This PDF is a selec on from a published volume from the Na onal
Bureau of Economic
Research
Volume Title: Risk Topography: Systemic Risk and Macro Modeling
Volume Author/Editor: Markus Brunnermeier and Arvind
Krishnamurthy, editors
Volume Publisher: University of Chicago Press
Volume ISBN: 0‐226‐07773‐X (cloth); 978‐0‐226‐07773‐4 (cloth);
978‐0‐226‐09264‐5 (eISBN)
Volume URL: h p://www.nber.org/books/brun11‐1
Conference Date: April 28, 2011
Publica on Date: August 2014
Chapter Title: Challenges in Iden fying and Measuring Systemic
Risk
Chapter Author(s): Lars Peter Hansen
Chapter URL: h p://www.nber.org/chapters/c12507
Chapter pages in book: (p. 15 ‐ 30)
15
1.1 Introduction
Discussions of public oversight of financial markets often make reference
to “systemic risk” as a rationale for prudent policy making. For example,
mitigating systemic risk is a common defense underlying the need for mac-
roprudential policy initiatives. The term has become a grab bag, and its lack
of specificity could undermine the assessment of alternative policies. At the
outset of this essay I ask, should systemic risk be an explicit target of mea-
surement, or should it be relegated to being a buzz word, a slogan or a code
word used to rationalize regulatory discretion?
I remind readers of the dictum attributed to Sir William Thomson (Lord
Kelvin):
I often say that when you can measure something that you are speaking
about, express it in numbers, you know something about it; but when you
cannot measure it, when you cannot express it in numbers, your knowl-
edge is of the meagre and unsatisfactory kind: it may be the beginning of
knowledge, but you have scarcely, in your thoughts advanced to the stage
of science, whatever the matter might be.
1
1
Challenges in Identifying and
Measuring Systemic Risk
Lars Peter Hansen
Lars Peter Hansen is the David Rockefeller Distinguished Service Professor in Economics,
Statistics, and the College at the University of Chicago and a research associate of the National
Bureau of Economic Research.
In writing this chapter, I benefited from helpful suggestions by Amy Boonstra, Gary Becker,
Mark Brickell, John Heaton, Jim Heckman, Arvind Krishnamurthy, Monika Piazzesi, Toni
Shears, and Stephen Stigler, and especially by Markus Brunnermeier, Andy Lo, Tom Sar-
gent, and Grace Tsiang. For acknowledgments, sources of research support, and disclosure
of the author’s material financial relationships, if any, please see http:// www .nber .org/ chapters
/ c12507.ack.
1. From lecture to the Institution of Civil Engineers, London (3 May 1883), “Electrical Units
of Measurement,” Popular Lectures and Addresses (1889), Vol. 1, 80– 81.
16 Lars Peter Hansen
While Lord Kelvin’s scientific background was in mathematical physics,
discussion of his dictum has pervaded the social sciences. An abbreviated
version appears on the Social Science Research building at the University of
Chicago and was the topic of a published piece of detective work by Merton,
Sills, and Stigler (1984). I will revisit this topic at the end of this essay. Right
now I use this quote as a launching pad for discussing systemic risk by ask-
ing if we should use measurement or quantification as a barometer of our
understanding of this concept.
One possibility is simply to concede that systemic risk is not something
that is amenable to quantification. Instead it is something that becomes
self- evident under casual observation. This is quite diVerent from Kelvin’s
assertion about the importance of measurement as a precursor to some
form of scientific understanding and discourse. Kelvin’s view was that for
measurement to have any meaning requires that (a) we formalize the concept
that is to be measured, and (b) we acquire data to support the measurement.
The need to implement new laws with expanded regulation and oversight
puts pressure on public sector research groups to develop quick ways to
provide useful measurements of systemic risk. This requires shortcuts, and it
also can proliferate superficial answers. These short- term research responses
will be revealing along some dimensions by providing useful summaries
from new data sources or at least data sources that have been largely ignored
in the past. Stopping with short- term or quick answers can lead to bad
policy advice and should be avoided. It is important for researchers to take
a broader and more ambitious attack on the problem of building quantita-
tively meaningful models with macroeconomic linkages to financial markets.
Appropriately constructed, these models could provide a framework for the
quantification of systemic risk.
In the short run, we may be limited in our ability to provide meaningful
quantification. Perhaps we should defer and trust our governmental oYcials
engaged in regulation and oversight to “know it when they see it.” I have two
concerns about leaving things vague, however. First, it opens the door to a
substantial amount of regulatory discretion. In extreme circumstances that
are not well guided by prior experience or supported by economic models
that we have confidence in, some form of discretion may be necessary for
prudent policy making. However, discretion can also lead to bad government
policy, including the temptation to respond to political pressures. Second, it
makes criticism of measurement and policy all the more challenging. When
formal models are well constructed, they facilitate discussion and criticism.
