Faculty of Actuaries Institute of Actuaries

EXAMINATIONS

April 2004

Advanced Certificate in Derivatives:

Further Mathematics, Principles and Practice

EXAMINERS REPORT

Faculty of Actuaries
Institute of Actuaries
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

QUESTION 1
Syllabus: (a) (ii) 7 & (v) 1
Reading: Hull (5th Edition) Ch. 19, Briggs et al Ch 6.

(i)
[More detail is covered in Hull Ch 19.]
Asian Option (or Average rate option) is an option where pay off at expiry is based on the
average stock price during the life of the option (typically computed daily but other rules can
apply) in relation to the strike price.
Barrier Option is an otherwise standard option, which deletes itself if the underlying stock
touches a knock-out or activates if it touches a knock-in price.
The knock-out or knock-in price is usually out of the money but may be in the money.
Basket Option is an option where underlying "instrument" is a specified basket (or portfolio)
of assets.
Bermuda Option is an otherwise standard option, which can be exercised only on a limited
number of dates.
It lies in valuation between that of a European Option and an American Option.
Binary Option is an option with a discontinuous (digital) pay off.
The most common form is the "Cash or Nothing" in which the option pays out a fixed
amount if the option is in the money at expiry.
Compound Option is an option on an option. The four main types are call on a call, put on a
call, call on a put and put on a put.
The option has a first period and strike level, and a second period and strike level.
Forward Start Option is an option, which will start at an agreed date in the future. At this
time the strike level will be set here at the money. The option is then a standard one.
Lookback Option is an option where pay off at expiry uses the historical price evolution of
the underlying stock over the life of the option to create the most favourable payout (i.e. the
payout depends on the maximum or minimum stock price during the life of the option).
Shout Option is an otherwise Standard European option in which on an agreed set of dates or
on the basis of a limited number of opportunities (including expiry) the option holder can
lock in the then current stock price to create a positive or more positive payout at expiry.
It is in effect a limited form of lookback option and will lie in value between a standard
European Option and a lookback option depending upon the number of "shouts" involved.

(ii)
Legal risk is defined to be the exposure to financial loss arising from adverse legal, legislative
or regulatory action.
A major loss from legal risk occurred in the UK due to an adverse decision on local authority
entering swaps, where they were deemed to be acting ultra vires (outside their powers).

Page 2
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

Legal risks fall into the following four categories (Briggs et al split the fourth one):
(a) Enforceability
Whether derivatives can be deemed legal transactions, and not gambling
several jurisdictions (e.g. France) have recently amended their law to clarify this.
(b) Capacity
Whether entity entering into the transaction, and the person actually arranging the deal, have
the power and authority so to do
which can be particularly problematic with government agencies
Capacity is a broad terms it covers limitations on types of instruments, their use (e.g. in
hedging or speculation), method of authorisation, accounting methods etc
(c) Documentation
The status of written documentation is important (e.g. ISDA agreement).
Some jurisdictions allow verbal confirmation
and there can be uncertainty when verbal and written confirmations disagree.
(d) Credit risk mitigation
A collateral agreement would help reduce credit risk ...
... but the trade valuations will need to be agreed with the counterparty not easy if they are
complex.
The ability to net exposures across similar contracts with a counterparty when it defaults on
one contract
reduces credit risk loss potential by offsetting all payments due (both to and from the
counterparty) against each other.
This should ideally apply across all derivative contracts between the counterparty and the life
office
and also across all branches and subsidiaries of the counterparty.

Page 3
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

QUESTION 2
Syllabus: (b) (i) 6 & 7, (iii) 2
Reading: Rebonato Ch 10 & 17, plus Hull (5th Edition) Ch 22.

