss imposed
on the economy by a potentially ineﬃcient bankruptcy procedure against the potential incentives
for excessive risk—taking from oﬀering an assisted resolution or a bailout. Taking this moral hazard
cost into account, a firm will be considered resolvable if the estimated impact stemming from its
failure is smaller or equal under bankruptcy than under any other method that involves government
support. Hence, in our framework, only if bankruptcy yields the lowest impact score will a firm’s
structure be deemed acceptable. We focus our analysis on how a firm’s structure translates into
cost to society in the event of its failure, and we do not consider the parts of a LW that relate to
the wind—down plan, since these plans are confidential and also lend themselves less naturally to a
quantitative assessment.
In practice, constructing this impact score requires two main ingredients: a list of firm characteristics that can be measured and compared across firms, and weights for each of these characteristics
that reflect their relative importance to the overall impact on the economy of the failure of the firm.
Though in this paper we stop short of providing a score that is usable for policy, we suggest a list
of firm characteristics that could be included in such a score, and we collect some real firm data on
these characteristics. Because some of the content in the LWs is confidential, we focus on a score
that uses only publicly available information. However, the framework we provide could also be
employed by supervisors using non public information to construct a confidential impact score.
There are two ways to think about our impact score. One is normative, in which empirical
research on the ties between firm characteristics and social costs from failure informs our weights
and helps determine the firm structure that could maximize welfare to society. Unfortunately,
research on this topic is still limited and far from being able to provide a quantitative basis for
our weights.7 The second way to think about the score is in a positive way, describing preferences
or beliefs of policymakers about this mapping of firm characteristics into externalities. In this
latter case, the score is analogous to a Taylor rule for monetary policy (which maps inflation and
macroeconomic indicators into a desired interest rate) or to capital requirement rules (which map
capital ratios to the odds of failure). If we had data on all the characteristics included in our impact
score for the large firms that were operating during financial stress times such as the 2007 crisis,
5

As a response to the GAO report, the Board of Governors and the FDIC published a description of the living
will review process and a summary of commonly found deficiencies in the plans (see Board of Governors and FDIC,
2016.)
6
See Jarque and Athreya (2015) for a discussion of the role of living wills in resolution.
7
For a review of both models and evidence on firesales eﬀects, for example, see Schleifer and Vishny (2011).

we could estimate the preferences of the policymakers who managed their failures. In other words,
we could learn from past policymakers’ actions (to provide assistance or allow failures) what their
beliefs about the contribution of each firms’ characteristics to impact of failure must have been at
the time when they decided on their resolution. In the absence of this data, we instead provide
hypothetical weights to illustrate the sensitivity of resolvability determinations to variation in the
beliefs of policymakers.
Within the positive approach, similar to Taylor rules or capital requirements, making public
the weights that determine the resolvability score may provide some credibility. Even with the
understanding that some flexibility will be needed to tackle the particular circumstances at the
time of a firm’s failure, publishing the details of the score, ex—ante, amounts to announcing a
“rule”; one could argue that announcing the rule makes it costly to deviate from it and may
provide some commitment to not bail out, helping curb moral hazard problems. More specifically,
our quantitative evaluation of the acceptability of a firm’s structure, which can be understood
as a proxy for the LWs, but using only publicly available information, could be valuable in two
main respects. First, it would allow regulators to more easily compare how resolvable diﬀerent
firms are or track progress toward resolvability for a given firm over time. Second, it may make
communication to the market, about firm resolvability, more transparent and credible.
We are going to argue in this paper that a good starting point for an impact score could be the
quantitative score used for the designation of Global Systemically Important Banks (GSIBs).8 The
GSIB score collects information on firm characteristics that are thought to lead to economy—wide
disruptions in the event of GSIB failure. In this article, we evaluate to what extent this GSIB score
captures resolvability, and how it could be modified to provide a better evaluation of failure costs.
To illustrate how an impact score would work, we use real firm data, in combination with several
sets of weights representing views of diﬀerent hypothetical policymakers, to compute examples. We
use the diﬀerent sets of weights as mere examples –which are, for now, without empirical foundation
– that allow us to illustrate the sensitivity of the designation of firms as resolvable given diﬀerent
policymakers’ preferences. Our exercise also highlights how, to be able to use an impact score
in practice, we would need empirical data from a number of financial crises to determine the
links between firm characteristics and impact from failure. These links would provide empirical
foundation for the weights used in the impact score so that it could be used as a normative tool
(evaluating the merits of diﬀerent bailout policies). On the other hand, gathering detailed firm
data from past crises would also allow us to use the score as a positive tool if the weights were set
to replicate actual bailout decisions for troubled firms in the past.
In the next section, we lay out our framework and introduce our impact score. In Section 3 we
evaluate whether the existing GSIB score is a good starting point for this impact calculation. We
discuss how each characteristic in the GSIB score relates to expected losses imposed on society by
a firm’s failure. In Section 4 we discuss other characteristics that would be useful for our purpose
but are missing in the current GSIB score. Last, in Section 5 we discuss reasonable restrictions on
the weights and provide our computed examples of the score. Potential extensions of our score are
8

The capital surcharge for GSIBs is a recent regulatory requirement for large, systemic institutions. This capital
surcharge is, in the U.S., part of enhanced prudential standards mandated by the Dodd-Frank Act.

suggested in Section 6. We conclude in Section 7. Details on the construction of our examples are
relegated to the Appendix.

A quantitative approach

Discussions of how to deal with the too—big—to—fail problem suggest that a large financial institution
is “resolvable” if policymakers are willing to let it file for unassisted bankruptcy in the event of
failure. This decision trades oﬀ the ability of bailouts to lower the loss imposed on the economy
from a potentially disruptive bankruptcy procedure against the potential incentives for excessive
risk—taking produced by an assisted resolution or bailout. The losses imposed on the economy
(the “impact” of the failure) may include disruptions in the provision of key services (payments,
asset custody, lending relationships, brokerage, counterparty provision for derivatives, hedging),
as well as contagion to other financial institutions, either because they are direct creditors of the
failing firm, or because the value of their assets is aﬀected by the failure. These costs are typically
associated with “externalities” because the distress in the financial system may aﬀect not only
the parties contracting directly with the failing firm but also the functioning of the economy (for
example, increasing the cost of funds for investment).
It is important to realize that a firm will be considered resolvable when the estimated costs
stemming from its failure are smaller or equal under bankruptcy than under any other method
that involves government support, i.e., a relative comparison of costs, and not when these costs
are necessarily “low” in absolute terms. In this section, we expand on this idea by formalizing
the decision of policymakers in the context of the LW review process. A LW will pass the review
process only if the plan and the structure it outlines imply that, in the event of failure, resolving the
firm through bankruptcy will be the preferred option (lowest estimated impact on the economy).
If financial markets understand this process, for a firm whose LW passes the review process, expectations of a bailout will be reduced and creditors will monitor risk—taking and charge a premium
for it, eﬀectively controlling excessive risk—taking. Our proposal is to complement the LW review
process by constructing an impact score that evaluates whether a firm’s structure is acceptable for
bankruptcy. We start by providing some definitions.
Definition: A firm is resolvable if policymakers are comfortable with it filing for bankruptcy in
the event of failure, i.e., the consequences for the economy of its liquidation without government
support are deemed preferable to the potential incentives for excessive risk—taking that a bailout may
provide to the financial system.
The DFA gave regulators the task of crafting measures that would make large financial firms
resolvable. The LW review process is part of the regulatory implementation of this objective. The
FDIC and the Board have worked together to provide a template for LWs that establishes the
information needed on the firm’s structure (financial structure, legal structure, and other relevant
characteristics) and its wind—down plan for unassisted resolution in a pre—established set of crisis
scenarios. Within this context, regulators evaluate the LWs annually. We can think about the
evaluation of these LWs as consisting of two distinct, but related, evaluations: of the structure of
the firm, and of the plan.
5

Definition: A firm’s structure is acceptable if, under the best plan for resolution within a bankruptcy procedure, this structure does not imply important diﬃculties that may make a bailout or
other government involvement preferable to bankruptcy.
For example, a LW may not pass the review process if the firm is considered too large or too
complex to be dealt with in bankruptcy with the proposed strategy. It is plausible, for example,
that policymakers are likely to be comfortable allowing a relatively small firm to file for bankruptcy,
but they may feel compelled to bail out a very large firm, and may decide instead to intervene in
the failure of a middle—sized firm by separating its assets across a “good bank” and a “bad bank,”
eﬀectively bailing out only some of its debt holders. Another example would be a firm having an
inadequate legal structure.9
Definition: A plan is adequate if it oﬀers a feasible means of liquidating a firm of a given structure,
in a given crisis scenario, within a bankruptcy procedure without resorting to public funds.
Supervisors analyze the proposed plans to ensure that they comply with requirements. For
example, some plans have been deemed deficient because their resolution strategies rely on cooperation among international regulators to permit recapitalization (i.e., no “ring—fencing” of assets),
though regulators think such cooperation should not be counted on (see Goldman Sachs’ feedback
letter, 2015). Another deficiency identified by supervisors is the lack of appropriate models to
evaluate recapitalization needs. Firms are required to estimate the resources needed to recapitalize
systemic subsidiaries across diﬀerent crises scenarios. They also should set triggers for bankruptcy
filing that ensure that there is a large enough capital buﬀer remaining in the firm to help with the
recapitalization of such subsidiaries.
Using both the concept of an adequate plan and an acceptable structure of the firm, we define
the criteria for a LW to pass the review process as follows:
Definition: A LW passes the review process if the structure of the firm that it describes is acceptable and the plan it proposes for each scenario is adequate.

2.1

An impact score

Plans are confidentially submitted to supervisors as part of the LWs, so market participants (outside
of employees of the submitting firm) must rely on the determination of supervisors to learn about
their adequacy. In contrast, in regard to the GSIB’s structure, while some details provided in the
LWs are also confidential, market participants have much better information thanks to regulatory
documents that are made publicly available. We propose using this public information to construct
a narrow evaluation of the firm’s structure that can be a proxy for the nonpublic assessment of
supervisors about the adequacy, for resolution, of the firm’s structure. We label this evaluation, or
calculation, an impact score.
Definition: An impact score is a mapping from measurable characteristics of a firm into a quantification of the disruption in the economy from the failure of such firm given a certain scenario
9

The December 2016 feedback letter to Wells Fargo cited “deficiencies related to Legal Entity Rationalization and
Shared Services” as one of the reasons for their LW not passing the review process.

(i.e., circumstances surrounding the firm’s failure) and a method of resolution (bankruptcy versus
other strategies that may involve various levels of government support).
An impact score will use information on a subset of firm characteristics, such as size, volume
of payments, or the structure of its short—term funding.10 We will denote a generic characteristic
as  , where the subscript  denotes a given firm in our sample,  ∈ {1   }, and the subscript
 denotes a given characteristic on our score:  ∈ {1  }. A firm  will be represented by a
vector X = {1    } of firm characteristics. The score collects the raw value of these firm
characteristics (typically reported in $) across a relevant group of global financial institutions and
constructs a normalized version, denoted  , by dividing each raw amount by the sum of that
characteristic’s entry across all firms in the group. Hence,  is a normalized value that can be
interpreted as a relative importance measure, or a market share in that characteristic. We can
think of a firm as a collection of normalized values for each characteristic. Then, the impact score
of a firm is a weighted sum of these normalized characteristics, where  represents the weight for
characteristic  :
X
X

P 
 =
  
0 6= 0 


2.2



Using the impact score

In order to use the score for our purpose of determining the acceptability of a firm’s structure,
we need to compute this impact for alternative crisis scenarios and diﬀerent resolution methods,
each of which may aﬀect the potential consequences that the firm’s resolution have on the broader
economy.
Claim: An impact score may vary under alternative crisis scenarios.
A given bank structure may have very diﬀerent systemic impacts in the event of failure depending on whether the failure is due to an idiosyncratic bank shock or to an aggregate financial crisis.
For example, the failure of an important provider of credit may translate into higher disruptions
if other firms are also in distress, and these other firms do not have enough capital or liquidity to
increase their lending to take on the initial firm’s borrowers.
Claim: An impact score needs to be contingent on a resolution method.
It is the comparison of the impact score under alternative resolution methods that allows us to
establish that a given structure is acceptable, because the eﬀect that the failure of a firm will have
on the economy will depend on how its failure is managed and its characteristics.
To illustrate this idea, in this article we consider a stylized set of three resolution methods: the
firm files for bankruptcy, is resolved using the Orderly Liquidation Authority (OLA) process, or
receives a bailout. We use  to denote a generic resolution method and use the following notations to
denote the three alternatives:  (bankruptcy court), ,  (bailout). For simplicity, we do not
distinguish between filing for bankruptcy (Chapter 7) or reorganization (Chapter 11). Also, when
10

For a recent paper on the eﬀect of short—term financing on the economy through the possibility of runs, see Covitz
et al. (2013).

we refer to a bailout, we consider both situations in which there are explicit capital injections with
public funds that allow the firm to continue to operate (such as the financing of the reorganization
of GM in 2009) and interventions that may not result in the survival of the failing firm, such as the
assisted purchase of Bear Stearns by JPMorgan Chase in March 2008, since these can also involve
a government subsidy.
For our analysis, the main relevant dimensions in which these methods will diﬀer are the expected duration of the wind down, the availability of funds to finance a bridge company, the possibility of delaying the exemptions to the automatic stay for qualified financial contracts (QFCs),
the possibilities for international coordination among the authorities involved in the wind—down, or
the adherence to bankruptcy’s established debt priorities. More generally, our proposed framework
can accommodate any number of relevant features across resolution methods that we may not be
considering here. Hence, our methodology can be used to evaluate any bankruptcy reform proposals or any other new institutionalized resolution method that may replace OLA. For example,
some bankruptcy reform proposals introduce the idea of specialized judges or the possibility of
government subsidized borrowing (Jackson, 2014).
In terms of scenarios, we consider in our analysis two types of shocks: idiosyncratic and aggregate. The framework, however, can accommodate any number of crisis scenarios. For example,
some specific aggregate shocks we could consider include a high domestic unemployment rate, a
bubble bursting in the U.S. housing market, a sustained recession in China, or any other internationally driven crisis. Specific idiosyncratic shocks may include damage to computer systems, rogue
traders, or subsidiaries of a BHC that specialize in diﬀerent business lines.
Allowing the weights ( ) of each characteristic  in the score to diﬀer according to the state
of the economy,  and the resolution method,  , results in our impact score, which is indexed by
the scenario  and the resolution method  , for a firm  with characteristics summarized in a vector
X :
X
  +  
 (X |  ) =


where  represents the moral hazard cost of method  in a shock of type . In our framework,
this impact score captures the relevant information needed by policymakers in order to decide on
the appropriate resolution method to employ once a particular scenario has materialized:
Assumption: The preferred resolution method for a firm  in each scenario  denoted  ∗ () will
be the one that has the lowest impact on the economy:  ∗ () is such that ( |  ∗ ) ≤ ( |  )
for all  other than  ∗ .
Note that the impact score under a bailout ( = ) deserves special consideration. Given
that firms are not liquidated if they receive a bailout, the costs of this resolution method are not
necessarily best captured with contingent weights given to specific characteristics but instead with
a lump—sum cost ( ) representing future ineﬃciencies in the economy due to increased moral
hazard, or a combination of the two. In our analysis, we will impose that all the weights 
be zero for all  0 s and shocks, though this could be generalized to have the total cost of a bailout
depend as well on the firm structure. For OLA, we will assume a combination of costs, with

   capturing the fact that OLA implies a liquidation and will always impose higher
costs on shareholders than a bailout, where the firm survives.
In summary, under the assumption that the impact score captures all the relevant implications
for the economy of a firm’s liquidation, and that resolvability is defined as a requirement that needs
to be satisfied in each scenario described in the LW review process, we re—state, here, our definition
of an acceptable structure and express it in terms of our score.
Definition: A firm’s structure is acceptable if, under the best plan for resolution within a bankruptcy procedure, the preferred resolution method for a firm with this structure is bankruptcy, i.e.,
if the impact of its failure when resolved using bankruptcy is less than or equal to the (moral hazard)
cost of bailing it out or of resolving it using some kind of government support, in each of a list of
given scenarios: ( | ) ≤ ( |  ) for all  diﬀerent than , and for all .
The mapping between this definition and the criteria for a LW to pass the review process is not
direct, because, as we noted earlier, our score does not account for the adequacy of the “plan.”11

