Failures in

Risk Management

R
isk management has received increasing attention in recent years,
both from academics and from practitioners. The number of
academic articles indexed in the Journal of Economic Literature
containing the key word “risk” in the subject description increased from
545 per year in 1988 to 859 in 1998, and the membership of the Global
Association of Risk Professionals (GARP) has grown to over 8,500 since
its founding in late 1996.
The heightened interest in risk management is the result of a number
of coincident secular trends. Globalization of trade and production have
increased financial and direct investment in volatile emerging markets. In
addition, in both developed and emerging economies, capital markets
have become more important as a means of allocating resources. As a
result, both banks and nonfinancial firms find that the number, type, and
extent of their exposures have increased significantly. Finally, a spate of
volatile financial innovations are simultaneously a source of risk and a
means to mitigate it.
But risk management has also attracted attention as a result of the
repeated and well-publicized failures associated with its implementation.
Despite the increased academic and professional attention paid to risk
management, frequent instances still occur when sophisticated investors
or firms experience sudden, unexpected, and devastating losses. In cases
such as Barings, Metallgesellschaft, and Long-Term Capital Management,
losses have been in the billions of dollars. In such cases the sophistication
Ralph C. Kimball of the risk management processes in place was clearly inadequate for the
level and type of risks assumed.
What is the source of these failures in risk management? Is it just
Economist, Federal Reserve Bank of extreme bad luck, similar to being struck by lightning while out jogging?
Boston, and Associate Professor of If so, it is hard to conclude that the victims were negligent; nor do such
Finance, Babson College. The author freakish outcomes say much about the desirability of either risk manage-
wishes to thank Richard Kopcke and ment or regular exercise. On the other hand, if such failures occur because
Peter Fortune for helpful comments. of flaws in conceptual approaches or in the way these approaches are
implemented, then one can expect such failures to be fined as the relative proportions of debt and equity in
repeated. the firm’s balance sheet, and the distribution of re-
This article discusses failures in risk management, turns is shown in Figure 1. The horizontal axis is
why they occur, and what can be done to reduce their denominated in terms of percent of total assets. The
occurrence. Part I discusses the nature of risk and the distance OA represents the ratio of debt to assets,
objectives of risk management. Part II argues that while the distance AB represents the ratio of equity to
intuitively attractive conceptual simplifications often assets. Their sum, represented by the distance OB,
create significant errors in risk measurement. Part III equals 100 percent of assets. The distance BC repre-
then describes failures in risk mitigation. Part IV sents the expected pretax return on assets, approxi-
discusses implications, both for managers and for mated by the historical mean of pretax return on assets
regulators. (ROA). The curve RR’ represents the potential varia-
tion in pretax ROA around the expected value.
Book insolvency occurs when operating losses
exceed the firm’s equity capital; it is represented by
I. Risk and Risk Management
the shaded area under the curve to the left of A on the
To an economist, risk is defined as the existence of horizontal axis. This shaded area represents the risk
uncertainty about future outcomes. Risk is a key factor borne by the debt holders and is equivalent to the
in economic life because people and firms make irre- probability of the firm’s insolvency. The area under
vocable investments in research and product develop-
ment, plant and equipment, inventory, and human
capital, without knowing whether the future cash
flows from these investments will be sufficient to The possibility of catastrophic but
compensate both debt and equity holders. If such real
investments do not generate their required returns,
extremely low-probability
then the financial claims on these returns will decline outcomes always exists. Firms can
in value. never hold enough equity to
A key conceptual advance in risk analysis oc-
curred when analysts began to describe the risk of an guarantee their solvency, but
investment as being equivalent to the distribution of they can estimate the amount
potential outcomes, where the distribution consists of
all possible outcomes weighted by their relative prob-
needed to reduce the probability
ability of occurrence. The more extreme the distribu- of insolvency to a socially
tion of outcomes, the riskier the project. Two projects acceptable small number.
could have the same expected return (the weighted
average of all possible outcomes) but differ in their
risk, if one project had a broader range of outcomes
or a higher probability of extreme outcomes than the the curve to the right of A represents the risk borne by
other. the equity holders. Clearly, the division of risk be-
A second key insight was that while individual tween debt and equity holders will be determined by
outcomes were not predictable, their distribution of- the position of point A relative to the distribution of
ten was. That is, distributions often could be described outcomes. The risk borne by debt holders can be
by mathematical models that depended on a few key decreased if the proportion of equity capital is in-
parameters, such as the mean and the standard devi- creased, shifting point A to the left, or increased if the
ation for the well-known normal distribution. If the proportion of equity is decreased, shifting point A to
appropriate type of distribution could be established, the right. Similarly, an increase in risk, signified by a
then an analyst could use relatively sparse historical flatter distribution, will require a higher proportion of
data to forecast the key parameters and thus the future equity to generate the same probability of insolvency.
distribution of returns. This ability to predict the Figure 1 oversimplifies matters, since the distri-
distribution of returns was critical both to the quanti- bution of outcomes actually extends to the left past
fication of risk premiums and to the determination of that shown by the curve RR’. That is, the possibility of
the capital structure of the firm. catastrophic but extremely low-probability outcomes
The relationship between capital structure, de- always exists. Firms can never hold enough equity to

