The real world is often an inchoate swirl of actions, occurrences, facts

and figures (Derman, 1996) and thus we treat with caution any claims
about the untrammelled success of a particular modelling formulation. As
Dempster (1998) writes: Model assumptions are reflective of objective re-
ality yet are not fully part of that reality. The logicist approach to choos-
ing and justifying a model fairly balances objective and subjective and
puts aside an operationally spurious concept of true model. We empha-
sise here the fundamental role of subjectivity within quantitative financial
modelling in particular and scientific endeavour more generally. While
there exists a widespread value-judgement that science ought to be ob-
jective and that subjective elements are to be viewed with suspicion, the
scientific process inevitably involves subjective elements.
2
The term sub-
jective covers diverse activities of problem-formulation and problem-solv-
ing carried out by individuals, drawing on memory, making choices and,
most especially, attempting to reason about specific phenomena and is-
sues under analysis (Dempster, 1998).
So what can we say about interest rate modelling? In particular, we
know there is only one (for example, dollar Libor) yield curve and thus
there can only be one underlying process driving the prices of market in-
struments related to this curve. This process may be stochastic or other-
wise
3
, but is and remains unknown we do not hope to identify the true
process. Rather, we demand a consistent pricing and hedging framework
for derivatives of the yield curve. This modelling framework must be suf-
ficiently general to incorporate reasonable dynamic features. The associ-
ated analytics must be advanced enough to develop tractable pricing and
risk management tools. Indeed, the demands on analytics and systems can
be non-trivial: these obstacles are often given as reasons that a consistent
approach is not implemented more widely.
Applied financial judgement. Any analysis with significant subjective
elements depends to a large degree on the judgement of the investigator.
Tukeys (1986) discussion of applied statistical reasoning is informative
here. Tukey stated that one requires judgement based on: mathematical
knowledge of the particular techniques; experience of the particular field
I
n recent years, there has been heightened interest in quantitative relative-
value trading within the interest rate derivatives markets. Inferior returns
in the equity markets have resulted in significant capital flows into fixed-
income hedge funds. As a consequence, there has been increased focus on
(and competition over) the identification of relative-value trade opportuni-
ties as funds strive to deliver superior performance to their investors.
Much of the interest rate modelling literature has concentrated on is-
sues related to market-making: in particular, calibration of models to a set
of relevant liquid market instruments and the subsequent interpolation of
prices for more illiquid instruments. In this article, we attempt to establish
a consistent framework for relative-value volatility trading that enables iden-
tification of value within the universe of liquid option products and in-
forms relative-value decisions between different option product classes.
First, we outline a set of guidelines that informs our modelling and trad-
ing judgements. We borrow approaches to modelling and applied quanti-
tative reasoning from modern applied statistics, in particular the logicist
approach described by Dempster (1998).
We then use these foundations in a detailed example, the construction
and use of forward-rate volatility surfaces. We describe how this frame-
work can identify relative-value trading opportunities in the volatility mar-
kets. Finally, we outline briefly how a similar approach can be extended
to further dimensions of the volatility markets.
Modelling foundations
Subjectivity and the logicist approach. Dempster (1998) contrasted
the logicist and proceduralist approaches to statistical modelling and in-
ference. The logicist paradigm is concerned with reasoning about a spe-
cific situation under analysis, as opposed to the rote application and simple
reporting of defined procedures. In the context of relative-value model-
ling, the logicist is concerned with what we can say about the derivatives
markets rather than simply applying procedures or analysing properties of
specific quantitative models.
1
The logicist approach requires choices necessarily subjective, driven
by the judgement of the investigator about which objective features to
include or omit in the modelling formulation. Thus, careful assessment is
required to determine which features should be included in any formal in-
terest rate model. We believe this type of analysis is sometimes incom-
pletely carried out within market-making operations: instead, a pragmatic
approach is adopted whereby the simplest model that gives a price for the
relevant instrument is used for pricing and risk management. This inevitably
leads to the adoption of models with different underlying dynamics, even
when these models have been calibrated to the same liquid instruments.
WWW.RISK.NET