2.3

Limitations of the score

There are three important limitations to any attempt to use our impact score as a means of
evaluating the structure of a firm. First, the score does not provide a framework to determine
the optimal choices of firm structure. Second, it can never capture relevant information about the
structure of the firm that is not measurable or verifiable by regulators. And, third, the score cannot
capture all possible shock scenarios or include an exhaustive list of the relevant characteristics
needed to determine the impact of a firm’s failure. This last limitation is particularly important,
because it leaves the door open to moral hazard and time inconsistencies (i.e., it does not solve the
too—big—to—fail problem even when the score indicates that all firms are resolvable). We discuss
each of these limitations in the next paragraphs.
First, we discuss the issue of the optimal choice for the firm characteristics, X. Given our
framework, one might imagine that a simple means of ensuring that firms have an acceptable
structure is to directly constrain their choices of X – for example, require firms to restrict their
short—term debt to less than a set percentage of all liabilities, or mandating a cap on firm asset
size. However, policymakers face important constraints on their ability to do so. Policymakers
must account for the costs of regulations or constraints, because such actions are likely to introduce
distortions, increasing banks’ costs of providing socially valuable financial services. GSIB failures
are expected to occur very infrequently, so that when taking actions to influence firm choices,
regulators will be forced to trade oﬀ long—lived distortionary costs that a conservative limitation on
11

Under these definitions, having a structure that is acceptable according to the impact score is a necessary (but
not suﬃcient) condition for a LW to pass the examination of supervisors. This is because we make the convenient
assumption that the structure is evaluated assuming the best plan possible is in place. One could think, perhaps
more realistically, that the determination of whether the structure is acceptable depends on the plan. In practice,
when calculating our score, this would mean that a characteristic that could, in principle, bear a high impact weight,
would be compensated by a LW that deals in its plan with the main diﬃculties that the characteristic would imply
in bankruptcy. However, this would not make it possible for us to evaluate firm characteristics independently of the
confidential information contained in the LWs.

firm choices of Xs would impose in normal times (noncrisis scenarios) against the lower expected
impact in the event of failure.
Expected impact of a firm’s failure can be calculated by assigning probabilities to the diﬀerent
crisis scenarios being considered:
X
 [ (X )] =
Pr ()  (X |  ∗ ()) 


The calculation of these expected distortions in order to evaluate the trade—oﬀ faced by policymakers
is outside the scope of this paper. For our analysis, hence, we take the firm’s choices of X as given.
In other words, our impact score is simply meant to be a tool that, for given X, determines whether
the firm’s structure is acceptable.
It is worth noting that, as the LW review process is described in the DFA, policymakers are
required to ensure that no LWs have deficiencies. As we have mentioned previously, for those firms
that don’t meet this requirement, revisions to the documents are requested and, if that does not
solve deficiencies identified in the review, even changes to firm structure (X) may be required.
In light of our model, this Dodd—Frank requirement implies that lawmakers evaluated the trade—
oﬀ between distortions and benefits that we just described and decided that all GSIBs should be
required to have an acceptable structure. In principle, however, it is possible that such evaluation
would determine that the costs of resolvability in terms of eﬃciency are too high for some firms
and, hence, they should be allowed to remain nonresolvable (for example, on the basis of their
economies of scale or scope). This determination is, as we have pointed out, outside of our model.
A second limitation when using a tool such as our impact score — or, more broadly, when
supervisors are reviewing a firm’s LW – is the problem of asymmetric information: policymakers do
not have all the relevant information needed to evaluate a firm’s business strategies. In other words,
many considerations go into choosing a firm’s structure (X ) and business strategies. Supervisors
must rely on limited firm disclosures (for example, from financial statements and discussions with
firms), which are likely to be insuﬃcient to fully evaluate a firm’s choices. This complicates further
the calculation of the relevant trade—oﬀ of costs and benefits of X’s choices.
A third limitation would arise even if the first two limitations were surmountable: the inability
of any practical score to consider all possible shock scenarios and to gather information from the
firms on all the characteristics that could potentially be relevant for determining the impact of a
failure. Unfortunately, game theory tells us that any room for a bailout (in an unforeseen scenario,
or because of revelations of a firm characteristic that makes its failure unexpectedly costly) will
translate into a window to moral hazard: if the firm knows there are some states of the world
in which it will receive a bailout, it will take more risk than society would want it to. If the
policymaker understands this, he may want to commit to not bail out. That is, when the firm fails,
he decides a bailout is best, despite the moral hazard cost, because of the unforeseen impact of
failure. However, because he anticipates that the firm is choosing riskier projects because it knows
it will be bailed out in this scenario, he wishes he could commit not to bail out, i.e., he suﬀers
from a time—inconsistency problem. In this case, a firm that was deemed as resolvable or having
an adequate structure may end up being “too messy to fail,” in a manner that the impact score or
the LW review process was unable to anticipate.
10

2.4

A proposed impact score based on the GSIB score

In this section, we evaluate whether the current score used to designate firms as GSIBs is a good
candidate for an impact score. We argue that the GSIB score, as measured, captures several metrics
of BHC structure and activity that are considered informative about the potential size of losses the
economy may suﬀer in the event of a crisis. As a result, the GSIB score should be a good starting
point to measure losses imposed in the economy in the event of failure.
The Basel Committee on Banking Supervisions (BCBS) has developed a methodology for designating global banks as systemically important. The BCBS recommends a capital surcharge to
make these banks more resilient. The U.S. implementation of this recommendation can be found in
“Regulatory Capital Rules: Implementation of Risk—Based Capital Surcharges for Global Systemically Important Bank Holding Companies; Final Rule.”12 As part of this U.S. regulation, Form FR
Y-15 collects from BHCs of $50 billion or more the information needed to construct a GSIB score.
This GSIB score is meant to capture the “systemic footprint” of an institution, i.e., the “loss
given default” that the institution would impose on society.13 The GSIB rule seems to equate
systemic footprint, or loss given default, to “externalities.” In policy circles, externalities from the
failure of a large financial institution are mostly believed to stem from contagion eﬀects or firesale
eﬀects.14 Contagion eﬀects occur when the failure of one institution has an impact on the health
of its counterparties. Firesale eﬀects occur when the failing institution, in the process of trying to
meet repayment demands, sells a large quantity of assets of one particular class, lowering the price
of that class of assets in the market — and all other firms holding this asset class, as a consequence,
suﬀer a diminished value of their assets. Higher externalities imply higher loss given default.
In order to calibrate capital requirements for GSIBs, an extra buﬀer of capital is required only
for global banks that have been designated as systemic, regulators estimate a probability of loss and
the implied expectation of loss given default for given firm characteristics and a level of capital:15
[Loss] =  (default | capital requirements) ∗ Loss given Default
where higher capital requirements translate into a lower probability of default but are costly for
the firm and hence the economy. Regulators calibrate the GSIB capital surcharge to keep expected
loss to a common level across systemic and nonsystemic institutions:
12

A link to the Federal Register notice containing the rule and associated documents can be found at:
https://www.federalreserve.gov/newsevents/press/bcreg/20150720a.htm. The surcharges imposed by the rule were
phased in beginning January 1, 2016, and become fully eﬀective January 1, 2019.
13
See page 1 of “The G-SIB Assessment Methodology — Score Calculation,” Basel Committee on Bank Supervision,
November 2014.
14
For discussions of firesale eﬀects see Diamond and Rajan (2011) and Allen and Gale (1994); for contagion eﬀects
see Allen and Gale (2000).
15
The rule determines the GSIB surcharge according to the “bucket” into which a firm’s score falls (i.e., a step
function that relates capital requirements to probability of default, as estimated using historical data on bank failures).
GSIB surcharge designations are available for 2013-15, with the same eight U.S. BHCs having been designated as
GSIBs in each of those years. JPMorgan Chase and Citigroup have repeatedly been the highest scoring firms, each
receiving a 3.5% surcharge in 2015. Bank of America, Goldman Sachs, Morgan Stanley, and Wells Fargo have received
surcharges between 1% and 2% in each year. Of the eight banks, Bank of New York Mellon and State Street have
consistently scored the lowest, receiving 1% surcharges each year.

 (default |Basel III capital requirements + GSIB surcharge) ∗ Loss given Default for GSIB

=  (default | Basel III capital requirements) ∗ Loss given Default of NON—GSIB

This implicitly assumes that without the extra capital buﬀer that the GSIB regulation imposes,
two diﬀerent institutions, one qualifying as a GSIB and the other not, may receive the same capital
requirements under Basel III even though the failure of the GSIB may have a larger systemic eﬀect
(i.e., loss given default) than the other. That is, there are firm characteristics that determine loss
given default that are not captured by the standard capital requirement calculations. The form
FR Y—15 collects information on these characteristics. The GSIB score aggregates the information
about systemic characteristics that is collected in form FR Y—15 into a unique number that can
be compared across firms internationally.16 Firms having a score above a certain threshold are
designated as GSIBs. In the 2016 exercise, there were 30 GSIBs; eight were U.S. based.17

The GSIB score

In this section we review in detail the firm characteristics included in the GSIB score to understand
the impact measured by each. The GSIB score tracks 12 indicators across six conceptual categories
shown in Table 1.18 In order to have a normalized value of the indicator, it divides each firm’s
individual entry by the sum of that entry across the largest 75 global banks according to a measure
of size.19 The final score, under the GSIB rule, is the weighted sum of each normalized indicator
(most of the indicators are, themselves, made up of several “items,” which correspond to what we
have been referring to as “firm characteristics”).
The U.S. implementation of the GSIB score calculates both the internationally consistent score
(Method 1) presented in Table 1, as well as an alternative (Method 2) that replaces the three Substitutability indicators with the Short—Term Wholesale Funding indicator. Note that the weights
imply each schedule is given equal importance in the final score. Method 2 has slightly modified
weights to better capture improvements common to all banks but is harder to interpret as relative
contributions to the score. The GSIB surcharge applied in the U.S. is the maximum of the surcharges that would be applied by either of these methods, which, in practice, is frequently Method
2.
16

The FR Y-15 is the basis of information collected from U.S. banks. Other countries collect information with their
own similar forms.
17
See FSB, “2016 list of global systemically important banks (G-SIBs),” November 2016, for a list of the GSIBdesignated firms in the 2016 exercise. On top of the extra capital buﬀer (the GSIB surcharge), GSIBs need to
comply with added capital and long—term debt requirements as mandated in Total Loss Absorbing Capacity (TLAC)
regulation as well as extra resolvability and internal control requirements. The TLAC regulation is available in the
Federal Register: Federal Reserve System 12 CFR 252 [Regulations YY; Docket No. R-1523] RIN 7100-AE37 (2017)
18
See Board of Governors of the Federal Reserve System, “Report Forms: FR Y-15, Banking Organization Systemic
Risk Report.” Available at: https://www.federalreserve.gov/apps/reportforms/reportdetail.aspx?sOoYJ+5BzDaRHakir9P9vg==
, accessed August 8, 2017.
19
The value of this denominator, for each characteristic, is published annually by the Bank for International
Settlements (see http://www.bis.org/bcbs/gsib/index.htm ).

Category

Indicator

Weights

Size

Total Exposures

20%

Interconnectedness

Intrafinancial System Assets
Intrafinancial System Liabilities
Securities Outstanding

6.67%
6.67%
6.67%

Substitutability

Payments Activity
Assets Under Custody
Underwritten Transactions in Debt and Equity Markets

6.67%
6.67%
6.67%

Complexity

Notional Amount of Over—the—Counter Derivative Contracts
Trading and Available—for—Sale Securities
Level 3 Assets

6.67%
6.67%
6.67%

Cross—Jurisdictional Activity

Cross—Jurisdictional Claims
Cross—Jurisdictional Liabilities

10%
10%

Table 1: Firm characteristics included in the GSIB score, and corresponding weights
Schedule A: Size
Total Exposures

Derivative exposures
Securities financing transactions exposures
Other on—balance—sheet exposures
Other oﬀ—balance—sheet exposures

Table 2: Components of Schedule A of the GSIB score.
In the next paragraphs, we discuss in some detail the specific items in each category: we study
what are the features of a firm’s structure that they capture, we conjecture why the designers of
the Y-15 form thought that these may be informative about losses stemming from failure, and we
evaluate whether they belong in our impact score.
Schedule A: Size, or “Total Exposures” (TE) This schedule attempts to estimate the volume
of financial activity undertaken by each institution (see Table 2 for details). Even though the title
of this schedule is “Size,” we see this schedule as a way of capturing the “value” that the firm brings
to the economy. To capture that value, one needs a measure of both the direct lending provided
by the firm and the financial services it provides. The former is captured in a straightforward way
from the balance sheet. The latter is harder to value, but regulators have developed measures that
capture some of the services produced by banks by performing some adjustments to balance sheet
items: the “Total Exposures” (TE) measure from capital requirement regulation (Basel III).20
For example, when counting securities financing transactions, the measure adjusts the value of
20

“Total Exposures” is the size measure in the Supplementary Leverage Ratio that applies to Advanced Approaches
organizations. Interestingly, the order of the individual entries in the two forms is diﬀerent — perhaps indicating
that the Y-15 wants to emphasize the importance of derivatives and securities financing transactions for systemic
importance of firms.

JPMORGAN CHASE & CO.
BANK OF AMERICA CORPORATION
CITIGROUP INC.
WELLS FARGO & COMPANY
GOLDMAN SACHS GROUP, INC., THE
MORGAN STANLEY
U.S. BANCORP
HSBC NORTH AMERICA HOLDINGS INC.
PNC FINANCIAL SERVICES GROUP, INC., THE
BANK OF NEW YORK MELLON CORPORATION, THE
CAPITAL ONE FINANCIAL CORPORATION
TD GROUP US HOLDINGS LLC
STATE STREET CORPORATION
BB&T CORPORATION
SUNTRUST BANKS, INC.
AMERICAN EXPRESS COMPANY
CHARLES SCHWAB CORPORATION, THE
FIFTH THIRD BANCORP
ALLY FINANCIAL INC.
CITIZENS FINANCIAL GROUP, INC.
REGIONS FINANCIAL CORPORATION
BMO FINANCIAL CORP.
SANTANDER HOLDINGS USA, INC.
MUFG AMERICAS HOLDINGS CORPORATION
M&T BANK CORPORATION
NORTHERN TRUST CORPORATION
KEYCORP
DISCOVER FINANCIAL SERVICES
BANCWEST CORPORATION
BBVA COMPASS BANCSHARES, INC.
COMERICA INCORPORATED
HUNTINGTON BANCSHARES INCORPORATED
ZIONS BANCORPORATION
DEUTSCHE BANK TRUST CORPORATION

3126.87
2808.78
2377.39
2146.08
1349.77
1107.95
518.5
449.65
422.642
402.213
372.483
296.676
252.733
234.321
232.861
189.415
187.588
170.345
162.112
160.112
146.354
144.569
142.11
138.633
138.459
136.548
125.288
104.891
104.469
101.787
88.1699
78.5274
66.8687
56.6695

500

1,000

1,500

2,000

2,500

3,000

3,500

Total Exposures (Billions of $)

Figure 1: Total Exposure of U.S. BHC that file the Y-15 form (from largest to smallest), for the
year 2015.
repurchase agreements, or “Repos,” but also includes the following items, which are part of the
activity of the firm but are not included in the balance sheet under current accounting standards:
• Derivatives exposure that is not on balance sheet. Only derivatives with a positive fair value
are on the balance sheet as assets — or, if their value is negative, on the balance sheet as
liabilities. However, derivatives that currently have a fair value of zero (for example, an
interest rate swap that was just created) still are valuable to both counterparties, so the TE
measure adds an estimate of their value.
• Other oﬀ—balance—sheet items (credit lines, etc.), measured using credit—equivalent adjustments.
It is reasonable to think that the larger the size (including services) of the firm, the more market
disruption its failure will cause.
We might worry about double counting the eﬀect of size, since disruptions through contagion
are also captured in Schedule B of the score (“Interconnectedness”). However, the focus of Schedule
B is on how much of the activity is within the financial system or through very tradable instruments
(securities). The focus of Schedule A, instead, is on providing an estimate of sheer size, so we think
it is important to look at both measures.
We consider the TE measure to be a valid method of measuring an important feature of a firm’s
structure, and it should be part of our impact score. Figure 1 presents the firms in the 2015 sample
for the Y-15 form, with their values of total exposures, ranked from largest to smallest. When
14

