REVIEW OF FINANCIAL MARKETS

This definition highlights some key points. Quantitative estimates of

DISSECTING PORTFOLIO STRESS-TESTING stress test outcomes are required, in monetary terms, but stress tests do
Quintin Rayer, Chartered FCSI, Chartered Wealth Manager, Fort Grey not necessarily provide statistical estimates of outcome likelihoods. The
Consulting, Guernsey scenarios indicate potential future outcomes under extreme conditions;
a scenario is not a stress test unless the outcome is adverse2. Portfolio
[Link]@[Link] investment scenarios that do not anticipate adverse outcomes are not
stress tests. For an example see [2].
Stress-testing only identifies potential problems, without resolving them.
ABSTRACT Thus stress-testing may be palliative (reducing pain but not offering a
Attempting to put meaningful numbers to portfolio risks is always cure) by reassuring a practitioner if no outstanding issues are detected,
challenging. Conventional risk measures are often considered not to fully but leaving unresolved questions as to what to do about problems that
capture all risks inherent in a portfolio, particularly under difficult market have been identified, or even whether the selected stressed scenarios are
conditions. Under such conditions, stress-testing against significant sufficient to identify all key portfolio weaknesses.
historical market events, or using invented scenarios may help identify
and quantify risks within a portfolio. Stress tests also help reassure a CLASSIFICATION OF PORTFOLIO STRESS-TESTING
portfolio or risk manager as to how a portfolio might respond to specific
market outcomes or other concerns. Stress-testing covers a wide range of methodologies, and various terms
are used in the literature rather loosely [3], thus a full classification may
This paper introduces stress-testing a portfolio of conventional assets be difficult. The classification below frames the current discussion and
against market risks using historical and artificial scenarios. It includes a may help other practitioners. Often historical events provide a source
definition of stress-testing and a classification to aid ongoing discussions, of stressed conditions; however, practitioners are free to imagine any
as well as thoughts on practical implementation. Four stress-testing damaging situation and attempt to quantify its portfolio impact3.
methodologies are explored: two ‘historical’ stress tests and two A key distinction is between historical scenarios (re-enactments of
‘hypothetical’ stress tests. Examples illustrate key concepts, drawing out particular market events with a defined start and end date) and artificial
strengths and weaknesses of the stress tests, which are then discussed scenarios (invented to capture a particular concern and often involving
with recommendations. assumptions), see Figure 1. Thereafter, classification divisions may
become more judgmental. This classification follows aspects of [1] by
INTRODUCTION splitting artificial scenarios into hypothetical and algorithmic scenarios.
The main types of stress tests are described in Table 1, together with
Portfolio stress-testing may be used when attempting to identify and advantages and disadvantages.
quantify risks that are not particularly well captured by more conventional
measures, particularly relating to the impact on a portfolio of difficult
market conditions. This paper discusses portfolio stress-testing using
historical and artificial scenarios, after commencing with a definition and
classification of stress-testing methods. Four approaches are explored,
two historical and two hypothetical stress tests. Examples are included
and the advantages and disadvantages discussed.

DEFINITION OF PORTFOLIO STRESS-TESTING

Portfolio managers associate a number of activities with stress-
testing, including looking at the potential downside risk of portfolios,
or methods to see what response might be expected under difficult Figure 1: Stress-testing classification. The stress test examples in this paper do
(‘stressed’) conditions. Although stress-testing cannot be guaranteed to not include the italicised types
identify actual impacts on a portfolio of future events, it is another tool
in the portfolio or risk manager’s armoury. Stress tests are designed to Historical stress-testing’s strength is that assets actually behaved in the
determine how a portfolio might respond to adverse developments, so way captured by the scenario, adding credibility. Although, if markets
that weak spots can be detected early and preventative action taken, have changed since the historical scenario period, perhaps due to
typically focusing on key risks such as market risk, credit risk and liquidity regulation changes, or other reasons, such response may no longer
risk1. be possible. Also, historical events can be ‘messy’; numerous knock-on
effects and proxy shocks can make it hard to isolate individual aspects
Consider the following definition of portfolio stress-testing [1]: for application to a particular portfolio.
• A method of the quantification of potential future extreme, adverse Artificial stress tests raise the question as to whether the proposed
outcomes in a portfolio of financial instruments. scenario is even possible; it can be difficult to make artificial stress
• A palliative for the anxiety that is experienced by managers with tests realistic. How can the designer possibly include all responses,
significant risk exposures. direct and indirect, to portfolio assets? However, artificial stress tests

