Physica A 389 (2010) 3503–3507

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

Risk externalities and too big to fail

Nassim N. Taleb, Charles S. Tapiero ∗
Department of Finance and Risk Engineering, The Research Center for Risk Engineering, New York University Polytechnic Institute, 11201 New York, United States

article info abstract

Article history: This paper establishes the case for a fallacy of economies of scale in large aggregate
Received 14 May 2009 institutions and the effects of scale risks. The problem of rogue trading and excessive risk
Received in revised form 1 November 2009 taking is taken as a case example. Assuming (conservatively) that a firm exposure and losses
Available online 20 March 2010
are limited to its capital while external losses are unbounded, we establish a condition for
a firm not to be allowed to be too big to fail. In such a case, the expected external losses
Keywords:
second derivative with respect to the firm capital at risk is positive. Examples and analytical
Risk management
Tail risks
results are obtained based on simplifying assumptions and focusing exclusively on the risk
Corporate finance externalities that firms too big to fail can have.
Quantitative finance © 2010 Elsevier B.V. All rights reserved.

1. Introduction

‘‘Too Big to Fail’’ is a dilemma that has plagued economists, policy makers and the public at large. The lure for ‘‘size’’
embedded in ‘‘economies of scale’’ and Adam Smith factories have important risk consequences that have not always been
assessed at their proper costs or properly defined. The presumption that the manufacturing sector has convex production
functions has fueled the growth of enterprises to sizes that may be both too large to manage, and have losses too large to
sustain. This is the case for industrial giants such as GM that have grown into a complex and diversified global enterprise
that have accumulated costs too large to maintain. This is also the case for banks that are strategically focused and bear risks
that are often ignored. Banks draw their legal rights from a common trust, to manage the supply and the management of
money for their own and the common good. The consequences of such failures, overflowing into the commons, far outstrip
their direct losses. When banks are perceived too big to fail, they have a greater propensity to assume risks, to ‘‘rule the
commons’’, price their services unrelated to their costs or quality and exercise unduly their market power.
Size may lead such firms to assume leverage risks that are unsustainable. This is the case when banks’ bonuses are indexed
to short term performance, at the expense of hard to quantify risk externalities. These risks arise when all costs and benefits
are not incorporated by the market. Externality is therefore an expression of market failure. For banks that are too big to fail,
these risk externalities are acute. For example, Frank Rich (The New York Times, Goldman Can Spare You a Dime, October
18, 2009) has called attention to the fact that ‘‘Wall Street, not Main Street, still rules Washington’’. Similarly, Rolfe Winkler
(Reuters) pointed out that ‘‘Main Street still owns much of the risk while Wall Street gets all the profits’’. Further, a recent
study by the National Academy of Sciences has pointed out to extremely large hidden costs to the energy industry—costs
that are not accounted for by the energy industry, but assumed by the public at large.
Banks and Central Banks rather than Governments, are entrusted to manage responsibly the monetary policy—not to be
used for their own and selfish needs, not to rule the commons, but to the betterment of society and the supply of the credit
needed for a proper functioning of financial markets. A violation of this trust has contributed to a financial meltdown and
to the large consequences borne by the public at large. In this case, ‘‘too big to fail banks’’ have contributed to an immense
negative externality—costs experienced by the public at large. Thus, banks have been endowed with this trust without being

∗ Corresponding author. Tel.: +1 212 9881421; fax: +1 717 2603653.

E-mail address: ctapiero@poly.edu (C.S. Tapiero).