(i)
[The following is an example of one of a number of approaches which could be adopted.
Similar marks could therefore be earned for alternative answers.]
An accrual swap can be decomposed into an ordinary fixed-floating swap plus a set of daily
binary options ( cash-or-nothing options) to cancel out the fixed payment for that day.
Hence the valuation would be that of an ordinary swap, less the value of the binary options
which have effectively been sold by the holder of the accrual swap.
In the accrual swap example given, there are two options on each day: one to cancel the fixed
payment if rates go above 5%, the other to cancel it if rates go below 4%.
Binary options are valued via the Black formula, suitably adjusted for whether the option is a
call or a put, using the N(d2) term the formula book calls this (d2) [which is actually the
risk-neutral probability that exercise occurs].
Theoretically, the binary option are valued daily and then summed over each accrual period,
but in practice they are usually grouped weekly or fortnightly without any loss of accuracy.
The risk profile of the given accrual swap is that it behaves like an ordinary fixed-floating
swap whilst rates lie well inside the 4% - 5% boundary, with small gamma.
However, as soon as rates approach either boundary, the gamma increases dramatically and
the option effect becomes very pronounced. This makes accrual notes a bit nasty to hedge
when rates get near the boundaries.
[Actually, on a boundary, the theoretical gamma of a binary option is infinite.]

(ii)
[Rebonato covers this point in Ch. 17.]
The chooser feature dramatically changes the valuation problem for accrual swaps.
Instead of having simple binary options dependent on market level, the market level is now
virtually irrelevant, as the owner of the chooser feature can move the strike level as he
wishes ...
... but the binary options now become very dependent on the slope of the yield curve, i.e. the
imperfect correlation of forward rates.
Hence the valuation depends on more than one factor ...
... and thus needs a more complex valuation model, such as Monte-Carlo or the two-factor
Hull-White model discussed in another question.

Page 4
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

(iii)
Definition
The Monte Carlo method uses a random number generator to provide potential values for the
stochastic term of an interest-rate model. One set of random numbers gives one simulated
path.
Method: turn time horizon into discrete time by small time steps t ...
... sample known probability distribution to give a path over all the time steps ...
... gather large number of paths, say 10,000, of equal probability ...
... value the derivative (option or whatever) on each path, and take the average.
rT
Technically it can be thought of as a forward induction method of calculating E e f T , the
(risk-neutral) expectation of the current value of the time T payoff fT .
Used to value complex derivatives that have no analytic solution, such as exotic options
which are path dependent or have non-linear payoffs.

Advantages
Simple idea conceptually, albeit rather brute force in style.
Extension of MC to multiple factors is simple, as the samples are taken e.g. from the
multivariate Normal rather than the simple Normal distribution. For the same level of
accuracy, the number of samples needed grows approximately linearly with the number
of factors. In tree methods, it grows exponentially, so is much more costly.
Binomial trees are not very efficient. They have large numbers of very extreme (large
and small) nodes which contribute almost nothing to the value of the derivative.
MC time steps can easily be tailor-made for the problem. Trees and lattices are much less
flexible.
Increasing the number of time steps in a tree increases the number of calculations by the
square of that number, whereas for MC it is roughly linear.
Successful for valuing many path-dependent options, because the path is always
generated for each sample.

Disadvantages
Needs complex and generally large-scale computer program to work effectively.
Convergence is slow. The variance of the set of independent random variables decreases
1
with as , where n is the number of sample paths, so accuracy (as measured by the
n
1
standard error) is proportional to .
n
Deltas and gammas are hard to obtain directly, due to compounding of the approximation
errors for these second and third order quantities.

Page 5
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

Often hard to get unbiased enough random or quasi-random sequences, so results nearly
always have some hidden bias from the true answer.
American options are tricky or impossible to value under MC because the method does
not enable one path to know about other paths, so as to compare values for the American
feature. Trees and lattices do this readily.
Intermediate values along the paths (for e.g. short rates or bond prices) are almost
impossible to calculate in MC, certainly for common log-normal distributions. Trees and
lattices are much better for this.