MAY 2004 RISK 91
Quantitative trading
l
Cutting edge
Practical relative-value
volatility trading
The rapid growth of fixed-income hedge funds has resulted in an increased focus on the
identification of relative-value trade opportunities. Here, Stephen Blyth presents a consistent
framework for identifying value within interest rate options markets. His approach stresses the
fundamental importance of subjective judgement within quantitative financial modelling, and
adopts elements of the logicist paradigm of modern applied statistics. A series of examples is
given, in particular the detailed analysis of parametric and non-parametric forward-rate
volatility surfaces in the dollar and euro options markets
1
Dempster (1998) quotes Niels Bohr: It is wrong to think that the task of physics is to find
out how nature is. Physics concerns what we can say about nature. We can think of the
financial markets as nature, and quantitative finance as physics
2
A referee makes the important comment that subjectivity is vulnerable to abuse and,
misapplied, it becomes detrimental to theory. This viewpoint is similar to that of Dempster
(1998), who warns that an obstacle to granting subjectivity a rightful place at the table
is the radical informal subjectivity of an anything goes variety
3
Financial markets involve the interaction of many individual agents and thus, as one of
the referees commented, it is an interesting philosophical question whether there exist
truly stochastic mechanisms within finance
of subject matter; and experience of how these techniques have worked
out in practice.
These guidelines are directly applicable to applied financial reasoning.
Technical expertise with regard to, for example, sophisticated yield curve
models, is certainly necessary. However, trading decisions are always tem-
pered by supply/demand dynamics that is, experience and knowledge of
the underlying market. In addition, quantitative trading strategies are in-
formed by experience over time of the performance of modelling approaches.
Our view is that the quantitative finance literature has tended to concen-
trate perhaps too heavily on the mathematical properties and behav-
iour of the underlying models.
In this article, we employ a range of mathematical models, but leave
most technical details to (generally well-known) references. Similarly, de-
tailed description of underlying market features is beyond the scope of the
article. We focus instead on our experiences with the practical implemen-
tation and use of the modelling framework.
To summarise the elements underpinning our modelling approach: we
strive to work with a consistent modelling framework, given that only one
process can drive the yield curve; and we use subjective judgement to con-
struct a framework that captures realistic yield curve dynamics, rather than
adopt the simplest or quickest model. We then use the modelling frame-
work to inform relative-value decisions: pricing inconsistencies between
the model and the market can indicate market features, modelling errors
or trading opportunities. We use market experience and judgement to at-
tempt to identify the trading opportunities.
Volatility surface analysis
We discuss in detail an example of the practical implementation of our ap-
proach: identification of value within the interest rate options markets via the
construction and analysis of volatility surfaces. When analysing volatility mar-
kets, it makes little sense to consider, for example, caps separately from swap-
tions, or individual swaptions as independent instruments with their own
volatilities. These and other products are functions of the same yield curve
and depend often in non-trivial ways on the same underlying dynamics;
thus one requires a consistent model to assess relative value between them.
We choose the Brace-Gatarek-Musiela (BGM) market model refinement
of the Heath-Jarrow-Morton (HJM) forward-rate model as our basic frame-
work. The BGM model has general specification and thus offers the po-
tential to fulfil the criterion of allowing realistic dynamics.
We write the multi-factor HJM framework in the form:
(1)
where f (t, T) is the instantaneous forward rate from time T seen at time t,
W(t) is a standard n-dimensional Wiener process,
n
i

2
i
(t, T) = 1, and the
drift a(t, T) is given by:
(2)
For further details, see Heath, Jarrow & Morton (1992). Let L(t, T, T + )
be the Libor rate from time T to time T + ; in this article, we will usually
set = 0.25 and adopt the simplifying notation L(t, T) := L(t, T, T + ).
Using the relationship:
(3)
and Its lemma, we obtain the BGM formulation:
(4)
which we can write in the general form:
(5) dL t T t T dt t T t T dW t
i
i
n
i
, , , , ( ) ( ) + ( ) ( ) ( )

1
( )
+ ( )
( ) ( ) + ( ) ( )
+

dL t T
L t T
t u du t u dudt t T dW t
T
T
i i
i
,
,
, , ,
1
1

nn
T
T

_
,

L t T f t u du
T
T
, exp , ( )