0378-4371/$ – see front matter © 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.physa.2010.03.014
3504 N.N. Taleb, C.S. Tapiero / Physica A 389 (2010) 3503–3507

party to the transactions that have produced such a financial meltdown. If a firm’s negative externalities are not compensated
by their positive externalities or appropriately regulated, then the social risks can be extremely damaging. In a recent New
York Times article (Sunday Business, section, October 4, 2009), Gretchen Morgenson, referring to a research paper of Dean
Baker and Travis McArthur, indicated the effects of selective failures, letting selected banks grow larger and ‘‘subsidized’’ at
a cost of over 34 Billion dollars yearly over an appreciable amount of time.
A naive optimization to size that does not recognize the nonlinearities of the risks of scale, the risks of dependence they
induce and convex their risk externalities, may lead to firms which cannot be economically sustainable [1,2]. Rather, we may
experience a risk of blowup. In fact, under any form of loss or error aversion, and concave execution costs, gains from an
increase in size should show a steady improvement in performance, punctuated with large and more losses, with a severe
increase in negative skewness [3,4].
Under a nonlinear loss function, increased exposure to rare and latent events may have the effect of raising costs of
aggregation while giving the impression of benefits — since costs will be borne during rare, but large-impact events. This
result is general. It holds not just for economic systems, but for biological, industrial and mechanical ones as well. For
example, Fujiwara [5], using an exhaustive list of Japanese bankruptcy data in 1997 (see also Refs. [6–9,4]) pointed out
to firms failure regardless of their size. Further, since the growth of firms has been fed by debt, the risk borne by large firms
seems to have increased significantly—threatening both the creditor and the borrower. In fact, the growth of size through
a growth of indebtedness combined with ‘‘too big to fail’’ risk attitudes has ushered, has contributed to a moral hazard
risk, with firms assuming non-sustainable growth strategies on the one hand and important risk externalities on the other.
Furthermore, when size is based on intensely networked firm (such as large ‘‘supply chains’’) supply chain risks (see also
Refs. [10–12]) may contribute as well to the costs of maintaining such industrial and financial organizations. Saito [13]
for example, while examining inter-firm networks noted that larger firms tend to have more inter-firms relationships
than smaller ones and are therefore more dependent, augmenting their risks. In particular, they point out that Toyota
purchases intermediate products and raw materials from a large number of firms; maintaining close relationships with
numerous commercial and investment banks; with a concurrent organization based on a large number of affiliated firms.
Such networks have augmented both dependence and supply chains risks. Such dependence is particularly acute in some
firms where one supplier may control a critical part needed for the proper function of the whole firm. For example, a small
plant in Normandie (France) with no more than a hundred employees could strike out the whole Renault complex. By the
same token, a small number of traders at AIG could bring such a ‘‘too big to fail’’ firm to a bankrupt state. This networking
growth is thus both a result and a condition for the growth to sizeable firms of scale free characteristic (see also Refs. [9,8]).
Simulation experiments to that effect were conducted by Alexsiejuk and Holyst [14] while constructing a simple model of
bank bankruptcies using percolation theory on a network of cooperating banks (see also Stauffer on percolation theory [15]).
Their simulation have shown that sudden withdrawals from a bank can have dramatic effects on the bank stability and may
force a bank into bankruptcy in a short time if it does not receive assistance from other banks. More importantly however, the
bankruptcy of a simple bank can start a contagious failure of banks concluded by a systemic financial failure. As a result, too
big to fail and its many associated moral hazard and risk externalities is a presumption that while driving current financial
policy and protecting some financial and industrial conglomerates (with other entities facing the test of the market on their
own and subsidizing such a policy), can be extremely risky for the public at large.
Size for such large entities thus matters as it provides a safety net and a guarantee by public authorities that whatever
their policies, their survivability will be ascertained for the greater good and at the expense of public funding. The rationality
‘‘too big to fail’’ is therefore misleading, based on a fallacy that negates the risk of size and does not account for the
omnipresent effects of latent, dependent and rare risks as well as their dependent moral hazard and risk externalities.
Scale is neither necessarily robust, in particular with respect to off-model risks. Under loss aversion, the gains from a
merger may show a steady improvement in performance, punctuated with large losses, with severe increases in skewness.
The essential question is therefore can economies of scale savings compensate their risks. Such an issue has been implicitly
recognized by Obama’s administration proposal in Congressional committees calling for banks to hold more capital with
which to absorb losses. The bigger the bank, the higher the capital requirement should be (New York Times, July, 27, 2009,
Editorial). However such regulation does not protect the ‘‘commons’’ from the risk externalities that banks create and the
common sustains.
To assess the effects of size and their risk externalities, this paper considers a particular and simple case based on rogue
traders’ risks and their effects on both a firm’s loss and their risk externalities. An example is used to demonstrate that rogue
trading or excessive risk taking can have significant impact on a firm risk exposure and on external losses in case of failure
— risks that augment significantly, the larger the size of the firm.