(iv)
[There are several possible methods - a few are listed below. The aim of all of these is to
improve the effectiveness of the sample distribution, by giving the same coverage of the
sample space with fewer sample paths.]

Antithetic Variables
For each sample path from the distribution, create another valid path by taking the opposite
sign on each random element. This creates an antithetic path.
The odd moments of the sample distribution are therefore zero, which works well with non-
skew distributions like the Normal distribution.

Moment Matching
The n sample paths are generated and stored, then adjusted by a scaling factor to ensure that
the variance (second moment) of the sample exactly matches that of the initial distribution.
Together with antithetic variables, this produces a tighter fit to the probability distribution
without adding further samples, so should be more accurate.
The fourth moment can also be matched in the same way if kurtosis is important.

Stratified Sampling
Divide the range of outcomes into bands according to probability, and then take more sample
paths from those with the higher probability.
This does not affect the overall accuracy, but removes the need to sample those outcomes
where there is a low contribution to the valuation.

Contravariates (or Control variates)

If an exotic derivative is being valued, there is probably an analytical expression available for
a similar, more simple, derivative (e.g. if calculating an average-rate caplet, the formula for
the plain vanilla caplet is known).
Then use MC to calculate the simple and exotic derivatives, and adjust the exotic value by the
same amount as the simple value needs to adjusted to agree with the known formula.

Page 6
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

Alternatively, the paths themselves can be adjusted (probably a few times) until the values
agree.

Quasi-Random Sequences
The MC approach does not need random values to succeed - all it needs is to have
representative enough paths that the approximation is not biased by the samples chosen.
There are some sequences which can be shown to be non-random, but which nevertheless
will not bias the approximation. The advantage of this type of series is that its standard error
1 1
is proportional to , rather than for the raw MC.
n n

Page 7
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

QUESTION 3
Syllabus: (b) (i) 6 & 7
Reading: Rebonato (2nd Edition) Ch. 8, 12, 13.

(i)
(a) Calculate 4-year cap and floor
The marking may need to be adjusted if candidates make a persistent slip. We assume the
strike does not need to be adjusted for day-count, though candidates may do this if consistent.
The basic discounting process in BDT is used throughout. This assumes a probability of ½
for up and down nodes, and V(t) = ½ [Vup(t + 1) + Vdn(t + 1)] / [1 + r(t)].
Caplet prices: final payoff at node j = max {rj(t) 2,0}
t = 1:
0.2923
0.1440
0.0000

t = 2:
1.0957
0.6030
0.3305 0.1380
0.0679
0.0000

t = 3:
2.0577
1.3642
0.8476 0.7560
0.5073 0.3701
0.1821 0.0000
0.0000
0.0000

Page 8
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

Floorlet prices: final payoff at node j = max {2 rj(t),0}

t = 1:
0.0000
0.1920
0.3897

t = 2:
0.0000
0.0000
0.1281 0.0000
0.2600
0.5283

t = 3:
0.0000
0.0000
0.0324 0.0000
0.1370 0.0663
0.2456 0.1355
0.4328
0.7427

Hence cap price = 0.1440+ 0.3305 + 0.5073 = 0.9818 and floor price = 0.1920 + 0.1281 +
0.1370 = 0.4571. These prices are in % of nominal.
(b) Check put-call parity of prices
If we take any caplet price minus the floorlet price, the final payoff is r(t) 2 at each node.
Hence, since the tree creates the present value of the final payoff, the value at t = 0 will be the
present value of the forward rate less the present value of the strike, which is put-call parity.
[Numerical examples could be given instead to show put-call parity for full marks.]