( )
( )
+

1
1
a t T t T t u du
t
T
, , , ( ) ( ) ( )

df t T a t T dt t T t T dW t
i
i
n
i
, , , , ( ) ( ) + ( ) ( ) ( )

1
where in general (t, T) and (t, T) depend on L(t, T). For fixed T,
(t, T) is the volatility of a specific eurodollar (Euribor) contract over its
life, and thus has a simple intuitive interpretation.
For further details, see Brace, Gatarek & Musiela (1997) or Jamshid-
ian (1997); there is also significant associated work on analytic approx-
imations for derivatives (for example, Rebonato & Jackel, 2003) and
Monte Carlo valuation (for example, Longstaff & Schwartz, 2001) with-
in a BGM framework.
To implement BGM, the user must specify the correlation structure
(t, T) and the volatilities (t, T) for all t and T. In our implementation,
we require a minimum of three underlying factors. Fewer are unable to
capture with a reasonable volatility structure the decoupling of forward
rates across the curve. There is debate about whether a higher number of
factors is appropriate (see Andersen & Andreasen, 2001, and Longstaff,
Santa-Clara & Schwartz, 2001, for an example). Our opinion is that most
statistical analyses cannot identify with any consistency more than four fac-
tors. We use three factors and take the correlation structure to be station-
ary, that is, a function of the form (T t). Although it is possible to calibrate
both factor structure and volatility surface together to market instruments
(Rebonato, 1999), we prefer to estimate the correlation structure from yield
curve data, and focus on the calibration and analysis of volatility surfaces.
Unlike prices of three-month Libor caps, prices of swaptions are not in-
dependent of the correlation structure, and thus comparison of volatility
surfaces can be difficult unless the correlation structure remains the same.
Furthermore, estimates of correlation between points on the forward curve
often have diffuse sampling distributions and thus large standard errors.
This makes it hard to differentiate statistically between different factor struc-
ture estimates. We advocate, however, a formal sensitivity analysis using a
number of reasonable correlation structures to confirm that inferences from
the volatility surface are robust.
We develop two ways of constructing and manipulating volatility sur-
faces: a (sparsely-) parametrised smooth surface for (t, T), and a non-
parametric surface where each discretised element (t, T) can be perturbed
independently of its neighbours. Our parametric surface is constructed with
12 parameters. Essentially, the columns (t, T) for fixed T (the volatility
of a specific eurodollar future over its life in other words, the volatility
of an individual caplet), the top row (0, T) (the instantaneous volatility
of the futures strip) and the diagonal (t, t) (the volatility of successive
front contracts, in some sense describing future expectations of central
bank activity), each of which is an univariate function, are described by
simple linear functions with exponential decay to an asymptotic value, of
the form (x) =
0
+ exp

3
x
(
1

0
+
2
x). We use a non-linear solver to
minimise the least-squares deviation between market and calibrated im-
plied volatilities.
The non-parametric calibration seeks an exact fit to the calibration in-
struments by perturbing a starting surface (often in practice the paramet-
ric fit) by the least distance possible. We can also perform local smoothing
on the exact-fit surface to obtain a non-parametric smooth surface: we ob-
tain a smoothed value
~
(t, T) by calculating a weighted average of (t, T)
and its immediate neighbours in the discretised surface. Smoothing is done
iteratively through the volatility surface. To simplify comparisons between
the original and smoothed surfaces one can impose, for example, the scale
restriction that sums of squares of columns (that is, cap prices) remain con-
stant in the two surfaces.
We advocate the use of both parametric and non-parametric volatility
surfaces to inform relative-value decisions. Clearly, we would like to im-
pose some smoothness criteria on our surfaces since it is unlikely that the
volatility of a particular futures contract is implied to jump around deter-
ministically. However, it is also interesting to analyse how an exact fit to
market instruments deforms the volatility surface and to investigate which
instruments cause the deformation. Our analysis of relative value can in-
clude, but is not limited to, examination of:
Residuals between the market and smooth-fit prices of instruments used
in the calibration.
92 RISK MAY 2004