Schedule B: Interconnectedness
Intra—Financial System
Assets

Funds lent to financial institutions (FIs)
Unused Portion of committed lines extended to FIs
Holdings of securities issues by other FIs
Net positive current exposure of securities financing transactions with FIs
OTC derivative contracts with FIs that have a positive fair value

Intra—Financial System
Liabilities

Deposits due to FIs
Borrowings obtained from FIs
Unused portion of committed lines from FIs
Net negative current exposure of securities financing transactions with FIs
OTC derivative contracts with FIs that have a negative fair value

Securities Outstanding

Secured debt securities
Senior unsecured debt securities
Subordinated debt securities
Commercial Paper
Certificates of deposit
Common Equity
Other subordinated Funding

Table 3: Components of Schedule B of the GSIB score (Interconnectedness).
discussing the next items in the score, we will compare the ranking of firms in these subsequent
characteristics to that of size by including the corresponding scatter plots.
Schedule B: Interconnectedness The Interconnectedness schedule is made up of three separate
indicators: intrafinancial system assets, intrafinancial system liabilities, and securities outstanding
(see Table 3 for details). The likely objective behind tracking intrafinancial system assets and
liabilities is to measure how much of a firesale eﬀect the failure of the reporting firm would have in
the system (assets) or how much contagion it would spread (liabilities). As for securities outstanding
(a subset of all liabilities issued by the reporting firm), regulators may be using this measure to
capture the contagion eﬀect that the failure of the institution would have for the rest of the economy:
the value of its securities decline, and the balance sheets of all the holders of these securities are
negatively aﬀected.
It is worth mentioning that, in our sample, the indicator for securities outstanding is highly
correlated with TE, suggesting that the amount of a firm’s outstanding securities is roughly proportional to its TE. Note also that the securities outstanding indicator captures all the borrowing that
the firm is doing in the form of securities, regardless of who is holding them (inside or outside of
the financial system). So it is a look at the liabilities side without the narrow focus on the financial
system alone.
Because securities outstanding is only a subset of liabilities (deposits, fed funds borrowings,
and repo liabilities are not included), the focus of the GSIB score appears to be on tradability and

Interconnectedness, 6 Largest Banks

450
Interconnectedness, $Billions

JPM
400

Citi

350
BoA

300
GS

250
200
MS
150
1000

1500

2000

2500

3000

3500

Bank Size, $Billions
Interconnectedness, All but Top 6

Interconnectedness, $Billions

140

BoNY

120
100

State St.

80
60

20
0

S BCorp

HSBC
Ally

Amex

Cap. One
PNC

DiscoverN. Trust
BMO
C. Schwab BB&T
Santander
Suntrust
DB
M&T
Fifth
MUFG
KeyCorpCitizens
Comerica
Huntigton
BBVA
Bancwest Regions
Zions

100

150

200

250

TD
300

350

400

450

500

550

Bank Size, $Billions

Figure 2: Scatterplot of Interconnectedness and Size.
Schedule C: Substitutability
Payments Activity

Payments made in the last four quarters

Assets Under Custody

Assets held as a custodian on behalf of customers

Underwritten Transactions

Equity underwriting activity
Debt underwriting activity

Table 4: Components of Schedule C of the GSIB score (Substitutability)
therefore firesale concerns related to securities. It is worth noting that certificates of deposit (CDs)
are included in the item, even though small—denomination CDs are typically not traded. However,
according to FDIC information, for the largest eight U.S. banks, 66 percent of their CDs are large
denomination (over $250,000), suggesting that including CDs in securities outstanding is consistent
with the idea of capturing potential firesale eﬀects. Also, these are individual banks’ securities, so
we wonder if there is enough exposure in the economy to each of these individual GSIBs for the
failure of one of them to cause large losses for other firms. Perhaps the Y-15 designers have in mind
a domino eﬀect, in which modest losses prompt further liquidation of other assets, and contagion
occurs.
Overall, we think that the Interconnectedness measure provides a valid way of measuring an
important feature of a firm’s structure, and it should be part of our impact score. As Figure 2
shows, the ranking of firms according to their interconnectedness data in 2015 diﬀers from the
ranking of the firms by Size, suggesting there is extra information to be gained from including
Interconnectedness in the score.
Schedule C: Substitutability Indicators Schedule C gathers information on each GSIB’s
volume of payments activity, its volume of assets under custody, and its volume of underwritten
transactions in debt and equity markets (see Table 4 for details). We agree that these measures of
financial services are a good indicator of the level of impact that the failure of a firm may have on
the financial system and the economy. Some refinements of this information may be valuable but
16

104

Substitutability, 6 Largest Banks

JPM

Substitutability, $Billions

Citi

8
6
4

BoA

2
WF
MS

0
1000

GS
1500

2000

2500

3000

3500

Bank Size, $Billions

Substitutability, All but Top 6

Substitutability, $Billions

BoNY
6

4
State St.
DB

N. Trust

0
50

BMO
Schwab Suntrust
Citizens
M&T
FifthC.
KeyCorp
MUFG
Regions
BB&T
Bancwest
Huntigton
Zions
Comerica
Discover
BBVA
Ally
Amex
Santander
100
150
200
250

TD
300

350

Cap. One
400

PNC HSBC
450

S BCorp
500

550

Bank Size, $Billions

Figure 3: Scatterplot of Substitutability and Size.
may be diﬃcult to implement.
One refinement would be to include information that helps regulators evaluate firms that are
important in these three categories also in terms of the risk they take in their business model.21
For example, a tri—party broker could choose two diﬀerent business models: in one, it would simply
be a broker that performs custodian activities and derives its income mainly from service fees; in
another model, it would exploit its client relationships to also engage in providing intraday credit
to its clients, earning interest on such loans. We expect that loss given default would be much
higher in the second scenario.
Another refinement could be to capture information on the credit services that are associated
with underwriting of debt and equity issuance. It seems relevant to evaluate potential impact by
recognizing whether these “dealers” or “brokers” are also often the default buyers of the equity or
debt that they underwrite. As the importance of credit provision increases, loss given default is
amplified.
We recognize, however, the diﬃculties of measuring risk—taking and these two forms of credit
provision.22 Hence, overall, we think the characteristics captured in this schedule are relevant as
they are measured and should be part of our impact score. Figure 3 presents the scatterplot of
Substitutability and Size in 2015 for the Y-15 filers. It shows clearly that the ranking of firms
according to their substitutability data diﬀers from the ranking of the firms by Size. In particular,
firms like State Street and BoNY, both of which have relatively small TEs but are responsible for a
significant fraction of the provision of custodian services (high payments and assets under custody
scores), are clearly outliers in the scatterplot. This suggests there is extra information to be gained
from including substitutability in the score.
Schedule D: Complexity Indicators There are three components in the complexity score:
OTC derivatives, trading and available for sale (AFS) securities, and level 3 assets (see Table 5 for
21
22

For an argument along these lines, see Duﬃe (2014).
See Begenau, Piazzesi and Schneider (2015) for a study of risk—taking by the main broker-dealers in the U.S.

Schedule D: Complexity
Notional OTC Derivatives

OTC derivative contracts cleared through central counterparty
OTC derivative contracts settled bilaterally

Trading and AFS Securities

Trading securities
Available—for—sale (AFS) securities

Level 3 Assets

Assets valued for accounting purposes using Level 3 measurement inputs

Table 5: Components of Schedule D of the GSIB score (Complexity).
Trading

Available—for—Sale (AFS)

Held to maturity

Selling restrictions

Expected to be sold
within, approximately,
one year

Meant to be held longer
than a year but expected
to besold before they
reach maturity

Meant to be held to
maturity, not sold

Accounting valuation
method
Accounting of
unrealized changes in
value

Fair value

Book value

Recorded in the income
statement as earnings

Do not aﬀect earnings and are
only recorded in a separate
form (shareholders’ value)
until they are realized

Not reported

Table 6: Classification of securities
details). We discuss each of those in turn.
OTC derivatives OTC Derivatives are bilateral contracts (even if cleared through a CCP),
and hence they may be used to hedge trades and risks specific to the firm’s counterparty. They
likely are counted toward loss given default in the GSIB score, because, if the institution fails, its
counterparty in these derivatives may have diﬃculty finding a new counterparty that desires an
equivalent position.
Trading and available—for—sale securities Securities are classified, at the time of purchase,
into one of the following three categories for accounting purposes: trading, available—for—sale (AFS),
and held—to—maturity. Mark—to—market changes in the valuation of trading securities and AFS
securities have diﬀerent consequences for income: changes in the value of trading securities appear
as expenses/income in the income statement, while changes in the value of AFS securities aﬀect
only equity. Changes in the value of held—to—maturity securities do not aﬀect either income or
equity. Table 6 summarizes the diﬀerences.
Note that classification occurs at the time of purchase and can only be changed at the next
reporting period. Moreover, FASB guidance states that only reclassifications from AFS to held—to—
maturity are viewed as normal. Other reclassifications require special reporting and justification
(acceptable reasons for the reclassification of held—to—maturity include, for example, the risk of a
18

credit rating downgrade).
The score adds together trading and AFS securities but subtracts the subset that gets a high
liquidity rating in the Liquidity Coverage Ratio (LCR) regulation (Level 1 and Level 2 assets).23
It seems reasonable that the less liquid assets in these categories are harder to value, or “complex.”
Despite the diﬀerences described in Table 6, the GSIB score treats AFS and trading securities
equally. The score does not penalize (count) securities classified as held—to—maturity (the outstanding amount of which is recorded only as a memoranda item). Two possible explanations come
to mind for the inclusion of trading and AFS securities in the score, both a consequence of the
sensitivity of security prices to market developments.24
First, the price of securities in trading and AFS may be more likely to be sensitive to news
about the financial health of the firm or institution issuing them, or industry developments that
may aﬀect profitability. This sensitivity could be the reason why the firm plans to actively trade
them, hoping to benefit from changes in their value. But this also implies that prices may react
to news about distress of a firm that is holding large quantities of a certain security in its trading
book; if the market concludes that a decrease in the value of that security may be behind the
financial distress of the firm, the price of this security is likely to decrease as a reaction to this
“news,” aﬀecting other holders of that security. Note that there is no need for the firm to sell these
assets for this eﬀect to occur, and hence this is not a firesale eﬀect. It is reasonable to assume that
in some instances the market’s inference about the value of some securities may be incorrect; in
such instances, any decrease in the price of these securities will be caused by the failure, will not
reflect a change in the fundamental value of the securities, and therefore is an impact of the failure.
Second, we conjecture that regulators are concerned with potential loss given default caused
by the price declines triggered by a firm in distress selling oﬀ securities in trading and AFS to get
liquidity. These potential firesale eﬀects would hurt other companies that hold the same type of
securities. The idea behind considering only AFS and trading securities is that in the days leading
up to failure, such securities are more likely to be first in line to be liquidated. According to FASB
recommendations, held—to—maturity securities can also be sold in such a financial distress condition,
but a justification needs to be provided, since sales of held—to—maturity securities are not typically
expected. Thus, there is a positive cost for tapping the held—to—maturity pool of securities, and
hence the firm is likely to sell trading and AFS securities first. We find support for this conjecture
of a firesale concern in the fact that, to produce the score, securities with very broad markets (such
as Treasuries) are subtracted from both AFS and trading securities — given that firm’s sale of large
amounts of such securities are unlikely to move market prices, due to the size of the market for
such securities, and hence less likely to produce firesales when sold.
One may think that assets that the firm classifies as held—to—maturity should perhaps be counted
23
The LCR will establish an enhanced prudential liquidity standard consistent with section 165 of the DFA. Large
and internationally active banking organizations will be required to hold high—quality, liquid assets (HQLA) such as
central bank reserves and government and corporate debt that can be converted easily and quickly into cash in an
amount equal to or greater than its projected cash outflows minus its projected cash inflows during a 30-day stress
period. The ratio of the firm’s liquid assets to its projected net cash outflow is its "liquidity coverage ratio," or LCR.
U.S. firms will be required to be fully compliant with the rule by January 1, 2017.
24
See Basel Committee on Banking Supervision (2011) for support of these views.

in the impact score given that their value could be more opaque (since they are accounted at book
value and traded less often). As it turns out, this is already taken care of by the “Level 3 assets”
item in the score, which we discuss next.
Level 3 assets25 “Level 3” is an accounting classification for assets that are deemed complex
to evaluate (because there is no clear market price or a standard valuation model). The score
likely penalizes them because complexity implies less value in bankruptcy, or in a rushed sale
when attempting to avoid bankruptcy, and hence implies more losses to debt holders. Because the
discount that these assets would suﬀer in a rushed sale is due to complexity, and not due to a large
quantity of them being sold in the market, we do not consider this a firesale eﬀect.
The FASB rule specifying which types of assets are classified as Level 3 (FAS 157) has a somewhat tenuous classification criterion: those measured “using significant unobservable inputs.”26 For
example, if a market price is available, the asset is considered Level 1; if an option can be priced
using the Black—Scholes model, that option is considered a Level 2 item, because a model is used
but data on volatility can be gathered from past behavior of prices. In contrast, the valuation
of the stock of a private company requires making assumptions about the future profitability of
that company — needing what accountants call “assumptions about market assumptions”: what
the company thinks the market will use as the valuation for an asset, etc., so is classified as a Level
3 item.
Note that the score attempts to gather information that is comparable (or harmonized) internationally; hence, rather than choosing a list of specific asset classes that regulators view as complex,
they pick an international classification of assets (Level 3) that is typically applied to complex
assets.
Overall, we find that schedule D accounts for important firm characteristics and that all of its
items should be included in our impact score. As Figure 4 shows, the ranking of firms according
to their complexity data in 2015 diﬀers from the ranking of the firms by Size, suggesting there is
extra information to be gained from including complexity in the score.
Schedule E: Cross—Jurisdictional Activity Indicators As Table 7 indicates, this schedule
essentially collects just two figures: 1) The sum of all amounts owed — including positive values of
derivatives — to the filing bank by foreign persons, public institutions, financial firms, and nonfinancial firms; and 2) the sum of all amounts owed by the filing bank to foreigners (including amounts
owed by the filer’s foreign oﬃces – amounts owed in both local and nonlocal currencies of these
oﬃces).27 The idea here is that assets and liabilities that cross national borders are more complex
to collect if the borrower becomes troubled, because of diﬀerences in bankruptcy treatment across
countries and the chance that assets might be ring—fenced. We think this is a valid concern, and
25

Note that Level 1,2,3 in FASB is diﬀerent than Level 1 and 2 liquid assets in the previous point (that classification
is instead derived from the LCR calculation in 12 CFR 249.20(a)).
26
See Financial Accounting Standards Board, “Original Pronouncements, As Amended, Statement of Financial Accounting Standards No.
157, Fair Value Measurements.” P. 29, available at:
http://www.fasb.org/jsp/FASB/Document_C/DocumentPage?cid=1218220130001&acceptedDisclaimer=true.
27
For U.S. Y-15 filers, these data are reported on FFIEC Form 009 and the Treasury International Capital reports.