1. For definitions of types of risk see [7], [3].

2. Should a proposed scenario that is expected to have an adverse outcome turn out actually to have a benign outcome, this would demonstrate that the scenario is
of little concern.
3. Since many stressed scenarios will be motivated by consideration of past events, those interested in stress-testing might be well advised to take a keen interest in
historical market crashes, ranging from the classics ( [8], [13], [10], [9]) to more contemporary events [11], [12].

2 [Link]/academic
RFM length [Link] 2 09/10/2015 [Link]
 REVIEW OF FINANCIAL MARKETS

Approach Summary Description Advantages Disadvantages

Historical Replay Re-enactment of a particular historical • It actually happened • Proxy shocks may be numerous
crisis event market event of significance. Scenario that way • No probabilistic interpretation
shocks. It must be reasonable since it • No guarantee of ‘worst case’
actually occurred

Hypothetical • Covariance matrix Modify covariance matrix to reflect • Relatively easy • Empirical support mixed
• Create event higher asset correlations. Specify • Very flexible • No guarantee of ‘worst case’
• Sensitivity analysis hypothetical shocks to market • Can be detailed • Limited risk information
factors (often historical events can
be a guide). Definition of a systemic
liquidity event. Shock specific
identified risk factors while neglecting
correlation. Explore a mixture
Algorithmic • Factor push Attempt to systematically identify • Minimal qualitative • No guarantee of ‘worst case’
• Maximum loss the worst outcome within a defined elements • Ignores correlations
feasible envelope. Push each risk • Attempts to identify • Assumes data from normal
factor a number of standard deviations ‘worst case’ in feasible set periods are relevant
in a direction that results in losses. • Computationally intensive
Identify the set of changes in market
risk factors that results in the greatest
loss

Table 1: Stress test types with advantages and disadvantages.

can attempt to include the impact of changes (or anticipated changes) Implementing stress-testing can be seen as a four-step process [4]:
on markets, perhaps due to regulatory developments, new currencies
1. Risk identification: historical events or anticipated concerns
and so on. An artificial test can also isolate specific concerns in a
portfolio. 2. Definition of stressed scenarios: involvement of stakeholders,
support of senior management, integration within investment
decision-making
IMPLEMENTING PORTFOLIO STRESS-TESTING 3. Execution of stress-test scenarios: derivation of portfolio value
Stress-testing tends to be an ad hoc practical activity rather than 4. Analysis of results: commentary in periodic reporting.
theoretically based [3]. A balance between art and science is required,
with the identification and imagining of dangerous scenarios followed The definition of stress test scenarios cannot be regarded as a ‘once
by efforts to examine their impacts on a portfolio. The definition of and forever’ activity. Existing scenarios should be constantly reviewed,
stress test scenarios requires judgment, even if implementation of the re-evaluated and possibly adjusted to maintain their usefulness, with a
selected scenarios can become more scientific. Selection of scenarios policy established to review stressed scenarios periodically to assist in
will depend on various assumptions, which should be broadly regarded establishing good discipline and to learn from experience4.
as ‘unlikely but plausible’ [3].
The judgmental aspects of defining stressed scenarios means HISTORICAL STRESS-TESTING USING VAR
involvement of stakeholders (including portfolio managers) is
essential, with unequivocal support by senior management. This Historical scenarios comprise a period with defined start and end dates
will likely be better achieved if stress-testing is an integral part of that span an interval when the asset or portfolio of interest performed
poorly. The asset price behaviours over the period are applied to the
portfolio management rather than an add-on. Indeed, a portfolio
current portfolio to see how it would respond.
manager’s input is likely to be critical in identifying issues of concern,
as well as determining the appropriate severity of a stressed scenario, Under stressed conditions, parametric Value-at-Risk (VaR) might be
which requires a balance between being challenging but possible. inadequate due to the assumption of normally (or log-normally)
Stress-testing should not be seen as an inconvenience, but as a distributed returns, making historical VaR more appropriate. Historical
reassurance to managers of the quality of their investment decisions. VaR takes actual period returns over some interval, assigning an equal
probability to each [1], so can be seen as a scenario analysis. Further, one
Robust stress-testing may also be seen from a corporate social could add selected ‘stressed period’ returns, equally-weighted with the
responsibility perspective. By making investment outcomes more robust, non-stressed returns and recalculate the VaR, thereby creating a stress
clients should benefit and management reputation should be enhanced. test with a stressed historical VaR.