4 January/February 2000 New England Economic Review

guarantee their solvency, but
the expected distribution of
outcomes can be used to esti-
mate the amount of equity
capital needed to reduce the
probability of insolvency to
a socially acceptable small
number.1 This number repre-
sents a compromise among
different providers of capital
(debt, equity, and trade credit
and, for some institutions,
regulators) who have differ-
ent tolerances for risk.
In addition to varying
the proportions of equity and
debt in their capital structure,
firms can also affect their
probability of insolvency by
mitigating the risk exposures
they face. That is, by pur-
chasing insurance, hedging,
screening customers, closely
supervising employees and monitoring suppliers, or earn higher returns or, alternatively, grow faster than
through diversification, firms can reduce the disper- its competitors. Thus, risk management, like technol-
sion of potential outcomes. A popular misconception ogy, distribution, or scale, is a source of competitive
is that the objective of risk management is to eliminate advantage.
risk. In fact, firms appear to pick and choose among If a firm fails because of “unexpected” losses, the
the types and degrees of exposures, assuming those failure is due to one of three causes. The firm may
that they believe they have a competitive advantage in have accurately estimated the loss distribution (its
managing and laying others off into the capital mar- exposure), but has insufficient capital to absorb the
kets, or accepting small or moderate exposures while draw from the distribution, usually because the losses
insuring against catastrophic ones ( Stulz 1996). Thus, are catastrophic and exceed the “socially acceptable”
a commercial bank may accept credit risk but avoid hurdle for insolvency.3 This can be termed the “hun-
interest rate risk, while an investment bank does the dred-year-storm” explanation and it has been ad-
opposite. vanced by the management of Long-Term Capital
Other aspects of the firm’s operations, such as the Management to explain the near-insolvency of that
convexity of its tax schedules, the ability of owner/ entity in the summer of 1998.4
managers to diversify their portfolios, and the propor- While “hundred-year-storms” do occur, firms fail
tion of intangible assets, can also affect the extent to
which managers attempt to mitigate risks (Tufano 1
Even a firm with 100 percent equity can fail if it incurs
1996). Increasingly, both economists and strategic liabilities that exceed its assets. This sometimes occurs if products,
planners are coming to view risk management as such as asbestos, prove harmful to the customer or the environment
being related to the issue of the boundaries of the and create liabilities for damages.
2
For a discussion of the boundaries of the firm, see Besanko,
firm.2 That is, each firm must decide which activities Dranove, and Shanley (1996), Chapters 2 & 5.
are to be carried out within the firm, and which ones 3
Insolvency can also occur if owner/managers deliberately,
are to be externalized to suppliers or customers. In this and without disclosing it, choose a probability of insolvency that is
higher than the socially acceptable small number. This situation
framework, the decision to assume or mitigate partic- can arise if the owners are attempting to maximize the value of
ular risks is analogous to the decision to integrate the implicit call option on the assets of the firm represented by
backwards or to outsource a particular function. To their equity.
4
See “Investors May See ‘LTCM, the Sequel,’ ” Wall Street
the extent a firm has a competitive advantage in Journal, May 20, 1999, p. C1. For an alternative explanation of
assuming and managing a particular exposure, it can LTCM’s losses, see Section II, below.

