Escolar Documentos
Profissional Documentos
Cultura Documentos
ralue at risk is Wall Street's latest advancement dealers use either an internal methodology or a
V in risk measurement. Simply defined, VAR is Bank for International Settlements (BIS) standard
an estimate of maximum potential loss to be ex- methodology to compute VAR and that the results
pected over a given period a certain percentage of be multiplied by a factor of three to determine the
the time. Its simplicity is seductive. Used to the amount of capital to be set aside for market risk.
extreme, in a single statistic, a firm can measure its Our research indicates that this amount may be too
exposure to markets worldwide. VAR enables a high or too low, depending upon the method used.
firm to determine which businesses offer the great- The need for a uniform VAR methodology or for
est expected returns at the least expense of risk. differing multiplication factors according to the
When one considers that risk management in the type of VAR is paramount to establish a common
early 1970s consisted almost entirely of the evalua- ground for comparative purposes.
tion of credit risk, VAR's power in the context of the In our analysis, historical simulations present
galaxy of risks we track, analyze, and manage quite different views of risk relative to Monte Carlo
today is breathtaking to consider. simulations. This difference is attributable to the
VAR can be dangerous, however. A review of extreme dependence of historical simulations on
dozens of dealers' and end-users' VARs revealed the underlying data set and the value of the relative
radically different approaches to the calculation. In randomness of key variables in Monte Carlo simu-
this study, eight common VAR methodologies were lations compared with sample-specific values. The
applied to three hypothetical portfolios. As illus- results also reveal the exceptional time sensitivity of
trated in Figure 1, the magnitude of the discrepancy certain portfolio risks and highlight the potential
among these methods is shocking, with VAR re- failure of VAR, even when bolstered by stress
sults varying by more than 14 times for the same testing. In sum, although VAR and stress testing are
portfolio. These results illustrate the VAR's extreme necessary, they are not sufficient to contain risk.
dependence on parameters, data, assumptions, and The differences in common VARs emphasize
methodology. the fact that no single set of parameters, data,
The implications of these discrepancies for assumptions, and methodology is accepted as the
capital adequacy standards are significant, espe- "correct" approach. Even if two firms use the same
cially given the Basle Committee on Banking Su-
quantitative technique, they often apply different
pervision's treatment of VAR in its proposed
assumptions in implementing the technique. For
amendment to The 1988 Basle Capital Accord, "The
example, some firms calculate the global VAR of
Supervisory Treatment of Market Risks," published
the firm over a one-day time horizon, using histor-
on April 12, 1995. This amendment proposes that
ical data series on markets and a specific set of
mathematical models. Others calculate regional
Tanya Styblo Beder is a principal of Capital Market Risk Advisors, VAR of product areas over a monthly or annual
Inc., in New York. time horizon, using random or implied data series
w i t h i n a n d b e t w e e n m e t h o d o l o g i e s . T h e V A R sta- s u m p t i o n s m a d e to p e r f o r m h i s t o r i c a l s i m u l a t i o n
tistics to the r i g h t of each b a r m a y b e i n t e r p r e t e d as o v e r the p r i o r 2 5 0 - d a y p e r i o d , the p r o b a b i l i t y is 1
follows: U n d e r the a s s u m p t i o n s specific to the p e r c e n t t h a t a loss e q u a l to or e x c e e d i n g 1.29
p a r t i c u l a r V A R c a l c u l a t i o n , the p r o b a b i l i t y is 5 p e r c e n t of the $1 m i l l i o n p o r t f o l i o i n v e s t m e n t w i l l
p e r c e n t (1 p e r c e n t ) t h a t the p o r t f o l i o w i l l suffer a occur over a one-day time horizon.