Delineating assumptions required to justify conclusions disciplines the com-
munication and commentary necessary to nurture improvements in models,
methods, and measurements. This leads me to be sympathetic to a longer-
term objective of exploring the policy- relevant notions of the quantification
of systemic risk. To embark on this ambitious agenda, we should do so with
Challenges in Identifying and Measuring Systemic Risk 17
open eyes and a realistic perspective on the measurement challenges. In what
follows, I explore these challenges, in part, by drawing on the experience
from other such research agendas within economics and elsewhere.
In the remainder of this essay: (a) I explore some conceptual modeling and
measurement challenges, and (b) I examine these challenges as they relate to
existing approaches to measuring systemic risk.
1.2 Measurement with and without Theory
Sparked in part by the ambition set out in the Dodd- Frank Act and
similar measures in Europe, the Board of Governors of the Federal Reserve
System and some of the constituent regional banks have assembled research
groups charged with producing measurements of systemic risk. Such mea-
surements are also part of the job of the newly created OYce of Financial
Research housed in the Treasury Department. Similar research groups have
been assembled in Europe. While the need for legislative responses puts pres-
sure on research departments to produce quick “answers,” I believe it is also
critical to take a longer- term perspective so that we can do more than just
respond to the last crisis. By now, a multitude of proposed measures exist
and many of these are summarized in Bisias et al. (2012), where thirty- one
ways to measure systemic risk are identified. While the authors describe
this catalog as an “embarrassment of riches,” I find this plethora to be a bit
disconcerting. In describing why, in the next section, I will discuss briefly
some of these measures without providing a full- blown critique. Moreover,
I will not embark on a commentary of all thirty- one listed in their valuable
and extensive summary. Prior to taking up that task, I consider some basic
conceptual issues.
I am reminded of Koopmans’s discussion of the Burns and Mitchell
(1946) book on measuring business cycles. The Koopmans (1947) review has
the famous title “Measurement without Theory.” It provides an extensive
discussion and sums things up saying:
The book is unbendingly empiricist in outlook. . . . But the decision not
to use theories of man’s economic behavior, even hypothetically, limits
the value to economic science and to the maker of policies, of the results
obtained or obtainable by the methods developed. (172)
The measurements by Burns and Mitchell generated a lot of attention
and renewed interest in quantifying business cycles. They served to motivate
development of both formal economic and statistical models. An unabash-
edly empirical approach can most definitely be of considerable value, espe-
cially in the initial stages of a research agenda. What is less clear is how to
use such an approach as a direct input into policy making without an eco-
nomic model to provide guidance as to how this should be done. An impor-
18 Lars Peter Hansen
tant role for economic modeling is to provide an interpretable structure for
using available data to explore the consequences of alternative policies in a
meaningful way.
In the remainder of this section, I feature two measurement challenges
that should be central to any systemic research measurement agenda. How
do we distinguish systemic from systematic risk? How do we conceptualize
and quantify the uncertainty associated with systemic risk measurement?
1.2.1 Systematic or Systemic
The terms systematic and systemic risk are sometimes confused, but their
distinction is critical for both measurement and interpretation. In sharp
contrast with the former concept, the latter one is well studied and sup-
ported by extensive modeling and measurement. “Systematic risks” are
macroeconomic or aggregate risks that cannot be avoided through diversi-
fication. According to standard models of financial markets, investors who
are exposed to these risks require compensation because there is no simple
insurance scheme whereby exposure to these risks can be averaged out.
2
This
compensation is typically expressed as a risk adjustment to expected returns.
Empirical macroeconomics aims to identify aggregate “shocks” in time
series data and to measure their consequences. Exposure to these shocks is
the source of systematic risk priced in security markets. These may include
shocks induced by macroeconomic policy, and some policy analyses explore
how to reduce the impact of these shocks to the macroeconomy through
changes in monetary or fiscal policy. Often, but not always, as a separate
research enterprise, empirical finance explores econometric challenges
associated with measuring both the exposure to the components of sys-
tematic risk that require compensation and the associated compensations
to investors.
“Systemic risk” is meant to be a diVerent construct. It pertains to risks
of breakdown or major dysfunction in financial markets. The potential for
such risks provides a rationale for financial market monitoring, intervention,
or regulation. The systemic risk research agenda aims to provide guidance
about the consequences of alternative policies and to help anticipate pos-
sible breakdowns in financial markets. The formal definition of systemic risk
is much less clear than its counterpart systematic risk.
Here are three possible notions of systemic risk that have been suggested.
Some consider systemic risk to be a modern day counterpart to a bank
run triggered by liquidity concerns. Measurement of that risk could be an
essential input to the role of central banks as “lenders of last resort” to
prevent failure of large financial institutions or groups of financial institu-
2. A more precise statement would be that these are the risks that could require compensa-
tion. In equilibrium models there typically exist aggregate risks with exposures that do not
require compensation. Diversification arguments narrow the pricing focus to the systematic
or aggregate risks.
