Finance and Economics Discussion Series

Divisions of Research & Statistics and Monetary Affairs

Federal Reserve Board, Washington, D.C.

A Review of Backtesting and Backtesting Procedures

Sean D. Campbell
2005-21

NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS)
are preliminary materials circulated to stimulate discussion and critical comment. The
analysis and conclusions set forth are those of the authors and do not indicate
concurrence by other members of the research staff or the Board of Governors.
References in publications to the Finance and Economics Discussion Series (other than
acknowledgement) should be cleared with the author(s) to protect the tentative character
of these papers.
 A Review of Backtesting and Backtesting Procedures
Sean D. Campbell∗
April 20, 2005

Abstract
This paper reviews a variety of backtests that examine the adequacy of Value-at-
Risk (VaR) measures. These backtesting procedures are reviewed from both a sta-
tistical and risk management perspective. The properties of unconditional coverage
and independence are defined and their relation to backtesting procedures is discussed.
Backtests are then classified by whether they examine the unconditional coverage prop-
erty, independence property, or both properties of a VaR measure. Backtests that ex-
amine the accuracy of a VaR model at several quantiles, rather than a single quantile,
are also outlined and discussed. The statistical power properties of these tests are
examined in a simulation experiment. Finally, backtests that are specified in terms of
a pre-specified loss function are reviewed and their use in VaR validation is discussed.

∗
Board of Governors of the Federal Reserve System. I would like to thank Mike Gibson, Jim O’Brien,
Pat Parkinson and Hao Zhou for useful discussions and comments on an earlier draft. The usual disclaimer
applies. The views expressed in this paper are those of the author but not necessarily those of the Board
of Governors of the Federal Reserve System, or other members of its staff. Address correspondence to Sean
Campbell, The Federal Reserve Board, Mail Stop 91, Washington DC 20551.
1 The Current Regulatory Framework
Since 1998, regulatory guidelines have required banks with substantial trading activity to set
aside capital to insure against extreme portfolio losses. The size of the set-aside, or market
risk capital requirement, is directly related to a measure of portfolio risk. Currently, portfolio
risk is measured in terms of its “value-at-risk”. The value-at-risk or V aR of a portfolio is
defined to be the dollar loss that is expected to be exceeded only α × 100% of the time over a
fixed time interval. If, for example, a financial institution reports a 1% V aR of $10,000,000
over a 1 day horizon then this means that 1% of the time the institution would be expected
to realize a loss in excess of $10,000,000. The current regulatory framework requires that
financial institutions use their own internal risk models to calculate and report their 1%
value-at-risk, V aR (0.01), over a 10 day horizon.
Market risk capital requirements are directly linked to both the estimated level of portfolio
risk as well as the V aR model’s performance on backtests. Specifically, the risk based capital
requirement is set as the larger of either the bank’s current assessment of the 1% V aR over
the next 10 trading days or a multiple of the bank’s average reported 1% V aR over the
previous 60 trading days plus an additional amount that reflects the underlying credit risk
of the bank’s portfolio. In precise terms the market risk capital is defined as,
!
1 X
59
MRCt = max V aRt (0.01), St V aRt−i (0.01) +c (1)
60 i=0

where St reflects the multiplication factor that is applied to the average of previously reported
V aR estimates. When a V aR model indicates more risk, the risk based capital requirement
rises. What may be less clear from the above expression is that the risk based capital
requirement also depends on the accuracy of the V aR model.
Importantly, the multiplication factor, St , varies with backtesting results. St is deter-
mined by classifying the number of 1% V aR violations in the previous 250 trading days, N,
into three distinct categories as follows,


 3.0 if N ≤4 green
St = 3 + 0.2(N − 4) if 5 ≤ N ≤ 9 yellow (2)


4.0 if 10 < N red

so that a V aR measure which is violated more frequently results in a larger multiplication

factor and accordingly a larger risk based capital requirement. Specifically, as long as the 1%
V aR has been violated four times or less in the previous 250 trading days or 2 % of the time
the multiplication factor remains at its minimal value of three. As the number of violations,

1
N, increases beyond four so too does the multiplication factor that determines the market
risk capital. In the event that more than ten violations of the 1% V aR are recorded in a 250
day span, corresponding to 4% of the sample period, the V aR model is deemed inaccurate
and immediate steps are required to improve the underlying risk management system.
This “traffic light” approach to backtesting represents the only assessment of V aR ac-
curacy prescribed in the current regulatory framework. This paper will suvey backtesting
methods and will discuss how these backtesting methods might improve on the current reg-
ulatory framework.

2 The Statistical Approach to Backtesting

2.1 The Definition of Value-at-Risk, V aR
Before describing V aR backtests in any detail it is important to have a concrete definition
of V aR in mind. In mathematical terms, a portfolio’s value-at-risk is defined to be the α
quantile of the portfolio’s profit and loss distribution,

V aRt (α) = −F −1 (α|Ωt ) (3)

where F −1 (·|Ωt ) refers to the quantile function of the profit and loss distribution which varies
over time as market conditions and the portfolio’s composition, as embodied in Ωt , change.
The negative sign is a normalization that quotes V aR in terms of positive dollar amounts, i.e.
“losses”. As previously mentioned, if today’s 1% V aR, V aRt (0.01), is $10,000,000 then this
amounts to saying that 1% of the time we should expect to observe a loss on this portfolio in
excess of $10,000,000. Since financial institutions adhere to risk-based capital requirements
by using their own internal risk models to determine their 1% value-at-risk, V aRt (0.01),
it is important to have a means of examining whether or not reported V aR represents an
accurate measure of a financial institution’s actual level of risk.1

2.2 The Statistical Framework of V aR Backtests

Since the late 1990’s a variety of tests have been proposed that can be used to gauge the
accuracy of a putative V aR model. While these tests differ in their details many of them focus
on a particular transformation of the reported V aR and realized profit or loss. Specifically,
1
In general, an institution’s V aR refers to a historical sequence of V aR measurements at a given level of
α, V aRt (α) , t = 1, 2, ..., T . To economize on notation we will refer to this object simply as V aR.