January/February 2000 New England Economic Review 5

far more frequently because they have estimated the distribution. Because good estimates of the mean and
distribution of outcomes incorrectly. Such failures are standard deviation can be obtained from relatively
due to errors in risk measurement. This is sometimes few observations, the use of the normal distribution
called model error, but it can also represent a form of allows risk managers to extrapolate the probability of
management myopia when managers fail to recognize extreme outcomes from relatively few data points.
that an exposure exists. The latter might be termed Moreover, while it has long been recognized that
risk ignorance, while model error usually results be- returns on most assets are not exactly normally dis-
cause some parameter of the loss distribution or the tributed, the discrepancy is often so small that many
analysts find it convenient to ignore the differences.
Unfortunately, while the assumption of normality
in returns is beguiling, it is not justified by empirical
While “hundred-year storms” do research, and basing risk management on such as-
occur, firms fail far more sumptions can lead to serious risk-measurement er-
rors. Many studies have concluded that most asset
frequently because they have returns are not normally distributed but instead are
estimated the distribution of fat-tailed and skewed to the left. (See Fama 1965,
Duffie and Pan 1997.) The use of the normal distribu-
outcomes incorrectly, through tion to estimate frequency of outcomes in such circum-
model error or management stances will result in estimates of the frequency of
major losses that are too low.
myopia, or risk ignorance. This is illustrated in Table 1, which shows sum-
mary statistics for monthly total returns for both large
stocks and long-term government bonds over the
covariance among different risks is misestimated. Both period from 1926 through 1997. The first two columns
forms of measurement error are discussed in Part II. show the mean and the standard deviation, based on
Finally, the firm may accurately measure its exposure, the monthly data. If returns on both stocks and bonds
but then be ineffective in its efforts to mitigate the risks are normally distributed, then we can use these sum-
it assumes. Such errors in risk mitigation are discussed mary statistics to estimate the frequency of extreme
in Part III. losses. For example, because the mean and standard
deviation of monthly large stock returns are 1.03
percent and 5.70 percent, respectively, and because 98
percent of the area under the normal distribution is
II. Issues in Risk Measurement encompassed by 2.33 standard deviations around the
mean, then in only one month out of a hundred
Are Returns Normally Distributed?
should the percentage loss on large stocks exceed
Using the distribution of potential outcomes to 12.25 percent (1.03 percent ⫺ 2.33 ⫻ 5.70 percent).
measure risk is a great conceptual advance, but it is a Column 3 shows the expected frequency for losses at
difficult one to implement. Estimation of a potential the 1 percent level under the assumption of normality,
return distribution is usually based on historical data, while column 4 shows the actual frequency of losses
but the availability of such data is often limited, and for the same range. The actual frequency of extreme
even when available, older data may have little fore- losses is almost twice that expected under the assump-
casting value because of institutional or structural tion of normality, for both large stocks and long-term
changes in the environment. In particular, estimation government bonds. In effect, the assumption of nor-
of the tails of the distribution, the area of special mality will result in a serious underestimate of the
interest for risk managers, is difficult, since by defini- frequency of major losses. “Hundred-year storms”
tion the number of observations in the tails is limited. will occur much more frequently than once every
Given these limitations on the availability of data, hundred years.
it is not surprising that the normal distribution became If the assumption of normality results in down-
the paradigm for risk management. The normal dis- ward-biased estimates of risk, then alternative models
tribution has one very strong advantage: The proba- are needed if actual distributions of returns are to be
bility of any given outcome can be estimated, given estimated. Is there an alternative statistical model that
the mean and standard deviation of the underlying can be used to estimate the actual distribution of

6 January/February 2000 New England Economic Review

 helpful, since it permits the use
Table 1
of relatively few observations to
Extreme Losses on Large Stocks and Long-Term U.S. extrapolate ex ante the distri-
Treasuries, 1926 to 1997 bution of outcomes over a much
longer period.
Number of
Standard Expected Actual
Second, the assumption of
Deviation of Extreme Losses Number of serial independence does not
Mean Monthly Mean Monthly if Distribution Extreme mean that runs of bad luck will
Return (%) Return (%) Was Normal Losses not occur, but it does mean that
Large Stocks 1.0291 5.7036 8.64 16 such runs will be normally dis-
Long-Term U.S. tributed. Thus the chances are
Treasuries .4469 2.2194 8.64 15 only one in four that a firm will
Source: Ibbotson Associates, Stocks, Bonds, Bills and Inflation: 1998 Yearbook. experience adverse outcomes in
two consecutive periods, and
only one in eight that a firm will
experience adverse outcomes in
returns on an ex ante basis? To date, our ability to three consecutive periods. The practical implication of
address this issue is limited. One possibility is to use this is that a firm can neglect the probability of adverse
non-parametric models that do not depend upon the runs in computing its required equity capital.
assumption of a particular type of distribution, but Like the assumption of normality in returns, the
usually such non-parametric models require more assumption of serial independence in returns is not
data than are available. Alternatively, one could as- justified by the empirical evidence (Campbell, Lo, and
sume other distributions or return-generating func- McKinlay 1997). Instead, returns tend to be positively
tions that produce fat-tailed and skewed distributions. serially correlated, so that an adverse outcome in this
For example, one might assume some form of the period is likely to be followed by an adverse outcome
stable distribution, which is bell-shaped, symmetrical, next period. Table 2 shows estimates of serial correla-
and fat-tailed. Or, one might assume a jump diffusion tion based on monthly return data for both large
model, in which the returns usually behave as if stocks and long-term U.S. Treasuries over the period
drawn from a normal distribution but are periodically from 1926 through 1997. Both series show a positive
“jumped” up or down by adding an independent, serial correlation of around 0.09.
normally distributed shock. Or one might assume that If returns are assumed to be serially independent
volatility is not constant but itself changes randomly but actually are positively correlated, then estimates
over time. (See Simons 1997 and Fortune 1999.) But of long-period standard deviations extrapolated from
while these alternative statistical models might fit short-period returns will be too low. Once again, the
the data better in a conceptual sense, their use often frequency of seriously adverse outcomes will be un-
requires the estimation of some specific parameter for derestimated, and a firm will require additional equity
which we have little confidence. capital if it is to achieve the same probability of
insolvency.
Serial Correlation
A second attractive simplification when analyz-
ing risk is to assume serial independence, that is, that
outcomes are not correlated over time, so that the Table 2
outcome next period does not depend on the outcome Serial Correlation for Large Stocks and
this period. The assumption of serial independence Long-Term U.S. Treasuries, 1926 to 1997
has two major implications. If outcomes are serially Coefficient of
independent, then the standard deviation of returns Serial Correlation
increases with the square root of time. That is, daily Large Stocks .0864
data can be used to estimate weekly, monthly, or Long-Term U.S. Treasuries .0894
annual volatility by multiplying the standard devia- Source: Ibbotson Associates, Stocks, Bonds, Bills and Inflation: 1998
Yearbook.
tion of the daily data by the square root of the number
of trading days in the longer period. This is very