2. Too big to fail and hidden risks

Consider the event, known as the Kerviel affair, which we simplify as follows. Societe Generale lost close to $7 Billions
dollars, $6 Billions of which came mostly from the liquidations costs of the (hidden) positions of Jerome Kerviel, a rogue
trader. In addition, it contributed to external losses that we estimate something around $65 Billions, coming from the
liquidation costs of other firms reacting to the meltdown. The former are risks that the bank sustained while the latter is a
cost 10 times larger which points out to the systemic risk externalities. These externalities are side effect of the liquidation
caused the collapse of world markets by close to 12%!! These extraordinary losses did not put in question the continuity
 N.N. Taleb, C.S. Tapiero / Physica A 389 (2010) 3503–3507 3505

of Societe Generale but put an important and disproportionate strain on the financial system. This situation has generated
consequential externality losses because they were signaling a lack of controls in a bank too big to fail, unable to manage
its hidden risks and at the same time created a lack of confidence reverberating in the financial supply chain of Societe
Generale. Uncertainty regarding the system as a whole, dominated by banks presumed too powerful and too big to fail was
put in question as well. Banks, and in particular large banks, are privy to a trust to maintain the safe operation of the financial
system for the betterment of the economy. When such trusts is violated, explicitly through the behavior of their managers
or implicitly, by an unreasonable risk taking policy, uncertainty sets in, producing costs commensurate with the size of these
firms.
Consider traders’ hidden positions defined as risks that are unanticipated or of unknown exposure and resulting in
stochastic shocks. These shocks can be called ‘‘Black Swan’’ effects, as they are not part of the common (statistical) risks
foreseen by the institution or the entity involved. These shocks are assumed to be both unpredictable in a statistical sense
and therefore with large variance or jump processes with important consequences that transcend the bank. These may be
hidden risks by rogue traders, miscalculation of risk positions discovered or booking errors, or action taken underscoring an
uncontrollable risk taking culture. An ‘‘unintended position’’ is thus a hidden risk from the activities of a rogue trader that
escapes the detection by the bank officials, and needs to be liquidated as it makes the total risk larger than allowed by the
capital of the institution while at the same time contributes to a risk overflow to the financial system. This risk can be later
generalized to any form of unintentional risk – errors commonly known in the business as ‘‘long v/s long’’ or ‘‘short v/s short’’
– positions that were carried on the books with a wrong sign (and constitute the nightmare for operational risk). The vicious
aspect of these ‘‘unintended positions’’ is that the sign (long or short) does not matter; it is necessary to reduce that risk
unconditionally. Given the multiplying risk factor of large banks, these failings—even if small, assume large consequences
for the common financial system they presumably ‘‘rule’’.
Given the nature of a hidden or speculative position, we assume that the positions (in absolute value) has a potential
loss probability distribution bounded above by the firm capital (its size) W or f (x : W ) = x ∈ [0, W ). In some cases, the
risk exposure of such trades may be larger than the firm capital and therefore our assumption may be assumed to be a
conservative one. Thus, given a firm loss due to a rogue trader or due to uncontrolled risk of its trading department, we let
the total loss, including external losses be given by g (y|x), y ∈ [x, ∞) , y ≥ 0. As a result, the joint probability distribution
of global financial and firm losses is f (y, x) = g (y|x)f (x : W ), y ∈ [x, ∞) , 0 ≤ x ≤ W . The external loss of a firm whose
capital is W has thus probability and cumulative distributions:
Z W Z W Z Y
g (y) = g (y|x)f (x : W )dx and G(Y ) = g (y|x)f (x : W )dx.
0 0 x
The effects of size on the aggregate loss are thus a compounded function of the probabilities of losses of the firm and their
external costs. If a firm has a loss whose external consequences (the loss y are extremely large), then they may be deemed to
be ‘‘too big to fail’’. ‘‘Too big to fail’’ entails therefore a responsibility by the firms that ought to be regulated and controlled
extensively. In this context, ‘‘too big to fail’’ is an issue whose relevance may be measured by its risk externalities. For
example, a bank ‘‘too big to fail’’ that assumes risks for the sake of excess short terms profits that are not sustainable is in
fact misusing its charter to serve the ‘‘commons’’. Such banks are thus irresponsible ‘‘polluters’’.