2
consider the event that the loss on a portfolio exceeds its reported V aR, V aRt (α). Denoting
the profit or loss on the portfolio over a fixed time interval, i.e. daily, as xt,t+1 then define
the “hit” function as follows,
(
1 if xt,t+1 ≤ −V aRt (α)
It+1 (α) = (4)
0 if xt,t+1 > −V aRt (α)

so that the hit function sequence, e.g. (0, 0, 1, 0, 0, ..., 1), tallies the history of whether or not
a loss in excess of the reported V aR has been realized.
Christoffersen (1998) points out that the problem of determining the accuracy of a V aR
t=T
model can be reduced to the problem of determining whether the hit sequence, [It+1 (α)]t=1 ,
satisfies two properties.

1. Unconditional Coverage Property - The probability of realizing a loss in excess of the

reported V aR, V aRt (α), must be precisely α × 100% or in terms of the previous
notation, Pr (It+1 (α) = 1) = α. If it is the case that losses in excess of the reported
V aR occur more frequently than α ×100% of the time then this would suggest that the
reported V aR measure systematically understates the portfolio’s actual level of risk.
The opposite finding of too few V aR violations would alternatively signal an overly
conservative V aR measure.

2. Independence Property - the unconditional coverage property places a restriction on

how often V aR violations may occur. The independence property places a strong re-
striction on the ways in which these violations may occur. Specifically, any two elements
of the hit sequence, (It+j (α) , It+k (α)) must be independent from each other. Intu-
itively, this condition requires that the previous history of V aR violations,
{..., It−1 (α) , It (α)}, must not convey any information about whether or not an ad-
ditional V aR violation, It+1 (α) = 1, will occur. If previous V aR violations presage
a future V aR violation then this points to a general inadequacy in the reported V aR
measure. As an example, suppose that 1% V aR violations always occur in pairs so
that one 1% V aR violation is immediately followed by another. Accordingly, the prob-
ability of observing a loss in excess of the 1% V aR after one has already been observed
is 100% and not 1%. Hence, the reported V aR does not accurately reflect the loss
that can be expected to be exceeded 1% of the time. In this example, the reported
1% V aR would have to be increased, i.e. a larger loss is to be expected following a
V aR violation, in order to satisfy the independence property. In general, a clustering
of V aR violations represents a violation of the independence property that signals a
lack of responsiveness in the reported V aR measure as changing market risks fail to

3
 be fully incorporated into the reported V aR measure thereby making successive runs
of V aR violations more likely.

It is important to recognize that the unconditional coverage and independence properties of

the hit sequence are separate and distinct and must both be satisfied by an accurate V aR
model. In principle, a particular V aR model could result in a hit sequence that satisfies
the unconditional coverage property but not the independence property. Likewise, another
V aR model could result in a hit sequence that satisfies the independence property but not
the unconditional coverage property. Only hit sequences that satisfy both properties can be
described as evidence of an accurate V aR model. Each property characterizes a different
dimension of an accurate V aR model.
In the case of the unconditional coverage property, it is clear that a V aR model which
does not exhibit any predictability in violations but exhibits violations well in excess of the
prescribed level, e.g. 1%, is problematic. Consider, for example, a bank that reports its
actual V aR at the 10% level instead of the 1% level prescribed in the current regulatory
framework. The independence property represents a more subtle yet equally important
characteristic of an accurate V aR model. A V aR model that results in a predictable hit
series implies that the underlying V aR model does not react quickly enough to changing
market conditions. Consider a case in which V aR violations tend to cluster in groups. This
clustering would imply that market risk capital requirements are underfunded for prolonged
periods during episodes of increased risk. The consequences of being exposed to succesive
periods of heightened risk may be as problematic as a systematic under reporting of risk
exposure. In particular, it may be even more difficult to recover from a string of unanticipated
and large losses than it would be to recover from a higher than expected number of large
losses that are spread out evenly over time. In this sense, it is important to recognize that
the unconditional coverage and independence property jointly determine the accuracy of a
given V aR model. A V aR model that satisfies one property or the other will result in an
inaccurate description of the bank’s risk exposure.
t=T
These two properties of the “hit” sequence, [It+1 (α)]t=1 , are often combined into the
single statement,

i.i.d.
It (α) ∼ B (α)

which reads that the hit sequence, It (α), is identically and independently distributed as
a Bernoulli random variable with probability α. These two properties of the hit sequence

4
represent a complete characterization of an accurate V aR measure at a given level, α. Put
differently, any series of V aR measures which are accurate in the sense that the reported
V aR coincides with the α quantile of the profit and loss distribution must produce a hit series
with the unconditional coverage and independence properties. Conversely, if a given V aR
series produces a hit sequence that exhibits the unconditional coverage and independence
properties then it is accurate in the sense previously described.
Ultimately, the main contribution of Christoffersen’s (1998) insight is that the problem
of determining whether a given V aR measure is accurate can be reduced to examining the
unconditional coverage and independence properites of the resulting hit sequence. Accord-
ingly, many of the V aR backtests that have been proposed in the last few years seek to
examine whether the hit sequence satisfies one or both of these properties. We now turn to a
discussion of these backtests. First, we discuss tests of the unconditional coverage property,
then tests of the independence property and finally tests that seek to jointly examine the
unconditional coverage and independence properties.

3 Tests of V aR Accuracy
3.1 Unconditional Coverage Tests
Some of the earliest proposed VaR backtests, e.g. Kupiec (1995), focused exclusively on the
property of unconditional coverage. In short, these tests are concerned with whether or not
the reported V aR is violated more (or less) than α × 100% of the time. Kupiec (1995), for
example, proposed a proportion of failures or POF test that examines how many times a
financial institution’s V aR is violated over a given span of time. If the number of violations
differs considerably from α × 100% of the sample, then the accuracy of the underlying risk
model is called into question. Using a sample of T observations, Kupiec’s (1995) test statistic
takes the form,
 T −I(α)  I(α) !
1−α
b α
b
P OF = 2log
1−α α
1
α
b = I (α) (5)
T
XT
I (α) = It (α)
t=1