January/February 2000 New England Economic Review 7

 probability of a significantly adverse return on one
Correlation among Outcomes
security being accompanied by a significantly adverse
In measuring the risk exposure of a firm or return on the other is substantially greater than one
financial institution, the estimation of the correlation would expect, based upon the coefficient of correlation
among asset returns is as important or more important calculated using data from all periods. For example, in
than the estimation of the distribution of the individ- a study of common stocks and REITs, Booth and
ual asset returns. This is so because the risk of a Broussard found a positive correlation for negative
portfolio of assets depends not only on the stand-alone returns of 0.834. However, if the 1 percent of the most
risks (standard deviations) of the individual assets but negative returns are eliminated from both series, then
also on the correlation (covariance) among them. the correlation coefficient for the remaining data pairs
Unless the different assets are perfectly positively declines to 0.214.5
correlated, then the assets will act as partial natural Like positive serial correlation and non-normality
hedges for each other, so that diversification of the in returns, the tendency for asset returns to be more
portfolio among different asset types provides an in- positively correlated during financial crises tends to
expensive and readily available means to mitigate risk. reduce estimates of the frequency of seriously adverse
Indeed, the existing paradigm for the quantifica- outcomes. In particular, diversification may be an
tion of risk premiums, the capital asset pricing model effective way to mitigate portfolio risk in “normal”
(CAPM), distinguishes between two types of risks: periods, but may be ineffective in extreme periods.
independent and systemic. Independent risks are
Risk Ignorance
Assumptions of normality, serial independence,
or non-varying return correlation are all examples of
The probability of a significantly “model error.” Model error occurs when the potential
adverse return on one security exposure is recognized but misestimated because
being accompanied by a some parameter of the distribution of outcomes is
misestimated, or because the correlation between dif-
significantly adverse return on ferent risks is misestimated. Model error often results
the other is substantially greater either because managers make inappropriate ex ante
assumptions concerning the shape of the distribution,
than one would expect, or because the conceptual models fail to capture some
based on historical data. important aspect of reality.
But a second and extreme form of risk measure-
ment error occurs when the firm fails to recognize it
has any exposure whatsoever. Such a case might be
measured by the standard deviation of returns, while termed “risk ignorance.” Risk ignorance is often asso-
systemic risks are measured by the extent of correla- ciated with the introduction of new products or pro-
tion in returns. Independent risk can be diversified cesses where the technical aspects or environmental
away through a broadly diversified portfolio, while effects of such products or processes are not fully
systemic risk cannot. Indeed, because independent understood, or with changes in the legal environment
risk can be so easily and cheaply hedged through through legislation or litigation. A classic example of
diversification, the CAPM implies that risk premiums risk ignorance was provided by the losses experienced
depend only on systemic risks, the risks that cannot be by Chase Manhattan Bank in connection with Drys-
diversified away. dale Securities, a government securities dealer, in the
Calculations of correlation among asset returns early 1980s. Chase’s loans to Drysdale were executed
usually use historical data, with the implicit assump- via repurchase agreements, then a relatively new form
tion that correlation is stable across periods. But of borrowing, using the latter’s government securities
several recent empirical studies contradict this as- as collateral. Because it was holding as collateral
sumption: Correlation among asset returns is much highly liquid and default-free securities sufficient to
higher during periods of extreme returns than during cover the principal of the loan, Chase believed its
“normal” periods. (See Roll 1988; Longin and Solnik
1995; and Booth and Broussard 1998.) That is, the 5
See Booth and Broussard (1998), Table 3.