Challenges in Identifying and Measuring Systemic Risk 19
tions. Others use systemic risk to describe the vulnerability of a financial
network in which adverse consequences of internal shocks can spread and
even magnify within the network. Here the measurement challenge is to
identify when a financial network is potentially vulnerable and the nature of
the disruptions that can trigger a problem. Still others use the term to include
the potential insolvency of a major player in or component of the financial
system. Thus systemic risk is basically a grab bag of scenarios that are sup-
posed to rationalize intervention in financial markets. These interventions
come under the heading of “macroprudential policies.” Since the Great
Recession was triggered by a financial crisis, it is not surprising that there
were legislative calls for external monitoring, intervention, or regulation to
reduce systemic risk. The outcome is legislation such as the rather cumber-
some and still incomplete 2,319 page Dodd- Frank Wall Street Reform and
Consumer Protection Act. The sets of constructs for measurement to sup-
port prudent policy making remain a challenge for future research.
Embracing Koopmans’s call for models is appealing as a longer- term
research agenda. Important aspects of his critique are just as relevant as a
commentary on current systemic risk measurement as they were for Burns
and Mitchell’s business cycle measurement.
3
1.2.2 Systemic Risk or Uncertainty
There are important conceptual challenges that go along with the use of
explicit dynamic economic models in formal ways. Paramount among these
is how we confront risk and uncertainty. Economic models with explicit sto-
chastic structures imply formal probability statements for a variety of ques-
tions related to implications and policy. In addition, uncertainty can come
from limited data, unknown models, and misspecification of those models.
Policy discussions too often have a bias toward ignoring the full impact of
uncertainty quantification. But abstracting from uncertainty measurement
can result in flawed policy advice and implementation.
There are various approaches to uncertainty quantification. While there
is well- known and extensive literature on using probability models to sup-
port statistical measurement, I expect special challenges to emerge when we
impose dynamic economic structure onto the measurement challenge. The
discussion that follows is motivated by this latter challenge. It reflects my
own perspective, not necessarily one that is widely embraced. My perspective
is consonant, however, with some of the views expressed by Haldane (2011,
2012) in his discussions of policy simplicity and robustness when applied to
regulating financial institutions.
3. One way in which the systemic risk measurement agenda is more advanced than that
of Burns and Mitchell is that there is a statistical theory that can be applied to many of the
suggested measurements of systemic risk. The ability to use “modern methods of statistical
inference” was one of the reasons featured by Koopmans for why formal probability models
are valuable, but another part of the challenge is the formal integration with economic analysis.
20 Lars Peter Hansen
I find it useful to draw a distinction between risk and alternative concepts
better designed to capture our struggles with constructing fully specified
probability models. Motivated by the insights of Knight (1921), decision
theorists use the terms uncertainty and ambiguity as distinguished from risk.
See Gilboa and Schmeidler (1989) for an initial entrant to this literature and
Gilboa, Postlewaite, and Schmeidler (2008) for a recent survey. Alternatively,
we can think of statistical models as approximations and we use such models
in sophisticated ways with conservative adjustments that reflect the poten-
tial for misspecification. This latter ambition is sometimes formulated as a
concern for robustness. For instance, Petersen, James, and Dupuis (2000) and
Hansen and Sargent (2001) confront a decision problem with a family of
possible probability specifications and seek conservative responses.
To appreciate the consequences of Knight’s distinction, consider the fol-
lowing. Suppose we happen to have full confidence in a model specifica-
tion of the macroeconomy appropriately enriched with financial linkages
needed to capture system- wide exposure to risk. Since the model specifies
the underlying probabilities, we could use it both to quantify systemic risk
and to compute so-called counterfactuals. While this would be an attractive
situation, it seems not to fit many circumstances. As systemic risk remains
a poorly understood concept, there is no “oV- the- shelf ” model that we
can use to measure it. Any stab at building such models, at least in the near
future, is likely to yield, at best, a coarse approximation. This leads directly
to the question: how do we best express skepticism in our probabilistic mea-
surement of systemic risk?
Continuing with a rather idealized approach, we could formally articulate
an array of models and weight these models using historical inputs and sub-
jective priors. This articulation appears to be overly ambitious in practice,
but it is certainly a good aim. Subjective inputs may not be commonly agreed
upon and historical evidence distinguishing models may be weak. To make
this approach operational leads naturally to a sensitivity analysis for priors
including priors over parameters and alternative models.
A model by its very nature is wrong because it simplifies and abstracts.
Including a formal probabilistic structure enriches predictions from a model,
but we should not expect such an addition to magically fix or repair the
model. It is often useful to throw other models “into the mix,” so to speak.
The same limitations are likely to carry over to each model we envision.
Perhaps we could be lucky enough to delineate a big enough list of possible
models to fill gaps left by any specific model. In practice, I suspect we can-
not achieve complete success and certainly not in the short term. In some
special circumstances, the gaps may be negligible. Probabilistic reasoning
in conjunction with the use of models is a very valuable tool. But too often
we suspect the remaining gaps are not trivial, and the challenge in using
the models is capturing how to express the remaining skepticism. Simple
models can contain powerful insights even if they are incomplete along
Challenges in Identifying and Measuring Systemic Risk 21
some dimensions. As statisticians with incomplete knowledge, how do we
embrace such models or collections of them while acknowledging skepticism
that should justifiably go along with them? This is an enduring problem in
the use of dynamic stochastic equilibrium models and it seems unavoidable
as we confront the important task of building models designed to measure
systemic risk. Even as we add modeling clarity, in my view we need to aban-
don the presumption that we can measure fully systemic risk and go after
the conceptually more diYcult notion of quantifying systemic uncertainty.