3. An example

The example we consider below assumes a mixture Pareto power conditional probability distribution for all losses,
including both the firm and external losses. External losses are bounded by the firm losses from below, assumed to be
fractional in the hazard rate and bounded by its capital. In particular we have used a truncated Weibull distribution. Such an
approach differs of course from the Copula approach that models the co-dependence of losses by the marginal distribution
of each distribution. It also differs from a generalization of the Pareto distribution that accounts for a potential correlation
between the firm and the external losses. Both approaches are not be applicable as external losses depend necessarily
on the firm losses but not vice versa. Further, the use of fractal models, based on modeling a process volatility growth
with additional parameters, is for the same reasons, not applicable. Although, a firm loss can be modeled as a truncated
Wiener–Levy or fractal model, used to randomizes external losses (in conjunction with other factors, such as market liquidity
and other macroeconomic variables). The example we thus consider is of course selected for simplicity and to highlight the
effects of a firm size on the external losses.
Explicitly, say that:
γx γx ∂γx
 
g (y|x) = (y)−γx −1 , y ≥ x, E (y|x) = x , 0 < γx < 1, > 0.
(x)−γx 1 − γx ∂x
In other words, the distribution parameter may be interpreted as the ‘‘odds’’ that a firm loss has on external and global
losses. The larger the ‘‘odds’’ the larger the risk externalities. In case of the Kerviel affair, a firm loss of 7 Billion dollars had
γ7
an external loss of 65 Billion dollars. In this case, the parameter equals 7 1−γ = 65 + 7 or γ7 = 72/79 = 0.911 and
7
∂ E (y|x)
∂γx
> 0. By the same token since,
∂ E (y|x) γx ∂γx /∂ x
   
= +x > 0 and ∂γx /∂ x > 0.
∂x 1 − γx (1 − γx )2
3506 N.N. Taleb, C.S. Tapiero / Physica A 389 (2010) 3503–3507

For example, say that γx has a logit distribution, or:

eSx 1 ∂γx e−Sx
γx = F (Sx ) = = , =
2 > 0.
1 + eSx 1 + eSx ∂ Sx 1 + e−Sx
γ
Then: 1−γx
x
= eSx and E (y|x) = xeSx with Sx a score for a firm ‘‘too big to fail’’. Such a score may be defined as a function
of both the loss and economic environmental conditions. A bank whose internal loss is its capital, contribute then to an
expected loss of:
E (y|W ) = W eSW .
In other words, in case of the Kerviel affair, assuming a loss of capital of 50 Billion dollars, the external (total loss) is
E (y|50) = 514.28 Billion dollars.
The loss probability distribution is then:

γx  y −γx
Z W Z W
 −γx
Y
g (y) = f (x : W )dx and G(Y ) = f (x : W )dx − 1.
0 y x 0 x
The probability of a loss greater than Y and its hazard rate are therefore,
RW γx Y −γx
(x : W )dx
W
 −γx

f
Z
Y
2 − G(Y ) = 2 − f (x : W )dx and h(Y ) = 0 Y x
W Y −γx
.
x (x : W )dx
R

0 2− 0 x
f

If a firm’s expected external loss is E (y : W ) then if ∂ E (y : W )/∂ W > 0 and ∂ 2 E (y : W )/∂ W 2 > 0, then ‘‘size’’ contributes
to a nonlinear and increasing growth in external losses—losses that are risk externalities.
For demonstration purposes, say that the probability distribution f (x : W ) is a constrained extreme (Weibull) distribution
defined by,
 c −1
e−(x/ζ )
x c
f (x) c ζ
f (x : W ) = = .
F (W ) ζ 1 − e−(W /ζ )c
The loss probability distribution and its cumulative distribution function are then:
c ζ 1 −c W
Z
γx y−γx −1 xγx +c −1 e−(x/ζ ) dx and
c
g (y) =
ζ 1 − e−(W /ζ ) c


0

c ζ 1−c W
Z
Y −γx xγx +c −1 e−(x/ζ ) dx − 1.
c
G(Y ) =
ζ 1− e−(W /ζ )
c


0

With expected losses:

c ζ 1 −c W ∞
Z Z
γx y−γx xγx +c −1 e−(x/ζ ) dydx
c
E (y) =
ζ 1−e −(W /ζ )c


0 x

c ζ 1 −c W
x1−γx c ζ 1 −c W
γx c −(x/ζ )c
Z Z
xγx +c −1 e−(x/ζ ) dx =
c
= γx x e dx.
ζ 1− e−(W /ζ ) 1 − γx ζ 1− e−(W /ζ ) 1 − γx
c c



0 0

The effects of the firm capital size on the expected losses are thus:
∂ E (y) γW γW
 
1
= c ζ −c W − E (y)
> 0 since c ζ −c W > E (y).
∂W 1 − γW W 1−c e(W /ζ ) − 1
c
1 − γW

The second derivative leads to:

∂γW /∂ W c ζ −c γW
 
W − c ζ −c W c −1 e−(W /ζ ) +
c
−(W /ζ )c 2 c ζ −c
∂ E (y)


 (1 − γW )  c −1 −(W /ζ )c (1 − γW ) 
1−e 2  2 
=
W c −1 e−(W /ζ )
c
∂W 2 γ W W e
− c ζ −c W − E (y)
 
1 − γW 1 − e−(W /ζ )
c


   (c − 2) 
−(W /ζ )c
+ 1−e − ζ cW
−c c −1
W
or

2
1 − e−(W /ζ )
c
∂ 2 E (y) γW ∂γW /∂ W ∂ E (y)
 
c−2  −(W /ζ )c
 1
c −1 −(W /ζ )c
= c ζ −c + W + 1 − e − 1−c − .
W e ∂W 2 (1 − γW ) (1 − γW )2
W W ∂W
 N.N. Taleb, C.S. Tapiero / Physica A 389 (2010) 3503–3507 3507

Since
∂γW /∂ W c−2  −(W /ζ )c
 1
W  1 − e − 1 −c
(1 − γW )2
W W
The condition for a positive second derivative is:
!
γW 1 ∂γW /∂ W E (y)
cζ −c
1− + c ζ −c + 1−c (W /ζ )c
>0
(1 − γW ) e /ζ ) − 1
( c
(1 − γ W )

2
W − c W W e −1
which is guaranteed if W c > ζ c ln (1 + W c )
These conditions establish therefore the conditions for an accelerating loss the larger the firm—a loss that may be far
larger than the firm capital loss.