WWW.RISK.NET
Cutting edge
l
Quantitative trading
crease as the 2013 contracts eventually become blues, greens and reds.
(In terms of swaption volatilities, this high-volatility environment manifests
itself by, say, three-month into 10-year swaptions having significantly high-
er volatility than one-year into 10-year swaptions. See Rebonato & Kainth,
2003.) The volatility then decreases slightly as the contracts expiring in
2013 become the front contracts (in 2012).
In table A, we give a subset of residuals between the market prices and
the smooth model fit. We use Black implied volatilities as our proxy for
the option price. Note, for example, that five-year into 10-year swaptions
look somewhat expensive and 10-year into 20-year look cheap to the
smooth surface. This is a familiar feature of the dollar options market: mort-
gage account buying of 10-year-tailed swaptions and supply of 30-year cor-
porate callable deals often distort prices.
Figure 3 shows an exact fit to the calibration instruments. The smooth
surface in figure 1 has been perturbed by the minimum distance possible
to fit all prices exactly. Note, for example, how (t, T) for 2 < t < 7,
25 < T < 30 are driven low. This suggests that, for example, five-year into
25-year or 10-year into 20-year swaptions are cheap compared with, for
instance, two-year into 30-year swaptions. Figure 4 shows the columns for
this exact fit; these are fairly smooth considering we are fitting the entire
set of calibration instruments. The upturn in (t, T), T = 3 as t T may
be an indication of two-year by three-year caps being rich compared with
two-year into one-year swaptions, although one must be cautious of stale
or inaccurate input prices and investigate accordingly. Figure 5 shows a
The shape of the exact-fit surfaces, for example, identification of signif-
icant peaks or troughs.
Pricing using a post-calibration smoothed version of the exact fit.
Forward volatility surfaces, that is, the volatility surface rolled forward
in time.
We give specific examples for the dollar and euro volatility markets
below.
Volatility surface inference for the dollar. Figure 1 gives a 12-para-
meter fit to at-the-money dollar cap and swaption markets from October
2003; time t and tenor T are given in years. We use a Gaussian specifica-
tion that assumes that
T +
T
(t, u)du does not depend on L(t, T). Under
such a formulation, the short rate is normal and Libor rates for small are
approximately normal. While short rates can go negative, bond prices are
non-negative. (t, T) is given in annualised standard deviation. The cali-
bration instruments are a grid of 50 swaptions and eight forward-starting
caps. The average absolute difference between model and market Black
(1976) implied volatilities is approximately 0.3%. Figure 2 displays the
columns of this surface.
Consider the volatilities of the 2013 eurodollar contracts (that is, the cur-
rent copper pack), given by the column (t, T), 9 < T < 10, 0 < t < T. Cur-
rently, we are in a high-volatility environment with elevated instantaneous
volatility across the futures strip: the copper pack has instantaneous volatil-
ity of around 160 basis points. The volatility is implied to decrease over
the next 18 months to a steady state of just under 100bp, then slowly in-
WWW.RISK.NET