See Haldane (2012) for a further discussion of this point.
What is at stake here is more than just a task for statisticians. Even though
policy challenges may appear to be complicated, it does not follow that
policy design should be complicated. Acknowledging or confronting gaps
in modeling has long been conjectured to have important implications for
economic policy. As an analogy, I recall Friedman’s (1960) argument for a
simplified approach to the design of monetary policy. His policy prescrip-
tion was premised on the notion of “long and variable lags” in a monetary
transmission mechanism that was too poorly understood to exploit for-
mally in the design of policy. His perspective was that the gaps in our knowl-
edge of this mechanism were suYcient that premising activist monetary
policy on incomplete models could be harmful. Relatedly, Cogley et al.
(2008) show how alternative misspecification in modeling can be expressed
in terms of the design of policy rules. Hansen and Sargent (2012) explore
challenges for monetary policy based on alternative specifications of incom-
plete knowledge on the part of a so-called “Ramsey planner.” The task of
this planner is to design formal rules for implementation. It is evident from
their analyses that the potential source of misspecification can matter in the
design of a robust rule. These contributions do not explore the policy ramifi-
cations for system- wide problems with the functioning of financial markets,
but such challenges should be on the radar screen of financial regulation.
In fact, implementation concerns and the need for simple rules underlie
some of the arguments for imposing equity requirements on banks. See,
for instance, Admati et al. (2010). Part of policy implementation requires
attaching numerical values to parameters in such rules. Thus concerns
about systemic uncertainty would still seem to be a potential contributor to
the implementation of even seemingly simple rules for financial regulation.
Even after we acknowledge that policymakers face challenges in forming
systemic risk measures that could be direct and explicit tools for policy, there
is another layer of uncertainty. Sophisticated decision makers inside the
models we build may face similar struggles with how to view their economic
environments. Why might this be important? Let me draw on contributions
from two distinct strands of literature to speculate about this.
Caballero and Simsek (2010) consider models of financial networks. In
such models financial institutions care not only about the people that they
interact with, say, their neighbors, but also the neighbors of neighbors, and
22 Lars Peter Hansen
so forth. One possibility is that financial entities know well what is going on
at all nodes in the financial network. Another is that while making probabi-
listic assessments about nearby neighbors in a network is straightforward,
this task becomes considerably more diYcult as we consider more indirect
linkages, say, neighbors of neighbors of neighbors and so forth. This view
is made operational in the model of financial networks of Caballero and
Simsek (2010).
In a rather diVerent application Hansen (2007) and Hansen and Sargent
(2010) consider models in which investors struggle with alternative models
of long- term economic growth. While investors treat each of the models as
misspecified, they presume that the models serve as useful benchmarks in
much the same way as in stochastic specifications of robust control theory.
Historical evidence is informative, but finite data histories do not accurately
reveal the best model. Important diVerences in models may entail subtle
components of economic growth that can have long- term macroeconomic
consequences. Concerns about model misspecification become expressed
more strongly in financial markets in some time periods than others. This
has consequences for the valuation of capital in an uncertain environment
and on the market trade- oVs confronted by investors who participate in
financial markets. In the example economies considered by Hansen (2007)
and Hansen and Sargent (2010), what they call uncertainty premia become
larger after the occurrence of a sequence of bad macroeconomic outcomes.
In summary, the implications of systemic uncertainty whether in contrast
or in conjunction with systemic risk are both important for providing policy
advice and understanding market outcomes. External analysts, say, statisti-
cians, econometricians, and policy advisors, confront specification uncer-
tainty when they build dynamic stochastic models with explicit linkages to
the financial markets. Within dynamic models with micro foundations are
decision makers or agents that also confront uncertainty. Their resulting
actions can have a big impact on the system- wide outcomes. Assessing both
the analysts’ and agents’ uncertainties are critical components to a produc-
tive research agenda.
1.3 Current Approaches
Let me turn now to some of the recent research related to systemic risk.
Just the wide scope of the Bisias et al. (2012) survey reminds us that there
is not yet an agreed upon approach to this measurement. To me, it suggests
that identifying what measurements will be the most fruitful to support our
understanding of linkages between financial markets and the macroecon-
omy is an open issue. In a superficial way, the sheer number of approaches
would seem to address the Kelvin dictum. The problem is complex and it
has many dimensions to it and thus requires multiple measurements. But I
am doubtful that this is a correct assessment of the situation. Alternative
Challenges in Identifying and Measuring Systemic Risk 23
measures are supported implicitly by alternative modeling assumptions and
it is hard to see how the full array of measurements provides a coherent set
of tools for policy makers. Many of the measurements to date seem closer
in spirit to the Burns and Mitchell approach and fall way short of the Koop-
mans standard. From a policy perspective, I fear that we remain too close to
the Potter- Stewart “we know it when we see it” view of systemic risk.