5
and close inspection of the test statistic reveals that if the proportion of V aR violations,
α
b × 100%, is exactly equal to α × 100% then the POF test takes the value zero, indicating
no evidence of any inadequacy in the underlying V aR measure. As the proportion of V aR
violations differs from α × 100%, the POF test statistic grows indicating mounting evidence
that the proposed V aR measure either systematically understates or overstates the portfolio’s
underlying level of risk. In the case of a 1% V aR, if the test is carried out using one year,
i.e. T = 255, of daily data then a common threshold for determining that a V aR measure
is under reporting the actual 1% V aR is 8 violations or roughly 3% of the sample.2
At this point is useful to draw a connection between tests of unconditional coverage and
the current regulatory framework. Recall that the market risk capital multiplier, St , is solely
determined by the number of times the 1% V aR has been violated in the past 250 trading
days. Similarly, Kupiec’s POF test is only a function of the number of V aR violations during
the sample period, αb. Accordingly, there is a close connection between movements in the
market risk multiplier and the POF test of unconditional coverage. For example, consider
the fact that the market risk capital requirement begins to increase after four violations of
the 1% V aR have occurred in the previous 250 days. This threshold is consistent with a
value of the Kupiec POF test of 0.76. A realization of the POF test of this magnitude or
larger would be expected to occur roughly 38% of the time in the case that the V aR model
under consideration is in fact accurate. At the other extreme, observing ten V aR violations,
in which case the multiplier is set to its maximal value of 4.0 and the underlying V aR model
is deemed inaccurate requiring immediate steps to improve its accuracy, is equivalent to a
POF test value of 12.95 which would be expected to occur much less than 1% of the time
in the case that the V aR model under consideration is accurate. Accordingly, the market
risk capital multiplier can be interpreted as a test of unconditional coverage that mandates a
larger market risk capital set-aside as the evidence that the V aR model under consideration
is inaccurate mounts.
Kupiec’s POF test of unconditional coverage is a well known example of a V aR back-
test. There are, however, a variety of statistical tests that could be employed to assess the
unconditional coverage property of a given V aR model. One alternative would be to simply
base a test directly on the sample average of the number of V aR violations over a given
time period, α
b. Under the assumption that the V aR under consideration is accurate then a
2
This is the threshold that is consistent with a Type I error rate of 5%. Put differently, if this threshold
is used in judging the adequacy of a proposed V aR measure, the test will identify an accurate V aR measure
as being inaccurate 5% of the time.

6
scaled version of α
b,
√
T (b
α − α)
z= p , (6)
α (1 − α)

has an approximate standard normal distribution. Moreover, the exact finite sample distri-
bution of z is known and so hypothesis tests can be conducted in exactly the same way that
hypothesis tests are conducted in the case of Kupiec’s POF statistic.3
While tests of unconditional coverage provide a useful benchmark for assessing the ac-
curacy of a given V aR model, these tests are hampered by two shortcomings. The first
shortcoming is that these tests are known to have difficulty detecting V aR measures that
systematically under report risk. From a statistical point of view these tests exhibit low
power in sample sizes consistent with the current regulatory framework, i.e. one year. Ku-
piec (1995), for example, reports that when the threshold of 8 violations in one year is used
in testing then the chance of detecting a V aR measure that reports the 3% V aR instead of
the required 1% V aR, thereby under reporting its portfolio risk and under funding its market
risk capital, is only 65%. In this case the odds of detecting an institution that systematically
under reports their V aR is slightly better than a coin flip. Moreover, the size of the under
reporting can be quite substantial. Consider a case in which portfolio profits and losses are
normally distributed. In this situation, reporting the portfolio’s 3% V aR instead of its 1%
V aR would result in a capital charge that is 20% too small. Specifically, the 1% quantile
of the normal distribution corresponds to 2.33 standard deviations while the 3% quantile
corresponds to 1.88 standard deviations. If portfolio returns are more prone to very large
profits and losses, as in the case of the t distribution, then the understatement will be even
more severe. In the case that portfolio returns are distributed according to a t(5) distri-
bution a similar calculation reveals that reporting the 3% V aR instead of the required 1%
V aR would result in a capital charge that is 27% too small. The problem of low power may
be ameliorated to some extent by increasing the sample over which the test is constructed.
In the case of the above example, Kupiec (1995) reports that increasing the sample period
from one year to two years increases the chance of detecting the systematic under reporting
of V aR from 65% to roughly 90%. This increase in detection, however, comes at the cost of
reducing the frequency with which the adequacy of the V aR measure can be assessed.
A second shortcoming of these tests is that they focus exclusively on the unconditional
3
The z statistic is actually the Wald variant of the likelihood ratio statistic proposed by Kupiec (1995).
One potential advantage of the Wald test over the likelihood ratio test is that it is well-defined in the case
that no V aR violations occur. Kupiec’s POF test is undefined in this case since the log of 0 is undefined.
Moreover, the possibility that no violations occur in a period as short as one year is not trivial. Accordingly,
it may be advisable to employ an unconditional coverage test which is well-defined in this event.

7
coverage property of an adequate V aR measure and do not examine the extent to which
the independence property is satisfied. Accordingly, these tests may fail to detect V aR
measures that exhibit correct unconditional coverage but exhibit dependent V aR violations.
As discussed previously, V aR models that violate the independence property may result in
losses that exceed the reported V aR in clusters or streaks. A streak of large unexpected
losses may result in even more stress on a financial institution than large unexpected losses
that occur somewhat more frequently than expected but are spread out over time. As
an example, over a one year period, four losses in a row that exceed an institutions 1%
V aR may be a larger signal of inadequate risk management practices than eight losses that
occur evenly over the two year period. Moreover, it is interesting to note that while the
second scenario would result an increase in the institutions market risk capital multiplier,
the first scenario would not. To the extent that dependent V aR violations signal a lack of
responsiveness to changing market conditions and inadequate risk reporting, relying solely
on tests of unconditional coverage in backtesting would appear problematic.