MAY 2004 RISK 93
1y
2y
3y
5y
7y
10y
20y
10y
7y
5y
4y
1.0
0.8
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.8
1.0
Tenor
E
x
p
i
r
y
Volatility differences
A. Sample dollar relative-value indications
Forward volatility through time
0.00
0.20
0.40
0.60
0.80
1.00
1.20
1.40
1.60
1.80
2.00
0
.
0
0
1
.
0
0
2
.
0
0
3
.
0
0
4
.
0
0
5
.
0
0
6
.
0
0
7
.
0
0
8
.
0
0
9
.
0
0
1
0
.
0
0
1
1
.
0
0
1
2
.
0
0
1
3
.
0
0
1
4
.
0
0
1
5
.
0
0
1
6
.
0
0
1
7
.
0
0
1
8
.
0
0
1
9
.
0
0
2
0
.
0
0
Time
1.0
3.1
5.2
7.3
9.4
11.6
13.7
15.8
17.9
20.0
%
2. Dollar columns
0.00%
0.20%
0.40%
0.60%
0.80%
1.00%
1.20%
1.40%
1.60%
1.80%
2.00%
Time
Tenor
HJM volatility surface
0
5
10
15
20
25
0
5
10
15
20
25
1. Parametric dollar surface
0.00%
0.20%
0.40%
0.60%
0.80%
1.00%
1.20%
1.40%
1.60%
1.80%
2.00%
Time
Tenor
HJM volatility surface
0
5
10
15
20
25
0
5
10
15
20
25
3. Exact fit non-parametric dollar surface
Swaption Mkt vol Model vol Diff
4y1y 27.02 27.35 0.33
4y5y 22.42 22.41 0.01
4y10y 19.63 19.09 0.54
4y20y 15.97 15.63 0.34
4y30y 14.76 14.26 0.50
5y1y 24.20 24.32 0.12
5y2y 23.06 23.09 0.03
5y5y 20.44 20.49 0.05
5y7y 19.30 19.24 0.06
5y10y 18.02 17.54 0.48
5y20y 14.81 14.64 0.17
5y30y 13.51 13.36 0.15
7y1y 19.75 19.61 0.14
7y3y 18.77 18.79 0.02
7y7y 16.57 16.63 0.06
7y10y 15.49 15.39 0.10
7y20y 13.04 13.23 0.19
10y1y 17.29 17.15 0.14
10y5y 14.79 14.97 0.18
10y10y 13.11 13.50 0.39
10y20y 11.06 11.89 0.83
smoothed version of this exact fit, after local weighted smoothing holding
cap prices constant. This enables us to determine the effects on swaption
prices of eliminating extreme peaks or dips; for example, under this
smoothed grid the 10-year into 20-year swaption increases in value by ap-
proximately 0.25% volatility. We note also that such a grid offers a com-
promise between smoothness and exact fitting, appropriate perhaps for
exotic derivatives pricing.
Volatility surface inference for the euro. Figure 6 shows a 12-para-
meter calibration to 60 euro swaptions and eight forward caps; the columns
(t, T) for fixed T are shown in figure 7. Once again, we are in a high-
volatility environment with elevated instantaneous volatilities across the
futures strip. In contrast with the dollar surface, however, volatilities
(t, T) increase noticeably as t T. This phenomenon is driven by caps
being richer relative to swaptions than in the US market; selling caps ver-
sus swaptions in euro perhaps with the opposite trade in the US is a
potential trade if one believes such a volatility dynamic is unlikely. Figure
8 shows this volatility surface rolled forward two years. The cap richness
manifests itself here in volatilities (t, T) which for given t are maximised
at t = T; in particular, the instantaneous volatility of the front futures con-
tract is higher than that of any other contract. This situation usually only
arises in times of high central bank activity. Table B gives some sample
relative value indicators in euro. One relative-value trade executed by cer-
tain market participants, buying two-year into two-year swaptions versus
two-year into one-year swaptions, is apparent. Furthermore, swaptions
with two-year tails generally look cheap relative to swaptions with 20-year
tails. This may arise due to dealers paying the 20-year constant maturity
swap (CMS) and receiving two-year CMS through structured steepener
transactions in euro, and thus becoming short volatility on forward 20-year
swaps and long volatility on two-year swaps.
We mentioned earlier carrying out sensitivity analysis to the correlation
specification. Figure 9 shows a volatility surface calibrated to the same set
of market instruments but using a factor structure with generally higher
correlations between forward rates. Comparisons between this and the first
calibration are most easily analysed in figure 10, which gives the differ-
ence in the columns (t, T) for fixed T between the high and standard cor-
relations. An increase in correlation with an unchanged volatility surface
will increase swaption prices while leaving cap prices unchanged. Thus in
order, for example, to keep the two-year into two-year swaption price un-
changed having increased correlation, we need to reduce the volatilities
(t, T), 0 < t < 2, 2 < T < 4. To keep the two-year by four-year cap price
unchanged, we must then increase (t, T), 2 < t < 4, 2 < T < 4. Under a
smooth calibration, this has the effect of pushing volatility from the top of
a column to the bottom, a phenomenon clearly seen in figure 10. We dis-
play the correlation structures used for the base and high correlation cases
in figure 11. Figure 11(a) shows the instantaneous correlation between for-
ward three-month rates dL(t, T) and the spot three-month rate dL(t, 0), for
t = 0. Figure 11(b) displays the correlation between dL(t, T) and dL(t, 5)
for t = 0.
94 RISK MAY 2004