What follows is a discussion of a few specific approaches for assessing
systemic risk along with some modeling and data challenges going forward.
1.3.1 Tail Measures
One approach measures codependence in the tails of equity returns to
financial institutions. Some form of codependence is needed to distin-
guish the impact of the disturbances to the entire financial sector from
firm- specific disturbances. Prominent examples of this include the work of
Adrian and Brunnermeier (2008) and Brownlees and Engle (2011). Measur-
ing tail dependence is particularly challenging because of limited historical
data. To obtain estimates requires implicit extrapolations from the historical
time series of returns because of the very limited number of extreme values
of the magnitude of a financial crisis. While codependence helps to identify
large aggregate shocks, all such shocks are in eVect treated as a conglomerate
when extracting information from historical evidence. The resulting mea-
surements are interesting, but they put aside some critical questions that are
needed to understand better policy advice. For example, while equity returns
are used to identify an amalgam of aggregate shocks that could induce cri-
ses, how does the mechanism by which the disturbance is transmitted to the
macroeconomy diVer depending on the source of the disturbance? Not all
financial market crises are macroeconomic crises. The big drops in equity
markets on October 19, 1987, and April 14, 2000, did not trigger major
macroeconomic declines. Was this because of the source of the shock or
because of the macroeconomic policy responses? Understanding both the
source and the mechanism of the disturbance would seem to be critical to the
analysis of policy implications. Further empirical investigation of financial
linkages with macroeconomic repercussions should be an important next
step in this line of research.
It is wrong to say that this tail- based research is devoid of theory, and in
fact Acharya et al. (2010) suggest how to use tail- risk measures as inputs
into calculations about the solvency of the financial system. Their paper
includes an explicit welfare calculation, and their use of measurements of
tail dependence is driven in part by a particular policy perspective. Their
theoretical supporting analysis is essentially static in nature, however. The
macroeconomic consequences of crises events and how they unfold over
time is largely put to the side. Instead, the focus is on providing a mea-
sure of the public cost of providing capital in order to exceed a specific
threshold. This research does result in model- based measurements of what
24 Lars Peter Hansen
is called marginal expected shortfall and systemic risk. These measurements
are updated regularly on the V-Lab web page at New York University. The
use by Acharya et al. (2010) is an interesting illustration of how to model
systemic risk and may well serve as a valuable platform for a more ambi-
tious approach.
The focus on equity calculations limits the financial institutions that can
be analyzed. The so-called shadow banking sector contains potentially
important sectors or groups of firms that are not publicly traded. One could
argue that if the monitoring targets are only SIFI’s (so- called systemically
important financial institutions), then the focus on publicly traded financial
firms is appropriate. But system- wide policy concerns might be directed at
the potential failure of collections of nonbank financial institutions includ-
ing ones that are not publicly traded and hence omitted by calculations that
rely on equity valuation measures.
1.3.2 Contingent Claims Analysis
In related research, Gray and Jobst (2011) apply what is known as contin-
gent claims analysis. This approach features risk adjustments to sectoral bal-
ance sheets while featuring the distinct roles of debt and equity. It builds on
the use of option pricing theory for firm financing where there is an under-
lying stochastic process for the value of the firm assets. Equity is a call option
on these assets and debt is the corresponding put option. Gray and Jobst
(2011) discuss examples of this approach extended to sectors of the econ-
omy including the government. In their applications, they measure sectoral
balance sheets with a particular interest in financial crises. This approach
neatly sidesteps statistical challenges by using “market expectations” and
risk- adjusted probabilities in conjunction with equity- based measures of
uncertainty and simplified models of debt obligations. Extending contingent
claims analysis from the valuation of firms to systems of firms and govern-
ments is fruitful. Note, however, if our aim is to make welfare assessments
and direct linkages to the macroeconomy, then the statistical modeling and
measurement challenges that are skirted will quickly resurface. Market
expectations and risk- neutral probability assessments oVer the advantage
of not needing to distinguish actual probabilities from the marginal utili-
ties of investors in financial markets, but this advantage can only be pushed
so far. A more fundamental understanding of the market- based “appetite
for risk” and a characterization of the macroeconomic implications of the
shocks that command large risk prices require further modeling and a more
prominent examination of historical evidence. Such an understanding is
central when our ambition is to engage in the analysis of counterfactuals
and hypothetical changes in policies.
4
4. The potential omission of firms not publicly traded limits this approach for the reasons
described previously.
Challenges in Identifying and Measuring Systemic Risk 25
1.3.3 Network Models
Network models of the financial system oVer intriguing ways to sum-
marize data because of its focus on interconnectedness. These models open
the door to some potentially important policy questions, but there are some
critical shortcomings in making these models fully useful for policy. A finan-
cial firm in a network may be highly connected, interacting with many firms.