3.2 Independence Tests

In light of the failure of tests of unconditional coverage to detect violations of the indepen-
dence property of an accurate V aR measure, a variety of tests have been developed which
explicitly examine the independence property of the V aR hit series, It (α). An early and
influential test in this vein is Christoffersen’s (1998) Markov test. The Markov test examines
whether or not the likelihood of a V aR violation depends on whether or not a V aR viola-
tion occurred on the previous day. If the V aR measure accurately reflects the underlying
portfolio risk then the chance of violating today’s V aR should be independent of whether
or not yesterday’s V aR was violated. If, for example, the likelihood of a 1% V aR violation
increased on days following a previous 1% V aR violation then this would indicate that the
1% V aR following a violation should in fact be increased. The test is carried out by creating
a 2 × 2 contingency table that records violations of the institution’s V aR on adjacent days as
in Table 1. If the V aR measure accurately reflects the portfolio’s risk then the proportion of
violations that occur after a previous violation, It−1 = 1, should be the same as the propor-
tion of violations that occur after a day in which no violation occurred, It−1 = 0. In terms
of the elements in Table 1 it should be the case that N1N+N 1
3
= N2N+N
2
4
. If these proportions
differ greatly from each other then this calls the validity of the V aR measure into question.
A more recent independence test that has been suggested by Christoffersen and Pelletier
(2004) uses the insight that if V aR violations are completely independent from each other
then the amount of time that elapses between V aR violations should be independent of the
amount of time that has elapsed since the last violation. In this sense, the time between

8
V aR violations should not exhibit any kind of “duration dependence”. The chance that a
V aR violation occurs in, say, the next 10 days should not depend upon whether the last V aR
violation occurred 10 or 100 days ago. Unlike, the Markov test, this duration based test can
not be constructed simply by computing a 2 × 2 contingency table. Carrying out the test
requires estimating a statistical model for the duration of time between V aR violations by
the method of maximum likelihood which must be done using numerical methods. Despite
the lack of transparency in the test statistic, Christoffersen and Pelletier (2004) provide some
evidence that this test of independence has more power than the Markov test to detect a
V aR measure that violates the independence property. They show, for example, that in a
setting in which the Markov test identifies an inaccurate V aR measure 28% of the time their
duration based test identifies the inaccurate V aR measure 46% of the time.
While independence tests provide an important source of discriminatory power in detect-
ing inadequate V aR measures, they are subject to one main drawback. All independence
tests start from the assertion that any accurate V aR measure will result in a series of in-
t=T
dependent hits, [It (α)]t=1 . Accordingly, any test of the independence property must fully
describe the way in which violations of the independence property may arise. In the case
of the Markov test, for example, the independence property may be violated by allowing
for the possibility that the chance of violating tomorrow’s V aR depends on whether or not
yesterday’s V aR was violated. There are, however, myriad ways in which the independence
property might be violated. For example, it might be the case that the likelihood of violat-
ing tomorrow’s V aR depends not on whether yesterday’s V aR was violated but whether the
V aR was violated one week ago. If this is the way in which the lack of the independence
property manifests itself then the Markov test will have no ability to detect this violation of
the independence property.
In statistical terms, any independence test must completely specify the alternative hy-
pothesis that the independence property is being tested against. Intuitively, an independence
test must describe the types of anomalies that it is going to look for when examining whether
or not the independence property is satisfied. Violations of the independence property which
are not related to the set of anomalies defined by the test will not be systematically detected
by the test. As a result, independence tests are only likely to be effective at detecting inac-
curate V aR measures to the extent that the tests are designed to identify violations of the
independence property in ways that are likely to arise when internal risk models fail to pro-
vide accurate V aR measures. This kind of information would likely come from a thorough
understanding of the types of situations in which common risk models fail to accurately
describe portfolio risk. To the extent that risk models can be slow to react to changing
market conditions, tests that examine the amount of clustering in V aR violations such as

9
the Markov test may be useful in identifying inaccurate V aR models.

3.3 Joint Tests of Unconditional Coverage and Independence

An accurate V aR measure must exhibit both the independence and unconditional coverage
property. Accordingly, tests that jointly examine the unconditional coverage and indepen-
dence properties provide an opportunity to detect V aR measures which are deficient in one
way or the other. Both the Markov test of Christoffersen (1998) and the duration test of
Christoffersen and Pelletier (2004) can be extended to jointly test for both independence
and unconditional coverage. In the case of the Markov test, it is particularly simple to
characterize how the joint test examines both properties of the V aR measure.
Recall that the Markov test proceeds by constructing a 2 × 2 contingency table that
records the frequency of V aR violations and non-violations on successive days as in Table 1.
The Markov independence test then examines whether the proportion of violations following
previous violations is equal to the proportion of violations following previous non-violations.
In terms of the notation in Table 1 this is equivalent to examining whether N1N+N 1
3
= N2N+N
2
4
.
If the V aR measure also exhibits the unconditional coverage property then these proportions
+N2
should match the total proportion of violations, N1N , and this should be identical to α, i.e.
N1 +N2
N1
N1 +N3
= N2 +N4 = N = α. Accordingly, the joint Markov test examines whether there is
N2

any difference in the likelihood of a V aR violation following a previous V aR violation or non-

violation and simultaneously determines whether each of these proportions is significantly
different from α.
The discussion of joint tests so far might seem to suggest that joint tests are universally
preferable to tests of either the unconditional coverage property or independence property
alone. While joint tests have the property that they will eventually detect a V aR measure
which violates either of these properties, this comes at the expense of a decreased ability to
detect a V aR measure which only violates one of the two properties. If, for example, a V aR
measure exhibits appropriate unconditional coverage but violates the independence property
then an independence test has a greater likelihood of detecting this inaccurate V aR measure
than a joint test. The joint test is hampered by the fact that one of the two violations it is
designed to detect, namely violations of unconditional coverage, is actually satisfied by the
V aR measure. The fact that one of the two properties is satisfied makes it more difficult for
the joint test to detect the inadequacy of the V aR measure. As an example, consider a 5%
V aR measure which exhibits the unconditional coverage property but not the independence
property. Specifically, suppose that a V aR violation occurs 20% of the time after a previous
V aR violation but that a V aR violation only occurs 4.2% of the time after a period in which

10
no V aR violation occurs.4 In a one year sample of daily observations the joint Markov test
detects this inaccurate V aR measure 50% of the time while the independence Markov test
detects this inaccurate V aR measure 56% of the time. The increased power of detection
that comes from choosing a test that focuses one of the two properties indicates that either
unconditional coverage or independence tests alone are preferable to joint tests when prior
considerations are informative about the source of the V aR measure’s potential inaccuracy.