Page 9
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

Delta
There are really several separate options, so we calculate for each caplet or floorlet, i.e. don t
add the deltas.
Delta for any caplet or floorlet is (Pup - Pdown)/(rup - rdown) at t = 0.
[Numerically not required for the solution:
Given that rup - rdown = 2.299 1.604 = 0.695, we get
Caplet for t = 1: 0.9576 t = 2: 0.7699 t = 3: 0.4205
Floorlet for t = 1: -0.3067 t = 2: -0.3740 t = 3: -0.5608]

(ii)
Both BDT and Black use a log-normal model so using the caplet volatilities for each time
step should give very close answers.
BDT should give exact same answers as Black for caps and floors if properly calibrated and
small t (i.e. t 0).
Our BDT tree is very coarse, so won t get very precise answers (though amazingly they are
very close). Need a smaller time step to be accurate.
Need to be sure day-count methods are same.
Generally our deltas will be poor due to the large time step.
Black deltas are always w.r.t. the forward rate, not the short rate as we calculated.
BDT deltas can be converted into forward rate deltas by first calculating, then dividing by,
the sensitivity of the forward rate to the short rate but this is a messy calculation and needs a
small time step.
[It is not correct to say that BDT gives only an approximate answer in any circumstances.]

(iii)
Comparison with HW 1-factor model.
Both models can fit all shapes of yield curve perfectly ...
... but both are single factor models, so only have limited ability to model simultaneously the
yield curve and volatility structure.
Both fail to give explicit imperfect instantaneous correlation of forward rates.
Both achieve a good fit by compromising on the future evolution of the term structure over
time ...
... with HW generally having a poorer snapshot but better evolution.
HW uses explicit mean reversion ...
... whereas BDT this is achieved by a declining volatility.
BDT uses binomial tree to value ...

Page 10
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

... HW a trinomial because of the extra degrees of freedom.

BDT fits the curve by adjusting the bottom rate ...
... whereas HW s trinomial is much more efficient because it avoids areas of low probability,
thereby not wasting calculation on unimportant areas of the tree.
BDT is log normal, HW normal ...
... which makes BDT easier to calibrate to Black-derived caps and/or swaptions.
Also, theoretically HW could give negative rates, although only with extremely low
probability.
In fact, very hard to calibrate HW, as at-the-money options are priced in both normal and log-
normal distributions but the differences come away from the current level of rates. Any cap
prices will include a range of at-the-money, out-of-the-money and in-the-money caplets.
Need to disentangle these effects, otherwise HW will not have correct balance between
volatility and mean-reversion parameters. In some cases, declining volatility as observed in
caplet pricing can lead to failure of the trinomial process further out in the tree.

Page 11
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

QUESTION 4
Syllabus: (c) & (d) (ii) 2
Reading: Kemp 11, Dowd 12, GN25, DTI 94/6

[The report should at least cover these points. If a candidate makes a valid point not shown
below, it can attract further marks.]

Internal management
Senior management must decide what sort of risks they want to take on, e.g. how much
do they want to be involved in derivatives?
There has to be congruence between the intended risk appetite and the actual risks taken
on. Sometimes this means encouraging more risk-taking, other times less.
Ensure effective delegation of risk control at both local and global level.
Beware hidden risks, for example those that cannot yet be reported on the firm s risk
systems. These are often the biggest risks, as they usually involve illiquid or
unquantifiable positions.
Everything should be reported even if it can t be risk-measured (and if not, management
may wish to ask why the position was taken in the first place).
Check the accounting methodology, particularly if business looks very profitable.
Check the documentation, that it is industry standard and robust.
Check the models. Don t always believe the rocket scientists - must have reasonability
check e.g. beware if they claim to be making money if other firms models are wrong .

Regulatory framework (banking)

Regulation for derivatives in banks in UK is covered by FSA (Financial Services Authority).
Their principles include:
Settlement and documentation departments to be separated from trading, so a trader
cannot settle his own trades (else e.g. he could bottom draw bad trades).
An independent risk management department to be set up within the bank, reporting to
senior management and not the head of trading ...
... to measure and report both market and credit risk ...
... and verify the financial models used in the trading areas, that they are robustly applied
and, if not industry standard, then soundly based theoretically.
Traders must be set trading limits expressed in risk-related terms, and any exceptions
must be reported daily.
Daily risk reporting to senior managers, who must be qualified to understand the risks
involved and aware of what action to take to mitigate them.