Complexity, 6 Largest Banks

Complexity, $Billions

JPM
1.5

Citi

BoA

1
MS
0.5
WF
0
1000

1500

2000

2500

3000

3500

Bank Size, $Billions
Complexity, All but Top 6

2500

HSBC

Complexity, $Billions

2000

1500

1000

500

State St.
N. Trust
MUFG
Regions
FifthC.
BB&T
Santander
Ally
KeyCorp
Citizens
Amex
Huntigton
BBVA
BMO
Schwab Suntrust
Comerica
M&T
Bancwest
Discover
DBZions
50
100
150
200
250

BoNY
TD
300

350

Cap. One
400

PNC

S BCorp

450

500

550

Bank Size, $Billions

Figure 4: Scatterplot of Complexity and Size.
Cross Jurisdictional Activity, 6 Largest Banks

Cross Jurisdictional Activity, $Billions

800

Citi
700
600

JPM

500
400
300

BoA

200
100
1000

WF
1500

2000

2500

3000

3500

Bank Size, $Billions
Cross Jurisdictional Activity, All but Top 6

Cross Jurisdictional Activity, $Billions

150

BoNY

State St.

100

50
N. Trust
Amex
0
50

HSBC

BMO FifthC. Schwab Suntrust
MUFG
BBVA
Santander
Comerica
Citizens
Bancwest
DBZions
KeyCorp
BB&T
Huntigton
Regions
Ally
M&T
Discover
100
150
200
250

Cap. One
300

350

400

S BCorp

PNC
450

500

550

Bank Size, $Billions

Figure 5: Scatterplot of Cross—Jurisdictional Activity and Size
that the items in this schedule should be part of our impact score. As Figure 5 shows, the ranking
of firms according to their cross—jurisdictional activity data in 2015 diﬀers from the ranking of the
firms by Size, suggesting there is extra information to be gained from including this characteristic
in the score.
Schedule G: Proportion of risk—weighted assets that are funded through short—term
debt This schedule is an attempt to quantify the potential losses to the economy that might arise
when a GSIB firm is a heavy user of certain types of short—term funding. Details on the information
collected are in Table 8. Use of short—term debt is measured as a percentage of risk—weighted assets.
GSIBs only began completing Schedule G in December 2016, and data will not be released until
November 2017, so as of this writing, the currently available GSIB scores do not incorporate data
from this new schedule.
The schedule collects information on how the debt of the firm is distributed across diﬀerent
characteristics including remaining maturity, liquidity of collateral, whether the debt is obtained
21

Schedule E: Cross—Jurisdictional Activity
Cross—Jurisdictional Claims
Cross—Jurisdictional Liabilities

Foreign claims on an ultimate—risk basis
Foreign liabilities

Table 7: Components of Schedule E of the GSIB score (Cross—Jurisdictional Activity).
Schedule G: Short—Term (ST) Wholesale Funding
ST Wholesale Funding secured by Level 1 (LCR) liquid assets
Funding
Retail brokered deposits and sweeps
Unsecured wholesale funding obtained outside the financial sector
Firm short positions involving Level 2B liquid assets or non—HQLA
Funding secured by Level 2A (LCR) liquid assets
Covered asset exchanges (Level 1 to Level 2A)
Funding secured by Level 2B (LCR) liquid assets
Other covered asset exchanges
Unsecured wholesale funding obtained within the financial sector
All other short—term wholesale funding
Table 8: Components of Schedule G of the GSIB score (Short—Term Wholesale Funding).
from inside or outside the financial sector, and whether it has FDIC insurance coverage. Weights are
then assigned to these characteristics (lines in the schedule), reflecting the beliefs of regulators about
how the characteristics translate into severity of losses. In particular, higher weights (implying
larger expected systemic losses) are assigned to:
• Shorter—term debt: the shorter the term, the greater the impact on creditors from not repaying. Additionally, shorter—term debt may be more likely to run prior to failure, which may
prompt the troubled GSIB to quickly sell assets, perhaps at firesale prices, thereby producing
an external impact, as well as weakening the firm and imposing larger losses on its remaining
long—term creditors.
• Unsecured Debt with counterparties within the financial sector: creditor losses will be higher
on uncollateralized debt, and losses in the financial system may have a greater impact in
the economy as a whole than losses outside of the financial system, through relationship—
based lending. Further, firms in the financial system may be more sensitive to the timing of
repayments than nonfinancial firms or individuals.
• Borrowing that is secured by less—liquid assets: the value of these assets would be more
uncertain in a distressed market. The degree of liquidity of collateral is determined using
HQLA liquidity classifications from the LCR regulation.
Note also that Schedule G does not assign diﬀerent weights depending on whether the debt
instruments are QFCs. Qualified financial contracts are repo loans, commodity contracts, forward
contracts, swap agreements, and securities contracts and are exempt from bankruptcy’s automatic
22

stay. A key component of bankruptcy, the automatic stay prevents, upon a debtor’s bankruptcy
filing, most creditors from attempting to collect on their claims. The primary objective of the stay
is to avoid the separation of complementary assets and to preserve the going—concern value of a
firm. QFCs, however, are exempt from the automatic stay so that investors who hold them have
the ability to immediately take possession (and sell, if they wish) any collateral that backs their
loans or derivatives. This is beneficial because it avoids contagion to counterparties. QFCs are,
hence, mostly used by counterparties who are particularly sensitive to the timing of repayments,
and would be particularly hurt by an automatic stay.28 As we have mentioned, a potential cost
of the exemption from the stay is the separation of complementary assets. However, the collateral
backing QFCs is typically not complementary to other assets of the firm, nor will QFC collateral be
important to the firm’s going—concern value. For example, QFC collateral often consists of highly
marketable or cash—like securities (for example Treasury debt instruments), which can be removed
from the firm without reducing the value of other firm assets. Broadly, therefore, QFCs are exempt
from bankruptcy’s stay because they are considered especially “systemic” instruments while, at the
same time, the collateral that backs QFCs consists mainly of assets that can be removed from the
failing firm without producing a large impact on the failing firm’s liquidation value.
The fact that Schedule G does not assign special weight to QFCs seems to implicitly assume
that there might be losses imposed on creditors even if they are lending to GSIBs through QFCs.
Perhaps the designers of the GSIB score considered the possibility that, while QFC—based lending
allows creditors to protect themselves by quickly seizing collateral in bankruptcy, creditors still
may suﬀer losses if they sell the seized collateral en masse, producing firesale losses. If that is the
case, losses might also extend beyond the creditors of a bankrupt GSIB to other holders of similar
assets.
Overall, we think that it is important for any impact score to include information about a firm’s
short—term debt structure. However, we have some suggestions on ways to capture this structure
in a way that may be more informative about the failure’s impact. Mainly, the weighted short—
term debt instruments are normalized in the GSIB score by dividing by risk—weighted assets, with
only the normalized ratio entering the GSIB score calculation. But because risk—weighted assets
generally discount assets like Treasuries, which are very liquid and hence less likely to imply losses
if they need to be liquidated to meet short—term debt obligations, we believe it would be more
appropriate to simply use assets as the denominator for the Schedule G indicator. In other words,
using risk—weighted assets to normalize may inappropriately penalize those GSIBs that have large
holdings of very liquid assets. In fact, a high proportion of short—term funding is less troublesome
if the GSIB matches this funding with an equivalent portion of liquid assets. As a result, a measure
that compares short—term debt with liquid assets (such as the one that the LCR rule requires),
which we propose to use (see next section), would be even more informative about the impact of
short—term debt use in a firm’s resolvability.
Moreover, we propose that the total amount of QFCs and of non—QFC short—term debt be
recorded and evaluated separately. In the next section, we discuss this idea and propose specific
28

See Pellerin and Walter (2012) for a discussion of the idea that QFCs are in a special class of financial instruments
(p. 20) and of the role of QFCs in resolution (p. 19-24).

New items: structure of short—term (ST) debt
QFCs

Fed funds and repo borrowing
Derivatives purchased with negative fair value
Derivatives sold with negative fair value

QFCs/ Assets

QFCs (as defined above) over total assets

Non—QFC ST debt

Commercial paper
Other borrowed money with remaining maturity of one year or less
Uninsured deposits

Non—QFC ST debt / Assets

Our measure of non—QFC ST debt above, over total assets

1/LCR

Our measure of non—QFC ST debt plus Fed funds and repo borrowing,
all divided by the amount of high quality liquid assets

Table 9: New items in the impact score.
ways to measure the impact of the short—term debt structure of firms.

New items for the score

In this section, we discuss some firm characteristics that we view as valuable additions our impact
score, above and beyond what the GSIB score considers. Because the Y-15 does not collect these
items, we assemble them from other regulatory filings, specifically from financial statements for
bank holding companies (FR Y-9C) and financial statements for banks (“Call Reports” — FFIEC
031). Table 9 lists these new items, which all relate to the structure of a firm’s short—term debt.
The appendix contains the details on how we construct each of them.
The amount and structure of short—term funding is important for resolution because these
features determine, to an important extent, post—failure outcomes: the DIP financing needs of the
firm; the potential for firesales; and possible losses to creditors. Importantly, diﬀering resolution
methods (bankruptcy, OLA, or bailout) will likely influence these outcomes because the resolution
method chosen may result in diﬀerent treatment of debtors. In order for our score to address these
important diﬀerences, in what follows we split the main impact of the structure of short—term debt
into (i) inability to perform financial functions essential to the economy because of the lack of DIP
financing, (ii) firesale eﬀects, and (iii) contagion eﬀects. We propose measurable firm characteristics
to be added as items in the score and discuss why each of them would be informative about each
of the three eﬀects.
Inability to perform financial functions essential to the economy because of a post—
failure lack of suﬃcient DIP financing
We expect the amount of short—term debt to be an indicator of DIP financing requirements
post—failure: if we assume all short—term creditors pull out prior to failure, the firm will require
DIP financing to replace this funding following failure and allow the firm to continue its former
activities.29 The larger the quantity of DIP financing required, the more challenging will be the job
29

It is diﬃcult to anticipate how much of a troubled firm’s short-term financing will remain at the time failure is

QFC Debt to Assets, 6 Largest Banks

QFC Debt, 6 Largest Banks

1200

JPM
1000

0.8

QFC Debt, $Billions

QFC Debt to Assets, Percent

MS
0.6
Citi

0.4

JPM
BoA

0.2

BoA
GS

800

Citi

600
MS
400
200

WF
0
1000

1500

2000

2500

3000

0
1000

3500

1500

Bank Size, $Billions
QFC Debt to Assets, All but Top 6

0.4
0.3
0.2
0.1
0
0

3000

3500

BoNY
State St.
BMO
DB
N.
Trust
Fifth Suntrust
M&T
Citizens
KeyCorp
MUFG
Bancwest
BBVA
Huntigton
S BCorp
Comerica
BB&TTD
Ally
Cap.PNC
One
Zions
Regions
Santander
Amex
Discover
C.
Schwab
100
200
300
400
500
600

HSBC

120

HSBC

0.5

2500

QFC Debt, All but Top 6

140

QFC Debt, $Billions

QFC Debt to Assets, Percent

0.6

2000

Bank Size, $Billions

100
80
60
40

BoNY

20
0
0

State St.
Suntrust
BMO
S BCorp
N.
Trust
Fifth
Citizens
M&T
Cap.PNC
One
BB&TTD
MUFG
DB
KeyCorp
Bancwest
Ally
BBVA
Huntigton
Regions
Santander
Comerica
Zions
Amex
Discover
C.
Schwab
100
200
300
400
500
600

Bank Size, $Billions

Figure 6: Scatterplot of QFC financing (right), and its ration to assets (left), and size.
of finding a lender with suﬃcient capacity to provide the needed DIP loan. Lack of DIP funding
will make it more diﬃcult for the firm to keep healthy subsidiaries working and performing financial
functions that may be essential to the economy. The firm characteristics that can be informative
about this eﬀect are:
• QFC financing: These financing sources will disappear even if they do not run prior to failure,
because they are exempt from bankruptcy’s automatic stay.
• Non—QFC short—term debt: These financing sources can run prior to failure, but not after,
so assuming all of this debt runs before failure provides an upper bound on impact.
Figure 6 presents the scatterplot of the ratio to assets (left) and the amount (right) of QFC
financing and size, while Figure 7 presents the non—QFC debt counterparts.
Contagion to creditors of the failing firm
There are two main types of debt holders: short—term creditors and long—term creditors. Because their treatment in a failure may diﬀer according to the resolution method, we discuss the
informativeness of short—term debt structure about contagion to each of these two classes of creditors separately.
declared because creditors often detect the firm’s problems and withdraw funding in the days leading up to a failure.
In our analysis, we consider both extreme scenarios (prior to failure either all debt runs or, alternatively, none of it
runs) in order for our score to capture the largest impact a failure might produce.

Non-QFC ST Debt to Assets, 6 Largest Banks

Non-QFC ST Debt, 6 Largest Banks

800

JPM

0.3
WF

Non-QFC ST Debt, Percent

Non-QFC ST Debt to Assets, Percent

0.35

BoA

0.25
Citi

0.2
0.15
0.1

BoA

600
WF
400
Citi

200

MS GS

0.05
1000

1500

MS GS
2000

2500

3000

0
1000

3500

1500

Non-QFC ST Debt to Assets, All but Top 6

0.8

2000

2500

3000

3500

Bank Size, $Billions
Non-QFC ST Debt, All but Top 6

200

0.6
Comerica
MUFG
BMO
Zions
Bancwest
State St.
KeyCorp
Fifth BB&T
Suntrust
Citizens
M&T
BBVA
N.
Trust
Regions
Huntigton
Santander
TD

0.4

0.2

S BCorp
BoNY
HSBC
Cap. One
PNC

C. Schwab
Ally
Amex
Discover

Non-QFC ST Debt, Percent

Non-QFC ST Debt to Assets, Percent

Bank Size, $Billions

S BCorp
150

BoNY

100

State St.

BMO
MUFG
Fifth
Citizens
DB Bancwest
M&T
KeyCorp
Regions
Santander
N.
Trust
Comerica
C. Schwab
BBVAAlly
Zions
Amex
Huntigton
Discover

0
0

100

200

300

400

500

600

Cap.PNC
One
HSBC

BB&TTD
Suntrust

100

Bank Size, $Billions

200

300

400

500

600

Bank Size, $Billions

Figure 7: Scatterplot of non—QFC short—term debt to assets (left) and amount (right) and Size.
First, we’d like to be able to estimate the contagion eﬀect to short—term debt holders due to a
delayed (and possibly partial) repayment of the short—term obligations of the failing firm. We would
like an estimate of the impact on counterparties, who are counting on timely repayment so they can
cover their own short—term obligations (obligations due sooner than the time taken to complete
a resolution) but will not receive such repayment (and, therefore, suﬀer from the “contagion”
eﬀect). If the assets of the failing firm are not very standard, as we might expect, borrowing
against bankruptcy claims may not be without a cost, and hence the short—term creditors may
face liquidity problems, on top of any capital losses that the expectation of partial repayment may
trigger.
When taking into account this contagion eﬀect, knowing the amount of short—term debt with
very short maturity (e.g., maturity of less than 30 days) would be useful but is not currently
available (this is information that will become available when the Y-15 is next revised). Using
instead financial information available in the Y-9C bank holding company report and in banks’
Call Reports, we propose to include in our score the following firm characteristics that can be
informative about the contagion eﬀect to short—term debt holders:
• QFC financing: QFC creditors (such the failing firm’s repo creditors) can seize their collateral
immediately upon default, so QFC financing is only a concern for contagion if some kind of
mini—stay is imposed (as in OLA).
• Non—QFC short—term debt: if we assume none of this debt runs before failure, we can use
26

this amount as an upper bound on this impact.
Second, we would like to estimate potential losses to long—term debt holders (and any short—
term creditors who do not run) from the liquidation of assets to repay short—term debt that has
run. For this objective, we need information about the relative importance of losses for the firm
stemming from the rushed liquidation of those assets; this will be informative about the decrease
in the value of total assets that the run of the short—term debt will impose and hence on the losses
imposed on long—term debt holders by decreasing the value of the firm’s assets in bankruptcy.
Note, as discussed below, that some policymakers may be more concerned with this firesale eﬀect
than others — in other words, some policymakers may believe it unlikely that asset values can be
driven below fundamental values. Using financial information from the Y-9C and Call Reports, we
propose to include in our score the following firm characteristics that can be informative about this
contagion eﬀect to long—term debt holders:
• Ratio of QFC financing to assets: for these debt instruments, typically, there will be margins
on derivatives, or haircuts on repo collateral, to compensate counterparties for the risk of
default on the contract. Such instruments imply losses for the failed firm when its QFC
contracts are terminated, which translates into lower recovery rates for the failed firm’s non—
QFC creditors.
• Non—QFC short—term debt to assets ratio: similar to the case of QFCs, but here losses, if any,
would come from a firesale eﬀect, where the liquidation of assets to cover running non—QFC
debt can drive the value of these assets below their fundamental values. If we assume all
short—term debt runs, this provides an upper bound for this impact.
• Inverse of LCR: the proportion of highly liquid assets to short—term debt (inverse of the LCR)
provides a measure of expected losses that might be suﬀered by long—term debt holders from
the rushed liquidation of assets to repay short—term obligations that are removed prior to
failure: a firm that relies heavily on short—term debt but has very liquid assets may avoid
large losses in the days leading up to failure, while one that has to sell illiquid assets is likely
to suﬀer losses on such sales. A higher ratio will mean lower losses on the proportion of the
balance sheet that is financed by short—term debt.
Figure 8 presents the scatterplot of the inverse of the LCR (short—term debt over assets) and
size.
Contagion to third parties through a balance—sheet eﬀect of firesales of assets
If creditors become unwilling to roll over the troubled firm’s short—term debt in the weeks or
days leading up to the failure of a firm, the amount of a firm’s short—term financing will be an
ex—ante indicator of potential firesale eﬀects: the larger the amount of short—term debt, the larger
the amount of assets that will have to be liquidated by the troubled firm (to repay creditors that
are unwilling to roll over) to try to maintain the firm’s solvency, and the greater the potential for
firesales. The firm characteristics that can be informative about this eﬀect are:
27

ST Debt HQLA ratio, 6 Largest Banks

ST Debt HQLA ratio, Ratio

JPM
1.8
BoA
1.6

1.4

Citi

1.2
1000

1500

2000

2500

3000

3500

Bank Size, $Billions
ST Debt HQLA ratio, All but Top 6

4
ST Debt HQLA ratio, Ratio

DB
3.5
HSBC

BMO Fifth
MUFG
KeyCorp
Comerica
Huntigton
Ally
BancwestSantander
Citizens
BBVA
Zions
M&T
Regions

2.5
2

BB&T
Suntrust

1.5

PNC

Amex

1
N. Trust
Discover

0.5
50

100

150

State St.