WWW.RISK.NET
Cutting edge
l
Quantitative trading
Forward volatility through time
0.00
0.20
0.40
0.60
0.80
1.00
1.20
1.40
1.60
1.80
2.00
0
.
0
0
1
.
0
0
2
.
0
0
3
.
0
0
4
.
0
0
5
.
0
0
6
.
0
0
7
.
0
0
8
.
0
0
9
.
0
0
1
0
.
0
0
1
1
.
0
0
1
2
.
0
0
1
3
.
0
0
1
4
.
0
0
1
5
.
0
0
1
6
.
0
0
1
7
.
0
0
1
8
.
0
0
1
9
.
0
0
2
0
.
0
0
Time
1.0
3.1
5.2
7.3
9.4
11.6
13.7
15.8
17.9
20.0
%
4. Exact fit dollar columns
0.00%
0.20%
0.40%
0.60%
0.80%
1.00%
1.20%
Time
Tenor
HJM volatility surface
0
4
8
12
16
20
0
20
16
12
8
4
6. Parametric euro surface
0.00%
0.20%
0.40%
0.60%
0.80%
1.00%
1.20%
1.40%
1.60%
Time
Tenor
HJM volatility surface
0
5
10
15
20
25
0
5
10
15
20
25
5. Smoothed non-parametric dollar surface
Forward volatility through time
0.00
0.20
0.40
0.60
0.80
1.00
1.20
1.40
0
.
0
0
0
.
5
0
1
.
0
0
1
.
5
0
2
.
0
0
2
.
5
0
3
.
0
0
3
.
5
0
4
.
0
0
4
.
5
0
5
.
0
0
5
.
5
0
6
.
0
0
6
.
5
0
7
.
0
0
7
.
5
0
8
.
0
0
8
.
5
0
9
.
0
0
9
.
5
0
1
0
.
0
0
Time
1.0
2.0
3.0
4.0
5.0
6.0
7.0
8.0
9.0
10.0
%
7. Euro columns
where f
^
t
is the most likely path from F to K. Figure 12 shows the ap-
proximate weighting function of local volatility ( f, t) for an at-the-money
option and for the same option under a market move. The local volatili-
ties of import to the value of the option change as the market moves, and
thus the implied volatility of the option must change also. Trading oppor-
tunities arise when one believes that a local volatility model is a reason-
able description of the underlying volatility process, and when market
prices do not readjust accordingly as the market moves.
Term structure of skew. The previous section addressed the skew dy-
namics of one fixed forward rate. It is natural, however, to investigate the
term structure of skew, that is, how skews of different term swap rates and
different option expiries interrelate. A local volatility surface for three-
month rates within the BGM formulation (plus a factor structure) deter-
mines the local volatility surface for all other term swap rates. The exact
form of this surface is complex and it is hard to obtain analytic approxi-
mations. However, skews can always be simulated and a local volatility
surface approximated. For example, a reasonable approximation for swap-
tions of intermediate expiry and tenor is a CEV model of the same order
as that underlying the BGM formulation. However, this approximation fails
in certain cases, and it is important to realise that certain market skews for
different term rates can be inconsistent under particular modelling formu-
lations. Recent research (for example, Joshi & Rebonato, 2003, and Meren-
er & Glasserman, 2003) provides alternative methodologies for examining
skew within the BGM framework.
Further examples
Skew. The examples above used a Gaussian specification. Similar analy-
sis can be carried out under a constant elasticity of variance (CEV) for-
mulation of BGM (see Blyth & Uglum, 1999, and Andersen & Andreasen,
2000). The CEV process allows a range of skew shapes, although it can
sometimes be too limited to obtain a reasonable representation of market
dynamics. We can in theory extend to a local volatility model (Dupire,
1994) where volatility is an explicit function of level ( f, t, T). This is prac-
ticable in the case where we consider one particular forward rate and op-
tion expiry and work in the appropriate forward measure, that is, we fix
T and write (f, t, T) = (f, t).
A consistent approach requires that the shape of the skew after mar-
ket movements is consistent with the underlying dynamic. The work of
Blacher (1998) and Gatheral (2001) shows how to calculate implied volatil-
ity as a function of local volatility, and thus gives the required consis-
tency restrictions under a local volatility model. We let
F, K, T
be the
Black implied volatility for option of strike K and maturity T when the
forward rate is F, and (f, t) be the Black gamma of the option at time t
and rate f when hedged at volatility
F, K, T
. Then from Gatheral we have
the approximation:
(6)