Page 12
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

Daily capital calculation, according to FSA rules, usually involving some sort of Value-
at-Risk methodology and stress testing for large market moves.
Adequate systems resources to be applied to valuation, P&L and risk reporting.
Regular discussions with FSA and occasional visits.

Regulatory framework (insurance)

[The candidate may choose to list some of the previous section s points under this heading, so
can receive those marks here instead.]
Regulation for derivatives in insurance subsidiaries is covered by DTI, based on the
Insurance Companies Act 1982.
Their principles include:
Prudential management controls must be in place, similar to the FSA s.
Appointed Actuaries must ensure that suitable reserves or provisions exist ...
... and report on these at appropriate intervals - continuous review ...
... including a daily schedule of derivative transactions undertaken, exposures and
sensitivities to large market movements.

Global vs. local considerations

Whenever businesses are managed in other locations, there is scope for unacceptable risk and
unexpected losses, especially if a crisis occurs in one timezone whilst other timezones are
closed.
Need to ensure adequate local supervision by trusted employees, who are empowered to
intervene swiftly if the local situation demands, but who will also provide more general
feedback to Head Office.

Comments on the specific businesses

(a) Swaps and MM in New York and Tokyo:
The business described is a fairly standard Treasury operation.
Documentation, settlement and accounting are crucial. If the business is fairly large
scale , these will need to be fully automated and robust.
Management will need to assess which parts of the business will require mark-to-market
valuation and which will required accruals valuation. Usually, Swaps are market-to-
market and MM instruments are accrued.
The setting and monitoring of risk limits for proprietary traders is crucial.
Risks should be approximately linear, as there are no options traded, so provided limits
are adhered to there should be no surprises.

Page 13
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

(b) Capital market trading in Hong Kong:

The key problem here is assessing the overall risk of the firm from all the different asset
classes ...
... and return on capital employed.
Value-at-risk would be useful here, to benefit from imperfect correlation between the
asset classes, hence reducing capital requirements ...
... but integrating equities and debt into the same framework needs careful thought, and it
may not be worth it as the business is small scale.

(c) Equity options in the insurance business:

Extent of hedging of liabilities important - should in theory be exact. There should be
little need to churn any of the positions.
Biggest risk is credit risk with respect to investment banks - don t want to be too
concentrated.
Careful adherence to schedule of positions with regards exercise of options.
Policy on managing lapses, i.e. closing out of options.

Page 14
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

QUESTION 5
Syllabus: (b) (i) 7
Reading: Rebonato Ch 17, plus Hull (5th Edition) Ch. 24

(i)
[The question does not ask for all the algebra, but some is helpful. Different approaches
could also be valid here.]
Consider the variable y(t, T) = ln P(t, T). Then, using Ito s lemma,
1 2
dy d (ln P ) r dt dz r 1
2 v 2 dt vdz (*)
2 P2 P

where v = / P depends (like ) on t, T and P.

But the time t forward rate spanning period T1 to T2 is given by:
y (t , T2 ) y (t , T1 )
f (t , T1 , T2 ) (**)
T2 T1
and, substituting equation (*) into the differential of equation (**), gives an expression for
df(t, T1, T2) of the form [...] dt + [...] dz involving v, the normalised volatility ...
... but not r or P directly, i.e. only on these variables in so far as v depends on them.
Letting T2 T1 = T, the LHS of the above equation becomes the instantaneous forward rate
F(t, T):
y (t , T )
F (t , T )
T
which leads to a risk-neutral process of the form [...] dt + [...] dz also for dF(t, T) ...
... where the [...] terms again depend only on v and its partial derivatives with respect to P and
T.
Therefore, once the volatility function v has been determined, this is enough to define both
the drift (dt) and the random (dz) term ...
... and so there is an explicit link between the volatility of zero-coupon bond prices and the
drifts of the forward rates.