3.4 Tests Based On Multiple V aR Levels-α

All of the backtests discussed so far have focused on determining the adequacy of a V aR
measure at a single level, α. In general, however, there is no need to restrict attention to a
single V aR level. The unconditional coverage and independence property of an accurate V aR
measure should hold for any level of α. Crnkovic and Drachman (1997), Diebold, Gunther
and Tay (1998) as well as Berkowitz (2001) have all suggested backtests based on multiple
V aR levels. These authors exploit the insight that if portfolio risk is adequately modeled
then the 1% V aR should be violated 1% of the time, the 5% V aR should be violated 5% of
the time, the 10% V aR should be violated 10% of the time and so on. Furthermore, a V aR
violation at any level should be independent from a V aR violation at any other level so that,
for example, a violation of a portfolio’s 5% V aR today should not portend a violation of the
portfolio’s 1% V aR tomorrow. In short, V aR violations at all levels should be independent
from each other.
Crnkovic and Drachman (1997) as well as Diebold, Gunther and Tay (1998) point out
that this insight can be formalized in the following manner. Again, consider xt,t+1 to be
the realized profit or loss from time period t to t + 1, i.e. over one day. Next consider
the quantile of the probability distribution that corresponds to the observation, xt,t+1 . In
particular define,

zt+1 = F −1 (xt,t+1 |Ωt ) (7)

where, as before, F −1 (·|Ωt ) refers to the conditional quantile function of the profit and loss
distribution. It is worth comparing the observed quantile, zt , to the hit indicator, It (α).
The reported quantile provides a quantitative and continuous measure of the magnitude of
realized profits and losses while the hit indicator only signals whether a particular threshold
was exceeded. In this sense, the series of reported quantiles, [zt ]Tt=1 , provides more infor-
mation about the accuracy of the underlying risk model. Moreover, a series of reported
4
These relative frequencies imply that a V aR violation occurs 5% of the time on average.

11
quantiles that accurately reflects the actual distribution of profits and losses exhibits two
key properties.

1. Uniformity - The series, [zt ]Tt=1 , should be uniformly distributed over the unit interval
[0, 1]. This property of zt is analogous to the statement that the α × 100% V aR should
be violated α × 100% of the time for each and every α and is a direct parallel to the
unconditional coverage property of the hit series, It (α), in the case of a single V aR
measure.

2. Independence - The series, [zt ]Tt=1 , should be independently distributed. This is anal-
ogous to the statement that V aR violations should be independent from each other
and that a V aR violation today at, say, the 1% level should not provide any informa-
tion about whether or not a V aR violation will occur tomorrow at the 5% level. This
property of the zt series directly parallels the independence property that is required
of the hit series, It (α), in the case of a single V aR measure.

These two properties are often combined into the single statement,

i.i.d.
zt ∼ U (0, 1) (8)

which reads that the sequence of quantiles is identically and independently distributed as
a Uniform random variable over the unit interval. A variety of statistical tests have been
proposed to assess whether a reported series of quantiles, [zt ]Tt=1 , accord with the uniformity
and independence properties. Just as in the case of assessing the accuracy of a single V aR
measure these tests may be conducted individually, either the uniformity or independence
property may be tested, or jointly.
The main advantage of these tests over and above tests based on a V aR measure at a
single α level is that they, in principle, can provide additional power to detect an inaccurate
risk model. By examining a variety of different quantiles, instead of a single quantile, these
tests can detect violations of the independence or unconditional coverage property across
a range of different V aR levels. This increased power to detect an inaccurate risk model,
however, comes at some cost. One component of the cost comes in the form of an increased
informational burden. In particular, in order to transform xt,t+1 to zt+1 one must have ac-
cess to the entire conditional quantile function, F −1 (·|Ωt ), rather than a single quantile,
F −1 (α|Ωt ). Risk models that assume a particular form of the distribution of portfolio losses
and gains may be reasonable models of extreme outcomes but may not be useful for char-
acterizing the frequency of more moderate outcomes. If the underlying risk model is more
focused on characterizing the stochastic behavior of extreme portfolio losses these models

12
may be misspecified over a range of profits and losses which are not highly relevant from a
risk management perspective. Accordingly, tests that employ the entire series of quantiles,
zt , may signal an inaccurate model due to this source of misspecification.
A second component of the cost relates to a broadening of the definition of portfolio risk.
Statistical tests that examine a variety of different quantiles implicitly broaden the definition
of portfolio risk relative to tests that focus on a V aR measure at a single α level. Current
regulatory guidelines dictate that portfolio risk be measured in terms of the 1% V aR for the
purpose of determining risk-based capital requirements. Accordingly, examining the behavior
of V aR measures at levels aside from the 1% level broadens the scope of the definition of
portfolio risk. The extent to which the definition of portfolio risk is broadened, however, may
be directly controlled by choosing to examine the uniformity and independence properties
of the z series over a particular range of α levels.
One could, for example, rely on a test that examines the frequency of VaR violations
over a range of predetermined α levels. An example of one such test is Pearson’s Q test
for goodness of fit. Pearson’s Q test is based upon the number of observed violations
at a variety of different VaR levels. The test is constructed as follows.5 First, partition
the unit interval into different sub-intervals. For example, one could choose the partition
[0.00, 0.01], [0.01, 0.05], [0.05, 0.10], [0.10, 1.00] which results in four separate regions or bins
on the unit interval. Once a partition is chosen, simply count the number of VaR violations
that occur within each bin. The number of VaR violations that occur in the [0.00,0.01]
range, for example, records the number of days on which a loss in excess of the 1% VaR
occurred. Likewise, the number of VaR violations that occur in the [0.10,1.00] range records
the number of days on which a loss less extreme than the 10% VaR occurred. With the
number of violations that have occurred within each bin in hand the Q test is computed
according to the formula,

2
Xk
N(li ,ui ) − N (ui − li )
Q= (9)
i=1
N (ui − li )

where N(li ,ui ) refers to the number of VaR violations in the ith bin and N refers to the total
number of days being used to construct the test. Also, li and ui refer to the lower and upper
bound of each bin. In the case that the VaR model being tested is accurate in the sense that
the model’s VaR coincides with actual VaR the test is approximately distributed according
to the chi squared distribution with k − 1 degrees of freedom.
As the equation for Q makes clear, the particular partition of the unit interval that is
5
Details concerning the foundations and derivation of this test statistic can be found in, DeGroot M.
(1989), Probability and Statistics, Addison-Wesley, Reading, Massacusetts.