loss g r e a t e r t h a n or e q u a l to the statistic s h o w n . For F i g u r e 3 c o m p a r e s the r e s u l t s of the h i s t o r i c a l
the t h i r d set of bars, for e x a m p l e , u n d e r the as- s i m u l a t i o n for Portfolio 1 w i t h the r e s u l t s of the
Hist. I Pr 100 d ld
~ 0 . 6 7
Hist. I Pr 100 d 2w 0 ' 2 0 ~ 0 . 8 2
Hist. I Pr 250 d ld
o 1.29
o Hist. I Pr 250 d 2w ] _ ~ / .7 2.14
,-C
MontC I RiskMetrics ld ~
0 . 8 8
MontC I RiskMetrics 2w , _ 1!82
~ 2 . 6 1
MontC I BIS/Basle ld • 10.68 0.97
MontC I BIS/Basle 2w
2.87
I I I I I
0.5 1.0 1.5 2.0 2.5 3.0
VAR (%)
[] 5% Probability [] 1% Probability
One-Da~' Returns
20
18
16
g 14
12 -
,.Q 10 -
,£a
2 8 -
,4 Z
6 -
4 -
2 -
I
0
-2.0 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2.0
Return (%)
Two-Week Returns
14
12
10
g2 8
4 -
2 -
0
-4.0 -3.2 -2.4 -1.6 -0.8 0 0.8 1.6 2.4 3.2 4.0
Return (%)
-- -- -- RiskMetrics 100 D a y s
Monte Carlo simulations for one-day and two-week rate in a trending market, they will be less accurate
returns. 3 As illustrated by the graphs, the historical when the trend changes•
simulations present a different view relative to the As summarized in the bar charts in Figure 2,
Monte Carlo simulations• This result is attributable the result for the prior 100 trading days, 5 percent
to their extreme dependence on the underlying data probability VAR equals 0.49 percent over a one-day
set. During the 100-day and 250-day periods in- horizon but then drops to 0.20 percent over a
cluded in the historical simulations, the value of two-week period• For all other VAR types, the VAR
Treasury strips largely appreciated• Had a period of result increases with the time horizon, as would be
rising interest rates been selected, the result would expected• This surprising result is explained by the
have been the opposite. The danger in basing VAR pattern of results during the specific historical pe-
estimates on relatively short periods of direct his- riods. Although the average return is positive dur-
torical observations is apparent--history must re- ing the first 100 trading days (the left side of Figure
peat itself for the results to predict the future• 4), negative returns are more common over the
Although historical estimates may be fairly accu- one-day time horizon than over the two-week time
I I I I I I I I I I I I I I I
/I
Two-Week Returns
4
3
2
1
0
2
-1
-2
-3
1 14 27 40 53 66 79 92 105 118 131 144 157 170 183 196 209 222 235 248
Day
horizon. Thus, VAR is higher for the one-day time potential capital exposure and expected profit over
horizon than for the two-week time horizon. a short-term or a long-term horizon? In terms of
Several other conclusions follow from this set risk-reward appetite, will the firm be satisfied if
of VAR results, as expected given interest rate losses m o u n t for two years but huge profits make
trends at the time. Monte Carlo simulations indicate u p for the losses in the third year? Or, is less erratic
higher expected losses than does the 100-day his- performance desired? Often, more-even perfor-
torical simulation but lower expected losses than mance is desired b y firms that report public finan-
for the 250-day historical simulation. Historical cial information, as well as by funds that must
simulations indicate that increasing the holding publish daily net asset values. Thus, two firms
period from one day to two weeks decreases the performing identical VAR calculations, other than
expectation of the highest expected profits and selection of time horizon, m a y have different but
increases the m a g n i t u d e of expected losses. Note not necessarily inconsistent VAR results.
that although Monte Carlo simulations indicate a Although a model m a y p r o d u c e adequate
similar change in the expectation of the highest views of capital at risk on an overnight or weekly
profits, they predict a larger increase in the magni- basis, it m a y p r o d u c e inadequate risk views over
tude of expected losses (three times versus two time horizons of several months, a year, or longer.