(ii)
Benefits of two-factor HJM:
HJM is a generalised framework for modelling the yield curve evolution, enabling a wide
range of valuations from one model, i.e. efficient and consistent.
Two factors enable the model to describe greater variations in the evolution of the yield
curve, since forward rates can be imperfectly correlated.

Page 15
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

Two factors also enable valuation of options which depend not just on the market level
but also on the yield curve slope, i.e. include correlation between forward rates on the
curve valuing a wider range of options.
HJM uses MC, which only needs twice the iterations of a single factor model, compared
with square of that number for e.g. tree or lattice methods.

Choice of factors:
The method of principal components can be used to identify the main drivers of the yield
curve ...
... which generally are the direction and the slope.

Implementation and calibration:

It is far from trivial to implement the HJM model based on market prices.
Would like to use prices of caps or swaptions to drive model parameters, but ...
... the problem is that these are valued in the market using the Black model, which is a log-
normal distribution ...
... and specifying a log-normal volatility for forward rates in HJM severely restricts the forms
of the HJM model that can be used.
Choice of correlation function is difficult - hard to be precise about it since so few reliable
option values to base it on.
Swaptions and caplets imply certain correlation effects but, considering the calibration
difficulties, are not likely to suggest a good correlation function.

(iii)
Equilibrium models are a particular class of models where a simple form of an entire
economy is described by the model.
To obtain a parsimonious model (few parameters) the economy is often simplified to consist
only of a single production source and a single good which can be consumed. Deferred
consumption (investing) enables higher future consumption, and is governed by a utility
function.
All securities and contingent claims are priced endogenously in this model. This gives a
world of absolute pricing.
However, in practice, the simplifications of the economy are so great that only certain yield
curve shapes are possible, and hence any attempt to map more complicated shapes at regular
intervals results in unstable parameters.

No-arbitrage models are a class of models which allow recovery of market prices of one set
of securities given prices of another set.
This gives a world of relative pricing.

Page 16
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

In a non-arbitrage-free model, securities could be priced using the model and then traded at a
different price in the real world, leading to persistent profit. In simplest terms, no-arbitrage is
the absence of a free lunch .

[Note: Hull describes equilibrium models as providing security prices as an output of the
model, whereas no-arbitrage models take security prices as an input.]

No-arbitrage is very important in yield curve models, since most complex structures are
limiting cases of simpler structures (such as swaps, caps, floors) and hence ideally the model
should recover the prices of the latter exactly.
Also, hedging is done using the simpler structures, so the absence of no-arbitrage would
mean the accounting process would be distorted by imaginary gains and losses.
Equilibrium models are no-arbitrage as far as the stylized economy they describe is
concerned, but the inadequacy of their formulation usually makes them poor at pricing
complicated structures (and sometimes simple structures). Hence the possibility of arbitrage
is introduced amongst instruments traded in the more complex ( real ) economy.

Page 17
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

QUESTION 6
Syllabus: (d) (i) 2 & (v) & (vi)
Reading: Dowd Ch 12, Hull (5th Edition) Ch 16

(i)
Value at risk (VaR) for a portfolio is a number which indicates the maximum loss which the
portfolio can sustain over a given timeframe (say 1 day) for a given confidence level (say
95% or 99% certainty).
It is normally calculated by taking into account:
future price movements (volatility)
the interdependencies of the portfolio constituents (correlations).
If normal returns are assumed, the VaR of the portfolio can be synthesized easily from the
VaR of the constituent individual holdings. If the portfolio has very many holdings, this can
lead to a large correlation matrix, so some benchmarking is necessary to first express
positions in terms of representative indices (FTSE100, FTSE250, SmallCap, AllShare etc)
before calculating VaR.
VaR is essentially a linear measure of risk, i.e. assumes the loss is always proportional to the
exposure, which is true except for option-type instruments. VaR can be adjusted for these
convex securities, but not if the convexity is too pronounced (e.g. very out-of-the-money or
exotic options).
[Note: as with many such questions, a shorter answer could attract full marks. More details
are given here to illustrate the range of possible points that could be covered.]