Perhaps these links are such that the firm is “too interconnected to fail.” A
critical input into a policy response is how quickly the networks structure
will evolve when such a firm fails. As is well recognized, in a dynamic setting
these communications links will be endogenous, but this endogeneity makes
modeling in a tractable way much more diYcult and refocuses some of the
measurements needed to address policy concerns.
1.3.4 Dynamic, Stochastic Macroeconomic Models
Linking financial market disruption to the macroeconomy requires more
than just using “oV- the- shelf ” dynamic stochastic equilibrium models, say,
of the type suggested by Christiano, Eichenbaum, and Evans (2005) and
Smets and Wouters (2007). By design, models of this type are well suited
for econometric estimation and they measure the consequences of multiple
shocks and model explicitly the transition mechanisms for those shocks.
Identification in these multishock models is tenuous. More importantly
they are “small- shock” models. In order to handle a substantial number
of state variables, they appeal to small noise approximations for analyti-
cal tractability. Since the financial crisis, there has been a rush to integrate
financial market restrictions into these models. Crises are modeled as times
when ad hoc financial constraints bind.
5
To use local methods of analysis,
separate approximations are made around crisis periods. See Gertler and
Kiyotaki (2010) for a recent development and discussion of this literature.
There is some promising recent research developing and applying compu-
tational methods that allow for a more global approach to analyzing non-
linear dynamic economic models. More application and experience with
such methods should open the door to a better understanding of stochastic
models with linkages between financial markets and the macroeconomy.
Enriching dynamic stochastic equilibrium is a promising research agenda,
but this literature has only scratched the surface on how to extend these
models to improve our understanding of the macroeconomic consequence
to upheaval in financial markets. It remains an open research question as
to how best (a) to model financial constraints, both in terms of theoretical
grounding and empirical importance; (b) to characterize the macroeconomic
5. I use the term ad hoc in a less derogatory manner than many other economists. I re-
mind readers of a dictionary definition: concerned or dealing with a specific subject, purpose,
or end.
26 Lars Peter Hansen
consequences of crisis level shocks that are very large but infrequent; and
(c) to model the origins of these shocks.
6
1.3.5 Pitfalls in Data Dissemination and Collection
Measurement requires data. Going forward, there is great opportunity
for the OYce of Financial Research in the United States and its counter-
parts elsewhere to provide new data for researchers. Some of the data in its
most primitive form will be confidential. Concern for confidentiality will
create challenges for sharing this information with external researchers. One
approach is to restrict the use of such data to be “in house.” This will limit the
value of the data collection. If the objective is to ensure the high quality of
research within government agencies, it is valuable to make important com-
ponents of the data available to external researchers. This external access
permits replication of results, and nurtures innovative modeling and mea-
surement.
7
Moreover, external analysis can provide a check against research
with preordained conclusions or inadvertent support for policies such as
“too big (or too something) to fail.” While providing external access requires
that data be distributed in manners that respect individual confidentiality,
the possibility of making such data available is a reality. The Census Bureau
has already confronted such challenges successfully.
There are additional data issues that require scrutiny. Distortions in the
collection of publicly available data can hinder the measurement of aggre-
gate risk exposures because of the temptation to disguise the problematic
nature of policies in place. Moreover, even when intentions are good, pre-
existing policies can make the assessment of risk using historical data more
challenging by partially mitigating risks in ways that are not sustainable in
the future. Brickell (2011) identifies this latter challenge and argues that it
may have contributed to errors in assessing housing market risk in the years
before the Great Recession. These types of concerns place an extra burden
on empirical researchers to model the biases in data collection induced by
both public and private incentives for distortion.
Given this state of econometric modeling and measurement, I see a big
gap to fill between statistical analyses measuring comovements in the tails
of financial market equity returns and empirical analyses measuring the
impact of shocks to the macroeconomy. This gap limits, at least temporar-
ily, the scope of the analysis of systemic risk. Closing this gap provides an
important opportunity for the future. The compendium of systemic risk
measures identified in Bisias et al. (2012) should be viewed merely as an inter-
6. For instance, the Macroeconomic Financial Modeling group funded by the Alfred P. Sloan
Foundation explores the challenges to building quantitatively ambitious models that address
these and other related challenges.
7. Andy Lo has made the related point that potentially relevant sectors, such as the insurance
sector, are not under the formal scrutiny of the federal government and hence there may be an
important shortfall in the data available to the OYce of Financial Research.
Challenges in Identifying and Measuring Systemic Risk 27
esting start. We should not lose sight of the longer- term challenge to provide
systemic risk quantification grounded in economic analysis and supported
by evidence. The need for sound theoretical underpinnings for producing
policy- relevant research identified by Koopmans many decades ago still
applies to the quantification of systemic risk. Policy analysis stemming from
econometric models aims to push beyond the realm of historical evidence
through the use of well- grounded economic models. It is meant to provide a
framework for the conduct of hypothetical policies that did not occur during
the historical observation period. To engage in this activity with the ambi-
tion to understand better how to monitor or regulate the financial sector to
prevent major upheaval in the macroeconomy requires creative adjustments
in both our modeling and our measurement.