4. Indeed, approaches for defining and maintaining a library of stressed scenarios could be seen as a large topic in its own right, which is beyond the scope of the
current article.

[Link]/academic 3

RFM length [Link] 3 09/10/2015 [Link]

 REVIEW OF FINANCIAL MARKETS

An example illustrates the process. Suppose for some asset, the 95%
historical weekly VaR is calculated over two years to current date (104 Historical Event Periods Hi
weekly returns). The historical VaR calculation comprises sorting returns 140 140
into ascending order and identifying the 5% lower quantile return. With 130 130
104 returns, the 5% limit would be the rank 5.2 lowest return5. Suppose a 120 -30.18 120
four-week period in 1987 has been identified with a severe impact on the 110 110

Value
-45.76
returns of our asset. The four additional weekly returns for the stressed 100 100
period can be added to the current returns already collected6. The new
Value

90 90
total of 108 weekly returns is re-sorted with the 5% lower quantile being
80 80
the rank 5.4 lowest return7. The resulting value would be the 95% weekly
historical stressed VaR under the scenario. 70 70
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0 1 2 3 4 5
The addition of a small number of stressed-period returns has only Time
slightly altered the 5% lower quantile rank (5.2 to 5.4), but since the
Asset A Asset B
stressed period, returns might reasonably be expected to comprise
returns lower (or amongst the lowest), compared with the 104 weekly
returns to current date. The resulting stressed VaR can be considerably Figure 2(a): The price histories of two assets are shown, Asset A and Asset B.
worse. The historical scenario lies between the two vertical lines from time periods
4–14. Asset A has a maximum value of 127.21 and a minimum of 81.45,
This identifies some strengths and weaknesses of the historical VaR stress resulting in a peak-to-trough fall of -45.76. Asset B has a maximum of 127.99
test. Recent returns were blended with a small sample of historical returns and a minimum of 97.81 with a peak-to-trough fall of -30.18.
from some stressed period that otherwise would have been excluded.
Instead of using a distribution of weekly returns over the period two years
Figure 2(a): The price histories of two assets are Figure 2(b): Price
to current date, we have arbitrarily added a further four weekly returns shown, asset A and asset B. The historical
Historical Event Periods the start of the sce
Historical Event Periods
140 140
from some period when the asset performed poorly. In the example, the scenario lies between the two vertical lines from value of 127.21, w
130 was much shorter than the usual period analysed, and
stressed period 130
- 30.18
120 effect on the rank used in the ordered returns to calculate
thus had little
time-periods
120 4-14. Asset A has a maximum -23.62 end; over the peri
110Broadly, if the stressed-period returns are all higher than the value110
historical VaR. of 127.21 and a minimum of 81.45, Asset B starts at 1
Value

-45.76
100historical VaR, the stressed VaR will be little different from resulting -9.24
non-stressedValue
100 in a peak-to-trough fall of -45.76. declines by -9.24
the non-stressed VaR. Equally, if the stressed-period returns are all rather
lower than the
90
Asset B has maximum of 127.99 and minimum
90
80 non-stressed VaR, then the value of the stressed historical 80
VaR will be largely determined by the stressed-period returns. Naturally, of 97.81 with a peak-to-trough fall of -30.18.
70 70
for a longer stressed-period
0 1 2 3 4 merged
5 6 7 with
8 9 a 10shorter
11 12 13non-stressed
14 15 16 17 18 current
19 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
historical VaR, the result will not be so clear-cut.
Time Time

Historical VaR uses a fixed periodAssetto date.