times). The difference in VAR driven b y the relative For example, the calculation of short-horizon VAR
r a n d o m n e s s of key variables in Monte Carlo versus m a y be misleading for customized or exotic prod-
sample-specific historical simulations is clear. ucts that cannot be liquidated u n d e r the assumed
Time horizon is clearly a crucial parameter in time horizon. To the degree that multiple horizons
VAR. Firms select quite different time horizons to are required, risk systems rarely are capable of
view their risk. Does the firm wish to analyze its incorporating them in aggregating portfolio risks,
e~ Hist. I Pr 250 d ld
O
a= Hist. I Pr 250 d 2w
MontC I RiskMetrics ld
X
MontC I RiskMetrics 2w
MontC I BIS/Basle ld
MontC I BIS/Basle 2w
I I I I I I I
0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
VAR (%)
[] 5% Probability [] 1% Probability
One-Day Returns
18
16 -
14 -
12 -
10 -
8 -
6 -
4 -
2 -
0 I I
-2.0 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2.0
Return (%)
Two-Week Returns
12
10
"~ 6
-g 4
_ .': '.. . ... .. /
2
0
M.0 -3.2 -2.4 -1.6 -0.8 0 0.8 1.6 2.4 3.2 4.0
Return(%)
tions. In the case of two-week returns, the distribu- time horizons. The top panel of Figure 7 shows tile
tion changes to display bimodal behavior. This return on Portfolio 2 on the day of construction as a
behavior is apparent in the lines that appear to be function of the underlying asset price (the S&P 500).
upside down "normal" distributions. A further Given the starting S&P 500 level of approximately
observation is that the historical simulations pro- 529, the portfolio reflects a small positive return
duce high-probability high-return expectations and (less than 1 percent) if the S&P 500 rises to 544, but
low-probability large-loss expectations relative to its greatest returns occur if the S&P 500 drops
the Monte Carlo simulations. below 510. Small downward moves of the S&P 500
The VAR calculations for Portfolio 2 expose (between 510 and 529), and larger upward moves
several weaknesses of VAR, which can be managed (above 544) produce losses.
with the addition of stress testing and limit policies. The center panel of Figure 7 shows the return
These weaknesses are illustrated by viewing the on Portfolio 2 at the end of the first two-week
differences in the portfolio's return over various holding period, again as a function of the underly-
~-2
o
-4
, ,q crisis, the Gulf War, and the Mexican peso crisis,
not only are key factors such as "maximum" vola-
tility difficult to predict but also correlation rela-
tionships often change substantially during ex-
treme market moves. The increasing complexity
and optionality of many derivatives makes relevant
After June 545 Call Expires scenario selection even harder. Given these chal-
15
lenges, many firms design stress tests to analyze the
10
impact of large historical market moves. In our
0 experience, portfolios do not necessarily produce
~-5 their greatest losses during extreme market moves.
-10
Whether asset based or asset plus liability based,
-15 I I__ I I I I I I I I I portfolios often possess Achilles' heels that require
501 505 509 513 517 521 525 529 533 537 541 545 549 only small moves or changes between instruments
S&P Price
or markets to produce significant losses. Stress
testing extreme market moves will do little to reveal
the greatest risk of loss for such portfolios. Further-
more, a review of a portfolio's expected behavior
over time often reveals that the same stress test that
ing asset price (the S&P 500). At about the S&P indicates a small impact today indicates embedded
value of 541, the magnitude of Portfolio 2's perfor- land mines with a large impact during future
mance changes by a factor of four from the first periods. This trait is particularly true of options-
one-day horizon to the first two-week horizon, as based portfolios that change characteristics because
illustrated by the amplitude of the graphs (1 per- of time rather than because of changes in the
cent versus 4 percent, respectively). The price inter- components of the portfolio. The need for other risk
vals under which Portfolio 2 loses and makes measures--for example, limits that restrict writing
money change as well. uncovered call options--is clear.