BoNY

C. Schwab
200

250

S BCorp

Cap. One

300

350

400

450

500

550

Bank Size, $Billions

Figure 8: Scatterplot of the inverse of the LCR ratio (short—term debt over assets) and Size.
• QFC financing: because these instruments are exempt from the automatic stay, even if QFC—
based funding doesn’t run prior to failure, this funding will be liquidated upon failure. As a
result of this liquidation, the collateral may land in the hands of a counterparty which values
the collateral less than the failed firm. Alternatively, the collateral may go to a creditor that
does not want to hold it, leading to the sale of a large quantity of the collateral. Either may
depress the asset’s market price.
• Non—QFC short—term debt: any of this type of debt may also have a firesale eﬀect if it runs
before the automatic stay is declared. Hence, assuming all of it runs we can use this amount
as an upper bound on this eﬀect.
As the discussion in this section makes clear, there are diﬀerent possible costs and benefits to
a GSIB firm’s heavy use of QFCs, and the sum of the costs and benefits, should the firm fail, is
unclear for the financial system or the economy as a whole. Further, as we will discuss in more
detail below, diﬀerent policymakers may weight each of these costs and benefits diﬀerently.

4.1

A snapshot of large U.S. firms using our new impact—relevant characteristics

Figure 9 presents the raw data on both the GSIB original firm characteristics and the new characteristics we have proposed as relevant to calculate impact, for the 34 BHCs in 2015. We use a color
scheme to illustrate the relative importance of each number (red corresponds to larger numbers,
yellow to intermediate, and green to smaller). Firms are ordered according to their size (as measured by Total Exposures), from largest to smallest. This allows us to compare all characteristics at
the same time, as opposed to one by one, to size, as we did in the scatterplot figures above. Green
28

colors in the top rows, or red colors on the bottom rows, indicate deviations from a proportional
relationship of the characteristic being measured and size. For example, Bank of New York Mellon
and State Street both score very high on substitutability given their size, because of their roles
as intermediaries in the Tri—party repo market (they have a disproportionately high number of
assets under custody). As another example, Wells Fargo scores relatively low in Substitutability,
Complexity and Cross—Jurisdictional activity, given that it is the 4th largest bank.
It is worth discussing the facts regarding some of our new items. QFC amount is fairly proportional to size, but this is less true for QFC/Assets. More importantly, even though the amount of
non—QFC short—term debt is proportional to size, the ratio of this type of debt to assets displays
an almost inverse relationship to size. This inverse relationship is also somewhat present in our
approximation of the inverse of the LCR, with many large firms showing a high proportion of liquid assets to short term debt, and instead many smaller firms showing high levels of illiquidity as
measured by our ratio.
Diﬀerences in these characteristics of the debt structure may translate into important diﬀerences
in impact following failure, as we have discussed. In the next section we will illustrate how diﬀerences of opinion regarding the mapping of these characteristics into impact may aﬀect judgement
calls on whether firms have an acceptable structure for resolution through bankruptcy.

Evaluating acceptable firm structures

We have discussed how a modified version of the existing GSIB score could be a good candidate for
a measure meant to establish whether a firm’s structure is acceptable for resolution in bankruptcy
– an impact score. We now illustrate how to use weights in the score to account for the fact that
the same objective quantity recorded for a given firm characteristic may translate into diﬀerent
levels of impact depending on the resolution method used to deal with the troubled firm or on the
crisis scenario being considered. We will use the score in a positive way, as an instrument to portray
diﬀerent opinions that hypothetical policymakers may have concerning the impact associated with
these characteristics. With this exercise we want to emphasize that these scores may be influenced
by subjective beliefs of policymakers.
In our examples, we use actual data on firm characteristics (collected by the Y-15, plus estimates
of the importance of short—term funding and of the use of QFCs that we calculate using publicly
available data from the Y-9C and the Call Report). We combine these data with three diﬀerent
sets of weights (see Figure 10), which we created in order to represent views of three “types”
of policymakers. The combination of actual firm data with weights that we devised produces
hypothetical scores that we use to study these hypothetical policymakers’ views about various
firms’ resolvability.30
In the next paragraphs we discuss, for several indicators, reasons that could be behind any
diﬀerence of opinion on how the impact of a given firm characteristic would diﬀer across diﬀerent
30

An Excel spreadsheet to calculate our impact score is available as an online appendix. It allows readers to adjust
policymaker’s weights, based on their own views of the likely impacts associated with various firm characteristics,
scenarios, and resolution methods, and examine how such changes shift scores for the 34 institutions required to file
the Y-15 in 2015.

Figure 9: Raw data for BHCs in 2015, in Billions of $US. Firms are ordered according to size (Total
Exposures), from largest to smallest. The colour scheme reflects relative size of the figures (red is
larger, green smaller).

resolution methods and scenarios. We then use these diﬀerent arguments to characterize the beliefs,
or weights, of our three hypothetical policymakers: we propose a ranking of the weights within each
characteristic and come up with numbers in our example (Figure 10) that will satisfy these rankings.
Policymaker 1 has two sets of weights, labeled (a) and (b), representing his views at two points
in time. The weights in (a) can be thought of as representing a Policymaker that currently agrees
with LW reviewers at the Federal Reserve and the FDIC, who have recently determined that all of
the large firms have acceptable LWs. As we will see when we present the results of our examples,
we set this first set of weights on firm characteristics such that, in fact, all the firms in our sample
have an acceptable structure. For contrast, the weights for Policymaker 1 in case (b), though
matching the relative importance of each characteristic with those in case (a), are twice as large
for each characteristic. This will imply, as we will see, that some of the firm’s structures are not
acceptable in a hypothetical case (a). With these two contrasting weights we want to capture:
1) some policymakers’ views that the TBTF problem has become less severe as a result of the
legislative and regulatory eﬀorts in the last 10 years geared toward assuring less costly large firm
resolution; and 2) views that policymakers might have held before these changes were implemented.
In particular, we are focusing on the shifts that were driven by international regulations such
as Basel III and several important legislative and regulatory changes introduced since the financial crisis of 2007 — 2008. These legislative and regulatory changes include TLAC, new liquidity
standards, and implementation of the living will process. Because of these measures, we model
Policymaker 1’s beliefs as shifting from (b) (before the crisis) to (a) (currently). Current weights
represent the belief that the failure of a bank is likely to lead to considerably less economic impact
than would have occurred prior to these changes.
When modelling the shift in beliefs of Policymaker 1 from (a) to (b), we also adjust the estimates
for the cost of government intervention. As we discuss in detail in the next subsection, these costs
can include not only eﬃciency losses due to risk shifting (i.e., moral hazard costs) but also bailout
“outrage” costs. In particular, in case (b), before the big bailouts of 2008, these outrage costs
may have been lower than today (case (a)). Intense public protests and legislator criticism of
bailouts and other forms of government assistance, following the crisis, imply that policymakers
can anticipate similar opposition in response to any future bank bailout or assistance. Moreover,
it seems possible that the last decade’s regulatory eﬀorts imply an increase in the moral hazard
consequences of a bailout: if, after all the costly LW process and other eﬀorts, a firm gets bailed
out, other firms in the financial system and their creditors are likely to increase expectations of a
future bailout out.
Alternatively, we could interpret case (b) for Policymaker 1 as representing the views of a current
policymaker who is not convinced that legislative and regulatory shifts have yet been suﬃcient to
significantly reduce the economic impact of a large bank’s failure. Therefore, for this policymaker,
the TBTF problem has not been reduced.
To further illustrate how our “acceptability” determinations depend on weights, we propose
two more policymakers (2 and 3) who share Policymaker’s 1(b) estimates of the cost of government
intervention, but disagree with him in some of their weights. This choice of the cost of government
intervention for Policymakers 2 and 3 is purely out of convenience (we need variation in the resolv-

Figure 10: Weights assigned to diﬀerent firm characteristics and to moral hazard costs by three
hypothetical policymakers in our examples. Policymaker 1 has lower weights (by a factor of 0.5) in
case (a) and higher in (b); Policymaker 1(b) and 2 agree on most of their weights, except 1 worries
about the eﬀect of firesales and Policymaker 2 does not; weights that are diﬀerent between them
as a result of this disagreement are in red. Policymaker 3 uses a simple rule based only on the size
measure and moral hazard costs.

ability determinations to illustrate how our score works), but in no sense do we mean to imply that
Policymaker 1’s weights in (b) are more realistic, or reasonable.
Policymaker 1 (a and b) worries about firesales. Policymaker 2, instead, does not believe that
the firesales are a significant concern. Hence, the weights on characteristics that may involve firesale
estimates will be typically higher for Policymaker 1 than for Policymaker 2 (the weights that diﬀer
are indicated in red in Figure 10). One can think of Policymaker 1 as being broadly representative
of those policymakers who are very worried about external eﬀects emanating from the failure of a
GSIB and Policymaker 2 as representative of those more skeptical about such eﬀects.31
Policymaker 3 wants a simple score based only on size. He has the same weights as Policymaker
1, but he only analyzes the individual firm characteristics for the largest five holding companies.
Then, he chooses alternative weights (in red in Figure 10) that will, on average, replicate these scores
for these five companies using only their size information. He uses these new weights to construct
scores for the rest of the companies looking only at their size information. This way, he does not
need to construct individual measures about short—term debt structure for all the firms in the
sample. Also, going forward, these same weights can be used in future years with only information
about size, even if filing a form such as the Y-15 is no longer required. Much systemic risk regulation
is size—based, meaning institutions smaller than a specified asset—size cutoﬀ are exempt, indicating
that the simplicity of such straightforward cutoﬀs can be attractive to policymakers.32
Determination of the cost of government support In order to determine whether the structure of a firm is acceptable, the impact score under bankruptcy and OLA needs to be compared to
the impact score of a bailout. In our examples, when we consider the impact score of a bailout, we
set the weight of each characteristic to zero. This choice is based on the fact that a troubled firm
is allowed to continue in business after it is bailed out, so there is no failure. However, we assume
that there is still an impact in the economy from a government intervention, mainly in the form of
a moral hazard cost, but also including redistributional distortions and political cost from public
outrage about the help received by financial institutions.
Throughout our examples, moral hazard cost depends on the resolution method and the crisis
scenario. Moral hazard costs are generated when a bailout causes market participants (financial
firms and creditors) to revise upward their expected probability of a future bailout, reducing par31

These contrasting views were apparent in several policymakers’ statements in the aftermath of the crises. For
example, see a comparison in Steve Williamson’s blog post (June 2010) of views expressed by Narayana Kocherlakota
and Jeﬀrey Lacker while they were both presidents of Federal Reserve Banks. Williamson quotes a May 2010 policy
piece by Kocherlakota, in which he defines firesales: “During financial crises, many financial institutions may have to
sell assets or collateral at the same time. These simultaneous sales will put downward pressures on the assets’ prices.
A given financial institution will not internalize the impact of its sales on the price of other institutions’ assets.”
On the other hand, Lacker is quoted from a speech he gave on May 26, 2010, in which he dismisses this price eﬀect
and contagion eﬀects as being caused by failures: “Arguments that one firm’s failure can spark costly runs at other
firms rely on the logic of panics as self-fulfilling prophecies. While this logic is correct as far as it goes, it provides
an unsatisfactory guide for policymakers, because it does not provide a means for determining whether creditors are
justified in pulling away from other firms. After all, news that one firm has failed can be genuinely informative about
fundamental prospects at other firms with similar exposures.”
32
An example of such a policy is the DFA’s $50 billion threshold for enhanced prudential standards.

ticipants’ incentive to monitor risk—taking. The policymakers in our example believe that a bailout
that occurs in an aggregate shock scenario implies less of a revision in this probability than a
bailout if the shock is idiosyncratic: because impacts are known to be larger in an aggregate shock,
a bailout in this state is not informative about the threshold for impact that would trigger a bailout
in an idiosyncratic shock. However, a bailout in an idiosyncratic shock scenario indicates that if
the same firm were to fail in an aggregate shock, this would also grant a bailout. Moreover, it is
reasonable to assume that the probability of an idiosyncratic shock is higher than that of an aggregate shock; hence, the eﬀect of an upward revision of a bailout in an idiosyncratic shock will aﬀect
the expected return of debt holders more than that of a bailout in an aggregate shock. Because of
these two arguments, we assume that the moral hazard cost we assign to a bailout is higher in the
idiosyncratic scenario (see Moral Hazard column in Table 10).
Our policymakers hypothesize that there might also be some moral hazard cost associated with
an OLA resolution (though smaller than the moral hazard cost created by a bailout), for two main
reasons: the possibility of subsidized borrowing from the OLF, and the possibility of violations of
the bankruptcy priority rights for creditors (i.e., the prioritization of short—term debt repayment for
systemic reasons). In particular, OLA may violate the bankruptcy priority rules by fully protecting
short—term debt holders, who, under bankruptcy resolution, might instead be expected to stand in
line to receive less—than—complete repayment with other general creditors. These types of assistance
may cause market participants to increase their estimate of future assistance, hence decreasing their
incentives to curb excessive risk—taking of the firm they are financing. We capture this moral hazard
cost as a lump—sum cost for OLA, but it could easily be generalized to be dependent on the amount
or proportion of short—term funding.
All policymakers agree that the moral hazard cost will be the highest if a troubled firm is
propped up with a bailout so that its creditors suﬀer no losses. They also agree that there is no
moral hazard cost (cost equals zero) if a troubled firm is resolved in bankruptcy, because no degree
of public assistance is typically forthcoming in bankruptcy.
To illustrate the sensitivity of our score to the moral hazard cost estimates, we present weights
for Policymaker 3 who cares only about size (as measured by total exposures). In this case, we
can see that the resolvability determination, for given weights on size, depends strongly on values
assigned to the moral hazard cost: because the impact is a linear function of size, by choosing
the moral hazard cost number we are eﬀectively deciding what is the threshold value of size that
will make a firm’s structure acceptable. For Policymakers 1 and 2, as we discuss after presenting
the results on Figure 11, a moral hazard cost does not imply a cutoﬀ in terms of size, since they
consider all 10 characteristics of the firms and these are not proportional to size.
Weights on Total Exposure Note that in this and the next sections discussing weights for other
characteristics, we will focus on the ranking of the weights across crisis scenarios and resolution
methods, rather than their values. Because of this, we will discuss case (a) and (b) for Policymaker
1 together: since his weights in case (b) are simply double those in case (a), the relative importance
he places on characteristics is constant throughout the two cases.