A One criticism
Asset B is that any market Asset A Asset B
event prior to the start of that period will be completely excluded. The
above adjusts the historical VaR to include the impact of a selected crisis Figure 2(b): Price histories of Assets A and B. At the start of the scenario
period that would otherwise lie outside the VaR window, addressing period, Asset A has a value of 127.21, with a value of 103.59 at the end. Over
this criticism. Additionally, the historical VaR uses actual returns, and the period, Asset A declines by -23.62. Asset B starts at 107.05 and ends at
therefore has a return distribution of arbitrary shape8. By adding crisis 97.81, so declines by -9.24 over the period.
period returns, which would likely lie deep in the negative tail of the
The two key approaches [1] are either to apply maximum peak-to-trough
Figure
distribution, it is 2(a):
probableThethatprice histories
the resulting of twowould
distribution assets
be are
more Figure 2(b): Price histories of assets A and B. At
movements in asset prices simultaneously (Figure 2a), in which case
negatively skewed than otherwise, which would seem desirable for a
stressedshown, asset
VaR analysis. A andthis
However, asset B. has
analysis The nothistorical
replicated the entire the
falls start ofinthe
of 45.75 scenario
Asset period
A and 30.18 asset
in Asset B are A has
used as a
occurring at the
returnsscenario
distributionlies between
for the the two
stressed period. Also,vertical lines
by using the from
distribution value of 127.21, with a value of 103.59 at theperiod (Figure
same time, or else to use actual movements over the full
2b), resulting in falls of 23.62 in Asset A and 9.24 in Asset B. In the case
time-periods
quantile, no path-dependency 4-14. hasAsset
beenAincluded
has a maximum
and no underlying end; over the period asset A declines by -23.62.
of Figure 2(b), the recovery in A reduces the impact, as does the initial
economic analysis has been conducted.
value of 127.21 and a minimum of 81.45, Asset
increaseBinstarts
B. at 107.05 and ends at 97.81, so
resulting in a peak-to-trough fall of -45.76. declines by -9.24
The simultaneous over
use of the period.
peak-to-trough movements captures the largest
HISTORICAL STRESS-TESTING USING EVENT PERIODS
Asset B has maximum of 127.99 and minimum moves in each asset, but ignores any delay between them. Putting these
Here, aof 97.81process
different withisaused
peak-to-trough fall
to apply the asset ofbehaviours
price -30.18. from shocks together may not make economic sense. Alternatively, using the
a historical period of poor performance to the current portfolio. For an movements over the entire period may be weaker if we have difficulty
individual market index, a crisis period might seem well-defined, however, defining the event window. The positive price movements in both A
in reality, historical scenarios may play out over extended periods due and B during part of the event window have decreased the magnitudes,
to market linkages and feedback. For a portfolio of varied instruments, making the resulting stress test less demanding. However, retaining the
defining a start and end date may be harder. This is illustrated in Figure 2, relative time behaviours of the assets makes the shocks economically
with two approaches identified. meaningful. On balance, the approach using the actual movements over

5. The fractional rank being obtained, by linear interpolation, say, as a weighted sum of 0.8 of the 5th worst weekly return and 0.2 of the 6th worst weekly return.
6. The stressed period returns would be expected to lie in a time period not included in the usual non-stressed historical VaR calculation.
7. Again, linear interpolation could be used to obtain the fractional rank return as a weighted sum of 0.6 of the 5th worst weekly return and 0.4 of the 6th worst
weekly return.
8. No assumption of normally or log-normally distributed returns.

4 [Link]/academic
RFM length [Link] 4 09/10/2015 [Link]
 REVIEW OF FINANCIAL MARKETS