13
chosen may be tailored to focus on a particular set of quantiles of interest. If only quantiles
more extreme than the 10th percentile are of interest, for example, one could use the partition
[0.00, 0.01], [0.01, 0.05], [0.05, 0.10], [0.10, 1.00]. If only quantiles more extreme than the 5th
percentile are of interest than the partition [0.00, 0.01], [0.01, 0.05], [0.05, 1.00] could be used.
In this way the particular partition employed can be used as a means of controlling how
broadly risk is defined in terms of the test.
In order to have a grasp of the quantitative nature of the tradeoff between increasing
power and broadening the scope of the definition of risk, it it useful to examine an example.
In order to better gauge this tradeoff an experiement is conducted in which profits and
losses are generated over a one-year period from a statistical model. The profits and losses
are then analyzed using three different risk models which are employed in the practice of
risk management to varying degrees. Each of the three risk models examined is inaccurate
in the sense that none of them accurately models the true underlying distribution of profits
and losses. Each of the three models is then used to produce a one-year sample of observed
quantiles, [zt ]255
t=1 . The quantiles are then used in conducting Pearson’s Q test. The power of
Pearson’s Q test is compared to the power of Kupiec’s (1995) test of unconditional coverage
at the 1% level to assess the increase in power that arises from considering other quantiles
besides the first percentile. Also, in addition to these three risk models we also consider the
power of Pearson’s Q test and Kupiec’s (1995) test in the case that VaR is systematically
under reported. We examine levels of under reporting that range between 5% - 25%.
The statistical model for portfolio profits and losses is one that assumes a fixed mean of
the profit and loss distribution and focuses on a model of time varying volatility or risk. The
model of time varying risk is one that incorporates three distinct features of volatility that
have been recognized to be salient features of a variety of different financial asset markets
ranging from stock markets to currency markets. First, the model provides for considerable
variation in volatility. Each period volatility changes and sometimes changes quite rapidly so
that a 25% change in volatility over a one month horizon is not particulalry unlikely. Second,
movements in volatility are assumed to exhibit a considerable degree of persistence. Much
research on financial asset volatility finds that increases and decreases in volatility can be
long lived. Volatility does tend towards a long run mean but shocks that either increase or
decrease volatility often persist for several months. Lastly, the model implies that volatility
tends to increase more after large portfolio losses than after large gains. The particular
model for volatility employed in the experiment is Nelson’s (1991) EGARCH(1,1) model.
This model is widely used in empirical studies of financial market volatility and also bears a
close resemblance to the model employed by Berkowitz and O’Brien (2002) in their empirical

14
study of bank trading revenues and VaR. The model’s particular specification takes the form,

xt,t+1 = vt,t+1


vt,t+1 ∼ N 0, σt2 (10)
   


 vt−1,t  v
ln σt = 0.02 + 0.94ln σt−1 + 0.22 
2 2  − 0.05 t−1,t
σt−1  σt−1

where the specific model parameters have been chosen to be consistent with the behavior of
monthly U.S. stock returns over the  period1927-1998. In particular note that the final term
, implies that volatility, ln (σt2 ), rises more after
v
in the volatility specification, −0.05 σt−1,t
t−1

a portfolio loss than a gain.

Three different models are used to measure portfolio risk ranging from naive to more
sophisticated. The first and most naive model assumes, correctly, that losses and gains are
normally distributed but assumes that the variance is P
constant. Accordingly, the V aR model
T −1 2
2 v
uses the standard sample estimate of variance, σ
bT = j=1T j,j+1 . With this variance estimate
in hand, the profit or loss, vt−1,t , on any given day can be sorted into its respective bin by
computing,
 
σt−1 Φ−1 (li ) , σt−1 Φ−1 (ui ) , (11)

for each bin and then assigning the profit or loss to the bin that brackets vt−1,t . As usual
Φ−1 (·) refers to the inverse of the standard normal distribution.
The second V aR model recognizes that volatility changes over time but is unaware of the
particular way in which it changes over time. As a result, the model uses an exponentially
weighted moving average of past squared profits and losses to estimate current volatility.
The weighted moving average volatility estimate takes the form,

2 2 2
σW M A,t = 0.97σRM,t + 0.03vt−1,t (12)

where the decay factor, 0.97, implies that volatility moves in a rather persistent fashion over
time. The exact value of the decay factor was chosen to be consistent with the values that
are employed in practice which often range between 0.94 and 0.97 (Simons, 1996). Again,
the profit or loss on any day can be sorted into its respective bin by computing,
 
σW M A,t−1 Φ−1 (li ) , σW M A,t−1 Φ−1 (ui ) ,

for each bin and then assigning the profit or loss to the bin which brackests vt−1,t .

15
 The third model employs the Historical Simulation approach. Unlike the previous two
models, the Historical Simulation approach makes no assumptions about the distribution of
the profits and losses and does not explicitly model volatility. Instead, Historical Simulation
uses the previous observed history of profits and losses and equates the distribution of future
profits and losses with the observed empirical distribution, (Fb).6 As is common in the
implementation of the Historical Simulation method one year of data is first used in order
to construct the empirical distribution function. Each realized profit or loss is associated
with a bin by ranking it among the previous 255 days’ trading experience. If, for example,
today’s loss was the 10th worst loss over the past 255 days then the loss would be associated
10
with the bin that brackests the 4th percentile ( 255 = 0.039 ≈ 0.04). More details on the
historical simulation method, its application and problems it presents for risk mangement
can be found in Pritsker (2001).
The final scenario we consider to assess the value of Pearson’s Q test as a backtesting
procedure is a seting in which the true VaR is systematically under reported. Specifically,
at each point in time the actual volatility of the future P&L distribution, σt , is known. The
reported VaR, however, is based on only a fraction,β, of true volatility. Accordingly, any
particular profit or loss, vt−1,t , is sorted into one of the bins,
 
(1 − β)σt Φ−1 (li ) , (1 − β)σt Φ−1 (ui ) ,

and varying amounts of under reporting are considered ranging from no under reporting,
β = 0, to a level of 25%, β = 0.25.
It is important to consider how this setting differs from the other three risk models. In
each of the previous three models, VaR is sometimes overstated and sometimes understated
depending on whether the estimated P&L volatility is above or below actual volatility. The
presence of systematic under reporting, however, implies that the reported VaR is always
below the actual VaR. Accordingly, this scenario may more realistically capture an envi-
ronment in which a bank’s risk model consistently fails to account for a portion of the risk
inherent in its trading activity.
Tables 2 and 3 display the power of Pearson’s Q test and Kupiec’s (1995) test of un-
conditional coverage at the 1% level to identify each of the three risk models as well as the
systematically under reported VaR as inaccurate. The power is computed by simulating
the statistical model for the P&L for one year and then computing both Pearson’s Q and
Kupiec’s (1995) backtest. This process was repeated 1,000 times and the fraction of test
statistics that exceed the 95% critical value is the estimate of the test’s power. In the case
PT
6
The empirical CDF, Fb in a sample of T observations is defined as Fb (c) = 1
T t=1 1(xt ≤ c).