Policymaker 1’s weights for Total Exposures (TE) satisfy the following ranking:

 
 



    

Recall that the TE indicator consists primarily of assets, derivative exposures, and credit commitments (e.g., undrawn lines of credit). Under an idiosyncratic shock, the impact of a firm with
a large TE is likely to be fairly large – the BHC has acquired valuable information about the
credit—worthiness of those to whom it lends, and this information is likely to be lost as a result of
its failure. Other BHCs will find it expensive to pick up this missing lending service as a result
of the absence of this private credit—worthiness information, increasing the total borrowing costs
faced by firms or preventing some of this lending from being replaced at all. Still, many of the failed
firm’s debtors could have similar relationships with other banks, partially oﬀsetting this eﬀect. In
the case of an aggregate shock, however, Policymaker 1 believes that this problem is considerably
worse — not only is it less likely that other lenders with a relationship with the failing institutions’
clients will be healthy enough to substitute for the disappearing relationship, but there may be no
banks willing or able to pick up the lending at all. As a result, TE receives a lower weight in an
idiosyncratic shock scenario than in an aggregate shock scenario.
The importance of the TE indicator may also depend on the resolution method and the shock
because of DIP financing concerns. This follows from the fact that larger firms are more likely
to need large DIP financing loans, which in turn can allow time to resolve the firm in a way that
maximizes the value of liquidated assets and also allow the firm to continue providing some of its
services. Then, on one hand, the shock is important because the state of the economy could be
very relevant in determining whether other firms are healthy enough to provide the private DIP
financing. On the other hand, if a BHC is resolved using OLA, DIP financing may be more readily
available: subject to certain limitations set forth by the DFA, an institution under receivership
(FDIC—overseen) can borrow from the Orderly Liquidation Fund (OLF).
Is size the only thing that matters? Asset size is typically thought of as an important
determinant of loss given default. In fact, all U.S. BHCs over $50 B are automatically subjected
to enhanced prudential standards.33 Therefore it seems natural to consider the possibility that
size is a suﬃcient indicator of the impact of a firm’s failure. As we have seen in our scatterplots
(Figures 1-9), changes in size do not always translate into similar magnitude changes in the other
firm characteristics. To evaluate the implications of only looking at size to determine the impact of
a firm’s failure, we include in our examples a “policymaker 3” who puts zero weights on indicators
other than TE.
Weights on Interconnectedness Policymaker 1’s weights for interconnectedness satisfy the
following ranking:




 
 
 


33

In the context of the GSIB score, Passmore and von Haﬀten (2017) ask this question and perform a principal
components analysis to establish the extra information added by indicators other than TE.

Interconnectedness measures two main firm characteristics. First, it measures the importance
of intrafinancial assets. These may be a source of disruption to the financial system through a
balance—sheet influence due to a firesale eﬀect on prices. Policymaker 1 cares about these and
believes that they are mainly a concern in an aggregate shock, in which case having OLA financing
would prevent the liquidation of assets and hence lower the cost. Policymaker 2 does not put weight
on this eﬀect.
Second, this item measures intrafinancial liabilities and securities outstanding, both being measures of borrowing by the failing firm. The main concern for disruption is through contagion to
counterparties expecting repayment. Policymaker 1 feels that the special financing possibilities
that are available in OLA to guarantee the debt of the failing firm, as well as the ability to violate
priority rules to prioritize repayment for systemic debt holders, will make the impact of failure lower
under this resolution method. On the other hand, if the shock is idiosyncratic, this policymaker
believes that the expected contagion eﬀect would be lower to start with, and therefore an OLA
resolution may imply the same impact on the economy as a bankruptcy resolution. Policymaker 2
agrees with this reasoning, and hence his weights satisfy the same ranking but will be lower because
he does not believe in firesales.
Weights on substitutability Policymaker 1’s weights for substitutability satisfy the following
ranking:




= 
= 
= 


Substitutability includes payments activity, assets under custody, and underwriting activity.
Contrary to the TE example, Policymaker 1 finds little reason to believe that the impact associated
with a firm scoring high on substitutability will vary drastically between an idiosyncratic and
aggregate shock scenario or between resolution methods. Policymaker 2 holds the same views —
given that firesale concerns are not at play here.
To illustrate why Policymaker 1 holds this view (similar impact regardless of type of scenario
or resolution method), consider the loss of payments services due to the failure of a BHC.34 The
failure certainly has an impact in the economy due to the fixed cost of recreating the client—specific
payments processes at a new bank. Still, the policymaker believes that this cost may not diﬀer
much under diﬀerent crisis scenarios, given that it seems likely that banks will be no more hesitant
to pick up additional payment clients during an aggregate shock than when just one bank fails
(an idiosyncratic shock). Since banks would be able to generate fee—based income by picking up
payment clients, and this extra business should not modify the riskiness of their portfolio, they
would likely be happy to take on the additional business. Therefore, under these policymaker’s
beliefs, the size of the eﬀect of payments activity will not vary considerably with these two types of
crises. Using a similar argument, since the main costs associated with replacing the lost payments
provider are the switching costs of clients having to shift from the failing firm to other payments
firms, this policymaker does not believe that alternative resolution methods would aﬀect these
34

See Lacker (2004) for a study of the eﬀects of payment disruptions following the September 11, 2001, attacks in
New York. In our paper, the focus is on whether the impact of the failure of one key firm in the payments system
would be diﬀerent across shocks and resolution methods, rather than on a quantification of the impact.

costs. For example, OLA’s readily available funding could allow a bridge company to continue
processing payments for some time, but Policymaker 1 does not believe that this extra time would
significantly decrease the switching costs of its clients.
Since the primary cost associated with the loss of custodial services is the fixed cost of transferring assets to another bank, a similar argument to the one used for payments applies.35 It is
worth pointing out, however, that if custodial services are packaged together with intraday credit
provision, as they were, to a large extent, in the tri—party repo business before the financial crises,
the considerations used by Policymaker 1 regarding TE (where credit provision was an important
component) would be relevant for Assets under Custody. To better judge whether weights for
Assets under Custody should mirror those of TE, it would be useful to require separate disclosure
of intraday credit provision (a point we have made earlier when discussing the details of the substitutability GSIB schedule). On the other hand, if custodial services are not exclusive relations,
the substitutability of intraday credit would be easier, and weights might not need to diﬀer across
scenarios and resolution methods after all.
For the last component of the substitutability item, underwriting services, again Policymaker
1 bases his weights on the fact that this is a service provision that needs to be substituted, paying
a cost that may not be very diﬀerent across shocks or resolution methods. Note that, even if
underwriting implies an evaluation of credit worthiness, and there is likely to be a cost from
substituting long—term relationships for which accumulated knowledge may be destroyed, again
this replication of the credit—worthiness monitoring cost is not likely to diﬀer significantly across
crises scenarios or resolution methods.
Weights on complexity Policymaker 1’s weights for complexity satisfy the following ranking:




= 
 
 



Complexity measures capture both concerns about losses from rushed liquidation of hard—to—
evaluate assets (as reflected in the sub—item “Level 3 assets”), contagion due to updated information
on the value of the assets that the firm typically trades, or through a firesale—like eﬀect (both
concerns captured in the sub—items “trading book” and “available—for—sale assets”). Policymaker 1
believes that, in an idiosyncratic shock, losses from the liquidation of complex assets are not likely
to depend on the type of resolution method. The impact on the economy of a revaluation of trading
book and AFS assets may be more severe if other firms are in trouble too, however. Hence, he sets
a higher weight on complexity in an aggregate shock.
Moreover, Policymaker 1 believes that if there were demand for DIP financing during liquidation,
in order to provide time for potential buyers to evaluate complex assets, meeting this demand could
be diﬃcult during an aggregate shock, when this financing is likely to be scarce in the market.
Hence, he believes that the availability of OLF funding — under an OLA resolution — could reduce
the failing GSIB’s impact on the economy.
35

For a historical example of a disruption in custodian services due to a computer problem at BoNY in 1986, see
Ennis and Price (2015). Here, again, our focus is on whether the disruption would be diﬀerent depending on the
shock and resolution and not on whether there would be a cost or not.

Policymaker 2 agrees with Policymaker 1 on the ranking of weights but attaches lower weights
overall because he discounts the existence of firesale eﬀects.
Weights on cross—jurisdictional activity Policymaker 1 has weights on the amount of cross—
jurisdictional activity that satisfy the following ranking:




 −
 −
 −

−

Cross—jurisdictional activity complicates resolutions because of the need for the U.S. authority
in charge of the resolution to coordinate with authorities in other countries. Policymaker 1 believes
that coordination among the U.S. supervisors in charge of an OLA resolution and any foreign
supervisory authorities would be more straightforward than among a U.S. bankruptcy judge and
foreign authorities, given that U.S. and foreign supervisors are likely to interact frequently with
one another, for example when setting international regulatory standards.36 Also, he believes that,
given the fixed costs of these interactions among policymakers across jurisdictions, in an aggregate
event it is more likely that eﬀorts may be put into eﬀect that facilitate any cross—border resolutions
under OLA, since policymakers may anticipate that several firms may benefit from them, compared
to when just one firm may benefit from these eﬀorts. Hence, under OLA Policymaker 1 gives a
lower weight to cross—jurisdictional activity in an aggregate shock than in an idiosyncratic one.
Under bankruptcy he believes this coordination would be diﬃcult, but it may be more likely in an
aggregate shock because of potential pressure of regulators for jurisdictions to coordinate. Since
this is not directly related to firesale concerns, Policymaker 2 agrees with the weights chosen by
Policymaker 1.
Weights on amount of QFCs As we discussed earlier, whether a firm’s short—term funding is in
the form of QFCs will determine potential contagion eﬀects to counterparties (lower value to long—
term creditors but speedier repayment to the counterparties of the QFCs) and potential firesale
eﬀects. Also, the total amount of QFC financing may be informative about DIP financing needs.
The relative importance of these eﬀects is a particularly controversial matter, and we illustrate how
diﬀerent opinions may aﬀect impact estimates by presenting weights for a second policymaker (who
we call Policymaker 2) that contrasts with that of Policymaker 1.
We assume that Policymaker 1 and 2 agree on the DIP financing needs that come from QFC
use and that these needs are more of a concern under bankruptcy than under OLA because of the
availability of OLF funding. They also agree that QFCs are useful in minimizing the contagion
eﬀect to short—term creditors. However, Policymaker 1 believes that these benefits to short—term
creditors are outweighed by the negative eﬀect that the liquidation of the assets backing the QFCs
will have on third parties that are holding these same assets. That is, Policymaker 1 believes that
there will be a firesale—like eﬀect on asset prices that will aﬀect the financial system through the
balance sheet channel. Policymaker 2 does not think that prices will be aﬀected by the sale of the
36

As an example of this idea, FDIC General Counsel Michael H. Krimminger noted in 2011 testimony that: “the
regulatory authorities who will administer the OLA are in a far better position to coordinate with foreign regulators
in the failure of an institution with significant international operations.”

assets per se. For illustration, we assume that this disagreement between the two Policymakers is
important enough that overall the combination of these three eﬀects results in a lower weight for
OLA for Policymaker 1, while Policymaker 2 estimated the impact to be lower under bankruptcy.
Policymaker 1 has weights on the amount of QFC short—term funding that satisfy the following
ranking:





          
Policymaker 2 has weights on the amount of QFC short—term funding that satisfy the following
ranking:





          
Again, these eﬀects are more important in an aggregate shock.
Weights on the ratio of QFC to assets Typically there are margins on derivatives, or haircuts
on repo collateral, to compensate counterparties for the risk of default on the contract. As a result,
when QFC contracts default and collateral (including margin and haircut amounts) is seized, the
defaulting firm suﬀers a loss equal to the amount of the margin or haircut, and fewer assets remain
for non—QFC creditors. Hence, a higher ratio of QFCs to assets normally means higher losses to
non—QFC creditors in the case of default. We assume that Policymaker 1 and 2 have equivalent
concerns for the contagion that flows from the seizure of collateral and haircuts to losses to long
term creditors. Policymaker 1 also believes that the liquidation of the collateral will lead to further
price decreases due to firesale eﬀects. This disagreement means that Policymaker 1’s weights on
this characteristic will be higher than those of Policymaker 2.
OLA’s mini—stay can imply smaller losses on QFCs because it allows for the transfer of the
contracts, rather than liquidation. The mini—stay can imply, thereby, less contagion to long—term
debt holders. For example, for contracts such as derivatives, the stay period means more time
to identify counterparties to take on the same side of a hedge, which may mean losses smaller
than the margin amount, since a similarly positioned firm may value the contract more than the
counterparty. Hence, OLA’s mini—stay may translate into a higher liquidation value of the failing
firm, benefiting non—QFC creditors. This eﬀect will be directly related to the fraction of assets that
are financed through QFCs, and hence the ratio of QFCs to assets, rather than the absolute QFC
amount. Therefore, the ratio is the relevant measure to consider here. Because of this benefit of the
QFC mini—stay, losses under bankruptcy (where QFC holders are allowed to seize collateral plus
margins and haircuts immediately) will be weighted as larger than under OLA. Moreover, these
losses to long—term creditors may be more of a concern in an aggregate shock. Policymaker 1 and
Policymaker 2’s weights, that reflect this thinking, will satisfy:





          

Weights on amount of non—QFC Short—term Funding The amount of non—QFC short—
term funding can be informative, as we described in previous sections, about three diﬀerent types
of impacts: DIP financing needs, direct contagion to the troubled firm’s short—term creditors and
indirect contagion to the financial system through firesale eﬀects.
39

Concerning the first impact (DIP financing needs), any short—term funding that would be
subject to the automatic stay will have incentive to run at the earliest sign of financial trouble.
Because of this, Policymaker 1 takes the amount of non—QFC short—term debt as indicative of
potential DIP financing needs (since any short—term debt that does not run before failure will be
subject to the stay, it will not translate into financing needs, so non—QFC short—term funding may
be less of a concern for this eﬀect than QFCs). DIP financing needs will be more of a concern under
bankruptcy than OLA because, again, of the availability of funds from the OLF.37
Second, Policymaker 1 believes there will be an impact of non—QFC short—term debt in the form
of contagion of financial troubles to counterparties that are not repaid. This applies to non—QFCs
that do not run and are subject to the automatic stay once the firm fails. Moreover, Policymaker 1
believes that the availability of a credit line from the Treasury under OLA would mainly be useful
in bailing out short—term debt. This implies that he gives non—QFC short—term financing a lower
weight under OLA than under bankruptcy.
Third, Policymaker 1 believes that whenever short—term funding does not roll over, the firm will
need to liquidate assets, potentially aﬀecting market prices for similar assets (meaning have a firesale
eﬀect). Again because not all short—term debt will necessarily run before failure, Policymaker 1
has lower firesale concerns for this debt than for QFCs.
DIP financing needs, contagion and firesale concerns are all larger problems in an aggregate
shock scenario, because financing may well be scarce (or expensive) in the market, on one hand,
and because firms likely to be hurt by firesales or nonrepayment (contagion) would be in a weaker
state in an aggregate shock.
Moreover, the diﬀerence between Policymaker 1’s OLA and Bankruptcy weights is larger in an
aggregate than idiosyncratic shock, because he believes that the political willingness to bail out
short—term debt will be stronger in an aggregate shock (or the potential opposition will be weaker).
Hence, Policymaker 1’s weights on non—QFC short—term debt will be ranked as follows:





              

Policymaker 2 agrees with Policymaker 1 on the problems that might arise when attempting to
raise DIP financing and about the risk of contagion to the failing firm’s short—term creditors if they
are unable to withdraw funds during a stay. However, Policymaker 2 diﬀers from Policymaker 1 in
that he does not worry about firesale eﬀects so the third impact (firesale eﬀects) is not relevant for
him.
In summary, the ranking of weights for Policymaker 2 coincides with those for Policymaker 1,
but the weights will be lower because he does not worry about firesale eﬀects.
37

There is nothing in the bankruptcy code that prevents the Treasury from providing DIP financing in a bankruptcy
process. There are, however, very few examples of this. Some notable cases were the GM and Chrysler bankruptcy
process in 2009. There are several reasons that should make this financing from the Treasury more diﬃcult and
slower to obtain than under OLA. First, trustees of the failing company need to argue in front of the bankruptcy
judge that ordinary course financing is not available in the market (Reference GM arguing in front of the courts).
Second, any funding approved by the Treasury would be subject to the usual scrutiny in the U.S. political system
and could come at a high political cost to the administration.