8 January/February 2000 New England Economic Review

loans to be fully secured. But Chase’s exposure in- tage. Of course the benefits from risk mitigation will
cluded not just the principal but also the accrued accrue only if such efforts are effective. If they are
interest, and when Drysdale failed, Chase was left not, then the firm’s actual distribution of returns
with an unsecured claim of $50 million.6 will be more volatile than envisaged and the proba-
bility of insolvency for any given level of equity will
be higher than planned. In the past few years, the
increased efforts by firms to mitigate risks have led to
III. Risk Mitigation more instances where poorly implemented risk miti-
A firm does not necessarily have to accept a gation efforts have led to major losses. (See Figlewski
particular distribution of outcomes, but often can 1994.)
modify the probability of adverse outcomes through A prominent example of poorly implemented risk
its own efforts. These efforts to alter the distribution of mitigation is the losses experienced by the German
outcomes can be termed “risk mitigation.”7 To the firm Metallgesellschaft. In the Metallgesellschaft case,
extent that a firm successfully mitigates its risks, then a U.S. subsidiary contracted to sell 154 million barrels
its distribution of outcomes will be less extreme, and it of oil on fixed-price contracts extending over a 10-year
will require less equity capital than if it had under- period, and then sought to hedge the resulting sub-
taken no risk mitigation. stantial exposure to oil price increases by taking a long
Risk mitigation can take a number of different position in oil futures and commodity swaps. When
forms. Perhaps the two most obvious are the purchase the spot price of oil subsequently fell, the long-term
of insurance, where the firm pays an unrelated third fixed-price contracts to deliver oil increased in value
party to assume the exposure, and hedging, where the while Metallgesellschaft’s futures and swap positions
firm takes an offsetting position in a security, com- declined in value. But because its futures positions
modity, or currency that is closely correlated with the were marked to market while its delivery contracts
exposure it wishes to mitigate. But firms also employ were not, Metallgesellschaft’s financial statements
a number of other measures to mitigate exposures, showed large losses and the firm experienced large
including market research, geographic and product cash outflows.9 Evidently convinced they could not
line diversification, screening and monitoring of cus- sustain further cash outflows, Metallgesellshaft’s man-
tomers, outsourcing, imposing risk premiums in pric- agement chose to liquidate the hedge, leaving them-
ing products, carrying inventories or slack in produc- selves exposed when oil prices subsequently rose
tive capacity, and imposing defined procedures de- again.10 (See Mello and Parsons 1995; Culp and Miller
signed to minimize operational risks.8 1995.)
To the extent a firm can mitigate its risks econom- Risk mitigation efforts may be ineffective for a
ically, it can earn higher returns for shareholders. A number of reasons. Perhaps the best known is agency
firm with less variable returns will require a lower risk, the risk that a manager or employee, inadver-
proportion of equity in its capital structure to obtain tently or purposefully, will fail to follow the policies or
the same probability of insolvency. On the other hand, procedures designed to mitigate risk. For example, a
risk mitigation usually involves real costs. If these real rogue trader whose compensation or tenure is depen-
costs of risk mitigation are less than the cost of the dent upon his trading results may fail to abide by
equity that would otherwise be required, the firm will position limits or hide cumulative losses, or mainte-
earn higher returns than competitors who do not
mitigate risks. Thus, risk mitigation, like scale econo- 9
In the futures markets capital gains and losses are computed
mies or superior technology or distribution, can be and paid at the end of each day, giving rise to continual cash
viewed as a potential source of competitive advan- flows. This process is called “marking to market.”
10
Another factor contributing to its losses was the structure
Metallgesellschaft chose for the hedge. Metallgesellschaft chose to
6
In many cases, risk ignorance creates model error. For exam- construct a rolling hedge consisting of short-term futures contracts.
ple, many of the more notorious failures in risk management In order to hedge the long-term delivery contracts, these short-term
involve a firm using an innovative financial instrument or hedging contracts needed to be rolled over at delivery. Usually this would
technique without fully appreciating the range of potential out- not have been a problem, since in petroleum markets spot prices
comes. See Figlewski (1994). are usually higher than near-dated futures and Metallgesellschaft
7
Risk mitigation implies that the efforts of the firm are aimed at would actually have made money each time it rolled the hedge.
reducing risk, but firms can also make an explicit decision to Unfortunately, soon after Metallgesellschaft established its position,
increase risk through activities such as increasing leverage or the usual relationship between spot and near-dated futures prices
reducing diversification. reversed itself and spot fell below the futures price. Thus, each time
8
See, for example, Biddle (1999). Metallgesellschaft rolled the hedge it experienced losses.