(ii)
In note format:
(a) matrix
standard method - statistically sound
too few factors not precise - too many and matrix too big so cumbersome to calculate
correlations not stable or even accurate
not forward looking - past not guide to future
problem in choosing correct amount of history to compute vols and correlations as far
past may not be relevant (could weight nearer observations but then have to choose
weights!)
normal model the only practical one - not realistic for outliers (kurtosis effect)
hence VaR does not really give the extremes - bit of a problem since it is supposed to
be risk capital
easy to calculate based on benchmarks (eg FTSE index) - not so easy for individual
securities

Page 18
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

(b) historical simulation

fast to compute
lots of data for a multi-currency portfolio
very easy to obtain distribution of losses from the data without relying on normal model
(but may be a weakness since distribution is implicit)
no problem with correlation assessment - all implicit (again, may be a weakness)
not forward looking
very affected by particular major events e.g. 87 crash
allows for specific risk as well as sectoral
problem in knowing how far back to go (as for (a))
difficult to cope with new securities since not available throughout history
so have to use sectors and hence thrown back on correlation problem/specific vs index

(c) Monte-Carlo & GARCH time series

Monte Carlo and GARCH are actually totally separate methodologies, but the question
has linked them for convenience
Monte Carlo - uses random number generator to provide potential values for the
stochastic term of an interest-rate model - one set of random numbers gives one simulated
path
can tackle path-dependent and non-linear instruments easily
extendible to many assets as can cope with several dimensions without excessive
calculation, since number of calculations is proportional to m, not m2 as in matrix method
(m = number of factors)
1
however, is slow to converge to solution (rate offor n iterations), unless random
n
numbers tweaked to give better coverage of domain by variance-reduction techniques
needs an explicit underlying model, which may have inadequacies
technically quite complicated to set up - needs skilled programming, especially to get a
truly random series

GARCH - predicts volatility and correlations based on evolution of previous

observations, using an autoregressive (i.e. implicitly fitting to itself) relationship.
good to try to look forward - VaR really should use implied vols not historical vols
problem is we are not really sure what model we are using
forecasting is unstable and maybe completely erroneous if trying to predict in a range
outside its experience
not so useful for predicting correlations - number of parameters required is enormous and
problems with keeping output values in range -1 to +1.

Page 19
 Advanced Certificate in Derivatives: Further Mathematics, Principles and Practice April 2004
Examiners Report

[There are many other points which could be raised. The above is only a selection.]

(iii)
Forward currency positions
Create two payoffs, one in each currency for the appropriate amounts at the agreed
conversion rate. These then act as zero coupon bonds without purchase price.
Dowd assumes that there is only one currency yield curve, but most banks run the DVaR
process for several currencies at once.

Structured notes
Mostly these can be decomposed into more basic securities, usually with a bond underlying.
For example, an inverse floater can be transformed into a FRN plus twice the amount of a
fixed-floating interest-rate swap.

Certain types of structured note have options which are very difficult to include within the
VaR framework, especially if they have barriers or path-dependent behaviour. Also,
additional factors such a dual-currency bonds can be hard to value and measure risk
sensitivities.
At the linear level, one could use option delta to give exposure at current level and enter this
into the VaR model.
However, non-linear characteristic of options (gamma) make VaR s linear approximation
only work for small range about the current market level hence no use for capital on
extreme move.
Exotic options also have very variable gamma and vega (volatility) profiles so risk profile at
current level is misleading. Some exotics are virtually riskless at current levels, but blow
up somewhere far away (e.g. barrier options or cancellation options).

END OF REPORT

Page 20