Portfolio 2 poses significantly different risks at
different points in time. For example, as shown in Portfolio 3
the bottom panel of Figure 7, prior to the expiration Portfolio 3 consists of the combination of Port-
of the short position in the June call (strike of 545), folio 1 and Portfolio 2. The portfolios are equally
the portfolio presents the possibility of huge loss. weighted, with the net investment in Portfolio 3
After expiration of the June 545 call, however, totaling $1 million.
Portfolio 2 no longer presents this possibility. Again, the VAR analyses, shown in Figure 8,
From management's perspective, VAR fails as reveal a wide range of risk profiles for the portfolio.
<> M o n t C I RiskMetrics 2w
] 2.51
3.00
M o n t C I BIS/Basle l d ::::~;:~;:~:~::;i;;:i!:;i:J!ii!;!i:10.80
:ii::iii:ii:i:i:i;1.08
:i::]
2.97
M o n t C I BIS/Basle 2w ii ~i: iji i:ii:~i::~i:i~:i!~i:i!:~:!i~i~i!:i~i:~i~i:i~:ii:~:j~::i:~!ii:j~i:i:~i~3.4
i~i:i:i~i~:i:]
I I I I I I
0.5 1.0 1.5 2.0 2.5 3.0 3.5
VAR (%)
[ ] 5% Probability [ ] 1% Probability
As in the case of Portfolios 1 and 2, historical correlated with movements in the Italian lira or tile
simulations present a different view of risk than do Mexican peso? Is the price of Saudi Light correlated
the Monte Carlo simulations, and Portfolio 3's VAR with movements in the price of natural gas? If so,
differences are magnified over the two-week hori- by how much? VAR requires that the user deter-
zon. One-day returns for this multiasset class port- mine correlations not only within markets (for
folio display more consistency under the VAR example, currency underlyings or commodity un-
methodologies than one-day returns for the single- derlyings) but also across markets (for example,
asset class portfolios that compose it. how do changes in the bond market in the United
The sensitivity to correlation assumptions is States affect the Australian equity market?). Given a
demonstrated by the difference in results between portfolio with multiple instruments within and
the Monte Carlo simulations under RiskMetrics and across markets, VAR varies significantly under al-
BIS/Basle factors. Under the RiskMetrics model, ternate correlation assumptions. Pension funds
positive correlation is assumed between the Trea- have addressed the issues of correlation for decades
sury strips in Portfolio 1 and the S&P 500 equity
in studying strategic versus tactical allocation of
index positions in Portfolio 2. Under the BIS/Basle
assets. Correlation issues are also a crucial compo-
method, the correlation is assumed to be 1 between
nent of performance measurement across asset
long positions and -1 between long and short
classes. A single approach to assessing correlation
positions. Not surprisingly, the BIS/Basle VAR
does not exist, and opposite views are common. For
factors are higher than RiskMetrics factors in all
example, what happens when a market breaks
cases.
Figure 9 compares the results of the historical through its historical or implied trading pattern
simulation for Portfolio 3 with the results of the and violates the correlation assumption in place'.,'
Monte Carlo simulations for one-day and two-week Recently, many currencies that previously had
returns. As with Portfolio 2, return patterns change displayed little or no correlation with the Mexican
significantly when the holding period is increased peso made sympathy moves during the devalua-
from one day to two weeks, but high-probability tion of the Peso. In some cases, the increased
extreme events no longer occur. volume of barrier options on spreads (also known
Correlation assumptions are an important as- as knock-out or knock-in options) has been blamed
pect of VAR. Firms select quite different answers to for unexpected high correlations during periods
which exposures are allowed to offset each other when market levels approach strike levels, with
and by how much. For example, is the Japanese yen both the writers and the buyers of the barriers
One-Day Returns
16
14
12
10 --
8 -
6 -
/.!
4 - ,/' ." '\
2 - ..'.Td ' ..