January/February 2000 New England Economic Review 9

nance personnel may overlook an incipient equipment swap. A may believe it has obtained protection against
failure. Agency risk has been responsible for a number interest rate risk, but such protection will only exist so
of notorious episodes, including the bankruptcies of long as B continues to perform on its side of the swap.
Orange County and Barings, and the large losses of That is, the swap does not really extinguish the
Sumitomo. exposure of A, but only transforms it from an expo-
Orange County operated a pooled short-term sure to interest rates to a credit risk exposure to B.
investment fund for local governmental units within Indeed, the greater the success of the swap in mitigat-
its jurisdiction. While the espoused objective of the ing interest rate risk, the greater the credit exposure of
pooled fund was to invest in risk-free or very low-risk A to B.12
securities, the Orange County official in charge of the The tendency for other types of risk to migrate
fund sought increased returns by borrowing to in- and reappear as counterparty or credit risk is illus-
vest in repurchase agreements and mortgage-backed trated by the experience of U.S. banks during the
securities. When interest rates rose abruptly, the Russian financial crisis of August 1998. Many U.S.
now leveraged fund experienced massive losses. (See banks had hedged their exposure to fluctuations in the
Jorion 1995.)
Both Barings, a British merchant bank, and Sumi-
tomo, a Japanese trading company, were victims of
rogue traders who successfully concealed huge cumu- A second factor that often
lative trading losses. In both cases the traders were
able to conceal such huge cumulative losses because causes risk mitigation efforts
they were responsible for reporting and settling their to fail is the tendency for
own trades. (See Fay 1997; The Economist 1996.)
risk management processes
Risk Migration
to fail incrementally over
a long period of time.
But risk mitigation efforts can also fail for more
subtle and indirect reasons. The first is the tendency
for risk to shift or change form. While an individual
firm may mitigate its risks by purchasing insurance or value of the ruble by executing foreign exchange
hedging, these actions do not reduce systemic risk in swaps with Russian banks that required the latter to
the economy, but only transfer it elsewhere.11 More- exchange dollars for rubles at a fixed rate. When the
over, in many cases hedging or purchasing insurance ruble was effectively devalued in August 1998, the
does not really transfer risk, but merely transforms the U.S. banks nevertheless experienced substantial losses
nature of the exposure. when a substantial number of Russian counterparties
For example, consider an interest rate swap in defaulted. (See Bomfim and Nelson 1999.)
which Party A pays a fixed rate and receives a variable Any loss, whether resulting from operations or
rate on some notional principal. The swap is a zero- from market fluctuations, must ultimately be absorbed
sum transaction, so Counterparty B receives the fixed by someone’s net worth. If Party A suffers a loss due
rate and pays a variable one. Party A’s motivation for to a market fluctuation, A’s net worth will decline. If
entering into the swap is to use it to hedge an the loss exceeds A’s net worth, then the excess must be
imbalance between its fixed-rate assets and variable- absorbed by the net worth of A’s creditors. If that net
rate liabilities. By entering into the swap, A protects worth is insufficient, any excess must be absorbed by
itself against a potential increase in short-term rates, the net worth of A’s creditors’ creditors. In short, like
since any increase in the cost of its liabilities will be
offset by increased revenue from the variable leg of the
12
Of course A can mitigate its credit exposure to B by enlisting
the assistance of swap dealer C. Both A and B contract with C. C
receives fixed payments from A and pays them to B, simultaneously
11
Independent risk is firm specific and can be reduced through receiving variable payments from B and paying them to A. Thus
diversification. Systemic risks cannot be affected through diversifi- neither A nor B has credit exposure to the other. C has no interest
cation. Because hedging is a form of diversification (the hedger rate risk exposure but assumes the credit risk that either A or B will
creates a portfolio of offsetting exposures), it only reduces indepen- fail to perform. But while this interpolation of a guarantor mitigates
dent risk, although it does transfer systemic risks elsewhere in the A’s credit exposure to B, it does so only by transforming it to a credit
economy. exposure to C.