4. Conclusion

The purpose of this paper was to to indicate that size matters in a nonlinear way and that the issues that pertain to
managing evaluating firms that are too big to fail require a far greater awareness and a regulation of the risk externalities
that these institutions represent. Firms that are ‘‘too big to fail’’ are ‘‘polluters’’ either by design when they over-leverage
their financial bets or their speculative positions or when they are struck by a Black Swan. This is the case because their
losses have far greater significance than their narrow well being affecting investors that had no part in their actions. In this
sense, their costs are a risk externality to be confronted and regulated as such. For this reason, regulation of firms that are
too big too fail, require that greater attention be given to their consequential external risks rather than application of VaR
techniques to protect their internal losses. The growth of economic units large enough to integrate their external risk is
of course not appropriate since the moral hazard risks resulting from their market power will be too great. Similarly, total
controls, total regulation, taxation, nationalization etc. are also a poor answer to deal with risk externalities. Such actions
may stifle financial innovation and technology and create disincentives to an efficient allocation of money. Coase observed
that a key feature of externalities are not simply the result of one CEO or Bank, but the result of combined actions of two
or more parties. In case of the financial sector, there are two parties, Banks that are ‘‘too big to fail’’ and the Government
as a stand in for the public. Banks are entrusted rights granted by the Government and therefore any violation of the trust
(and not only a loss by the bank) would justify either the removal of this trust or the takeover of the bank. A bargaining
over externalities would, economically lead to Pareto efficient solutions provided that banking and public rights are fully
transparent. However, the non-transparent bonuses that CEOs of large banks apply to themselves while not a factor in banks
failure is a violation of the trust signaled by the incentives that banks have created to maintain the payments they distribute
to themselves.

References

[1] V. Pareto, Le cours d’Economie Politique, Macmillan, London, 1896.

[2] N.N. Taleb, Errors, robustness, and the fourth quadrant, International Journal of Forecasting 25 (4) (2009) 744–759.
[3] Y. Ijiri, H.A. Simon, Skew Distributions and the Size of Business Firms, North Holland, New York, 1977.
[4] K. Okuyama, M. Takayasu, H. Takayasu, Zipf’ss law in income distribution of companies, Physica A 269 (1999) 125–131.
[5] Y. Fujiwara, Zipf law in firms bankruptcy, Physica A 337 (2004) 219–230.
[6] L.A.N. Amaral, S.V. Bulkdyrev, S.V. Havlin, H. Leschron, P. Mass, M.A. Salinger, H.E. Stanley, M.H.R. Stanley, Journal of Physics I (1997) 621.
[7] M.H.R. Stanley, L.A.N. Amaral, S.V. Bulkdyrev, S.V. Havlin, H. Leschron, P. Mass, M.A. Salinger, H.E. Stanley, Nature 397 (1996) 804.
[8] J.P. Bouchaud, M. Potters, Theory of Financial Risks and Derivatives Pricing, From Statistical Physics to Risk Management, 2nd ed., Cambridge University
Press, 2003.
[9] D. Garlaschelli, S. Battiston, M. Castri, V.D.P. Servedio, G. Caldarelli, The scale free nature of market investment network, Physica A 350 (2005) 491–499.
[10] C.S. Tapiero, Consumers risk and quality control in a collaborative supply chain, European Journal of Operations Research 182 (2007) 683–694.
[11] C.S. Tapiero, Risk Finance and Financial Engineering (tentative title), 2 volume, Wiley, 2010 (forthcoming).
[12] Konstantin Kogan, Charles S. Tapiero, Supply Chain Games: Operations Management and Risk Valuation, in: Frederick Hillier (Ed.), Series in Operations
Research and Management Science, Springer Verlag, 2007.
[13] Y.U. Saito, T. Watanabe, M. Iwamura, Do larger firms have more interfirm relationships, Physica A 383 (2007) 158–163.
[14] A Aleksiejuk, J.A. Holyst, A simple model of bank bankruptcies, Physica A 299 (2001) 198–204.
[15] D. Stauffer, Introduction to Percolation Theory, Taylor and Francis, London, Philadelphia, A, 1985.