0 I
-2.0 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2.0
Return (%)
Two-Week Returns
7
5 - / .." ~ ' % , 1.
(... ~",,,:
/, ..'. -,_ ",~ .'.
4 -
• .'Y.i .'". : "< "
:'....~..'. ..': :
3 -
o "-)P - . E"
2 - J/: '.." ::
1 -
-/?
0 ~'
-4. -3.2 -2.4 -1.6 -0.8 0 0.8 ] .6 2.4 3.2 4.0
Return (%)
s u s p e c t e d of t r a d i n g in l a r g e v o l u m e to i n f l u e n c e I n o u r r e v i e w of d i f f e r e n t a p p r o a c h e s to V A R ,
the outcome. s o m e f i r m s a s s u m e d t h a t all c a s h f l o w s w e r e c o r r e -
Correlation assumptions also can mask risks l a t e d a c r o s s all m a r k e t s a n d o t h e r s a s s u m e d a l o w e r
t h a t m a y b e s i g n i f i c a n t for m a n y firms. F o r e x a m - d e g r e e of c o r r e l a t i o n . S o p h i s t i c a t e d m e a n - v a r i a n c e
ple, m a n y p o r t f o l i o s d i s p l a y e m b e d d e d r o l l o v e r m o d e l s - - f o r e x a m p l e , t h e o n e u s e d to c o m p u t e t h e
risk created through timing mismatches. An exam- R i s k M e t r i c s d a t a s e t - - a l l o w c o r r e l a t i o n for all in-
p l e is t h e c o m m o n s t r a t e g y t h a t f u n d s u s e to h e d g e s t r u m e n t s a c r o s s all m a r k e t s t h a t a r e c o v e r e d . A t
long-dated foreign currency positions by rolling the other extreme are models that allow correlation
over short-dated forward foreign exchange con- o n l y w i t h i n a s s e t c l a s s e s (e.g., f i x e d i n c o m e , f o r e i g n
tracts. U n d e r c o m m o n t i m e h o r i z o n s for V A R a n d e x c h a n g e , e q u i t y ) a n d r e q u i r e p e r f e c t p o s i t i v e cor-
its c o r r e l a t i o n a s s u m p t i o n s , a flat c u r r e n c y r i s k r e l a t i o n a c r o s s r i s k - f a c t o r g r o u p s . A n e x a m p l e of
p o s i t i o n o f t e n a p p e a r s . This p a t t e r n c a n m a s k t h e this a p p r o a c h is t h e p r o p o s e d a m e n d m e n t to t h e
l o n g - t e r m r o l l o v e r r i s k i n t r i n s i c in h e d g i n g t h e 1988 Basle C a p i t a l A c c o r d o n m a r k e t risks.
c u r r e n c y r i s k of 10- o r 2 0 - y e a r s e c u r i t i e s w i t h o n e - V A R r e q u i r e s t h e u s e of m a t h e m a t i c a l m o d e l s
to t h r e e - m o n t h c u r r e n c y c o n t r a c t s . to v a l u e i n d i v i d u a l i n s t r u m e n t s , as w e l l as to v a l u e
1. Note that selection of key statistical parameters such as the 4. See Tanya Styblo Beder, "The Realities of Marking to Model,"
mean and variance can significantly affect distributions and, Bank Accounting & Finance, vol. 7, no. 4 (Summer 1994):4-12.
therefore, the results of simulations. 5. Note that a single RBC approach exists for life insurance
2. Duration is defined here as the change in price with respect to companies, and a single alternate approach exists for proper-
yield. ty/casualty companies.
3. The Monte Carlo simulations are based on the assumptions 6. The author wishes to thank Frank Iacono, Maarten Nederlof,
that the Treasury strip yields are lognormally distri- Tom Riesing, Anil Suri, and Charles Taylor for their invalu-
buted and that the average change, or "drift," in each yield able assistance and input during the preparation of this
is zero. article.