F K
T
T
t
T
E f t f t
E f f t
dt
T
f
,
, ,
,
2
2
2
0
2
0
1 1

( ) ( )

1
]
( )

1
]

(
(

,,t dt
( )
WWW.RISK.NET

MAY 2004 RISK 95
0.00%
0.20%
0.40%
0.60%
0.80%
1.00%
1.20%
Time
Tenor
HJM volatility surface
2
5
8
11
14
17
20
20
17
14
11
8
5
2
8. Forward parametric volatility surface
1y
2y
5y
7y
10y
20y
30y
5y
4y
3y
2y
1.0
0.8
0.6
0.4
0.2
0.0
0.2
0.4
0.6
0.8
1.0
Tenor
E
x
p
i
r
y
Volatility differences
B. Sample euro relative-value indications
0.00%
0.20%
0.40%
0.60%
0.80%
1.00%
1.20%
1.40%
Time
Tenor
HJM volatility surface
0
4
8
12
16
20
0
20
16
12
8
4
9. High correlation parametric euro surface
Forward volatility through time
0.10
0.08
0.06
0.04
0.02
0.00
0.02
0.04
0.06
0.08
0.10
0
.
0
0
1
.
0
0
2
.
0
0
3
.
0
0
4
.
0
0
5
.
0
0
6
.
0
0
7
.
0
0
8
.
0
0
9
.
0
0
1
0
.
0
0
1
1
.
0
0
1
2
.
0
0
1
3
.
0
0
1
4
.
0
0
1
5
.
0
0
1
6
.
0
0
1
7
.
0
0
1
8
.
0
0
1
9
.
0
0
2
0
.
0
0
Time
1.0
3.1
5.2
7.3
9.4
11.6
13.7
15.8
17.9
20.0
%
10. Column change due to correlation change
Swaption Mkt vol Model vol Diff
2y1y 25.61 25.16 0.45
2y2y 22.27 22.90 0.63
2y5y 17.36 17.97 0.61
2y10y 14.16 13.98 0.18
2y20y 11.84 11.25 0.59
2y30y 11.05 10.74 0.31
3y1y 22.27 21.99 0.28
3y2y 19.66 20.17 0.51
3y5y 15.39 15.99 0.60
3y7y 14.19 14.33 0.14
3y10y 12.94 12.74 0.20
3y20y 11.19 10.49 0.70
3y30y 10.51 10.12 0.39
4y1y 19.46 19.63 0.17
4y5y 14.07 14.52 0.45
4y10y 12.15 11.81 0.34
4y20y 10.63 9.96 0.67
4y30y 10.02 9.71 0.31
5y1y 17.40 17.60 0.20
5y2y 15.88 16.09 0.21
5y5y 13.07 13.39 0.32
5y7y 12.31 12.26 0.05
5y10y 11.41 11.06 0.35
5y20y 9.97 9.55 0.42
5y30y 9.53 9.42 0.11
Stochastic volatility. The distinct smile on out-of-the-money options in
some markets has led many practitioners to consider adding stochastic volatil-
ity overlays to, say, CEV yield curve models. Again, we emphasise the need
for a logicist rather than proceduralist approach. For example, while a He-
ston (1993) type scalar multiplier of overall volatility levels within the BGM
framework can allow reasonable fitting of the volatility smile in certain mar-
kets, the prices of other products dependent on stochastic volatility should
be consistent with the same process, otherwise trade opportunities can arise.
For instance, we can calibrate a Heston framework to recreate correctly the
market smiles on swaptions on five-year swap rates. We can then use this
calibrated model to price a compound option to buy in one year at fixed
price a five-year into five-year swaption whose strike is determined to be
the forward rate at expiry of the compound option. Inconsistencies between
the model and market prices offer potential opportunities trading smile (that
is, out-of-the-money options) against compound options. These opportuni-
ties ultimately arise from dealers pragmatic use of inconsistent stochastic
volatility models to price different derivatives.
Stephen Blyth is managing director and head of European arbitrage
trading at Deutsche Bank in London. He would like to thank two
anonymous referees for substantive and useful reviews of an earlier
draft, and Maciej Sawicki and Greg Merchant for helpful comments. The
opinions or recommendations expressed in this article are those of the
author and are not representative of Deutsche Bank AG as a whole.
Email: stephen.blyth@db.com
96 RISK MAY 2004