16
of Pearson’s Q test, the partition, [0.00, 0.01], [0.01, 0.05], [0.05, 0.10], [0.10, 1.00] was chosen
to focus on percentiles that are associated with relatively rare occurences but does expand
the scope of the definition of risk beyond the first percentile.
Table 2 reports the power of the Q test to identify each of the first three risk mod-
els, recursive, weighted moving average and historical simulation, as inaccurate. The table
suggests that with respect to the first three risk models, some modest gains in power are
attainable by expanding the range of quantiles that are analyzed in the backtest. In the
case of the most inaccurate VaR model, the recursive model, Pearson’s Q test identified it as
inaccurate 33.3 percent of the time while the Kupiec (1995) backtest only identifies the re-
cursive model in 25.5 percent of all simulations. The power gains are less impressive for VaR
models that are less egregiously misspecified. In the case of historical simulation, the Q test
only marginally increases power from 11.8% to 14.8%. The gain in the case of the moving
weighted average model is imperceptible as both tests identify the model as inaccurate in
5% of the simulations. The inability of either tests to identify the weighted moving average
model as inaccurate is due to the fact that while this model does not perfectly model P&L
risk, it provides a very close approximation to the underlying level of risk. These results
suggest that the power gains may be largest in cases where the degree of misspecification
is largest which is noteworthy since these are precisely the cases in which identifying an
inadequate risk model may be most important.
Table 3 displays the power of Pearson’s Q test and the Kupiec (1995) backtest in the
case that VaR is systematically under reported. The table displays the power results for
varying amounts of under reporting ranging between 5%-25%. The last row of the table
reports the ratio of the power of the Pearson Q to the Kupiec backtest. At lower levels
of systematic under reporting, between 5% and 10%, the benefit of Pearson’s Q test are
largest in terms of proportional increase in power. At the 5% level of under reporting, for
instance, Pearson’s Q test identifies the VaR model as being inaccurate 13.5% of the time as
opposed to only 6.3% of the time in the case of the Kupiec backtest. The proportional gain
in power declines as the degree of under reporting increases and the power of both tests rises
towards unity. In the case of a 25% under reporting in VaR, for example, the Q test only
provides a 10% improvement in power over and above the Kupiec test. Despite the smaller
proportional increase, however, the absolute gain in power is substantial. In the case of a
25% under reporting of VaR, Pearson’s Q identifies the model as inaccurate 94.2% of the
time as opposed to 79.7% of the time in the case of the Kupiec test. These results suggest
that Pearson’s Q test may provide more substantial power gains in settings where there risk
is systematically under reported. While the Q test does exhibit some ability to increase
power in the case of the other three risk models, the gain in power in the case of systematic

17
under reporting is considerably larger.

4 Loss Function Based Backtests

All of the backtests that have been discussed so far have focused on examining the behavior
of the hit function. Specifically, recall that the hit function is defined as follows,
(
1 if xt,t+1 ≤ −V aRt (α)
It+1 (α) =
0 if xt,t+1 > −V aRt (α)

where xt,t+1 represents the profit or loss between the end of day t and t + 1. Despite the
fact that the hit function plays a prominent role in a variety of backtesting procedures, the
information contained in the hit function is limited. One might be interested, for example, in
the magnitude of the exceedance rather than simply whether or not an exceedance occurred.
In general, a backtest could be based on a function of the observed profit or loss and the
corresponding model V aR. This would result in the construction of a general loss function,
L (V aRt (α), xt,t+1 ), which could be evaluated using past data on profits and losses and the
reported V aR series. Lopez (1999a,b) suggests this approach to backtesting as an alternative
to the approach that focuses exclusively on the hit series.
As an example, consider a loss function which measures the difference between the ob-
served loss and the V aR in cases where the loss exceeds the reported V aR measure. Specif-
ically, consider the loss function suggested by Lopez (1999b),
(
1 + (xt,t+1 − V aRt (α))2 if xt,t+1 ≤ −V aRt (α)
L (V aRt (α), xt,t+1 ) = (13)
0 if xt,t+1 > −V aRt (α)

which measures how well the model V aR predicts losses when they occur. A backtest could
then be conducted by using actual data on reported V aR and profits and losses to assess
whether or not the realized losses are in accord with what would be expected when the
reported V aR accurately reflects underlying risk. In particular, a backtest using this loss
function based approach would typically be based on the sample average loss,

1X
T
b
L= L (V ar (α) , xt,t+1 ), (14)
T t=1

and then the degree to which the observed average loss is consistent with an accurate V aR
model would have to be assessed. This example highlights the increased flexibility of the

18
loss function based approach. The loss function may be tailored to address specific concerns
that may be of interest when analyzing the performance of a V aR model.
Unfortunately this increased flexibility comes at a substantial increase in the informa-
tional burden associated with assessing the accuracy of the V aR model under consideration.
b is “too large relative to what
Specifically, in order to determine whether the average loss, L,
would be expected” it is necessary to understand the stochastic behavior of the loss func-
tion, L(V aR (α) , xt,t+1 ). Generally, this would require that the backtest make an explicit
assumption about the distribution of profits and losses, xt,t+1 . Lopez (1999b) suggests a
three step procedure for determining the range of values for average loss that are consistent
with an accurate V aR model.

1. Fit a statistical model to the underlying profit and loss data and determine a suitable
model of the distribution of profits and losses, f (xt,t+1 |Ωt ).