10 January/February 2000 New England Economic Review

a row of dominos toppling in sequence, losses will
IV. Implications
migrate through the financial system until completely
absorbed by net worth.13 Thus, a firm’s exposure is not
Implications for Managers
limited to its own counterparties, but also includes the
other exposures of these counterparties.14 The existence of non-normality in returns, posi-
A second factor that often causes risk mitigation tive serial correlation, and state-sensitive correlation
efforts to fail is the tendency for risk management in returns means that managers must view their
processes to fail incrementally over a long period of ability to forecast the distribution of future outcomes
time (Grabowski and Roberts 1997). Case studies of with some skepticism. Use of simplifying assumptions
major industrial accidents have noted that such acci- such as normality is likely to result in significant
dents are often preceded by a long incubation period underestimation of the probability of seriously ad-
marked by a gradual degradation of the risk manage- verse outcomes. Many institutions have recognized
ment processes. An initial failure to follow mainte- the danger of building their risk management pro-
nance schedules or to test backup systems does not cesses upon assumptions such as normality, and have
usually result in immediate harmful incidents, leading developed approaches that address model error.
managers to conclude that the probability of failure Perhaps the most common way to address this
has been overestimated, or that such processes are issue of model error is to allow for a margin of error.
redundant. In effect, the organization becomes de- For example, the normal distribution might be used to
sensitized to the existence of risk. As a result, efforts to generate outcomes, but then excess capital is held to
rectify the degradation of the risk mitigation process offset the model error that is believed to exist. This is
are actually less likely to occur as the degradation
continues. Over time these incremental failures accu-
mulate until some incident causes them to interact in
a fashion that results in a major loss—a loss that could As risk migrates through the
have been avoided, had the original mitigation pro- system, it tends to emerge in its
cesses been followed.
While the long incubation period associated with most basic form, as credit risk.
failures in risk mitigation processes has been most This means that those institutions
clearly identified with respect to industrial catastro-
phes, a similar process appears in financial crises.15 In that specialize in managing and
particular, financial crises are often preceded by a long absorbing credit risks, commercial
period of gradually increasing exposures to a partic-
ular customer or asset type, with an accompanying
banks, play a special role.
loss of diversification. Because returns are attractive
and losses minimal in the initial stages, managers
often convince themselves that they have learned how the procedure used in the application of value-at-risk
to manage the risks involved, or that potential losses models to estimate risk exposures and set capital
are less than previously believed, and that they are requirements for the trading books of commercial and
thus justified in increasing their exposures. investment banks. Currently, commercial banks are
permitted to use the value-at-risk methodology to
compute exposures, but then are required to hold
three times as much capital as that indicated by the
13
In severe crises where the losses exceed private domestic model. While the margin-of-error approach is direc-
equity, the losses must be absorbed by the government or by tionally correct in addressing the danger of underes-
providers of foreign capital. This often occurs in banking crises
when the banking system must be re-capitalized. timating the frequency of seriously adverse events, we
14
Because ex post risk migration is more likely to occur when do not understand whether the extra capital required
losses are large, this may explain why asset return correlation is is sufficient or excessive.
higher during extreme events.
15
Some macroeconomists have minimized the significance of One way to test the sufficiency of a margin of
lax regulatory policies as a contributing factor to the 1998 Southeast error is to “stress test” the firm’s current exposures.
Asia crisis because such regulations were in place over a sustained Historical data on returns and their correlation during
period of time and were not relaxed just prior to the crisis. But such
an argument ignores the long incubation period that occurs as risk some crisis period, such as the November 1987 stock
mitigation processes undergo degradation. See Whitt (1999). market crash or the 1998 Asian crisis, may be used to

January/February 2000 New England Economic Review 11

‘measure the potential losses in a worst-case scenario undertaken. Second, the greater the amount of risk
and to determine if the firm has sufficient equity to mitigation undertaken through hedging or the pur-
weather such an extreme event. For example, one chase of insurance, the more likely that unforeseen
large international re-insurer stress tests its portfolio losses will migrate quickly from one market to an-
under the assumption of simultaneous 8.5 Richter- other, or from one country to another. That is, while
scale earthquakes in Tokyo and Los Angeles. hedging acts to reduce independent risk, it can en-
Failures in risk mitigation most commonly arise hance systemic risk. Finally, as risk migrates through
due to agency risk. Such risk is best controlled through the system, it tends to emerge in its most basic form, as
active monitoring of exposures and employees, to credit risk. This tendency for errors in risk manage-
ensure compliance with limits and procedures. Often ment to ultimately emerge as credit exposures means
this involves specialized staff who in effect conduct that those institutions that specialize in managing and
real-time audits, both to confirm reported exposures absorbing credit risks, that is, commercial banks, play
and to rectify minor deterioration in risk mitigation a special role.
processes before they can incubate into major losses. Because commercial banks are in the business of
accepting and managing credit risk, they act as shock
absorbers, absorbing errors in risk management made
Implications for Regulators
elsewhere in the system. The capacity of the banks to
As noted above, risk tends to migrate in the act as buffers against errors in risk management de-
financial system. In particular, hedging does not re- pends on their ability to measure and mitigate their
duce systemic risk, but only transfers the exposure own exposures, as well as the sufficiency of their
elsewhere or transforms the type of the exposure. equity capital. A well-managed and well-capitalized
Thus, risk migration has three important implications. banking system is thus requisite for avoiding systemic
First, because risk mitigation activities such as hedg- economic and financial crises.
ing do not reduce the amount of systemic risk in the
system, they also do not reduce the aggregate amount
16
of equity capital needed to absorb this risk.16 That is, Once again, hedging can reduce independent risk, but not
systemic risk. To the extent that firms in the aggregate hold equity
the amount of equity capital needed systemwide is capital against independent risk, hedging can reduce the amount of
independent of the amount of risk mitigation that is equity capital held, but not that needed against systemic risk.