1.4 Conclusion
We should not underestimate the diYculty of measuring systemic risk in
a meaningful way. But success oVers the potential for valuable inputs into
policy making. Wearing my econometrician’s hat has led me to emphasize
measurement challenges and the associated uncertainty caused by limited
data or unknown statistical models used to generate the data. Of course
clever econometricians can always invent challenges, and in many respects
part of the econometrician’s job is to provide credible ways to quantify
measurement uncertainties. After all, quantitative research in economics
grounded by empirical evidence should be more than just reporting a single
number but instead ranges or distributions that include sensitivity to model
specification. Good econometrics is supported simultaneously by good eco-
nomics and good statistics. Exploring the consequences of potential model
misspecification necessarily requires inputs from both economics and statis-
tics. Economic models help us understand what statistical inputs are most
consequential to economic outcomes and good statistics reveal where the
measurements are least reliable. Moreover, such econometric explorations
will benefit discussions of policy by providing repeated reminders of why
gaps in our knowledge can have important implications.
Allow me to close by returning to the Kelvin dictum and drawing on some
intellectual history of it as it relates to social science research. The decision
to place this dictum on the Social Science Research building at the Univer-
sity of Chicago caught the attention of some distinguished scholars. This
building housed the economics department for many years and the Cowles
Commission for Research in Economics during the years 1939 to 1955 when
many young scholars came there to explore linkages between economics,
mathematics, and statistics.
8
Two of the original pillars of the “Chicago
8. After moving to Yale in 1955, the Cowles Commission was renamed the Cowles Foun-
dation.
28 Lars Peter Hansen
school,” Knight and Viner, had notable reactions to the use of the Kelvin
quote and proposed amendments:
9
Knight: If you cannot measure a thing, go ahead and measure it anyway.
Viner: and even when we can measure a thing, our knowledge will be
meager and unsatisfactory.
Perhaps just as intriguing as Knight’s and Viner’s scepticism are the major
challenges that were levied to Lord Kelvin’s own calculations about the age
of the sun. These challenges provide an object lesson in support of model
uncertainty. Kelvin argued that the upper bound of the sun’s age was twenty
to forty million years, although his earlier estimates included the possibil-
ity of a much larger number, up to 100 million years. Kelvin’s evidence
and that provided by others were used to question the plausibility of the
Darwinian theory of evolution. Darwin’s own calculations suggested that
much more time was needed to justify the evolutionary processes. In hind-
sight, Lord Kelvin’s estimates were over one hundred times lower than the
current estimate of 4.5 billion years. Kelvin’s understatements were revised
upward by substantive advances in our understanding of radioactivity as
an energy source. This historical episode illustrates rather dramatically an
impact of model uncertainty on the quality of measurement. While Knight’s
and Viner’s words of caution were motivated by their perception of social
science research several decades ago, their concerns extend to other research
settings as well. It is diYcult to fault Lord Kelvin for not anticipating the
discovery of a new energy source. Nevertheless, I do not wish to conclude
that the potential for model misspecification should induce us to abandon
earnest attempts at quantification. Instead quantification should be a val-
ued exercise, and part of this exercise should include a characterization of
sensitivity to alternative model specifications. Unfortunately, there are no
guarantees that we have captured the actual form of the misspecification
among the possibilities that we consider, but at least we can avoid some of
the pitfalls of using models in naive ways.
Quantitative ambitions have the virtue of providing clarity for what is
to be measured. Models provide measurement frameworks and facilitate
communication and criticism. While simple quantifications of systemic risk
may be a naive hope, producing better models to support policy discussion
and analysis is a worthy ambition. Building a single consensus model is
unrealistic in the near term, but even exploring formally the consequences
of alternative models adds discipline to policy advice. Without such model-
ing pursuits, we are left with a heavy reliance on discretion in governmental
course of action. Perhaps discretion is the best we can do in some extreme
circumstances, but formal analysis should provide coherency and transpar-
ency to economic policy.
9. See Merton, Stills, and Stigler (1984).
Challenges in Identifying and Measuring Systemic Risk 29
While systemic- risk modeling and measurement is a promising research
agenda, caution should prevail about the impact of model misspecification
on the measurements and the consequences of those measurements. A criti-
cal component to this venture should be to assess and guard against adverse
impacts of the use of measurements from necessarily stylized models. Com-
plete success along this dimension is asking too much, otherwise we would
just “fix” our models. Nevertheless, confronting the various components
of uncertainty with some formality will help us to use models in sensible
and meaningful ways. As our knowledge and understanding advance over
time, so will our comprehension and characterization of uncertainty in our
model- based, quantitative assessments.