2. Generate, from the model derived in the first step, a history of profits and losses and
bi .
the associated V aR, V aRt (α), and construct a value for the average loss, L

3. Repeat the above process for a hvery large number of trials, say 10,000, and use the
ii=10,000
bi
resulting set of average losses, L as an estimate of the distribution of the
i=1
average loss.
h ii=10,000
The quantiles of the empirical distribution of the simulated average losses, Lbi , may
i=1
be used just as in a standard hypothesis testing framework to determine when a proposed
V aR model is acceptably accurate.
As the description of this method makes clear, the backtesting procedure relies on a
precise description of the stochastic behavior of profits and losses, f (xt,t+1 |Ωt ). As a result,
a finding that the average observed loss is “too large” could either signal an inaccurate
risk model or an inaccurate assumption about the stochastic behavior of profits and losses.
This feature of the loss function based approach is at odds with tests that examine the
stochastic properties of the hit function, It+1 (α). Recall that regardless of the stochastic
properties of profits and losses, f (xt,t+1 |Ωt ), as long as the underlying V aR model accurately
reflects the true risk exposure, the hit function has the property that it is independently
and identically distributed as a Bernoulli random variable with a probability of success of
precisely α × 100%. Accordingly, inferences about the accuracy of the V aR model may
be made without assuming anything about the stochastic behavior of profits and losses.
This is a property of hit function based backtests which is not shared by the more general
loss function based backtests. This is not to say, however, that loss function based backtests
have no place in the backtesting process. Loss function based backtests, for example, may be

19
extremely useful for determining whether one V aR model provides a better risk assessment
than another competing V aR model. In this situation, the loss function could be tailored to
address specific concerns about how each V aR model behaves in different types of market
conditions. In this sense, loss function based backtests may be more suited to discriminating
among competing V aR models rather than judging the accuracy of a single model.

5 Conclusion
Verifying the accuracy of internal risk models used in setting market risk capital requirements
requires backtesting. The current regulatory framework uses a “traffic light” approach to
backtesting that is related to Kupiec’s (1995) proportion of failures test. In this paper, the
fundamental properties of an accurate VaR model, independence and unconditional coverage
were defined and their relevance from a risk management perspective was discussed. A variety
of extant tests that examine the validity of the independence property, the unconditional
coverage property or both properties were reviewed. Backtests that use pre-specified loss
functions to evaluate the accuracy of a VaR model were also reviewed. Also, a test that
focuses on other quantiles of the P&L distribution besides the first percentile was discussed
and its power properties were examined in a small simulation experiment. The results of
the simulation experiment suggest that moderate gains in statistical power relative to the
power of a test that only examines the first percentile can be achieved by examining other
quantiles besides the first percentile. Tests that examine several quantiles are most succesful
in identifying inaccurate VaR models in the presence of systematic under reporting of risk.

20
 BIBLIOGRAPHY

Berkowitz, J., “Testing Density Forecasts With Applications to Risk Management,” Journal
of Business and Economic Statistics, 19, 2001, 465-474.
Berkowitz, J. and O’Brien J., “How Accurate Are the Value-at-Risk Models at Commercial
Banks,” Journal of Finance, 57, 2002, 1093-1112.
Christoffersen P., “Evaluating Interval Forecasts,” International Economic Review, 39,
1998, 841-862.
Christoffersen P., and Pelletier D., “Backtesting Value-at-Risk: A Duration-Based Ap-
proach,” Journal of Empirical Finance, 2, 2004, 84-108.
Cassidy, C. and Gizycki M., “Measuring Traded Market Risk: Value-At-Risk And Back-
testing Techniques,” Research Discussion Paper 9708, Reserve Bank of Australia, 1997.
Crnkovic C., and Drachman J., “Quality Control,” in VaR: Understanding and Applying
Value-at-Risk, 1997, London, Risk Publications.
DeGroot M. (1989), Probability and Statistics, Addison-Wesley, Reading, Massachusetts.
Diebold F.X., Gunther T., and Tay A., “Evaluating Density Forecasts with Applications
to Financial Risk Management,” International Economic Review, 39, 1998, 863-883.
Haas M., “New Methods in Backtesting,” Financial Engineering Research Center, Working
Paper, 2001.
Kerkhoff J., and Melenberg B., “Backtesting for Risk-Based Regulatory Capital,” Tilburg
University, Working Paper, 2003.
Kupiec P., “Techniques for Verifying the Accuracy of Risk Management Models,” Journal
of Derivatives, 3, 1995, 73-84.
Lopez J.A., “Regulatory Evaluation of Value-at-Risk Models,” Journal of Risk, 1999a, 1,
37-64.
Lopez J.A., “Methods for Evaluating Value-at-Risk Models,” Federal Reserve Bank of San
Francisco Economic Review, 1999b, 2, 3-17.
Pritsker M., “Evaluating Value-at-Risk Methodologies: Accuracy versus Computational
Time,” Journal of Financial Services Research, 1997, 12, 201-241.
Pritsker M., “The Hidden Dangers of Historical Simulation,” Working Paper 2001-27, Board
of Governors of the Federal Reserve System, 2001.
Simons K. (1996), “Value at Risk - New Approaches to Risk Management,” VAR: Under-
standing and Applying Value-at-Risk, RISK publications, London.
Stahl G., “Backtesting Using a Generalisation of the Traffic-Light-Approach,” Neural Net-
work World, 1997, 4, 565-577.
Sullivan, J., Brooks, R., and Stoumbos Z., “Assessing the Accuracy of Value at Risk,”
Working Paper Rutgers University, 2003.

21
 Table 1: Contingency Table for Markov Independence Test

It−1 = 0 It−1 = 1

It = 0 N1 N2 N1 + N2

It = 1 N3 N4 N3 + N4

N1 + N3 N2 + N4 N

22
Table 2: Power of Pearson’s Q and Kupiec Tests (%)

Q Test Kupiec Test

Risk Model
Recursive 33.3 25.5
Weighted Moving Average 5.00 4.90
Historical Simulation 14.8 11.8

Table 3: Power of Pearson’s Q and Kupiec Tests (%)

Under Reporting Level (α) 5% 10% 15% 20% 25%

backtest

Pearson’s Q 13.5 35.9 63.8 86.0 94.2

Kupiec Test 6.30 19.4 43.8 69.0 79.7

23