Weights on the ratio of non—QFC Short—term Funding to assets As already noted, this
ratio is informative about losses to long—term debt holders; a troubled firm will be forced to liquidate
assets as short—term financing is withdrawn prior to failure (at least to the extent that short—term
creditors are aware of the firm’s troubles). Policymaker 1 believes that there might be a firesale
eﬀect that drives down the price of the assets being sold to cover short—term debt withdrawals.
As far as the eﬀect of a run on short—term debt in the days leading up to failure, there should
be little or no diﬀerence between bankruptcy and OLA in the amount of loss to long—term debt
holders, given that non—QFC short—term funding withdrawals can only occur before resolution and
both bankruptcy and OLA resolution impose a stay on withdrawals. On the other hand, once
the firm fails, assets will need to be liquidated in order to repay creditors. If OLA modifies debt
priority rules to favor short—term creditors to the detriment of long—term creditors (which Dodd—
Frank allows),38 then long—term debt holders may be worse oﬀ under OLA than under bankruptcy,
where debt priority rules may result in a split of losses that is more favorable to long—term debt
holders. However, if OLA makes use of the OLF to protect short—term creditors, less liquidation
may be needed. Depending on the extent to which OLF is used, this latter eﬀect may dominate,
making losses less of a concern under OLA. In our weights we assume that this latter eﬀect does
indeed dominate. As usual, it seems reasonable to assume these concerns of contagion are larger
in an aggregate state, and we assume for illustration purposes that the weights for both resolution
methods are the largest in an aggregate shock. Policymaker 1’s weights, then, satisfy the following
ranking:





              
The ranking of weights for Policymaker 2 will be the same, but Policymaker 2’s weights will be
lower, especially in an aggregate shock, because he is not concerned about firesale eﬀects.
Weights on the inverse of the Liquidity Coverage Ratio As we argued above, a firm with
a higher “inverse of the LCR” will suﬀer more losses if its short—term debt holders run prior to its
failure. Such a firm will also suﬀer more losses if its QFCs get liquidated thanks to an exemption to
the stay once the firm is in bankruptcy. Hence, a high inverse of the LCR translates into contagion
to long—term debt holders because there are fewer resources remaining in the firm. Note that we
are already penalizing firms that have high ratios of QFC/Assets and Non—QFC debt/assets. The
reason to include the inverse of the LCR is that, for two firms with similar values of these two
ratios, long—term creditors of the firm with the higher inverse will suﬀer more losses in the event
of failure. Hence the ranking weights here will mimic that of the other two characteristics:




 1
 1
 1

1

Also, Policymaker 1 will have higher weights than Policymaker 2 because of his extra concern
with firesales.
38

See Pellerin and Walter (2012), p. 11, 16-17.

5.1

Resolvability determinations under three types of policymakers

Figure 11 presents the results for our examples: they show how these diﬀerent sets of weights
would translate into diﬀerent designations of firms’ structures as acceptable. The actual scores
that determine these designations are included in a table in the Appendix. For this exercise, we
normalize each firm’s entry for each characteristic by the sum of that characteristic for the 34 firms
that we have in our sample for the year 2015.
These illustrative examples are constructed so that for each of the four policymakers a diﬀerent
list of firms is considered resolvable. We now discuss the commonalities and disagreements as a
way to better understand the sensitivity of the score to the policymaker’s weights. For Policymaker
1(a), by construction, all firms are considered acceptable. Hence, we will focus on disagreements
among the other Policymakers: P1(b), P2 and P3. An Excel spreadsheet that computes our score
is available as an online appendix, so that readers can modify weights themselves and explore
implications for the list of firms.
First, note that all disagreements between P1(b) and P2 (columns 2 and 3 in the table of Figure
11) can be tracked to disagreements about the weights in our score that capture impact due to
firesales (see weights in red in Figure 10). Most of these disagreements relate to debt structure.
Looking at the detailed scores for the firms (listed in Figure 13 in the appendix) we see that for U.S.
Bancorp, PNC Financial Services, and Capital One, OLA is the preferred choice for P1(b) in an
aggregate shock, while P2 chooses bankruptcy. Disagreements about the importance of firesales will
be most important for middle—sized firms, for which the scores on debt—related characteristics can
make a meaningful diﬀerence in their final score. In other words, P1(b) and P2 disagree about the
best method to resolve U.S. Bancorp, PNC Financial Services, and Capital One Financial because,
for the characteristics for which these policymakers agree on weights, the measured values are
low enough that a disagreement on the weights of interconnectedness, complexity, and new debt—
related terms tilts the comparison between costs of bankruptcy and costs of OLA. The policymaker
not concerned with firesales (P2) gives these three firms much lower partial scores based on their
interconnectedness, complexity, and debt structure in bankruptcy, and this makes the bankruptcy
score the lowest and the firms resolvable. For the largest six companies, on the other hand, their
high numbers on size, substitutability, and cross—jurisdictional activity drive the comparison of the
scores, even for the policymaker who discounts the fire—sale eﬀect from the rest of the characteristics.
This can also be understood by focusing on HSBC and Bank of New York Mellon (BoNY), which
are similar in size to PNC. While the partial score attached to short—term debt items is also low
for these two firms when fire sales are not considered, their measures of complexity (HSBC) and
substitutability (BoNY) are at least one order of magnitude higher than those of PNC and hence
the relative disadvantages of bankruptcy for dealing with complexity and substitutability end up
implying that these firms are not resolvable. Lastly, a firm such as Deutche Bank Trust Corporation,
of relatively small size in this sample, would normally be considered resolvable but it has very high
entries for non—QFC short—term debt to assets ratio, and for illiquidity of its assets, so P1 considers
OLA the best method to resolve it in an aggregate shock.
To see the diﬀerence that looking beyond size can make in resolvability determinations, we compare Policymaker’s (3) results with the rest of the columns. Policymaker 3 considers the structure of
42

Figure 11: Resolvability determinations. Firms are ordered from largest to smallest according
to Total Exposures. A firm’s structure is acceptable if the impact score under Bankruptcy is
smaller than under OLA or a bailout in both crises scenarios. TRUE = acceptable, FALSE =
non–acceptable.

TD Group non—acceptable (OLA is his preferred resolution alternative in an aggregate shock), while
Policymakers 1(b) and (2) think of it as acceptable. The reason for the diﬀerence is that the firm is
fairly large and therefore receives a high impact score from Policymaker 3, who discriminates only
based on size. At the same time, it has fairly low readings on their non—size characteristics (such as
Interconnectedness, Substitutability, and Non—QFC Short—Term Debt/Assets), which leads Policymakers 1(b) to view the firm as resolvable in bankruptcy. In general, for Policymaker 3, OLA will
be the preferred option for firms with a middle—of—the—range size: OLA has a lower impact weight,
but it does carry a moral hazard impact as well. A simple way to rationalize this is to think of the
parameters of the linear function of size that is behind the score: the impact of size under OLA has
a lower slope, but a higher intercept, while the bailout score is independent of size. This implies
two thresholds in size that determine the firms that will be best resolved under bailout, OLA, and
bankruptcy.
It is interesting to compare our Policymaker 1(a) determination of adequate structure with the
determinations of the FDIC and the Board about the credibility of LWs. For the 2015 review
process, either the Board or the FDIC (or both) determined that the LWs of most of the firms
being reviewed that year had deficiencies important enough to fail the review process.39 Neither of
the two agencies found deficiencies in Citigroup’s LW. To square our results, which indicate that for
Policymaker 1(a) all firms have an acceptable structure, with the fact that these LW reviews found
deficiencies, it is necessary to recall that one important limitation of our exercise is that we lack
information on the “plan” in the LWs. As a result of this limitation, some deficiencies that are not
necessarily associated with characteristics cannot be captured in our impact score. For example,
one deficiency found in Bank of America’s LW was the failure to specify clear triggers under which
the parent company would provide liquidity to its subsidiaries, which is a plan item rather than a
firm characteristic that we can pick up in our financial—report—based score.40
As a final comment on our results, we want to emphasize again that our main contribution in
this paper is conceptual. The exercise presented in Figure 11 is just for illustration and our score
is far from being ready for policy use. More work is needed both to expand the list of scenarios
considered in it, to include any missing relevant characteristics of the firms, and to have the weights
be empirically based.

5.2

Comparison to the GSIB score

Another way of understanding what the new firm characteristics add to the determination of resolvability is to compare the ranking of firms according to the Basel GSIB score with our impact
score. We choose the comparison with the bankruptcy aggregate impact because that is likely to
be the assumption under which Basel understands the weights they use. Figure 12 presents the
results. The main diﬀerences in ranking are for smaller firms: while the GSIB score ranks them low,
the impact scores assign them more impact due to their relatively high numbers in the short-term
debt ratios and liquidity measures. This means that the smallest firm in the sample, Deutche Bank
39

The list of firms whose LWs had deficiencies in 2015 is: Bank of America, Bank of New York Mellon, JP Morgan
Chase, State Street, Wells Fargo, Goldman Sachs, and Morgan Stanley.
40
Bank of America Living Will Feedback Letter (2016).

Trust Corporation, is raked 13 out of 34 according to our impact score, while it was 34 under the
GSIB.

Extensions

In this section, we highlight a few other firm characteristics that would be relevant to calculate
the impact of a failure. These are characteristics that are diﬃcult to measure and we leave their
inclusion in the impact score for future research: information on counterparties, data on network
structure, measures of impact outside of the financial system, legal complexity, core businesses and
services, and aspects of capital structure.

6.1

Information about counterparties

In the LWs, filers are asked to identify their “major counterparties . . . and describe the interconnections, interdependencies and relationships with such major counterparties.”41 Schedule B
(Interconnectedness Indicators) can be thought of as a “rough,” publicly available, measure of
these interconnections. Although privacy concerns must be taken into account, we think it would
be beneficial to construct, in the spirit of the LCR, a summary measure, based on average trades,
that captures the importance of the filing BHC to other financial institutions or to companies
outside of the financial system. For example, given that an individual firm knows to whom it is
lending and, and from these firms’ 10-Ks, the total amount of borrowing of each of its borrowers,
a firm could fairly easily estimate the percent of each of its major borrowers’ total funding that it
represents. Then, the firm could be required to report the total number of companies for which it
represents more than a threshold percent. Presumably, reporting of such a count would not reveal
private information.

6.2

Network structure

Even though there is information about the overall amount of interconnectedness in Schedule B, the
schedule ignores the network structure of counterparties, which is important for financial system
resilience in the event of failure (Allen and Gale, 2000). The recent literature on networks has
developed summary statistics that help characterize contagion propensity of diﬀerent classes of
networks (see, for example, Gai and Kapadia, 2007). Including such statistics in the score, if
feasible, would be informative about contagion eﬀects. Further work is necessary to establish the
optimal statistic and to deal with information revelation concerns.

6.3

Interconnectedness outside the financial system

We care greatly about contagion within the financial system because disturbances to the financial
system are thought to have a greater impact on economic output than disturbances in other sectors.
41

“Federal Reserve System Reporting Requirements Associated with Regulation QQ (Resolution Plans Required), Model Template for §165(d) Tailored Resolution Plan.” January 31, 2016, p. 18. Available at:
https://www.federalreserve.gov/bankinforeg/resolution-plans/tailored-resolution-model-template.pdf.

Figure 12: Comparison of the score and the ranking of firms according to the GSIB score and our
impact score, under two diﬀerent policymaker’s weights.

Accordingly, Schedule B is focused on capturing the direct impact of the failure of the reporting
firm for the financial system and holders of GSIB—issued securities. However, we think it would be
valuable to capture more generally the impact on the rest of the economy. At some level, the TE
measure can be thought of as a measure of potential contagion to all the economy, but nevertheless,
more specific information on the importance of a firm for the rest of the economy would be worth
collecting and including in the score. For example, we would estimate higher impact if the GSIB is
a major payments provider to a major nonfinancial firm or a major lender to the airline industry.

6.4

Legal complexity, critical operations, and core businesses

We analyzed the feedback letters arising from the 2016 LW review process to look for firm characteristics that were singled out as problematic for resolution (as opposed to shortcomings in the
plans for resolution). We found that regulators highlighted, in letters to several banks, an overly
complex legal entity structure (JP Morgan, State Street, and Wells Fargo). One concern, noted in
letters, is the location of entities in multiple jurisdictions. Another concern was the existence of
cross—guarantees, whereby a parent company bears unlimited liability for contracts created by its
subsidiaries. Such guarantees produce strong links between the solvency of diﬀerent subsidiaries,
which may even be in diﬀerent jurisdictions, and can create complications during resolution due
to the possibility of ring—fencing (whereby one jurisdiction prevents assets located within its border from being used to repay debt holders in a diﬀerent jurisdiction). Given these examples, we
would expect a measure of this legal complexity to be included in the GSIB score; but no such
measure is incorporated, probably due to the diﬃculty constructing quantitative and informative
measures that can be comparable across firms and over time. This is an area of work that should
be prioritized when thinking about resolution.42
LWs also require firms to identify both the core businesses of the company and the critical operations that it performs for the financial system. While it appears that information regarding critical
operations is already captured in the GSIB score, we argue that information on core businesses is
not captured.
Critical operations are those, which, if shut down due to firm failure, would produce a strong
negative economic impact. To a large extent, the designation of critical operations is independent of
the firm’s structure and instead reflects the services provided to the economy. In our view, Schedule
C, Substitutability, which records the volume of payments, assets under custody, and underwriting
of securities, constitutes a reasonable measure of the importance of a firm’s critical operations.
Core businesses are defined as business lines and associated services that are important sources
of revenue for the firm. In the event of a wind down, knowing these sources could be helpful, so that
they can be protected as much as possible. As an example, IT for a whole BHC may be located in
one single subsidiary. In a scenario in which this particular subsidiary becomes insolvent, the most
orderly way of winding down the firm may involve using the parent company as a source of strength
for the troubled subsidiary, rather than enforcing firewalls (that are intended to prevent the spread
of the troubles of one subsidiary to others). It is clear that information about core businesses
should inform the “plan” included in LWs. In our example, a resolution plan that proposes to
42

Schedule E provides minimal information on related ring-fencing concerns. See our discussion above.

sell a troubled subsidiary to minimize its impact on the remainder of the firm might not recognize
the importance of IT as a core business line or service. However, readiness for resolution as far as
core businesses protection is concerned could also be evaluated in the firm’s structure and hence
be captured in an impact score. In our example, if a BHC replicates IT services across subsidiaries
so that subsidiaries can be sold independently, in a resolution, this should translate into a lower
impact score. As with legal complexity, at this point, it is not clear in what way such information
could be incorporated into our quantitative impact score in an objective, general, and comparable
manner, but it may be an area for future attention.