the entire period is probably preferred, since it results in a more plausible only are targeted correlations changed, but also other correlations
scenario, although it may remain more vulnerable to correct identification in the same matrix rows and columns. Numpacharoen and Bunwong
of suitable start and end dates and neglect impacts within the window. (N&B) [6] propose an alternative, whereby the correlation matrix is
adjusted directly. Cholesky decomposition ensures that a positive semi-
definite correlation matrix is obtained, correlations are represented
HYPOTHETICAL STRESS-TESTING USING THE VARIANCE- using trigonometrical functions and changes made in correlative
COVARIANCE MATRIX angles. This ensures correlations lie within -1≤ρij≤+1 and the resulting
Volatility and VaR are often used to quantify risk, with de-correlated adjusted correlation matrix has the necessary mathematical properties.
assets to achieve diversification, thus reducing a portfolio’s volatility and These two approaches are not expected to give the same adjusted
parametric VaR. Accepting the intuition that correlations often increase correlation matrix, for example [6], with initial and target correlation
during market crashes9, to stress-test diversification we may increase matrices of:
correlations to quantify the impact this would have on portfolio volatility
and VaR.
[ ]; ̂ [ ].
For a multi-asset portfolio, we construct n×n volatility matrix v with the
volatilities of the n assets down the leading diagonal. Using correlation
matrix R, we obtain the variance-covariance matrix S=vRv. The asset
weight vector ⃑⃑ gives the portfolio variance ⃑⃑ T S ⃑⃑ =σ2, and portfolio Adjusted correlation matrices are generated:
⁄
parametric | |, where N is the number of standard
deviations for the confidence level we require. We can increase both
individual asset volatilities and correlations to reflect some stressed ̂ [ ] and ̂ [ ].
scenario.
Consider a four-asset portfolio, with assets A–D, weights wA=0.25,
wB=0.40, wC=0.30, wD=0.05 and annual volatilities σA=9.78%, σB=3.76%, It is not entirely clear which method should be preferred. Finger’s
σC=11.17%, σD=14.84%. Now suppose a non-stressed correlation matrix: approach has intuitive appeal, since returns are adjusted towards an
average to increase correlation. However, a goal-seek algorithm is
required and, for a large multi-asset portfolio, a long history of returns
[ ] has to be adjusted (potentially including rescaling for volatilities),
which might become cumbersome. In some cases, a suitable asset
return history may not be available. In this case, N&B’s approach seems
This leads to a portfolio volatility of 6.08%pa, and a 95% monthly practical, since only the correlation matrix is required, although the
parametric VaR of 2.89%10. mathematical sophistication may discourage some practitioners.
Now stress-test by increasing the volatilities to σ’A=14%, σ’B=5%, σ’C=16%, Although N&B’s method ensures the resulting correlation matrix has
σ’D=23% and correlations to: the correct properties, there is no guarantee of economic validity. In
practical terms, choice between the two methods may be dictated
by availability of asset returns for Finger [5], and access to a Cholesky
[ ] decomposition algorithm (and level of intellectual comfort) for N&B [6].

HYPOTHETICAL STRESS-TESTING USING CREATED EVENTS

We obtain a stressed portfolio volatility of 9.55%pa and stressed 95%
A hypothetical created event stress test is an invented scenario which
monthly parametric VaR of 4.54%. In fact, common practice would
attempts to capture a particular concern. One, several or many factors
suggest applying a multiplier of 4 to the portfolio volatility [1], increasing
that may impact the portfolio are selected and deliberately tweaked
the VaR to 18.14%11.
to assess portfolio response. The practitioner has almost complete
However, we are not at liberty to modify the correlation matrix freedom in identifying relevant factors to shock, revealing a weakness of
arbitrarily. Some combinations of correlations can result in implausible the approach, since it can be difficult to create economically meaningful
stressed returns and variance-covariance matrices that are not positive stressed scenarios. An envelope approach can be used [3], which helps
semi-definite, meaning that negative variances can arise. This can be ensure a degree of consistency and makes it easier to include important
circumvented by taking a correlation matrix from a stressed historical factors, although may not guarantee economic consistency12.
period, but it makes the stress test more like a historical scenario, and
Figure 3 illustrates the stress-envelope approach. Stress factors are
may not explore the asset correlations of primary concern. Alternatively,
identified and, for each, the worst possible shock determined. Individual
mathematical techniques can be used to construct the correlation matrix
scenarios are based on envelope values. Generally, not all of the factors
appropriately. Two such approaches are discussed here.
will be used, and the stressed scenario levels chosen will be somewhat
If return histories on portfolio assets are available, the correlation matrix lower than the envelope maximums. Multiple stressed scenarios will
can be revised following Finger [1], [5]. Correlations are adjusted by reflect differing concerns. Nothing in this process ensures the economic
modifying selected return vectors period-by-period, and must be rescaled consistency of individual scenarios thus created, so there is no guarantee
if the original asset variances are to be unchanged. Consequently, not that the scenarios created are realistic, possible or extreme enough.