WWW.RISK.NET
Cutting edge
l
Quantitative trading
Andersen L and J Andreasen, 2000
Volatility skews and extensions of the Libor market model
Applied Mathematical Finance 7, pages 132
Andersen L and J Andreasen, 2001
Factor dependence of Bermudan swaptions: fact or fiction?
Journal of Financial Economics 62, pages 337
Blacher G, 1998
A new approach for understanding the impact of volatility on option prices
Preprint, IIR Conference on Correlation and Volatility, July
Black F, 1976
The pricing of commodity contracts
Journal of Financial Economics 3, pages 167179
Blyth S and J Uglum, 1999
Rates of skew
Risk July, pages 6163
Brace A, D Gatarek and M Musiela, 1997
The market model of interest rate dynamics
Mathematical Finance 7, April, pages 127155
Dempster A, 1998
Logicist statistics I. Models and modelling
Statistical Science 13, pages 248276
Derman E, 1996
Model risk
Risk May, pages 3437
Dupire B, 1994
Pricing with a smile
Risk January, pages 1820
Gatheral J, 2003
Fitting the volatility skew
Case Studies in Financial Modelling, Courant Institute for Mathematical Sciences,
available at www.math.nyu.edu/fellows\_fin\_math/gatheral/lecture2\_2003.pdf
Heath D, R Jarrow and A Morton, 1992
Bond pricing and the term structure of interest rates: a new methodology for
contingent claims valuation
Econometrica 60, pages 77105
Heston S, 1993
A closed form solution for options with stochastic volatility with applications to
bond and currency options
Review of Financial Studies 6, pages 327343
Jamshidian F, 1997
Libor and swap market models and measures
Finance and Stochastics 1(4), pages 293330
Joshi M and R Rebonato, 2003
A stochastic-volatility displaced-diffusion extension of the Libor market model
Quantitative Finance 3, pages 458469
Longstaff F, P Santa-Clara and E Schwartz, 2001
Throwing away a billion dollars: the cost of suboptimal exercise strategies in the
swaptions market
Journal of Financial Economics 62, pages 3966
Longstaff F and E Schwartz, 2001
Valuing American options by simulation: a simple least squares approach
Review of Financial Studies 14, pages 113147
Merener N and P Glasserman, 2003
Numerical solution of jump-diffusion Libor market models
Finance and Stochastics 7, pages 127
Rebonato R, 1999
On the simultaneous calibration of multi-factor log-normal interest-rate models to
Black volatilities and to the correlation matrix
Journal of Computational Finance 2, pages 527
Rebonato R and P Jackel, 2003
The link between caplet and swaption volatilities in a Brace-Gatarek-
Musiela/Jamshidian framework: approximate solutions and empirical evidence
Journal of Computational Finance 6, pages 4159
Rebonato R and D Kainth, 2003
A two-regime stochastic volatility extension of the Libor market model: theory and
empirical evidence
Submitted to International Journal of Theoretical and Applied Finance
Tukey J, 1986
The collected works of John Tukey
Monterey: Wadsworth and Brooks
REFERENCES
At-the-money option After market move
9
8
7
5
4
3
Today
Maturity
9
8
7
5
3
Today
Maturity
Existing strike New ATM level ATM
12. Local volatility weightings
(a) Spot three-month rate v. forward three-month rates
0.40
0.50
0.60
0.70
0.80
0.90
1.00
Forward time T
C
o
r
r
e
l
a
t
i
o
n
Base
High
(b) Five-year forward three-month rate v. forward three-month rates
0.40
0.50
0.60
0.70
0.80
0.90
1.00
C
o
r
r
e
l
a
t
i
o
n
0 2 4 6 8 10 12 14 16 18 20
Forward time T
0 2 4 6 8 10 12 14 16 18 20
Base
High
11. Euro correlations