6.5

Measuring the GSIB’s capability to recapitalize systemically important
subsidiaries

Another important aspect when evaluating the impact of a failure is the ability of the firm to
recapitalize systemically important subsidiaries using either an appropriate trigger for bankruptcy
(i.e., filing with enough capital buﬀer) or through the conversion of long—term debt into capital, as
mandated by TLAC. This ability to recapitalize depends both on whether maturity transformation
takes place at the parent company or at the subsidiaries (whether the BHC is “clean” or not,
according to TLAC requirements) and on the “plan” of resolution that the firms develop. A lack
of such ability to recapitalize has been pointed out as a source of deficiency in the evaluation of
LWs.43 The guidance for preparation of LWs issued by the Board and the FDIC in April of 2016
emphasized the need for both external and internal TLAC, as well as the need to pre—position
capital and liquidity in certain subsidiaries and to have governance arrangements in place such as
early bankruptcy triggers. These are all measures aimed at ensuring the ability of the failing firm
to recapitalize all subsidiaries that perform key services in the financial system.44
Current capital ratios may be informative about a firm’s ability to recapitalize subsidiaries that
provide critical services. However, for the purpose of our impact score, specific measures of capital
needs of each particular subsidiary may be desirable. As of now, using Y-15 data, we can only
construct a score at the BHC level. More work is needed to find publicly available data to construct
indicators of the structure of liquidity and capital across subsidiaries and parent companies.

Conclusion

Making the resolution of large, systemically important firms feasible without public support (i.e.,
making the firms “resolvable”) has been a priority in recent regulatory changes. One important
eﬀort has centered on having firms describe in resolution plans (also known as living wills — LWs)
their structures and the means by which these structures, together with a wind—down strategy,
contribute to making unassisted bankruptcy the preferred method of resolution. The evaluation of
these LWs by regulators is complicated and has been deemed nontransparent to outsiders. In this
paper, we seek to develop tools, based on publicly available information, which can complement
the LW review process.
43
44

See Board of Governors and FDIC (2016a).
See Board of Governors and FDIC (2016c).

Our main contribution is to develop a conceptual framework to illustrate how to quantitatively
evaluate whether a firm is resolvable. We consider using a score that maps a firm’s financial data
into costs imposed on the economy – the impact – of resolving this failing firm via diﬀerent
resolution methods and in diﬀerent economic scenarios. For a given firm’s structure, bankruptcy
is preferable – from a certain policymaker’s point of view — to other resolution alternatives that
involve government support if the resulting impact calculated by the score is the lowest. Our
framework makes clear an important concept in the evaluation of LWs: though the failure of a
large firm is always going to be disruptive for the economy, the relevant question for resolvability
is whether the extra losses stemming from unassisted resolution in bankruptcy are larger than the
moral hazard costs of providing support.
We discuss the data needs and diﬃculties of constructing such an impact score by proposing a
simplified two—scenario version of it. For this we use as a starting point the current Basel—developed
GSIB score. This score was created as a means of designating global banks as systemic and for
setting supplemental capital requirements. It measures how much economic damage a firm might
impose by its failure based on firm characteristics (financial data). We discuss each of the firm
characteristics included in the GSIB score and evaluate what failure costs each is meant to capture.
We supplement the original list of firm characteristics in the GSIB score with information on the
structure of debt of the firm, with emphasis on short—term debt and QFCs, as well as liquidity
measures. While our objective here is to measure resolvability, our modified score can also be
useful in thinking about refinements to this macro—prudential tool. Indeed, in what we think is
a step in the right direction, the revisions to the GSIB score for the U.S. implementation of the
capital surcharge that are being implemented in 2017 somewhat expand required disclosures on the
structure of short—term financing.
The moral hazard problem behind TBTF implies that, as long as firms put a positive probability
on assistance in some states of the world, they will have incentive to change their structure or
operations in ways that exploit the safety net. It follows from our analysis that the more realistic
and complete the set of scenarios the score considers, the more useful this instrument will be in
curbing the TBTF problem. In other words, expanding the set of scenarios may help policymakers
signal credibly that a firm will not be bailed out in a wide enough range of financial distress
scenarios so that creditors believe that the safety net will not protect them from losses under most
circumstances
To illustrate the concept of an impact score we compute the simplified two—scenario version of
our score with actual firm data and examples of how policymakers may evaluate impact. With
these examples, we seek to emphasize how the decisions to bail out a failing firm depend on
policymakers’ beliefs about how a given firm’s characteristics may translate into costs to society. In
practice, to make our score a positive tool (describing actual bailout decisions) or a normative tool
(evaluating the merits of diﬀerent bailout policies) we would need empirical data from a number
of financial crises to determine the links between firm characteristics and impact from failure. Our
work highlights the need for future research along these lines.

References
[1] Allen, F., and D Gale (2000): “Financial Contagion,” Journal of Political Economy, 108(1),
1-33.
[2] Allen, F., and D Gale (1994): “Limited Market Participation and Volatility of Asset Prices,”
American Economic Review, 84(4), 933-955.
[3] Basel Committee on Banking Supervision, “Consultative Document. Global systemically important banks: Assessment methodology and the additional loss absorbency requirement”.
Available at https://www.bis.org/publ/bcbs201.pdf (2011)
[4] Board of Governors of the Federal Reserve System and Federal Deposit Insurance Corporation. “Bank of America Living Will Feedback Letter for 2015.” (2016a)
Available at: https://www.federalreserve.gov/newsevents/pressreleases/files/bank-of-americaletter-20160413.pdf
[5] Board of Governors of the Federal Reserve System and Federal Deposit Insurance Corporation.
“Resolution Plan Assessment Framework and Firm Determinations.” (2016b). Available at:
https://www.federalreserve.gov/newsevents/pressreleases/files/bcreg20160413a2.pdf
[6] Board of Governors of the Federal Reserve System and Federal Deposit Insurance Corporation. “Guidance for 2017 165(d) Annual Resolution Plan Submissions by Domestic Covered
Companies that Submitted Resolution Plans in July 2015.” (2016c)
[7] Board of Governors of the Federal Reserve System and Federal Deposit
Insurance
Corporation.
Joint
Press
Release.
“Agencies
announce
joint determinations for living wills.” December 19,
2017. Available at:
https://www.federalreserve.gov/newsevents/pressreleases/bcreg20171219a.htm
[8] Covitz, Daniel, Nellie Liang, and Gustavo A. Suarez (2013). “The Evolution of a Financial
Crisis: Collapse of the Asset-Backed Commercial Paper Market,” Journal of Finance, vol. 68,
no. 3, pp. 815-848.
[9] Diamond, D. and R. Rajan (2011): “Fear of Fire Sales, Illiquidity Seeking, and Credit Freezes,”
Quarterly Journal of Economics, 126(2) 557-591.
[10] Dodd—Frank Wall Street Reform and Consumer Protection Act; Regulatory Capital Rules:
Implementation of Risk—Based Capital Surcharges for Globally Systemically Important Bank
Holding Companies, 80 Fed. Reg. 49082 (August 14, 2015) (to be codified at 12 CFR pts. 208
& 217).
[11] Duﬃe, D. (2010) “How Big Banks Fail and What to Do about It”.
[12] Duﬃe, D. (2014) “Financial Market Infrastructure: Too Important to Fail” HOOVER INSTITUTION, Economics Working Paper 14101
50

[13] Ennis and Price (2015), Discount Window Lending: Policy Trade-oﬀs and the 1985 BoNY
Computer Failure, Economic Brief, Federal Reserve Bank of Richmond, May 2015, No. 15-05.
[14] Gai, P., and S. Kapadia: “Contagion in Financial Networks,” the Proceedings of the royal
Society of London, p.2401-2423 (2010).
[15] GAO. April 2016. “Resolution Plans”. http://www.gao.gov/assets/680/676497.pdf
[16] Hoenig, Thomas. March 2, 2015. “Remarks by Thomas M. Hoenig, Vice Chairman, Federal
Deposit Insurance Corporation: Financial Supervision: Basic Principles Presented to the Institute of International Bankers Annual Washington Conference; Washington, DC”
[17] Jackson, Thomas H. July 9, 2014. “Building on Bankruptcy: A Revised Chapter 14 Proposal
for the Recapitalization, Reorganization, or Liquidation of Large Financial Institutions.” Manuscript, Stanford University.
[18] Jarque, Arantxa, and K. Athreya. “Understanding Living Wills”. Federal Reserve Bank of
Richmond Economic Quarterly, Third Quarter 2015.
[19] Kocherlakota, Narayana. “Taxing Risk and the Optimal Regulation of Financial Institutions.” Economic Policy Papers. Federal Reserve Bank of Minneapolis. May 18, 2010. Available at: https://www.minneapolisfed.org/research/economic-policy-papers/taxing-risk-andthe-optimal-regulation-of-financial-institutions.
[20] Krimminger, Michael H. “Does the Dodd-Frank Act End Too Big to Fail?” Statement before the Subcommittee on Financial Institutions and Consumer Credit; Financial Services
Committee; U.S. House of Representatives; Washington, D.C. June 14, 2011. Available at:
https://www.fdic.gov/news/news/speeches/archives/2011/spjun1411.html.
[21] Lacker, Jeﬀrey M., “Payment System Disruptions and the Federal Reserve Following September 11, 2001.” Journal of Monetary Economics 51, no. 5 (July 2004): 935-965.
[22] Lacker, Jeﬀrey M. “Economics After the Crisis:
Models, Markets, and Implications for Policy.” Presentation before the Center for Advanced Study in Economic Eﬃciency, Arizona State University. February 21, 2014. Available at:
https://www.richmondfed.org/press_room/speeches/jeﬀrey_m_lacker/2014/lacker_speech_20140221.
[23] Passmore, Wayne, and Alexander H. von Haﬀten (2017). “Are Basel’s Capital Surcharges
for Global Systemically Important Banks Too Small?,” Finance and Economics Discussion Series 2017-021. Washington: Board of Governors of the Federal Reserve System,
https://doi.org/10.17016/FEDS.2017.021.
[24] Pellerin, Sabrina R. and John R. Walter. “Orderly Liquidation Authority as an Alternative to
Bankruptcy.” Federal Reserve Bank of Richmond, Economic Quarterly, First Quarter, 2012.
[25] Shleifer, Andrei, and Robert W Vishny. 2011. “Fire Sales in Finance and Macroeconomics.”
Journal of Economic Perspectives 25 (1): 29-48.
51

[26] Total Loss-Absorbing Capacity, Long-Term Debt, and Clean Holding Company Requirements
for Systemically Important U.S. Bank Holding Companies and Intermediate Holding Companies of Systemically Important Foreign Banking Organizations, 82 Fed. Reg. 8266 (January
24, 2017)
[27] Wells Fargo 2015 Living Will Feedback Letter. (2016)
[28] White, Phoebe and Tanju Yorulmazer, “Bank Resolution Concepts, Trade-oﬀs, and Changes
in Practices”. Economic Policy Review, December 2014 Volume 20, Number 2

Appendix

8.1

Construction of short—term debt items in score

For the purpose of our examples of the impact score in the remainder of the paper, and because
data on Schedule G is not yet available, we construct our own measure of short—term borrowing
using Y-9C and Call Report data.45 In the following description of our calculations we provide the
item number for Y-9C figures in parenthesis.
• Short-term debt = commercial paper (bhck2309) + “Other borrowed money with a remaining
maturity of one year of less,” (bhck2332) + “fed funds and repo borrowings” (bhck3353, an
average over the last quarter) + uninsured deposits (collected from FDIC data for each
institution). Compared to the level of detail on short—term debt borrowing that the schedule
G gathers, our measure is fairly coarse. We are not able to break down short—term debt
by maturity or identify the portion of short—term debt emanating from within the financial
system. These are relevant details in the evaluation of the impact of failure. Once the revised
Y-15 form goes into eﬀect, it may be feasible to refine our impact score using this detailed
information.
• QFCs = “fed funds and repo borrowings” (bhck3353, an average over the last quarter)
+ “derivatives purchased with negative fair value (“w.n.f.v.”)” (bhckc222) + derivatives
sold w.n.f.v. (bhckc220) + derivatives (interest rate swaps) w.n.f.v., traded (bhck8737)
and non—traded (bhck8745) + derivatives (foreign exchange) w.n.f.v., traded (bhck8738)
and non—traded (bhck8746) + derivatives (equity) w.n.f.v., traded (bhck8739) and non—
traded (bhck8747) + derivatives (commodities) w.n.f.v., traded (bhck8740) and nontraded
(bhck8748)
45

The Y-9C, Consolidated Financial Statements for Holding Companies, is a regulatory report, collected by the
Federal Reserve gathers “basic financial data from a domestic bank holding company (BHC), a savings and loan
holding company (SLHC), a U.S intermediate holding company (IHC) and a securities holding company (SHC) on a
consolidated basis in the form of a balance sheet, an income statement, and detailed supporting schedules, including
a schedule of oﬀ balance-sheet items.” The Call Report, or more formally, Consolidated Reports of Condition and
Income, is completed by all U.S. banks and savings associations and is submitted to the Federal Financial Institutions
Examination Council, an interagency body representing the various U.S. banking supervisors. The Call Report gathers
detailed information on depository institution balance sheets, income statements, and oﬀ balance sheet items. The
Report includes a line item for “Estimated amount of uninsured deposits.”

— Note that, ideally, we would want to leave out of the QFC measure the amount of fed
funds borrowings. However, measurements of only repo borrowings are only available
as an end-of-quarter measure, which is subject to window-dressing concerns (i.e., that
firms may substitute this form of borrowing only on the days in which their accounting
information is used for regulatory filings, in order to get a more favorable capital requirement). Hence, we use the entry for both fed funds borrowings and repos, which is
provided as an average over the quarter. Because in recent years fed funds borrowings
have been relatively low, we think this is a reasonable approximation.
• QFC/Assets: the above measure over “Total Assets” (bhck2170)
• Non—QFC—short—term debt = our above measure of short—term debt - “fed funds and repo
borrowings” (bhck3353, an average over the last quarter)
• Non—QFC—short—term debt ratio = our above measure of Non—QFC—short—term debt over
“Total Assets” (bhck2170)
• Inverse of Liquidity Coverage Ratio = our above measure of short—term debt over our measure
of high-quality liquid assets (HQLA), following the classification used to compute the LCR,46
where
— HQLA = HQLA class 1 assets + 0.85 * HQLA class 2a assets, where, in turn,
— HQLA class 1 assets = hqla_1_1 (bhckd958) + hqla_1_2 (bhckd962) + hqla_1_3
(bhckd967) + hqla_1_4 (bhckd977),
— HQLA class 2a assets = hqla_2a_1 (bhck1295) + hqla_2a_2 (bhck1298) + hqla_2a_3
(bhckg305) + hqla_2a_4 (bhckg307) + hqla_2a_5 (bhckg379)
About the Repo measure: the list of QFCs includes repos, derivatives contracts such as swaps,
forwards, futures, and options. Because the Y-15 does not record separately the importance of
QFCs, we calculate our own measure using data from the Y-9C. Our calculation includes only
liability-related QFCs, because only GSIB liabilities can produce losses for creditors; so we include
repo borrowings and credit derivatives (sold or purchased) with a negative net value.
Although there are other types of derivatives that constitute a liability of the firm, beyond
those that we include in our calculation, only for these credit derivatives we can find their current
net value. Hence, we are underestimating the amount of liabilities in QFCs. Moreover, we have
no information about the maturity of the borrowing, or the liquidity of the collateral in these
contracts, which could be useful in considering their relative importance for contagion eﬀects and
firesale concerns, as we argue below. This information will be collected by the revised Y-15, allowing
for future refinement of our impact score.
About the inverse LCR: while the mapping between liquidity classification (HQLA) from the
LCR regulation and firesale concerns is not unambiguous, it could act as a first approximation.
46

Data on the LCR for GSIBs and large foreign organizations are collected at high frequencies by regulators, but
they are not publicly available.

For Level 1 collateral (cash, central bank reserves, certain sovereign debt, and securities backed by
central banks) firesales are likely to be of little concern, because the total market volumes of Level
1 items are probably too large for a single institution to aﬀect their price through a rushed sale. For
Level 2a assets (securities or debt issued by government agencies and riskier sovereign debt) and
Level 2b assets (certain mortgage backed securities and corporate debt and equity), the concern
might be stronger, and probably, for our measure, we would be the most concerned with any debt
backed with assets with a rating below 2b. Because of this, we do not include them as HQLA.
The next tables present the scores we calculated for the 34 BHC in 2015, for each policymaker.

Figure 13: Scores according to the three policymakers in our examples.

Full text of Working Papers (Federal Reserve Bank of Richmond) : On the Measurement of Large Financial Firm Resolvability, Working Paper 18-06R : Working Paper

FRASER