9. A number of academic studies debate this point, a discussion can be found in [1].
10. N=1.645, δt=1⁄12, so VaR%=|-1.645×6.08×√(1⁄12)|=2.89%.
11. Calculated as VaR%=|-1.645×4×9.55×√(1⁄12)|=18.14%.
12. In an ideal world, one would have a complete global market model to which shocks could be applied and from which the responses of all portfolio assets could
be obtained. Since such a model does not exist, practitioners constructing a hypothetical scenario should try to make it as realistic as they reasonably can.

[Link]/academic 5

RFM length [Link] 5 09/10/2015 [Link]

 REVIEW OF FINANCIAL MARKETS

presented: two historical and two artificial. Table 1 lists advantages and
disadvantages of the main types, while Table 3 captures key differences
between the approaches.
FX
The selection of a stress-testing methodology will depend on the
requirements of the practitioner (consider Table 3). With concern to
how an historical event might impact the current portfolio, a historical
Volatility Equity stress test would be required, although history may be used as a guide
in generating hypothetical correlation matrices or created events. But
if the objective is to address concerns over new market developments,
regulations and so forth, hypothetical stress tests may be more
Scenario appropriate.
There are other considerations. If a stressed-VaR measure is desired,
then a choice between parametric or historical returns distributions may
Interest lead to either historical VaR or hypothetical variance-covariance matrix
Commodity
Rates approaches. When testing the diversification benefits of a portfolio,
then historical event-periods could be used, although hypothetical
variance-covariance matrix testing comes into its own when explicitly
Credit
exploring correlations and volatilities.
Spreads
Should economically meaningful scenarios be the primary consideration,
then the historical methods are likely to be preferred (although note
‘new market developments’ in Table 3). However, historical event-period
Figure 3: Illustration of stress-envelope scenarios may not be appropriate if maximum-peak-to-trough price
movements are used, and the variance-covariance matrix scenarios
Following [3], an example illustrates the process. Consider an envelope of could be based on historical correlations and volatilities, making them
four factors as follows13: economically realistic.
1. European equities fall by 25% Regarding flexibility in scenario creation, historical stress tests are
2. World ex-Europe equities fall by 20% limited to historical events, while hypothetical methods allow more
3. A parallel downward shift in the yield curve of 200bp freedom. For the ability to isolate specific concerns, historical events
tend to be ‘messy’ with many knock-on effects, while the hypothetical
4. Foreign exchange rates: EUR weakens relative to USD by 10%. methods permit a focus on individual portfolio aspects. Similarly,
to explore extreme events, the historical methods only permit this if
Based on this envelope, one scenario is created as: suitable events lie within the historical record, while the hypothetical
methods permit the option of pushing factors further.
• European equities fall by 20%
• World ex-Europe equities fall by 15%
Factor Maximum Maximum Scenario Scenario Scenario
• A parallel downward shift in the yield curve of 50bp.
stress stress shocks shock values
envelope envelope weights
Only a subset of factors has been selected and, in each case, the size of shocks values
the factor shock is not greater than that of the envelope. A judgment
must be made whether the shocks selected are economically feasible.
Europe -25% -€1000 -20% 20/25 = -€800
Implementing the stress test involves determining the impact on the equities 0.8
portfolio of the maximum shock for each factor individually, and then
pro-rating these for the overall impact, as shown in Table 2. While the World -20% - €800 -15% 15/20 = -€600
linear interpolation used to evaluate the impact of factors may appear ex-Europe 0.75
simplistic, [3] argues that it is actually conservative. equities
An advantage is flexibility to assess the impact of any imagined
scenario. However, its weakness is that there is no guarantee that the A parallel -200bp + €200 -50bp 50/200 = +€50
events created are realistic, possible or extreme enough. Elements downward 0.25
such as portfolio diversification and correlation are ignored. Historical shift in the
events may be used as a guide in creating such scenarios, which would yield curve
support credibility. However, the advantage of the created event is that
an historical event can be modified to incorporate new aspects, such as
changes to regulations, developments in markets, geopolitics and so Foreign -10% +€150 Not Not €0
forth, giving an opportunity to add real value. exchange used used
rates
Total -€1350
SUMMARY AND DISCUSSION
Following a definition of portfolio stress-testing and a classification of
stress-testing types, examples of four kinds of stress tests have been Table 2: Illustration of hypothetical created-event stress test

13. In reality, one would expect the envelope to contain many more than four factors, however this is sufficient to illustrate the example.