References
Acharya, Viral V., Christian Brownlees, Robert Engle, Farhang Farazmand, and
Matthew Richardson. 2010. “Measuring Systemic Risk.” In Regulating Wall
Street, edited by Viral V. Acharya, Thomas F. Cooley, Matthew Richardson, and
Ingo Walter, 85– 119. Hoboken, NJ: John Wiley and Sons, Inc.
Admati, Anat R., Peter M. DeMarzo, Martin F. Hellwig, and Paul Pfleiderer. 2010.
“Fallacies, Irrelevant Facts, and Myths in the Discussion of Capital Regulation:
Why Bank Equity Is Not Expensive.” Research Papers 2065. Stanford University,
Graduate School of Business.
Adrian, Tobias, and Markus K. Brunnermeier. 2008. “CoVaR.” Technical report,
Federal Reserve Bank of New York, StaV Reports.
Bisias, Dimitrios, Mark Flood, Andrew W. Lo, and Stavros Valavanis. 2012. “A
Survey of Systemic Risk Analytics.” Working Paper 0001, OYce of Financial
Research, US Department of Treasury.
Brickell, Mark. 2011. “Lessons Gleaned from Flawed Mortgage Risk Assessments.”
Financial Times, April 13.
Brownlees, Christian, and Robert Engle. 2011. “Volatility, Correlation and Tails for
Systemic Risk Measurement.” Technical report. Available at: http:// ssrn .com
/ abstract=1611229 .
Burns, Arthur F., and Wesley C. Mitchell. 1946. Measuring Business Cycles. New
York: National Bureau of Economic Research.
Caballero, Ricardo, and Alp Simsek. 2010. “Fire Sales in a Model of Complexity.”
Working Paper no. 09-28, Department of Economics, MIT.
Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans. 2005. “Nomi-
nal Rigidities and the Dynamic EVects of a Shock to Monetary Policy.” Journal
of Political Economy 113 (1): 1– 45.
Cogley, Timothy, Riccardo Colacito, Lars Peter Hansen, and Thomas J. Sargent.
2008. “Robustness and U.S. Monetary Policy Experimentation.” Journal of
Money, Credit and Banking 40 (8): 1599– 623.
Friedman, Milton. 1960. A Program for Monetary Stability. New York: Fordham
University Press.
Gertler, Mark, and Nobuhiro Kiyotaki. 2010. “Financial Intermediation and Credit
Policy in Business Cycle Analysis.” In Handbook of Monetary Economics, vol. 3,
edited by Benjamin M. Friedman and Michael Woodford, 547– 99. Amsterdam:
Elsevier.
30 Lars Peter Hansen
Gilboa, Itzhak, and David Schmeidler. 1989. “Maxmin Expected Utility with Non-
Unique Prior.” Journal of Mathematical Economics 18 (2): 141– 53.
Gilboa, Itzhak, Andrew W. Postlewaite, and David Schmeidler. 2008. “Probability
and Uncertainty in Economic Modeling.” Journal of Economic Perspectives 22 (3):
173– 88.
Gray, Dale F., and Andreas A. Jobst. 2011. “Modelling Systemic Financial Sector
and Sovereign Risk.” Sveriges Riksbank Economic Review 2:68– 106.
Haldane, Andrew G. 2011. “Capital Discipline.” Technical report, Bank of England.
Based on a speech given to the AEA meeting in Denver, Colorado. January.
———. 2012. “The Dog and the Frisbee.” Paper presented at the Federal Reserve
Bank of Kansas City’s 36th Economic Policy Symposium, Jackson Hole, Wyo-
ming. August.
Hansen, Lars Peter. 2007. “Beliefs, Doubts and Learning, Valuing Macroeconomic
Risk.” American Economic Review 97 (2): 1– 30.
Hansen, Lars Peter, and Thomas J. Sargent. 2001. “Robust Control and Model
Uncertainty.” American Economic Review 91 (2): 60– 6.
———. 2010. “Fragile Beliefs and the Price of Uncertainty.” Quantitative Econom-
ics 1 (1): 129– 62.
———. 2012. “Three Types of Ambiguity.” Journal of Monetary Economics 59 (5):
422– 445.
Knight, Frank H. 1921. Risk, Uncertainty, and Profit. Boston: Hart, SchaVner and
Marx; Houghton MiZin Co.
Koopmans, Tjalling C. 1947. “Measurement without Theory.” Review of Economics
and Statistics 29 (3): 161– 72.
Merton, Robert K., David L. Sills, and Stephen M. Stigler. 1984. “The Kelvin Dic-
tum and Social Science: An Excursion into the History of an Idea.” Journal of the
History of Behavioral Sciences 20:319– 31.
Petersen, Ian R., Matthew R. James, and Paul Dupuis. 2000. “Minimax Optimal
Control of Stochastic Uncertain Systems with Relative Entropy Constraints.”
IEEE Transactions on Automatic Control 45:398– 412.
Smets, Frank, and Rafael Wouters. 2007. “Shocks and Frictions in US Business
Cycles: A Bayesian DSGE Approach.” American Economic Review 97 (3): 586–
606.
Dostları ilə paylaş: |