References
Besanko, David, David Dranove, and Mark Shanley. 1996. Economics Fortune, Peter. 1999. “Are Stock Returns Different Over Weekends?
of Strategy. New York: John Wiley & Sons. A Jump Diffusion Analysis of the ‘Weekend Effect.’ ” New England
Biddle, Frederic M. 1999. “Can Kaufman & Broad Beat the Home Economic Review, September/October, pp. 3–19.
Building Cycle?” The Wall Street Journal, October 27, p. B4. Grabowski, Martha and Karlene Roberts. 1997. “Risk Mitigation in
Bomfim, Antulio N. and William R. Nelson. 1999. “Profits and Large-Scale Systems: Lessons from High-Reliability Organiza-
Balance Sheet Developments at U.S. Commercial Banks in 1998.” tions.” California Management Review, Summer, pp. 152–59.
Federal Reserve Bulletin, June, pp. 369 –95. Jorion, Phillipe. 1995. Big Bets Gone Bad: Derivatives and Bankruptcy in
Booth, G. Geoffrey and John Paul Broussard. 1998. “REIT Returns, Orange County. San Diego, CA: Academic Press.
Probability of Large Losses, and Asset Allocation.” Eli Broad Grad- Longin, F. and B. Solnik. 1995. “Is the Correlation in International
uate School of Management, Michigan State University, Processed. Equity Returns Constant? 1960 –1990.” Journal of International
Campbell, John Y., Andrew W. Lo, and A. Craig McKinlay. 1997. Money and Finance, vol. 14, pp. 3–26.
The Econometrics of Financial Markets. Chapter 2. Princeton, NJ: Mello, A. and J. E. Parsons. 1995. “Maturity Structure of a Hedge
Princeton University Press. Matters: Lessons from the Metallgesellschaft Debacle.” Journal of
Culp, Christopher and Merton H. Miller. 1995. “Metallgesellschaft Applied Corporate Finance, Spring, pp. 106 –20.
and the Economics of Synthetic Storage.” Journal of Applied Roll, Richard R. 1988. “The International Stock Market Crash of
Corporate Finance, Winter, pp. 62–76. 1987.” Financial Analysts Journal, vol. 44, no. 5, pp. 19 –35.
Duffie, Darrel and Jun Pan. 1997. “An Overview of Value at Risk.” Simons, Katerina. 1997. “Model Error.” New England Economic
The Journal of Derivatives, vol. 4, no. 3 (Spring), pp. 7– 49. Review, November/December, pp. 17–25.
The Economist. 1996. “Coming A Cropper in Copper,” June 22, pp. Stulz, René. 1996. “Rethinking Risk Management.” Journal of Applied
69 –70. Corporate Finance, Fall, pp. 8 –24.
Fama, Eugene F. 1965. “The Behavior of Stock Market Prices.” Tufano, Peter. 1996. “Who Manages Risk? An Empirical Examina-
Journal of Business, vol. 38 (January), pp. 34 –105. tion of the Risk Management Practices of the Gold Mining
Fay, Stephen. 1997. The Collapse of Barings. New York: W.W. Norton Industry.” Journal of Finance, September, pp. 1097–1137.
& Co. Whitt, Joseph. 1999. “The Role of Shocks in the Asian Financial
Figlewski, Stephen. 1994. “How to Lose Money in Derivatives.” The Crisis.” Federal Reserve Bank of Atlanta Review, Second Quarter, pp.
Journal of Derivatives, vol. 2, no. 2 (Winter), pp. 75– 82. 18 –31.

12 January/February 2000 New England Economic Review