6 [Link]/academic
RFM length [Link] 6 09/10/2015 [Link]
 REVIEW OF FINANCIAL MARKETS

Aspect Historical VaR Historical event-period Hypothetical variance- Hypothetical

covariance created event
Historical basis Yes Yes Maybe as a guide Maybe as a guide

New market developments No No Maybe Yes

Returns distribution Historical returns - Parametric -
Diversification - Actual period movements: Yes Yes -
Max-peak-to-trough: No
Economically meaningful Yes Actual period movements: Yes Can be if correlation realistic No
Max-peak-to-trough: No
Flexibility in scenario creation Any historical Any historical event Yes, in terms of correlation and Yes
event volatility
Ability to isolate specific concerns No No Yes Yes
Possibility to explore ‘extreme’ cases Only if historical Only if historical events, Yes, in terms of correlation and Yes
events although max-peak-to trough a volatility
possibility
Data availability Historical data Historical data required Full asset returns or just No
required correlation matrix

Table 3: Key aspects of stress tests.

A practical consideration may be data availability. The historical scenarios management-aifmd-private-equity-real-estate_24092014.pdf. [Accessed 6
that can be replicated will be limited by data availability on each asset, July 2015].
so for less recent events this could be a significant issue. Potentially, the
[5] C. Finger, ‘A methodology to stress correlations’, RiskMetrics Monitor,
hypothetical variance-covariance matrix test can get away with only the
Vols. Fourth Quarter, 3-11, 1997.
current portfolio correlation matrix, while hypothetical created events
probably have the least demanding data requirements of all, being [6] K. Numpacharoen and K. Bunwong, ‘An Intuitively Valid Algorithm for
essentially limited to the current portfolio. Adjusting the Correlation Matrix in Risk Management and Option Pricing,’
SSRN-id1980761, 2012.
Thus, in practice, the choice of stress-testing method used for a portfolio
would depend on the objectives and requirements of those setting the [7] M. Choudhry, An Introduction to Value at RiskWiley Publishing (with
stress-testing programme, as well as the resources and data available. CISI), 2013.

REFERENCES [8] M. Dash, Tulipomania: the story of the world’s most coveted flower and
the extraordinary passions it aroused, Phoenix, 1999.
[1] B. Schachter, ‘Stress testing’, in The Professional Risk Managers’
Handbook, vol. III, C. Alexander and E. Sheedy, Eds., PRMIA Publications, [9] G. Le Bon, The crowd, 2nd ed., Transaction Publishers, 1995.
Wilmington, DE, 2004. [10] C. Mackay, Extraordinary popular delusions and the madness of
[2] L. B. Chincarini and D. Kim, Quantitative equity portfolio management, crowds, Wordsworth Editions, 1995.
McGraw-Hill, 2006. [11] G. Blakey, A history of the London stock market 1945-2009, 6th Revised
ed., Harriman House, 2011.
[3] M. Crouhy, D. Galai and R. Mark, The essentials of risk management,
2nd ed., New York: McGraw-Hill Education, 2014. [12] C. P. Kindleberger and R. Z. Aliber, Manias, panics and crashes: a history
of financial crises, 6th ed., Palgrave Macmillan, 2011.
[4] Deloitte, ‘Risk management within AIFMD for private equity and real
estate funds’, 24 September 2014. [Online]. Available: [Link] [13] M. Balen, A very English deceit, the secret history of the South Sea
com/content/dam/Deloitte/lu/Documents/financial-services/lu_risk- Bubble and the first great financial scandal, Fourth Estate, 2002.

[Link]/academic 7

RFM length [Link] 7 09/10/2015 [Link]